0% found this document useful (0 votes)

3K views365 pages

Virtual Competition The Promise and Perils of The ...

Uploaded by

qrfnmcxzr9

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

3K views365 pages

Virtual Competition The Promise and Perils of The ...

Uploaded by

qrfnmcxzr9

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 365

Copyright © 2016. Harvard University Press. All rights reserved.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Virtual
Competition
Copyright © 2016. Harvard University Press. All rights reserved.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Copyright © 2016. Harvard University Press. All rights reserved.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Virtual
Competition
T H E P RO M I S E A N D P E R I L S O F T H E
A LG O R I T H M - D R I V E N ECO N O M Y

Ariel Ezrachi • Maurice E. Stucke

Cambridge, Massachusetts
London, England
2016

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Copyright © 2016 by the President and Fellows of Harvard College
All rights reserved
Printed in the United States of America

First printing

Library of Congress Cataloging-in-Publication Data

Names: Ezrachi, Ariel, 1971– author. | Stucke, Maurice E., author.

Title: Virtual competition : the promise and perils of the
algorithm-driven economy / Ariel Ezrachi, Maurice E. Stucke.
Description: Cambridge, Massachusetts : Harvard University Press, 2016. |
Includes bibliographical references and index.
Identifiers: LCCN 2016018188 | ISBN 9780674545472 (cloth)
Subjects: LCSH: Electronic commerce. | Pricing—Technological innovations.
Classification: LCC HF5548.32 .E996 2016 | DDC 381/.142—dc23
LC record available at https://lccn.loc.gov/2016018188
Copyright © 2016. Harvard University Press. All rights reserved.

Preface vii

PART I Setting the Scene 1

1 The Promise of a Better Competitive Environment 3

2 New Economic Reality: The Rise of Big Data and Big Analytics 11
3 Light Touch Antitrust 22
4 Looking beyond the Façade of Competition 27

PART II The Collusion Scenarios 35

5 The Messenger Scenario 39

6 Hub and Spoke 46
7 Tacit Collusion on Steroids: The Predictable Agent 56
8 Artificial Intelligence, God View, and the Digital Eye 71
Copyright © 2016. Harvard University Press. All rights reserved.

PART III Behavioral Discrimination 83

9 Price Discrimination (Briefly) Explained 85

10 The Age of Perfect Price Discrimination? 89
11 The Rise of “Almost Perfect” Behavioral Discrimination 101
12 Behavioral Discrimination: Economic and Social Perspectives 117
13 The Comparison Intermediaries 131

PART IV Frenemies 145

14 The Dynamic Interplay among Frenemies 147

15 Extraction and Capture 159
16 “Why Invite an Arsonist to Your Home?” Understanding
the Frenemy Mentality 178
17 The Future of Frenemy: The Rise of Personal Assistants 191

PART V Intervention 203

18 To Regulate or Not to Regulate 205

19 The Enforcement Toolbox 218

Final Reflections 233

Notes 251
Acknowledgments 345
Index 347
Copyright © 2016. Harvard University Press. All rights reserved.

Could digital commerce and new technologies actually harm us? Today,
the rise of the Internet, Big Data, computer algorithms, artificial intelli-
gence, and machine learning all promise to benefit our lives. On its surface,
the online world—with the growth of price comparison websites, dynamic
pricing, web promotions, and smartphone apps—seems to deliver in terms
of lowering prices, improving quality, widening the selection of goods and
ser vices, and hastening innovation.
And yet, could it be that, after the initial procompetitive promise, these
technologies lead to higher prices, poorer quality, fewer options presented
to us, and less innovation in things we care about, such as our privacy?
Our suggestions may sound heretical and counterintuitive. After all, in
many markets, data and technology have visibly stimulated entry, expan-
sion, and competition. We do not dispute these benefits. Technology and
Big Data can be beneficial, no doubt. However, once one ventures beyond
the façade of competition, a more complex reality emerges.
The dynamics of artificial intelligence, price algorithms, online trade, and
competition lead us to uncharted ground—to a landscape that ostensibly
Copyright © 2016. Harvard University Press. All rights reserved.

has the familiar competitive attributes to which we are accustomed, and

yet delivers far less than what we would expect.
The new market dynamic, new technologies, and start-ups have capti-
vated our attention and created a welfare mirage—the fantasy of intensified
competition. Yet, behind the mirage, there operates an increasingly well-
oiled machine that can defy the free competitive forces we rely on.
Our thesis concerns the implications of the rise of a new—algorithm-
driven—power, which changes several structural and behavioral pillars that
underpin traditional markets.
vii

Competition, as we knew it—the invisible hand that distributes the necessi-

ties of life—is being displaced in many industries with a digitalized hand. The
latter, rather than being a natural force, is man-made, and as such is subject to
manipulation. The digitized hand gives rise to newly possible anticompetitive
behaviors, for which the competition authorities are ill-equipped.
Of course, we agree that the rise of Internet commerce through sophisti-
cated computer algorithms can intensify competition in ways that increase
our welfare. But, importantly, this is not assured. Our book explores how
the paradigm shift can leave some of us better off, while leaving many in
society worse off. Moreover, competition authorities may need to reassess
and reinterpret the legal tools at their disposal to prevent and punish these
unusual new forms of anticompetitive restraints. Even basic questions, such
as “Can computers collude?” or “How much choice does the online environ-
ment offer?” may be challenging. At times, it may be difficult to see beyond
the façade of competition to the toll that the new paradigm has on us, our
welfare, and our democratic ideals.
In what follows we explore these dynamics. We consider the possible use
of sophisticated price algorithms and artificial intelligence to facilitate col-
lusion or conscious parallelism. We reflect on the expansion of behavioral
advertising and the possible use of advanced technology and tracking to
engage in “almost perfect” behavioral discrimination. The discussion also
explores information harvesting and analysis, the effects of intermediation
and price comparison websites, the rise of super-platforms, and their
“Frenemy” relationship with independent application developers.
Our exploration of these themes raises challenging questions as to the
true competitiveness of present and future online markets. We consider the
limits of competition, consumer protection, and privacy law in an advanced
algorithm-driven environment, and reflect on the enforcement gaps and
policy implications.
Copyright © 2016. Harvard University Press. All rights reserved.

This book was born of a question that challenged our minds during a
stroll along the River Thames: “What if computers could collude?” To para-
phrase T. S. Eliot, that led us on our journey:
Oh, do not ask, “What is it?”
Let us go and make our visit.

And so we did. Our research prompted additional questions and stimulating

discussions with competition officials, lawyers, economists, computer scien-
tists, philosophers, and engineers. We welcome you to the debate.

Setting the Scene

M UCH HAS BEEN WRITTEN about the transformative effects that recent
technological changes have had on our society and well-being. These
technological developments in e-commerce, computers, Big Data, and
pricing algorithms, have no doubt changed the way we shop and commu-
nicate. The dynamics of online commerce have freed customers from reli-
ance on local offerings. Gone are the days when many of our choices were
restricted to a few local retailers who controlled which products were placed
on the shelves, the deals we struck, and largely the information on which
we based our decisions. Advances in technology and changes in commu-
nications, transportation, and commerce are expected to further change
our environment and promise to increase competition and well-being.
Our discussion in this part presents two contradictory themes. We be-
gin with the commonly accepted promise of the algorithm-driven econ-
omy; then we switch gear and outline its perils—its darker and less charted
sides.
Chapter 1 explores the many alluring features of online markets and the
promise they carry—to increase efficiency, competition, and ultimately our
Copyright © 2016. Harvard University Press. All rights reserved.

prosperity. The new economic reality promises to be bright.

Chapter 2 looks at key technological developments—the rise of self-
learning algorithms and Big Data that are fueling these dynamic innovations—
every thing from books sold on Amazon to airplane tickets on Orbitz. We
illustrate how Big Data and Big Analytics are providing online retailers like
Amazon a competitive advantage over brick-and-mortar behemoths like
Walmart.
In Chapter 3 we summarize the enforcers’ typical approach to digital
markets. We note how, given the significant potential benefits of innovation
1

and technology, the rallying cries within the tech industry, and increasingly
the antitrust circles, are that only a light regulatory hand, if any, is needed.
Having explored the “promise” of a data-driven economy, we turn in
Chapter 4 to introduce its darker sides. Venturing behind the façade of
virtual competition, we question the conventional wisdom that the com-
petitive problems of the analog world—collusion, monopoly, and price
discrimination—are less likely to reappear in the digital world, where rivals
are simply a click away. As the remaining parts of this book explore, vari-
ants of these traditional anticompetitive scenarios may develop—with a
vengeance.
Copyright © 2016. Harvard University Press. All rights reserved.

The Promise of a Better

Competitive Environment

T ODAY, with a few taps on our smartphone, tablet, or computer, we can

discover an array of products, reviews, and prices. The Internet has
made our world smaller. After watching an entire season of Downton Abbey
on Amazon Prime, you could, without leaving your home—whether in
Oxford, Mississippi, or Oxford, U.K.—aspire to the British aristocracy,
buying your Barbour hunting jacket and Hunter boots from a U.K. mer-
chant, your Range Rover from a dealership several hundred miles away,
your Rhodesian Ridgeback from a California kennel, your summer rental
in the Lake District from a family through Airbnb, and Wordsworth’s
poems and a sketchpad from Amazon.com. You could find eager sellers on
eBay, Fiverr.com, or one of the many tradesmen’s advice websites, and join
a host of communities, chat rooms, and information websites. Indeed,
you could even find your future spouse online to accompany you to your
English manor, where you could post photos on Facebook to celebrate your
elevated social status. Freed from the restrictions and the tyranny of the
former gatekeepers—whether the local media, brick-and-mortar retailers,
or tastemakers—we must be better off.
Copyright © 2016. Harvard University Press. All rights reserved.

The competitive future appears so bright because the online data-driven

competition has many appealing economic features. You want to travel
next week to Las Vegas? Previously you would have gone to a travel agent
or searched the travel advertisements in the local newspaper. Today you
would likely turn to search engines and price comparison websites (PCWs).
The Internet has a constellation of platforms to reduce the time and ex-
pense of searching the World Wide Web for what we want. Consumers are
increasingly relying on these platforms for purchasing decisions.1 Indeed,
many platforms have established themselves as significant players in the
3

distribution chain.2 Among the promises of online commerce are greater

market transparency, efficiency, and ease of use. All of these, as we’ll see,
should increase competition in ways that promote our well-being.

Increasing Market Transparency and Flow of Information

If you shop in a store where the products are not clearly priced, the process
can be frustrating. Transparency enables us to readily compare products’
price and quality and to choose the product that matches our price/quality
requirements. Economists have long recognized that information is a key
component in promoting a competitive market, which in turn promotes
consumer welfare.3 Indeed, the undistorted flow of information is one of
the conditions of the theoretical economic model of “perfect competition,”
under which consumers benefit from lower prices, wider choice, and better
quality.4 Market transparency, the OECD noted, “increases efficiency by re-
ducing customers’ search costs and allowing suppliers to benchmark their
performance with that of their competitors.”5 Market transparency, besides
helping buyers, helps sellers “to save costs by reducing their inventories,
enabling quicker delivery of perishable products to consumers, or dealing
with unstable demand etc.”6
In general, increased transparency ameliorates the problems of “infor-
mation asymmetries.”7 This is when one party knows more key informa-
tion than the other party (such as the seller of a used car who knows more
about the car’s problems than the buyer).8 As the flow of information in-
creases and becomes more balanced, sellers and buyers are more likely to
make educated decisions, and markets become more efficient.9
We often see the benefits of increased transparency when we shop online.
Rather than trudge to different retail stores, we can quickly search for and
identify the particular product we want, compare prices among the major
Copyright © 2016. Harvard University Press. All rights reserved.

online and brick-and-mortar retailers, and have the item shipped to our
home or available for pickup at a nearby outlet. There are now even compa-
nies, such as Doddle in the U.K., that have set up delivery points for a range
of online retailers, such as Amazon and ASOS, allowing customers to pick
up all of their online shopping in one place. Many online platforms and re-
tailers provide user reviews and other information that consumers consider
important to their purchasing decisions.10 Sellers can easily inform cus-
tomers of new products or ser vices, their characteristics, and the price.
Customers, aware of the range of options available in the marketplace, can
intelligently choose the option that matches their preexisting preferences.

Lower Search Costs

Having more information and greater market transparency is not espe-
cially helpful if it takes too much time and effort for consumers to review
the information. New York City has over 5,000 grocery stores.11 Grocery
stores are often transparent in their prices. But consumers don’t have the
time to travel around town to compare prices for each grocery item. Thus
another procompetitive feature of online markets is their ability to reduce
users’ search costs.
The economic literature has long illustrated that increases in search costs
will likely lead to increases in the seller’s power and prices.12 Ill-informed
customers are more likely to be subjected to higher, even monopolistic,
pricing. When the seller’s market power is based on the presence of high
search costs related to quality or price, reducing those costs should, in
theory, decrease the seller’s market power and prices.13
Online platforms can help reduce search costs for both sellers and buyers
by facilitating information flow and enabling users to quickly compare a
range of products and relevant prices.14 The promise of online platforms,
including PCWs, in promoting competition lies not only in their provision
of price information, but in several other features that support customers’
decision making and reduce their search costs. For example, online shop-
ping platforms provide users with interactive tools to identify the products
or ser vices that match their preferences. The combination of user-defined
parameters, such as maximum price and average user rating, with a plat-
form’s own algorithms, such as matching accessories for the item that the
consumer is considering, mean the online platform can distill for con-
sumers a far greater volume of relevant information than they other wise
would have, enabling better purchasing decisions made more efficiently.15
By reducing our search costs, these platforms enable us to undertake
Copyright © 2016. Harvard University Press. All rights reserved.

multiple searches on multiple platforms, further enhancing the competi-

tive pressure on sellers—again, to our benefit.
Illustrative is the launch of a price comparison website in the U.K.
dedicated to extended warranties. This information website was created
following an investigation by the U.K. Competition Commission that
identified a deficiency in the availability of relevant information, which
undermined the competitive process.16
Another example is air travel. Suppose you want to fly next Friday from
London to Las Vegas. You can search each airline’s website for fares; but to
lower your search costs, you can use Orbitz or another web-aggregator.

Indeed, some sites, such as SkyScanner, search multiple web-aggregators,

which in turn search the web for airfares to Las Vegas next Friday. You can
quickly determine the best flight, based on price or non-price factors (such
as length of flight, number of transfers, and quality of airline ser vice). You
can also run the same search on multiple platforms (e.g., Orbitz and Kayak)
to quickly assess airfares to Las Vegas next Friday. The web-aggregators
also offer fare calendars that tell you the average prices for traveling to Las
Vegas on other days that month.

More Entry and Expansion

A third attractive feature of online markets is seemingly lower entry bar-
riers. Customers generally benefit when potential sellers can rapidly enter
and exit the market without incurring significant costs that they cannot
recover elsewhere. The belief is that companies cannot exercise market
power for long when entry into markets would be timely (generally under
two years), likely (profitable for the entrants), and sufficient (the entrants
would attain sufficient business to prevent the exercise of market power by
the incumbent firms).17 If a firm raises prices above, or degrades quality
below, competitive levels, entrants and incumbents would seize the oppor-
tunity to profit and competition would be fully restored. Whether this is
empirically true is another matter.18 But there is little dispute that market
power can be sustained in markets with significant barriers to entry and
expansion, and that “entry analysis constitutes an important element of the
overall competitive assessment.”19 Thus, as entry barriers significantly de-
crease, so too should concerns about likely anticompetitive effects.
In the online world, one doesn’t need brick-and-mortar retail outlets to
compete. One can easily design and create a website, offer ser vices online,
and reach customers with the help of online advertising and unbiased
Copyright © 2016. Harvard University Press. All rights reserved.

search engines. For example, you could compete against hotels and hostels
through the accommodations app Airbnb or Booking.com. These online
platforms facilitate entering the accommodation market in hosting guests
at our own residences. In the same vein, it is a lot easier for drivers to enter
the taxi market through a ride-sharing app, such as Uber, Lyft, or Didi
Chuxing, than to acquire a taxi medallion or a cab.
Online platforms can facilitate a competitive market by mitigating the
seller’s actual (or perceived) risk and costs of entering. The risks of renting
my place to a stranger may appear daunting. So platforms like Airbnb pro-

vide information and ratings of potential guests and a “Host Guarantee”

that reimburses eligible hosts for damages up to $1,000,000.20 In a similar
vein, Uber provides its drivers with a passenger rating, which is an average
rating of those provided by all of a passenger’s previous drivers, which is
not immediately available to the passenger. So for intoxicated passengers
who vomit in the car, their chances of getting picked up are slim; they may
not even be able to use the app.21 In providing advice, reviews, and guar-
antees, online platforms can attract individuals who would other wise be
apprehensive about transacting with unfamiliar parties.
Online platforms can also foster entry by reducing advertising expenses.
Suppliers who wish to advertise directly on search engines will bid for
search words, as on Google AdWords, and pay for any click on the adver-
tisement. That is an improvement over the old media model, where you
would pay to advertise, often not knowing how many people saw, listened
to, or read your ad. Price comparison websites (PCW), which benefit from
economies of scale and high conversion rates, can further lower these ad-
vertising costs and facilitate access to markets.22 Indeed, it has been re-
ported that consumers indicated that “they would only know to contact a
few companies for any given product or ser vice, but on [PCWs] they get a
wider range of options to choose from.”23

More Dynamic Disruption and Efﬁciencies

Reducing search costs, lowering entry barriers, and increasing information
flows can increase the competitive pressure to innovate.24 Thus the fourth
promise of online markets is to promote distinct dynamic and allocative
efficiencies. The disruptive technology—by increasing transparency and
reducing search costs—can more efficiently match buyers and sellers,
thereby promoting allocative efficiency. With these online tools, users can
Copyright © 2016. Harvard University Press. All rights reserved.

quickly identify the provider or product that better matches their needs.25
As the U.K. Office of Fair Trading (OFT) noted,26 “[t]he Internet allows for
a much swifter search and comparison across a wide variety of choice
factors including price, dates, quality and location.”27
The rise of Big Data and Big Analytics may yield other distinct economic
efficiencies. For instance, they can reduce costs by optimizing inventory
levels; “to have the right amount of stock in the right place at the right
time.”28 Manufacturers, distributors, and retailers can rely on sensors
to track products and components throughout the supply chain from

production to point-of-sale. Moreover, online platforms can unleash eco-

nomic value on several levels. The sharing economy, for example, promises
to increase efficiency through greater transparency and disintermediation.
People can immediately profit from assets currently being underutilized—our
cars, houses, power tools, or spare time. As more people rely on ride-sharing
apps, fewer people will need to buy cars. Fewer cars, or individual car trips,
mean less space devoted to garages and parking lots; space in high-rent
urban centers like San Francisco now can be used for housing and other
productive endeavors.
Online retailers are already employing complex pricing algorithms “that
take into account factors like an item’s popularity and what competitors
are charging for it” and “data about you—such as where you live, when
you shop, how often you’ve visited the site, and what you’ve bought in the
past.”29 These increasingly automated, digitized transactions could create
a more transparent marketplace in which resources are allocated more ef-
ficiently and in which the best product or ser vice, at the lowest price, tri-
umphs. The new market environment provides retailers with the capacity
to better identify their customers’ needs and react to market changes with
ever-increasing speed.

Reduction in Seller Power

Finally, old world antitrust problems seem less likely. If online markets in-
crease information flow, market transparency, and dynamic innovation,
and reduce entry barriers, then sellers should have less market power, and
monopolies should be even rarer. Importantly, the popularity of search en-
gines, PCWs, and shopping platforms like Amazon and eBay should make
it harder for suppliers to take advantage of ill-informed customers who are
subjected to high information costs.30 As Amazon notes, “The presence of
Copyright © 2016. Harvard University Press. All rights reserved.

many competing sellers on the same e-commerce site strengthens compe-

tition to provide the best offers and prices. It also enables customers to
easily compare competing offers by brand, quality, price, speed of delivery
or other attributes and select the offers that best meet their needs.”31
Suppose you are interested in buying a particular brand of coffee maker.
A web-aggregator can tell you the price of that coffee maker at different
online and brick-and-mortar retailers (intrabrand competition), but also
other manufacturers’ coffee makers, their specifications, features, war-
ranty, and customer reviews (interbrand competition). The increase in

both intra- and interbrand competition can further pressure manufac-

turers and retailers to reduce prices, increase quality, and enhance ser-
vices, such as free repairs. Hotels, travel agents, insurance brokers, and
other upstream providers compete on transparent platforms in which
price, ser vice, and other variables are visible to all.
The rise of web-aggregators in some markets has in fact led to lower
prices and consequently lower profit margins for upstream sellers.32 For ex-
ample, one empirical economic study found that the rise of Internet com-
parison shopping sites for life insurance reduced term life prices in the
1990s by 8 to 15 percent and increased consumer surplus by at least $115–
$215 million per year.33
Furthermore, with the rise of pricing algorithms, we arguably no longer
need to worry about collusion, where competitors agree in smoke-filled
hotel rooms to fix price, allocate markets, or reduce output. When each firm
relies on its own pricing algorithm, cartels may become less stable. Indeed,
the advance of pricing algorithms might suggest the end of cartels. Com-
puters do not exhibit trust, which is important for many cartels’ success.34
Nor is there any collusion among the computers. “Collusion is more likely,”
the U.S. Department of Justice noted, “if the competitors know each other
well through social connections, trade associations, legitimate business con-
tacts, or shifting employment from one company to another.”35 Pricing algo-
rithms will not “congregate in the same building or town,” thereby having
“an easy opportunity for last-minute communications.”36 Instead, it is often
assumed that algorithms, in engaging in cold, profit-maximizing calcula-
tions, won’t agree with, or trust, other computers; even if they did, they would
find ways to cheat on any agreements.
Price discrimination should also be less likely. The collation of informa-
tion makes it easier for consumers to compare the prices of advertised
goods—thus making it harder for sellers to selectively increase the prices
Copyright © 2016. Harvard University Press. All rights reserved.

or degrade the quality of goods.37 Armed with more information, con-

sumers become aware of the full range of substitutes, which they can take
into account when making purchasing decisions.38

On the Path to Better Competition

With the growth of online platforms—from search engines to price com-
parison websites—we are seemingly on the road to optimizing competi-
tion. Prices should steadily decline toward marginal cost. Fully informed

sellers and buyers easily enter and exit the market; such as Uber drivers
who are enticed by surge pricing to hop in their cars and meet the surge in
demand.
So the rise of the digital economy can be a good thing. Few hunger for
1970s fashions. Why then pine for the old competitive framework, with its
cartels, including the government-supported uranium cartels, and monop-
olies like Kodak and IBM? If online markets accelerate market forces, we’re
heading toward healthier competition, where entry and exit are easier, buyers
and sellers are numerous and better informed, prices are approaching mar-
ginal cost, and firms are innovating to remain relevant. Antitrust becomes
less relevant, as monopolies and cartels are less durable. In short, the promise
of online markets could free us from the monopolies and gatekeepers of old
and unleash tremendous value as resources are used more efficiently.
Copyright © 2016. Harvard University Press. All rights reserved.

New Economic Reality:

The Rise of Big Data and Big Analytics

O NLINE MARKETS have many attractive features that promise to increase

competition in ways that improve our well-being. So what is driving
this new economic reality?
In this chapter we examine how self-learning algorithms and Big Data
are providing online platforms, like Amazon.com, a competitive advantage
over brick-and-mortar behemoths like Wal-Mart Stores, Inc. (Walmart).
This intense competitive pressure is changing the nature of retail. Many
brick-and-mortar outlets face the reality of adapting or losing even more
sales. As the data arms race and shift to pricing algorithms intensify, the line
between online and brick-and-mortar retail will blur.

The Battle between Walmart and Amazon

A few years back, when you thought of market and buyer power, one re-
tailer that probably came to mind was Walmart.1 As many small and large
sellers can attest, Walmart is “powerful”; its purchasing agents “can make
you or break you.”2 One fear was that when Walmart moves in, small busi-
Copyright © 2016. Harvard University Press. All rights reserved.

nesses and jobs move out, and Main Street dies.3 As a 2003 BusinessWeek
cover story “Is Wal-Mart Too Powerful?” put it, “the more size and power
that ‘the Beast of Bentonville’ amasses, the greater the backlash it is stirring
among competing retailers, vendors, organized labor, community activists,
and cultural and political progressives.”4 Thwarting Walmart’s ambitious
expansion strategy into urban America, the 2003 article noted, was the “in-
tensifying grassroots opposition.”5
Let us fast-forward to January 2016. Walmart announced its closing of
269 stores globally, 154 of them in the United States.6 Why the retreat? The
11

threat was not from the grassroots progressives. Rather, the threat came
from online commerce. Its customers increasingly “are using computers,
tablets, and smart phones to shop online with [Walmart] and with [their]
competitors and to do comparison shopping.”7 Many of us bring our smart-
phones to stores, to review and compare store prices with online prices, read
online reviews, and so on.8 The result is that the likelihood of our purchasing
in brick-and-mortar stores, even once we are in them, is decreasing.
Walmart is now working hard to catch up in the accelerating shift to
online sales. Walmart’s goal is to position itself “to win at the convergence
of digital and physical.”9 To strengthen its e-commerce operations, in
2015–2016, Walmart planned to spend $2 billion, far more than the $700
million it spent on e-commerce in 2014.10
So when Walmart was slipping, who was gaining? Amazon. As one Wall
Street analyst observed in 2015, “With every passing year, it becomes harder
and harder for Wal-Mart to compete with Amazon.”11 Walmart’s reve-
nues in 2014 were five times greater than Amazon’s ($486 billion vs. $89
billion). But Amazon’s stock market value as of mid-2015 had eclipsed
Walmart’s by over $70 billion.12 Moreover, Amazon’s net sales have
accelerated—from $34 billion in 2010, to $48 billion in 2011, to $61 billion
in 2012, to $74 billion in 2013, to $88.9 billion in 2014, and $107 billion in
2015.13 Amazon was the fastest company ever to reach $100 billion in an-
nual sales.14
The sentiment is that Walmart’s distributional efficiencies from its brick-
and-mortar store model do not translate to the data-driven analytics and
dynamic pricing of the online world. To illustrate the significance of these
dynamics, and the way they affect competition, let us compare Amazon’s
business practices to those of the brick-and-mortar retailers.
First, Amazon.com has a far greater product assortment and inventory
than any brick-and-mortar retail outlet. Amazon and third parties sell mil-
Copyright © 2016. Harvard University Press. All rights reserved.

lions of unique products on the platform, across dozens of product catego-

ries.15 In 2014 Amazon sold over 2 billion products,16 and today it sells far
more books than any retail bookstore. Its power in books was illustrated
by the Apple antitrust case, in which the dominant publishers complained
of their inability to act unilaterally (or without “critical mass”) against
Amazon’s pricing practices.17 Amazon also is expected to be by 2017 the
largest clothing retailer. Thus even the ubiquitous retailer Gap is consid-
ering selling its clothing on Amazon’s super-platform. To not consider this
possibility, said Gap’s CEO, would be “delusional.”18

Second, as any retailer’s product assortment grows, so too does the

impracticability of manually adjusting pricing. Humans would have to
process vast reams of data to decide the price. Moreover, pricing, if done
manually, like the clerk stamping the price on the food tins, could take
months, if not years. Amazon uses computer algorithms that harvest per-
sonal and market data to constantly adjust its pricing for its millions of
products. Amazon’s pricing algorithms made headlines when they led
to an unintended escalation in price of Peter Lawrence’s book The Making
of a Fly.19 At its peak, Amazon priced the book at $23,698,655.93 (plus
$3.99 shipping).20 Notwithstanding that incident, Amazon “aggressively
changes prices, sometimes altering them more than once per day in reac-
tion to other retailers.”21 Its algorithms can adjust prices quickly to re-
spond to changes in market conditions, including its competitors’ prices.
Take the price of a frozen yogurt ice cream and sorbet maker, which, ac-
cording to CamelCamelCamel.com (a website which tracks Amazon’s
prices), fluctuated between $27.97 and $59.99.22 Some prices change dra-
matically. Amazon’s price for a ladies’ watch, for example, plunged from
$115 to $57.50 in just a few days.23
Third, as Amazon and other online retailers expand their pricing algo-
rithms to other product offerings, the competitive pressure on competing
online and brick-and-mortar retailers to use pricing algorithms will inten-
sify. Amazon epitomizes this increasingly intense pricing-algorithm arms
race. As one venture fund observed:
“In a world where companies like Amazon are changing price and cus-
tomer experience in real-time to optimize sales, retailers cannot afford
to [be] revisiting pricing decisions on a weekly or monthly basis, and
hope to survive,” said Scott Jacobson, Managing Director, Madrona Ven-
ture Group. “To compete, they need sophisticated technologies like Boo-
Copyright © 2016. Harvard University Press. All rights reserved.

merang’s that enable instantaneous updates based on changing market

data. Guru and his team have developed technology that helps level the
playing field, leveraging hundreds of millions of data points to help re-
tailers automate and accelerate their decision-making to drive profitable
growth.”24

As the venture fund noted, Amazon is not alone. Boomerang Com-

merce, for example, is a market leader in the field of computerized price
optimization. Its pricing algorithms examine over 100–150 data points on
a minute-by-minute basis in adjusting prices.25 “Amazon has hundreds of

millions of products with the ability to change prices every 15 minutes,”

the founder of Boomerang says. “The average retailer has far fewer items
but only changes price every one to three months.”26 Staples, one customer
of Boomerang, felt the competitive pressure to use dynamic pricing: “We
don’t have a choice. Prices are constantly fluctuating.”27 The competitive
pressure to switch to dynamic pricing has opened a new competitive front
between retailers. Pricing algorithms already dominate online sales in
hotel booking, and the travel, retail, sports, and entertainment industries—
optimizing the price based on available stock and anticipated demand.28
Fourth, online retailers cannot simply post their products on their web-
site and expect sales to surge. Data, and importantly, the scale of data, are
key. Companies that operate and control these online platforms can col-
lect a large volume and variety of personal data that may have significant
value. Having control over, and being able to quickly analyze, the personal
data can provide the platform operator a key competitive advantage. In-
deed, Amazon originally sold books as a way to gather personal data on
affluent, educated shoppers.29 Also, algorithms learn through trial and
error and finding patterns from a greater volume and variety of data. Am-
azon collects far more data on its users than many retailers possibly could.
Yet more ominous for its brick-and-mortar competitors is that, as Amazon
collects more data on its users, and as its algorithms have more opportuni-
ties to experiment (such as presenting items, suggesting other purchases),
its pricing will become even more dynamic and differentiated. Basically,
price changes will be quicker, product offerings will be increasingly
tailored to particular users’ tastes, and price optimization will occur.
Fift h, Amazon’s algorithms will increasingly be pitted against other
algorithms (rather than humans) for pricing decisions. For example, Jet
.com—an e-commerce site based on a subscription model—has raised over
$200 million “to take on Amazon with a dynamic pricing model” and
Copyright © 2016. Harvard University Press. All rights reserved.

“promises to offer prices that are 10-to-15% lower than anywhere else, in-
cluding Amazon.”30 As the industry-wide use of algorithms increases, the
algorithms, through learning by doing, will better anticipate and respond
to rival algorithms’ actions.
To better compete against online giants like Amazon, Boomerang offers
its retail clients a “Dynamic Price Optimizer” as part of its main software
service. The optimizer “starts by analyzing pricing data from a retail client
and its competitors. But the secret sauce is its proprietary algorithms, which
incorporate sophisticated game theory and portfolio theory models, fil-
tering the data for almost any variable or desired outcome.”31

Sixth, some of the drawbacks of online shopping are disappearing. Some

shoppers, for example, like the immediate gratification of walking out of
the store with the goods. Online sellers are now increasing the speed at
which goods arrive at your door. For instance, with a subscription to Am-
azon’s Prime ser vice, consumers can now have millions of goods delivered
to their door within a couple of days, if not the same day.32 For an extra
cost, some goods can be delivered within a one-hour window.33 With the
ser vice now boasting dairy, chilled, and frozen products, the online pro-
vider can satisfy almost all of one’s needs.34 In addition to fast delivery or
click and collect options, some online retailers have now invested in brick
and mortar shops, to support their online operations.

Rise of Big Data and Big Analytics

As our Amazon example shows, Big Data and Big Analytics are increasingly
fueling our online marketplace. Big Data has various definitions, many of
which are broad and inclusive.35 Although data is varied, we predominantly
focus here on personal data, which is generally defined as “any information
relating to an identified or identifiable individual (data subject).”36 Big Data
has commonly been characterized by four Vs: the volume of data; the ve-
locity at which data is collected, used, and disseminated; the variety of in-
formation aggregated; and finally the value of the data.37
The use of Big Data and its value have increased with the rise of Big Ana-
lytics: the ability to design algorithms that can access and analyze vast
amounts of information. Moreover, the introduction of machine learning
has propelled performance in this area even further.
Recent years have witnessed groundbreaking research and progress in
the design and development of smart, self-learning algorithms to assist in
pricing decisions, planning, trade, and logistics. The field has attracted sig-
Copyright © 2016. Harvard University Press. All rights reserved.

nificant investment in deep learning by leading market players.38

In 2011, International Business Machines Corp.’s Jeopardy!-winning
Watson computer showcased the power of its deep-learning techniques,
which enabled the computer to optimize its strategy following trials and
feedback.39 Since then, IBM has invested in widening the capacity and func-
tionality of the technology, with the aim of making it “the equivalent of a
computing operating system for an emerging class of data-fueled artificial-
intelligence applications.”40
Recently, the launch of the Deep Q network by Google showcased enhanced
self-learning capacity. The computer was designed to play old-fashioned

Atari games. Importantly, it was not programmed to react to any possible

move in the game. Rather, it relied on models that enabled it to “learn” the
game environment through trial and error and improve its performance
over time. The technology mimics human learning by “changing the
strength of simulated neural connections on the basis of experience. Google
Brain, with about 1 million simulated neurons and 1 billion simulated con-
nections, was ten times larger than any deep neural network before it.”41
Deep-learning techniques have also been implemented in day-to-day
technologies. Smart algorithms are increasingly used to support automated
customer support, e-commerce, and online communications, and to create
interactive experiences online. Already in 2015, the European Data Protec-
tion Supervisor observed, “algorithms can understand and translate lan-
guages, recognise images, write news articles and analyse medical data.”42
For instance, the technology has been used by Microsoft in its Windows
Phone and Bing voice search;43 by Google, Toyota, Apple, Audi, and Jaguar
in developing “driverless” cars;44 and also in stock exchange analysis and
other ser vices.45
Big Data and Big Analytics have a mutually reinforcing relationship. Big
Data would have less value if companies couldn’t rapidly analyze the data
and act upon it. Machine learning, in turn, relies on accessing large data
sets. As the European Data Protection Supervisor observed, “Deep learning
computers teach themselves tasks by crunching large data sets using
(among other things) neural networks that appear to emulate the brain.”46
The algorithms’ capacity to learn increases as they process more relevant
data.47 The belief is that simple algorithms with lots of data will eventually
outperform sophisticated algorithms with little data.48 Part of this is due
to the opportunity for algorithms to learn through trial and error. Another
is seeing correlations from big data sets.
Thus one thing IBM’s Watson and artificial intelligence (AI) generally
Copyright © 2016. Harvard University Press. All rights reserved.

need in order to “do meaningful work” is data.49 That is why IBM acquired
the digital and data assets of Weather Co., owner of the Weather Channel.
Watson could analyze the volume of weather data to refine its algorithms.50
Watson’s ser vices, in turn, can be sold to other parties, like insurance apps.
Octo Telematics, for example, uses IBM’s real-time weather data “as a crit-
ical input to its driver behav ior scoring app.”51 Octo’s free mobile app of-
fers personalized insurance quotes based on the driver’s behav ior.52 Octo’s
algorithm assesses not only the driver’s speed, braking, and acceleration,
but also “outside variables often directly affected by weather, such as road
and traffic conditions, to determine driver scoring.”53 Drivers with good

scores, as determined by Octo’s algorithm, are rewarded with the option of

a discounted insurance quote from a panel of insurers, which they can
choose to accept at their discretion. Here we see how IBM’s data-driven al-
gorithm helps its client “construct a more accurate and reliable scoring
algorithm based on the precise weather conditions at the place and time of
the driver’s trip.”54 We also see how insurers are migrating from historical
data (such as the number of speeding tickets one has received in the past few
years) to near-real-time data (how the driver performed on the icy roads
yesterday evening) in personalizing insurance pricing.
Another example concerns the combination of smart algorithms with
Facebook’s vast user base, to improve targeting of ads and promotions. In
its annual developer conference in 2016, the company discussed the way
artificial intelligence (AI) could interact with the rich flow of data from its
users. Facebook CEO, Mark Zuckerberg, noted how “with AI and natural
language processing combined with human help, people will be able to talk
to Messenger bots just like they talk to friends.”55 David Marcus, VP mes-
saging products, reported how the company is “testing if business bots can
re-engage people on threads with sponsored messages.”56 Not surprisingly,
Apple, Amazon, Google and Microsoft are also investing in voice-activated
digital assistants that “learn” to make decisions rather than simply follow
instructions.57 The future of instant and online communications will
heavily rely on the mutually reinforcing relationship between Big Data and
Big Analytics.
Another recent significant development concerns the ability of com-
puters to operate with limited information. Computer algorithms long ago
solved perfect information games—like the board game checkers—where
players know every thing that happened previously. The year 2015 marked
a significant advancement. Several computer scientists announced a new
computer algorithm capable of solving extensive-form, “imperfect infor-
Copyright © 2016. Harvard University Press. All rights reserved.

mation games” much larger than previously possible. Their new algorithm
“weakly solved” one popular game of poker.
Let us consider the significance of such advancements. In checkers, both
players know of all the past moves and the current state of play (based on
where each piece is on the board). In poker the players do not have full
knowledge of past events (the unobserved cards that the other player had
in earlier rounds) and the current round (the unobserved cards).58 Thus
solving the poker game is more complex, with 3.16 × 1017 possible states
and 3.19 × 1014 decision points (where a player must make a decision). The
algorithm, however, computed a strategy for two-player limit Texas Hold

’Em poker so that it cannot be beaten with statistical significance in a

human’s lifetime.59 For poker aficionados, we’ll mention that the algorithm
confirmed (for two-player limit Texas Hold ’Em) that the dealer has a sub-
stantial advantage, and the nondealer’s optimal strategy is more often to
play than to fold.60 The significance of this advancement lies in the com-
puter’s ability to address the “real-world” complexity of imperfect informa-
tion, unleashing the possibility for complex “ human-like” interaction and
decision making.

Cloud Computing and the Internet of Things

As the breadth and quality of data increase over the next decade, the posi-
tive feedback loop between machine learning and Big Data will accelerate.
One contributing factor will be the developments in cloud computing.
Amazon’s cloud division in 2015, for example, added an Amazon Ma-
chine Learning ser vice. Amazon’s algorithms help the client find patterns
in its existing data.61 Then Amazon creates models, which process the cli-
ent’s incoming data and generate predictions. The models could predict
likely fraudulent purchases, products or ser vices that might appeal to the
client’s customer, or consumer trends. As more data is processed, the pre-
dictive models are refined. Google and Microsoft likewise provide as part
of their cloud computing ser vices machine-learning algorithms to analyze
data and predict future outcomes.62 A positive feedback loop can ensue:
Clients will have an even greater incentive to collect data and use the cloud
computing ser vices if they can obtain a competitive advantage through
these predictive models. And access to the many different clients’ data will
improve Amazon’s, Microsoft’s and Google’s algorithms.
Another contributing factor will be the “Internet of Things,” that is, the
integration of soft ware and sensors embedded in everyday objects. This
Copyright © 2016. Harvard University Press. All rights reserved.

technology enables machine-to-machine communication (M2M), as well

as the collection and analysis of information gathered through sensors.
For instance, Amazon in 2015 launched its “IoT platform,” which “lets
connected devices easily and securely interact with cloud applications and
other devices.”63 The platform is designed to process trillions of messages
from billions of devices “and can process and route those messages to [Am-
azon Web Ser vice] endpoints and to other devices reliably and securely.”64
The research firm International Data Corp estimated the “global market for
Internet of Things” to nearly triple to $1.7 trillion by 2020.65 The firm also

notes how technology firms, like Google, Intel Corp, Cisco Systems, Sam-
sung Electronics and the major telecoms such as Vodafone and Verizon,
“are betting heavily on it to drive revenue and profit in the future.” 66
Whereas traditional data is harvested through our interaction with online
sellers and our digitalized environment, the Internet of Things would
widen the scope of data for the algorithms. As more products have sensors,
the interfaces will include anything from household appliances, clothing,
cars, and bicycles, to streetlights, airports, smart building materials, and
human-embedded sensors.

Emerging Trends
The relevance and usefulness of real-time data are becoming increasingly
difficult to ignore. Our “real” and “online” environments are converging,
and digitalization will seemingly track individuals before their birth to
their death.67
These developments may improve our welfare well beyond online com-
merce. For instance, health ser vices could provide faster response and
monitoring through automated data collection. Smart meters and appli-
ances can help optimize our electricity usage. Even our local authorities
can optimize their ser vices by carefully collecting and using data from
various sources.68
In the context of our discussion, one distinct trend is the shift from
brick-and-mortar stores to online sites. We see this already with Amazon’s
sprawling platform. E-commerce, as a percentage of total retail, is in-
creasing.69 As a recent White House report noted,
Americans are using the Internet to shop in rapidly growing numbers,
suggesting that consumers believe they are getting a good deal on the In-
Copyright © 2016. Harvard University Press. All rights reserved.

ternet, regardless of any differences in the pricing practices of online and

offline retailers. The U.S. Census estimates that e-commerce has in-
creased from 2 percent of total U.S. retail sales in 2004 to 6 percent in
2014. Moreover, electronic commerce revenue is currently growing at a
rate of 16 percent per year in the United States, more than three times the
5 percent growth rate in overall retail sales.70

You might have noticed this shift when shopping on Black Friday in any
of the large U.S. retail stores. In 2015, the big U.S. shopping day after
Thanksgiving witnessed fewer customers in many stores. People were

already online on Thursday buying presents.71 Amazon’s Thanksgiving sales

rose 29 percent compared with a year before.72 As the Wall Street Journal
reported, Walmart “made the majority of its Black Friday deals available
online in the wee hours of Thanksgiving morning—some 15 hours before
its stores opened. So many shoppers visited Wal-Mart’s website when the
door-busters went on sale early Thursday, that the site was overloaded and
checkouts were snarled.”73
As online markets cover an ever-increasing spectrum of commercial
activities, another noteworthy trend is how Big Data and Big Analytics can
offer firms “even greater opportunities for competitive advantage (online
businesses have always known that they were competing on how well they
understood their data).”74
The business literature highlights the following ways in which Big Data
and Big Analytics can transform industries:

• Companies are increasingly adopting business models that rely

on personal data as a key input. Data-driven business models, for
example, involve multisided markets; companies offer individuals
free ser vices with the aim of acquiring valuable personal data to
assist advertisers to better target them with behavioral advertising.75
• The four Vs of Big Data—volume, velocity, variety, and value—will
increase, as companies undertake data-driven strategies to obtain
and sustain a competitive advantage. Companies will offer products
and ser vices to harvest a greater volume of data that is not other wise
publicly available. With the Internet of Things, sensors, microphones,
and cameras will sweep in significantly more data on human be-
havior in their homes, cars, and work.76 The value of data may also
come from its variety. Data fusion “occurs when data from different
sources are brought into contact and new facts emerge.”77 Through
Copyright © 2016. Harvard University Press. All rights reserved.

data fusion, companies can identify and improve their profi les of
individuals; better track individuals’ activities, preferences, and
vulnerabilities; and better target individuals with behavioral
advertisements.
• As the competitive value of data increases, companies will strive to
acquire a “data advantage,” and thus a competitive advantage over
rivals. Companies will increasingly invest in computer algorithms to
analyze the volume and variety of data. Even for publicly available
data, velocity will be critical—namely, getting and analyzing the

data quickly to outmaneuver rivals.78 The velocity in which data is

generated, accessed, processed, and analyzed will accelerate,79 and
for some applications it is now approaching real time.80
• As the velocity of generating, accessing, processing, and analyzing
data increases, the velocity of adjusting prices will also increase.
With online trading platforms, computers can assess and adjust
prices—even for particular individuals at particular times—within
milliseconds.81
• As more online sellers use AI and pricing algorithms, their rivals, to
prevent being at a competitive disadvantage, will feel greater pressure
to develop “smart” pricing algorithms themselves.
• As more companies switch to pricing algorithms, algorithms will
increasingly determine industry pricing of goods and ser vices. The
distinction between online and offline pricing will blur and eventu-
ally disappear in many industries.
• As their industry-wide use increases, algorithms, through learning
by doing, will anticipate and respond to other algorithms’ actions.
Online trading platforms may also enable sellers to segment the
market by using dynamic, differential pricing.82
• Learning from the volume and variety of our personal data, com-
puters, using AI, will increasingly make decisions for us—with
digital personal assistants predicting our needs and wants.
If traditional powerhouses like Walmart are vulnerable to the dynamic
disruption of Big Data and Big Analytics, then arguably the competitive
gales within the online marketplace will intensify into hurricanes. One
might argue that market power should be fleeting, as the digital hand will
drive prices lower, and information flows will raise quality levels. Thus the
rallying cry in some circles, as the next chapter explores, will be for a light
Copyright © 2016. Harvard University Press. All rights reserved.

regulatory hand (if any).

Light Touch Antitrust

I N 1998, a banking merger wave was increasing substantially the measures

of U.S. national concentration. Many financial institutions were falling
into fewer hands. Alan Greenspan, chairman of the Federal Reserve System,
told the U.S. Senate not to worry. The antitrust thinking over bigness had
evolved:
In the 1970s and 1980s, there was a significant shift in emphasis from a
relatively deterministic antitrust enforcement policy to one based on the
belief (under the aegis of the so-called Chicago School) that those market
imperfections that are not the result of government subsidies, quotas, or
franchises would be assuaged by heightened competition. Antitrust ini-
tiatives were not seen as a generally successful remedy. More recently,
limited avenues for antitrust policy are perceived by policymakers to
enhance market efficiencies.1

Many antitrust enforcers in 1998 would have agreed. Remarkably, some

antitrust scholars and enforcers would agree even today—despite the eco-
nomic crisis and the U.S. government bailing out financial institutions
Copyright © 2016. Harvard University Press. All rights reserved.

that, as a result of the mergers, were deemed too big to fail.

While jurisdictions around the world exhibit varying levels of interven-
tion, the dominant voices in competition policy over the past thirty-five
years, as this chapter explores, have advocated lighter intervention, if any,
in many mergers and monopolies. One exception is the prosecution of
cartels that fi x prices, allocate markets or bids, or reduce output.2
Seemingly, from the discussion in Chapters 1 and 2, a light touch ap-
proach would appear justified. Governmental intervention in the online
world appears superfluous—markets are dynamic and competitive. New
22

online business dynamics, it is often argued, have changed markets for the
better. Companies’ algorithms, fueled by the increasing flow of data, are
seemingly perfect strategies to optimize profitability. These developments
appear to give rise to new forms of competition and commerce. The tradi-
tional competitive problems (collusion, monopoly, and price discrimina-
tion) should arguably appear infrequently in the digital world, where rivals
are simply a click away. With price algorithms analyzing and responding in
real time to far more market data than humans could consider in their life-
time, we appear on course to a more dynamic marketplace.
Before delivering the DOJ a stinging defeat, the U.S. Court of Appeals
for the Ninth Circuit first praised the invisible hand, while condemning
centrally planned economies (and price regulation):
Competition is the driving force behind our free enterprise system. Un-
like centrally planned economies, where decisions about production and
allocation are made by government bureaucrats who ostensibly see the
big picture and know to do the right thing, capitalism relies on decen-
tralized planning—millions of producers and consumers making hun-
dreds of millions of individual decisions each year—to determine what
and how much will be produced. Competition plays the key role in this
process: It imposes an essential discipline on producers and sellers of
goods to provide the consumer with a better product at a lower cost;
it drives out inefficient and marginal producers, releasing resources to
higher-valued uses; it promotes diversity, giving consumers choices to fit
a wide array of personal preferences; it avoids permanent concentrations
of economic power, as even the largest firm can lose market share to a
feistier and hungrier rival. If, as the metaphor goes, a market economy is
governed by an invisible hand, competition is surely the brass knuckles
by which it enforces its decisions.3
Copyright © 2016. Harvard University Press. All rights reserved.

Why the court digressed is anyone’s guess. But regulation has fallen on
hard times. Even communist countries are now touting the free market. The
Chinese government in 2007, for example, enacted an Anti-Monopoly Law
for “the purpose of preventing and restraining monopolistic conducts, pro-
tecting fair market competition, enhancing economic efficiency, safeguarding
the interests of consumers and the interests of the society as a whole, and
promoting the healthy development of socialist market economy.”4
In the United States and elsewhere, the push since 1980 has been to de-
regulate. The belief is that competition with a light touch enforcement of

competition laws yields better outcomes. As the DOJ’s Antitrust Division

warned, “regulation can be an imperfect and very costly substitute for
‘regulation’ by market forces. Accordingly, exceptions to the general rule
of free market competition, protected by antitrust enforcement, should be
permitted only on compelling evidence that competition cannot work or
is inimical to some overriding social objective.”5 Thus, the modern inter-
pretation of Adam Smith’s “invisible hand” 6 has been central to the
changing attitudes toward antitrust enforcement.
Many adherents of neoclassical economic theory assume competition to
be “a self-initiating process,”7 which, when left alone by government regula-
tors, will generally allocate resources efficiently toward users who value
them the most. Any company’s attempt to secure or maintain market power
would likely be defeated by other well-informed profit maximizers—either
new entrants or existing competitors. The key proponents were economists
and lawyers associated with the University of Chicago.8 They generally as-
sumed that market participants were rational, were self-interested, and had
strong willpower, that most markets were competitive, that mergers and ver-
tical arrangements often created efficiencies, and that market forces would
often defeat any attempt to exercise market power.
The government, under this theory, operates outside the free market, and
must justify the necessity of its intervening and “displacing” competition.
Any suggestion to improve or manage competition smacks of socialism and
industrial policy. Government intervention should be limited to clear and
sustained instances of market failure, of which “only explicit price fixing
and very large horizontal mergers (mergers to monopoly) [are] worthy of
serious concern.”9 For some, even then, the government must proceed with
caution. The spontaneous free market forces will eventually defeat, through
expansion or de novo entry, this temporary market power.
Under the Chicago School theory, the government will often cause more
Copyright © 2016. Harvard University Press. All rights reserved.

harm than good. In attempting to preempt the exercise of market power,

the government may chill procompetitive behav ior. The concern is
that, unlike market-created impediments, market forces may not readily
overcome these government-imposed impediments to competition. The
greater concern around governmental intervention lies with the risk of
false positives, which can chill procompetitive market behavior and which
market forces cannot readily redress, rather than false negatives, which
entry or expansion eventually corrects.10

The Chicago School’s neoclassical economic views in favor of removing

or minimizing governmental restraints on the free market took hold in the
Reagan administration. Coinciding with the Reagan administration’s view
of governmental institutions as a necessary evil,11 competition advocacy
underscored how government interference likely causes more harm than
good, by inhibiting the market’s efficient allocation of scarce resources.
Consequently, the debate among some neoclassical economists is whether
and when the government should intervene in certain markets.

Dynamic Markets Will Correct Themselves

Many online industries are dynamic and fast-growing. The European
Commission, for example, took account of the dynamic market character-
istics when it approved Microsoft’s acquisition of Skype. In upholding the
Commission’s decision, the General Court observed that the consumer
communications sector was “a recent and fast-growing sector which is
characterised by short innovation cycles in which large market shares
may turn out to be ephemeral.”12 In such a dynamic context, the Court
noted, “high market shares are not necessarily indicative of market power
and, therefore, of lasting damage to competition. . . .”13
Some argue that courts and agencies should rarely, if ever, intervene in
dynamic industries.14 They claim that governmental intervention in dy-
namic economic markets will often harm consumers. The antitrust bene-
fits are limited, they argue, because online markets are so dynamic that any
market power is fleeting. (At times this is true, but no empirical evidence
supports any such blanket assertion.)
Another concern is that with dynamic industries it may be hard to attack
anticompetitive practices while preserving incentives to innovate. As one
antitrust official noted, “This can mean bringing an action to prevent con-
Copyright © 2016. Harvard University Press. All rights reserved.

duct that reduces innovation or it can mean declining to act where overly
aggressive antitrust enforcement risks chilling the type of vigorous, innova-
tive competition that brings long-term benefits to consumers. In this regard,
we recognize that when innovation leads to dynamic efficiency improvements
and a period of market power, it is not a departure from competition, but it
is a particular type of competition, and one that we should be careful not to
mistake for a violation of the antitrust laws.”15 This is especially sensitive for
antitrust scrutiny of product designs.16 Some, like one FTC commissioner,

argue for a very light touch: “Although I am not arguing that antitrust has
no place in technology markets, with a statute as elastic as Section 5, I think
the Commission ought to tread extremely lightly in that space. Otherwise, it
runs a serious risk of chilling innovation in what are arguably some of the
most important industries in our economy.”17

Reﬂections
The Chicago School has not influenced the EU competition policy to the
same extent it has influenced U.S. policy. Even in the United States, the
Chicago School—before the recent economic crisis—had begun losing its
luster. But aside from cases of collusion, the common wisdom that con-
tinues to emerge is that the costs and harms of regulatory intervention in
online industries will often exceed the benefits. As one FTC commissioner
observed, “Where the Chicago School tends to advocate a hands off ap-
proach based on an over-riding concern about false positives, one could
characterize the post-Chicago scholars as counseling a ‘light touch.’ ”18
Because online markets fueled by pricing algorithms should increase
competition by lowering search costs and entry barriers, and increasing in-
formation flows and market transparency, market power is transient.
Thus, it is argued that most online markets should not possess the char-
acteristics that make antitrust intervention (or regulation) necessary. Any
claims for antitrust or regulatory intervention should be treated with sus-
picion. The intervention will likely be unnecessary and harm consumers, as
its aim will be to protect firms in the old economy from the new economy.
While the algorithm-driven economy may herald the decline of “tradi-
tional” competition, the era of machine learning fueled by Big Data will
unleash greater efficiencies that improve our welfare.
Of course, we accept and acknowledge these benefits. But once we look
Copyright © 2016. Harvard University Press. All rights reserved.

beyond the shiny outer layer, the emerging online markets reveal several
significant dangers. Accordingly, it may be too soon to celebrate the op-
timal competitive order. We may want to wait with the champagne, at least
for a while.

Looking beyond the Façade of Competition

B IG DATA and technological innovations are neither good, bad, nor neu-
tral. As we’ll explore, their nature depends on how firms employ them,
whether their incentives are aligned with our interests, and certain market
characteristics. We’ll see that at times, Big Data and Big Analytics can pro-
mote a competitive online environment where we benefit. However, we
cannot uncritically assume that we will always benefit. When we critically
examine the complex algorithm-driven environment, we witness the im-
perfections of the new market dynamics. Thus, the risks to our well-being
are greater than many would admit.

Controlled Ecosystems: The Truman Show

New technologies are changing the dynamics of competition as we know it
and are giving rise to a new environment, which displays the characteristics
of competitive markets but is driven by different forces. The good old invis-
ible hand of competition, which safeguarded our welfare when we shopped
in our local fruit market, is being displaced by the digitalized hand.
Copyright © 2016. Harvard University Press. All rights reserved.

Think of the 1998 American movie The Truman Show—a controlled en-
vironment which is nothing more than a façade, but has the potential to
deliver relative joy to its subjects. The main beneficiary, of course, is the one
who controls the ecosystem. Likewise, some online markets may appear to
be subject to ordinary free market forces. We, like Truman, may think that
we’re ordinary consumers with ordinary lives with unremarkable pur-
chases. We have no idea about how, and the extent to which, we are being
exploited.
27

On a consumer level, we are entering the age of datafication, which in-

volves “taking all aspects of life and turning them into data.”1 In a data-driven
economy, sophisticated players will strive to improve their capacity to mon-
itor our online and offline activities, accumulate data, target us during key
purchasing opportunities, and react to changes with ever-increasing speed.
Using sophisticated algorithms, companies are engaging in data mining,
data trade, pattern recognition,2 demand estimation, and price optimiza-
tion.3 Our behav ior and preferences trigger individualized promotions, all
meant to help us make the right choice.4 But right for whom?

The Cost of Free: Data as Currency

With the rise of Big Data and Big Analytics, firms will not merely passively
track us. Instead, as one White House report noted, there is the “growing
potential for big data analytics to have an immediate effect on a person’s
surrounding environment or decisions being made about his or her life.”5
As the European Data Protection Supervisor observed, “Governments and
companies are able to move beyond ‘data mining’ to ‘reality mining’, which
penetrates everyday experience, communication and even thought.” 6
Increasingly our identities, both personal and professional, are shaped
through online media—Facebook, Twitter, WhatsApp, and LinkedIn, to
name but a few. Indeed, such is the perceived importance of these media
platforms that we deploy, at times, sophisticated means to shape our per-
ceived selves.7
Today, data is the currency which provides us with “ free” online ser vices
and an advanced Internet environment. For these media outlets to be avail-
able, a price is paid. We accept the “cost” of “ free.” We are not surprised to
receive targeted promotions, coupons, and ads. We expect our web searches
to deliver the right results, swift ly. We have come to expect the benefits that
Copyright © 2016. Harvard University Press. All rights reserved.

flow from this tracking and harvesting.

Yet, increasingly, we have concerns that the “cost” is now too high, and
that we have lost control over it. Many of us do not know what data is col-
lected about us, how it is being used, when, by whom, and for what pur-
pose. Indeed, we have increasingly expressed concern as to the invasion
of privacy from the tracking, harvesting, and use of our personal data. It
has been reported that over “90 percent of Americans feel they’ve lost
control over how their personal information is collected and used on the
Internet.”8

These privacy concerns will intensify. The digitalization of information,

our increasing reliance on smart technologies, and the growth of online
markets have significantly increased the volume and variety of available
data. Our information, the data, serves as a valuable commodity that trans-
lates into targeted advertisements, sales, and money. Lots of money. It is
therefore no surprise that companies are investing many resources into
harvesting and analyzing such data, and many powerful tech firms, as we’ll
see, view privacy protection technologies as a threat. These trends create
new gatekeepers and new forms of market power. They may also give rise
to new forms of anticompetitive behav ior that reduce our welfare.

Anticompetitive Dynamics
Subsequent chapters identify how the rise of sophisticated computer algo-
rithms and the new market reality can significantly change our paradigm
of competition for the worse—with more durable forms of collusion (be-
yond the reach of enforcers), more sophisticated forms of price discrimi-
nation, and data-driven monopolies that, by controlling key platforms (like
the operating system of your smartphone), dictate the flow of your personal
data, and who gets to exploit you.
The scenarios below are not conjectural. Competition agencies are al-
ready grappling with the scenarios we identify. Officials from the United
States, U.K., France, and Germany, citing our earlier work, have publicly
recognized the potential harm from these scenarios and question the
adequacy of their current enforcement tools.9 Many enforcers have also
privately shared with us the concern that their tool kit at times will be in-
adequate to prevent and redress the harm.
Copyright © 2016. Harvard University Press. All rights reserved.

Collusion: From Smoke-Filled Hotel Rooms to Vapor-Filled Data Centers

As we saw in Chapter 2, industries are shifting from a pricing environment
where store clerks stamped prices on products, to dynamic, differential
pricing where sophisticated computer algorithms rapidly calculate and up-
date prices. At times dynamic pricing is good—one example we explore is
“smart” parking meters in San Francisco. But as pricing shifts from humans
to computers, so too will the types of collusion and behavioral exploita-
tion in which companies may engage. Part II considers a classic antitrust
mainstay—cartels—to explore the shift from a world where executives

expressly collude to one in which computers facilitate collusion. We illus-

trate how in some markets, the industry-wide use of pricing algorithms,
rather than increasing competition, may result in us paying more for goods
and services.

Approaching Perfect Behavioral Discrimination

Part III takes us in a different direction, to a market where the prices and
products you see differ from those offered to your neighbors, relatives,
friends, and families living on the other side of town. Companies collect
data about you and track your behavior to better predict what you are like-
lier to buy and how much you are willing to spend. Here we will see the
expansion of behavioral advertising and price discrimination across online
markets. We consider the means by which companies might approach, but
not achieve in the near future, perfect behavioral discrimination, and the
possible anticompetitive effects that may follow.
As part of our discussion of discrimination we also consider the emerging
role of intermediates, such as price comparison websites. We saw in Chapter 1
how these platforms can promote customers’ decision making and intensify
competition among suppliers. But we will consider in Part III how these
price comparison websites, in changing the competitive dynamics, may, at
times, actually harm consumers with fewer choices and higher prices.

Frenemy Dynamics
A growing, and seemingly appealing, part of the online marketplace is free
goods and ser vices. The proliferation of free mobile apps seemingly bene-
fits consumers (as well as advertisers, smartphone manufacturers, mobile
carriers, and independent application developers) by reducing search costs
Copyright © 2016. Harvard University Press. All rights reserved.

and increasing demand.

Part IV involves the dynamics of “Frenemy”—where a relationship of both
competition and cooperation exists between the super-platforms and inde-
pendent apps. We consider the world of mobile and tablet operating systems,
in which two super-platforms—Apple’s iOS and Google’s Android mobile
software platforms10—dominate mobile phones. Each super-platform, like a
coral reef, attracts to its ecosystem software developers, apps, and accessory
makers.
We reflect on the rise of super-platforms and the way in which they foster
a mix of competition and interdependence among market participants.

Firms cooperate to extract data from individuals and promote asymmetrical

information flows to foster behavioral exploitation, while simultaneously
competing among themselves over the consumer surplus. Extraction and
capture may be viewed from an evolutionary perspective: a den of lions co-
operates to circle the gazelle and they then compete over which of them gets
the choice cuts. They all benefit from the combined effort, yet the dominant
lion gets the best cut, which further enhances its power.

Possible Intervention
The three core dynamics discussed above are not always easy to trace. Mar-
kets may on their surface appear competitive—but these dynamics can
progressively hinder our autonomy, livelihood, and welfare. In other words,
despite the mirage of ordinary competitiveness, the emerging online markets
will at times reduce, rather than increase, our well-being, as the competitors’
pricing algorithms tacitly collude in a transparency-enhanced environment,
price discriminate, or collectively extract our personal data and compete over
how best to capture our wealth.
In this reality, in which algorithms and data pools provide the founda-
tion for possible unilateral, coordinated, and Frenemy behavior, is the “in-
visible hand” still a viable concept?
As we will show, markets may be dynamic but still be dominated by a
few firms. In this controlled ecosystem, the traditional signposts of greater
free competition—notably market transparency, entry, and choice—may
be merely a mirage.
The common competitive ideal is that we would want many companies
to compete to provide the best products and ser vices. But if the critical re-
source at this point is data—not merely to target advertising, but also to
optimize the products and ser vices themselves—the firms with the most
Copyright © 2016. Harvard University Press. All rights reserved.

data are not merely in the best position to dominate their own sectors—
they are also poised to take over adjacent fields. Further, to the extent that
such firms compile politically sensitive information about users, and me-
diate their experience of content, they are also power ful political actors.
With that in mind, careful intervention may be necessary to remedy
market failure and promote customer welfare.
From our experience, having mentioned possible intervention, we ex-
pect a roar of dismay (or outrage) from some stakeholders. Indeed, some
of the dominant tech players work hard through various channels to cap-
ture and frame the debate—conflating criticisms of the means by which

they increase their profitability with criticisms of technology itself, and

characterizing those criticisms as a threat to innovation, investment, and
competition.
Needless to say, our aim is not to argue in favor of intervention per se.
But the anticompetitive effects we identify can be significant and durable.
There is no reason to tolerate these market failures simply because they
occur in the new digital economy.
So where exactly does this leave us? Is competition law salvageable, or is
it simply a relic of a predigital economy? Or, perhaps paradoxically, is the
less technocratic, more political competition law of American antitrust
prevailing in the mid-twentieth century the right direction for current
policymakers to travel? Part V addresses several means by which govern-
ments may address the concerns raised in this book. It outlines the possible
costs and benefits of intervention and the ability to fine tune instruments to
improve the competitive landscape.
Importantly, our scenarios of possible collusion, behavioral discrimina-
tion, and Frenemies do not challenge innovation, technology, and efficien-
cies. Our aim is to go beyond the slogans and myths and examine the
relative costs and benefits of these phenomena, and to highlight the effects
that current and future dynamics may have on our welfare. The mirage of
competition will compound, rather than solve, the problems we identify
and worsen, rather than improve, our well-being. Competition officials with
their current tools can fix some but not all of the problems. Regulation should
no longer be a dirty word. Smart regulation, in a data-rich world, may prove
quite beneficial.

Food for Thought

Beyond the “laissez-faire competition good, regulation bad” refrain, chal-
Copyright © 2016. Harvard University Press. All rights reserved.

lenging questions await us. For instance—is the algorithm price the com-
petitive price, or merely a fiction created by the digitalized hand? Turning
to a famous economist, Friedrich A. Hayek, we inquire whether the
emergence of super-platforms— companies that dominate the digital
landscape—could indicate a monumental shift toward the attainment of all
knowledge. Platforms’ sophisticated computer algorithms could increas-
ingly determine the competitive market price. Data collection by leading
platforms like the car-sharing app Uber, and super-platforms like Google,
Apple, and Amazon, could create an economy which, for all purposes, is

planned not by bureaucrats or CEOs, but by the technostructure. If so, a

subsequent question arises: if private firms can harness Big Data and Big
Analytics to effectively set prices, can governments use the same tools to
monitor industry prices, or even determine a competitive price? If Uber,
which doesn’t own any cars or employ any drivers, can determine prices,
why can’t the government?
Copyright © 2016. Harvard University Press. All rights reserved.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Copyright © 2016. Harvard University Press. All rights reserved.

The Collusion Scenarios

W HEN HUMANS ARE PROSECUTED in the United States for price fi xing, they
generally go to jail. What happens with the rise of pricing algorithms,
when competitors’ computers fi x prices (or help fi x prices)? We explore this
issue here.
The antitrust community is accustomed to company executives fi xing
prices, allocating markets and bids, and reducing output. The fi lm The
Informant! dramatizes these real-life executives who every year conspire
around the world to fi x prices and reduce output. Cartels are generally
regarded in the antitrust world as “no-brainers.” The cartel agreement,
even if unsuccessful, is typically condemned as per se illegal—being anti-
competitive by object. The executives and companies have few, if any, legal
defenses. And in the United States, among other jurisdictions, the guilty
executives are often thrown into prison.
Cartel agreements are not always easy to establish and maintain. Neo-
classical economic theory would suggest that many cartels are unstable as
they are susceptible to distrust, cheating, or detection. Yet empirical obser-
vations suggest that in practice cartels are more durable than neoclassical
Copyright © 2016. Harvard University Press. All rights reserved.

economic theory posits.1

So why do firms collude? Often because it is easier than competing. By
agreeing to raise or stabilize prices, companies earn greater profits. In al-
locating markets, each cartel member can dominate its territory without
fear of its competitors entering.
Humans have for many years been the moving force behind these price-
fi xing activities; they have decided to what extent they should increase
prices, reduce output, allocate bids and markets, or eliminate other par-
ameters of competition. They may meet yearly or even monthly as the cartel
35

adjusts its activities. Humans have colluded on every thing from turtles2 to
packaged ice3 to rare banknotes.4 To deter cartels, the United States “has
steadfastly emphasized the importance of individual accountability and
stiff corporate fines.”5 Despite rising fines, prison sentences, and attrac-
tive leniency programs, cartels persist.6
So, as more firms and industries migrate to pricing algorithms, does that
spell the end of classic cartels, or does it create new ways to collude?
In this part we consider the role played by algorithms in facilitating car-
tels and illegal activity. Our focus goes beyond a simple collusion story
in which the co-conspirators use computers to support their cartel ac-
tivity. Rather, our interest is in new dynamics that could widen the circum-
stances in which anticompetitive activity may take place.
We note how Big Data and Big Analytics—in increasing the speed of
communicating price changes, detecting any cheating or deviations, and
punishing such deviations—can provide new and enhanced means to
foster collusion. The danger here is not express collusion where computers
limit competition through “agreement” or concerted practice, but more
elusive forms of collusion, achieved through subtler means, which do not
amount to a hard-core cartel, and are beyond the reach of the law. Altogether
we consider four scenarios in which computer algorithms may promote
collusion.
The first scenario—Messenger—concerns humans’ agreeing to collude
and using computers to execute their will. Th is is a simple extension of
human will—the use of the IT environment to enhance existing collusion.
Under this scenario, humans collude. They use computers to assist in im-
plementing, monitoring, and policing the cartel or to facilitate information
exchange and signaling; in the United States and elsewhere, they go to jail
if caught.
Our second scenario—Hub and Spoke—is more challenging. Here we
Copyright © 2016. Harvard University Press. All rights reserved.

consider the use of a single pricing algorithm to determine the market price
charged by numerous users. In this framework, a cluster of similar vertical
agreements with many of the industries’ competitors may give rise to a
classic hub-and-spoke conspiracy, whereby the algorithm developer, as the
hub, helps orchestrate industry-wide collusion, leading to higher prices.
The third scenario—The Predictable Agent—is even more challenging.
It explores how we are shifting from a world where executives expressly col-
lude in smoke-filled hotel rooms, to a world where pricing algorithms act
as predictable agents and continually monitor and adjust to each other’s

prices and market data. In this new world, there is no collusive agreement
among executives. Each firm unilaterally adopts its own pricing algorithm,
which sets its own price. The result is algorithm-enhanced conscious
parallelism—or as we call it, Tacit Collusion on Steroids.
Finally, we consider the most challenging collusion scenario—Digital
Eye. The computers, in learning by doing, determine independently the
means to optimize profit. Artificial intelligence operating in enhanced
market transparency leads to an anticompetitive outcome, with no evidence
of any anticompetitive agreement or intent. In this scenario we may not
even know when something is amiss. In the end, we may think the mar-
kets, driven by these technologies, are competitive. And yet, we’re not ben-
efitting from this virtual competition.
Copyright © 2016. Harvard University Press. All rights reserved.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Copyright © 2016. Harvard University Press. All rights reserved.

The Messenger Scenario

We will not tolerate anticompetitive conduct, whether it occurs in

a smoke-fi lled room or over the Internet using complex pricing
algorithms. American consumers have the right to a free and fair
marketplace online, as well as in brick and mortar businesses.
—Bill Baer, U.S. Department of Justice, 2015

U NDER OUR FIRST collusion scenario, Messenger, humans are the mas-
ters who agree to collude and map out the cartel. The computer algo-
rithms are the messenger, which the cartel members program to help ef-
fectuate the cartel and monitor and punish any deviation from the cartel
agreement.
From an enforcement perspective, this is a no-brainer. Competition law’s
concept of agreement can be applied straightforwardly. Prosecutors, with
sufficient evidence of the humans’ agreement or concerted practice, will
have little difficulty in condemning the use of computers to facilitate the
cartel.
To illustrate: in a classic cartel, executives from rival firms secretly agree
to fix prices, allocate markets or bids, or reduce output.1 Here, the executives,
after secretly colluding, leave it to their computer algorithms to monitor and
enforce the illegal agreement.
Copyright © 2016. Harvard University Press. All rights reserved.

The advancement of computer technology enables forms of collusion

that are just as pernicious and anticompetitive as the cartels of yesteryear.
It made the news in 2015 when the DOJ warned antitrust lawyers, econo-
mists, and scholars of the illicit use of complex pricing algorithms. The
DOJ charged members of a price-fi xing scheme involving posters sold
through Amazon Marketplace. David Topkins and his coconspirators
adopted specific pricing algorithms that collected competitors’ pricing in-
formation for specific posters sold online and applied the sellers’ pricing
rules. The competitors used the computer algorithms with the goal of
39

coordinating changes to their respective prices2 for the sale of their

posters.3 The cartel began as early as September 2013. Topkins pled guilty.
The ongoing DOJ investigation into price fi xing in the online wall décor
industry led to a price-fi xing charge against Daniel William Aston and
his U.K.-based company.4 Assistant Attorney General Bill Baer of the DOJ’s
Antitrust Division commented that “U.S. consumers deserve competitive
markets when they shop online,” and that the DOJ “will continue to pros-
ecute conspiracies that subvert online competition.”5
Another recent example of illicit use of computer algorithms was the
major financial firms’ manipulation of benchmark interest rates. In May
2015, five banks— Citicorp, JPMorgan Chase & Co., Barclays PLC, the
Royal Bank of Scotland plc, and UBS AG—all pled guilty to felony charges
of conspiring to manipulate the price of U.S. dollars and euros exchanged
in the foreign currency exchange spot market.6 The DOJ described how
traders at Citicorp, JPMorgan, Barclays, and RBS—self-described members
of “The Cartel”—used for five years an exclusive electronic chatroom and
coded language to manipulate benchmark exchange rates.7 The new U.S.
attorney general noted how the steep financial penalties “should deter com-
petitors in the future from chasing profits without regard to fairness, to
the law, or to the public welfare.”8 This is optimistic, given the limited suc-
cess that steep fines have had in curbing such behavior.9
Also illustrative is a Greek competition authority investigation into the
use of IT systems to facilitate anticompetitive practices. In 2010, the Greek
Hellenic Competition Commission fi ned Carrefour Marinopoulos €12.5
million for a number of infringements, including resale price maintenance
(RPM).10 The practice, which is illegal under EU competition law, was de-
tected within Carrefour’s franchise network.11 In its decision, the Compe-
tition Commission emphasized the role played by Carrefour’s joint IT
system in facilitating the infringement. This system, “which formed an
Copyright © 2016. Harvard University Press. All rights reserved.

integral part of the [franchise] network,”12 allowed the franchisor to mon-

itor any deviations by franchisees from its recommended resale prices,
“with the sole purpose of appraising the ‘appropriateness’ of such retail
prices in relation with the total pricing policy of the network.”13 Moreover,
the Competition Commission added, the nature of the IT system in ques-
tion “rendered the management of prices by the franchisees difficult and
time-consuming in practice, thereby facilitating price rigidity.”14
An earlier example of the use of computers to facilitate collusion is the
DOJ’s civil Airline Tariff Publishing case.15 The United States alleged that

the defendant airlines used their computerized fare dissemination ser vices
to freely negotiate among themselves supracompetitive fares in multiple
markets. No one questioned that the defendants’ computerized fare dis-
semination system had a procompetitive purpose in supplying travel
agents with basic information about airline fares for specific routes. How-
ever, the antitrust risks arose when the defendant airlines also used this
system as a forum to exchange information that was of limited or no use to
consumers, but was important to the other airlines in communicating and
agreeing upon supracompetitive fares.
The DOJ asserted that the defendant airlines essentially signaled their
concurrence in or disagreement with entreaties to raise fares and/or elimi-
nate discounted fares through the First and Last Ticket Dates. Essentially,
the defendant airlines communicated among themselves relatively costless
proposals to change fares through these footnote designators with First and
Last Ticket Dates. They employed sophisticated computer programs to pro-
cess all this fare information, which enabled them to monitor and analyze
their competitors’ responses to current and future fares on certain routes.
These negotiations at times would link fare changes among different routes,
and would continue for several weeks until all the airlines had indicated
their commitment to the fare increases by filing the same fares in the same
markets with the same First Ticket Date. Likewise, the airlines used the
Last Ticket Dates in connection with the footnote designators to commu-
nicate proposals to eliminate discounted fares currently being offered to
consumers. Not only did this computerized fare dissemination system en-
able the defendants to negotiate higher fares, it importantly enabled them to
verify that such fares would stick, and to signal retaliatory measures against
any airline that did not go along with specific fares for specific routes.16
In a modified scenario of this case, the airline executives could agree
broadly not to compete along certain routes and program their computers
Copyright © 2016. Harvard University Press. All rights reserved.

to ensure that each airline was allocated its set of customers, to monitor any
deviations, and to react automatically to any defections. Importantly, the
computers here are used to execute the task that they were set, using pre-
loaded data and orders. While faster than their creators, the computer
algorithms reflect—and are limited by—the amalgamation of human in-
structions. The computers simply help execute the humans’ anticompetitive
agreement.
These examples illustrate the way in which computers may be used to
facilitate and monitor anticompetitive agreements. Collusion, however,

may at times be established through weaker forms of communications,

which do not give rise to an agreement yet are still condemned under anti-
trust laws. Here as well, computers and algorithms may play a role as the
executors of human will—for instance, for the exchange of information or
signaling between companies.17
All in all, from an enforcement perspective, the Messenger collusion
scenario is relatively straightforward. The illegality inheres in the agree-
ment or collusion among humans. So, while pricing algorithms facilitate
the illegal behav ior, the noteworthy conduct can be analyzed through a
“ human” prism. Thus, executives who agree to fi x prices cannot blame
their computers.
At the administrative level, competition enforcers can rely on the case law
involving an illicit agreement or concerted practice and use the concept of
“object”18 or “per se” illegality to establish violations and impose fines on
the companies.19 The computers’ failure to effectuate or monitor the agree-
ment does not affect the agreement’s illegality.20 The stronger the evidence of
an anticompetitive agreement in the Messenger scenario, the less the need
for evidence of intent to establish the concurrence of wills or the agreement’s
purpose.

The Algorithm as an Intermediary

From a legal perspective, the use of computers to help execute the cartel’s
tasks does not change the “ human” prism. The use of algorithms facilitates
tasks which humans would other wise execute. Although the legal implica-
tions are the same—namely, that humans are guilty if they agreed to fi x
prices—the technology shift may have an important psychological impact
on prospective colluders.
The computer, by increasing the distance between the human and the
Copyright © 2016. Harvard University Press. All rights reserved.

illegal day-to-day activity, can reduce the guilt of wrongdoing. Executives

who fix prices often find excuses for their criminal behavior.21 By using
computer algorithms to fix prices, rather than secretly meeting and com-
municating with the other coconspirators, executives will likely feel less cul-
pable. The computer, in serving as an intermediary, may help individuals
wash their hands of the illicit conduct.
To explain how the presence of an intermediary may facilitate such
actions, we enter the area of behavioral experiments, which explore the
issue of distancing in decision-making.

One famous example—and the basis for Peter Gabriel’s song “Milgram’s
37 (We Do What We’re Told)”22—is Stanley Milgram’s electric shock ex-
periments.23 You might have seen the black-and-white videos24 in which
the test subject and a confederate of the experimenter were told that the
experiment tested the effects of punishment on memory. To determine
their assigned roles, the confederate and test subject drew lots, which were
rigged so that the test subject always received the teacher role. The teacher-
participant then administered a test in which the confederate-learner was
to memorize word pairs. Each time the confederate-learner answered in-
correctly, the teacher-participant was to administer an electric shock to the
learner. A “shock generator” had thirty clearly marked voltage levels,
ranging from fifteen to 450 volts, with designations from “Slight Shock” to
“Danger: Severe Shock.” Two switches after the last designation were simply
marked “XXX.” Unbeknownst to the teacher-subject, the confederate was
not actually receiving electric shocks. The confederate-learner gave stan-
dardized responses. In one variation of the experiment, the confederate-
learner pounded on the wall of the room in which he was bound to the
electric chair after the 300 volt shock was administered. The teacher-subject
could hear the pounding. Thereafter, the learner no longer responded;
the experimenter instructed the teacher-subject to treat the absence of a
response as a wrong answer, and to continue with the experiment. As the
experiment continued, the teacher-participant was told to administer in-
creasingly intense shocks to the now nonresponsive confederate-learner,
even to the levels marked “XXX.” These experiments actually sought to
measure at what voltage level the teacher-participant would disobey
and refuse to continue with the experiment. Milgram varied the situa-
tional factors to determine the extent to which they altered the degree of
obedience.
Before his famous experiment, Milgram asked college students, psychi-
Copyright © 2016. Harvard University Press. All rights reserved.

atrists, and middle-class adults for their predictions. No one predicted

that the experiment participants would administer shocks above 300 volts.
Nearly all the subjects, they predicted, would disobey the experimenter,
only 4 percent of the subjects would administer 300 volts, and only a path-
ological fringe (about one in a 1,000) would administer the highest shock
of 450 volts.25 They were wrong. In his primary experiment, all forty sub-
jects administered shocks up to 300 volts (when the learner-confederate
pounded on the wall), and twenty-six subjects complied until the end and
administered 450 volts.

Milgram’s experiments highlight the importance of situational factors in

explaining how ordinary blue-collar workers and white-collar professionals,
contrary to their own expectations, administered a lethal dosage of 450 volts
to an unresponsive, possibly dead, fellow test-subject.
One situational factor is to create opportunities for the diff usion of re-
sponsibility or abdication of responsibility for negative outcomes. Milgram
ran the same experiment but changed certain conditions. In one variation,
the teacher-participant administered only the test, while a confederate ad-
ministered the shock for every wrong answer. In this situation, the degree
of compliance was even higher: thirty-seven of the forty participants pro-
ceeded to the highest voltage level. Greater compliance may also be attrib-
utable to the reduced salience, as the teacher-participant was not directly
administering the shock.
Two other variations of Milgram’s experiment demonstrate the impact
of increased salience on compliance. When the confederate-victim was
in the same room as the teacher-participant, fewer teacher-participants
administered the maximum voltage; compliance was even less when the
teacher-participant had to force the victim’s hand onto a shock plate.26
Let us consider the implications of Milgram’s studies on computer-
assisted collusion. Would the use of an intermediate algorithm increase
the willingness to collude?
People generally perceive indirect harms as less problematic than direct
harms.27 Thus, price fi xing may already appear less problematic and be
easier for cartel members when they do not deal directly with the end cus-
tomer, such as cartels for intermediate manufactured goods and ser vices.28
Cartels’ perceived illegality will be further diminished if computers, rather
than humans, monitor and punish any deviations from the cartel agree-
ment. Moreover, as we will see with the next scenarios of collusion, price
fi xing may be easier to defend when the competitors’ algorithms not only
Copyright © 2016. Harvard University Press. All rights reserved.

monitor and punish any deviations, but tacitly collude.

Reﬂections
We see how computers, in facilitating communications, monitoring for any
cheating and punishing any defections, may help humans collude. We also
note how the use of the computer as an intermediary could weaken the in-
dividual’s sense of illegality and, in so doing, further facilitate the illegal
activity.

It is, however, impor tant to remember that just as computers may be

used to facilitate cartels, they may also be used by individual companies to
execute more aggressive competitive behav ior. For instance, they may pro-
vide a sophisticated tool in the hands of a maverick firm which would in
fact destabilize cartel activity.
We thus return to the heart of the Messenger scenario: algorithms may
facilitate collusion when such is the desire and intention of their operators.
As messengers, algorithms are neither a negative nor positive force; rather,
they are a technological extension of the human will.
Copyright © 2016. Harvard University Press. All rights reserved.

Hub and Spoke

H AVING CONSIDERED the “simple” Messenger scenario, we next consider

instances in which computer algorithms are used as the central “hub”
to coordinate competitors’ pricing or activities. While the competitors do
not directly contact or communicate with each other, the overall impact of
the practice is akin to horizontal collusion.

Traditional Hub and Spoke

Hub-and-Spoke conspiracies are not unique to the online environment or
antitrust. After all, both cocaine and price-fi xing cartels may be facilitated
by such arrangements. These conspiracies, as one court described, take
form when “a central mastermind, or ‘hub,’ controls numerous ‘spokes,’ or
secondary co-conspirators.”1 The spokes each “participate in independent
transactions with the individual or group of individuals at the ‘hub’ that
collectively further a single, illegal enterprise.”2 A common example is
where the mastermind recruits different coconspirators to carry out the il-
legal enterprise’s various functions, such as procuring the guns, stealing
Copyright © 2016. Harvard University Press. All rights reserved.

the get-away car, laundering the money, and so on.

The coconspirators need not communicate with, or even know, each
other. As the U.S. Supreme Court noted, “an unlawful conspiracy may be
and often is formed without simultaneous action or agreement on the
part of the conspirators.”3 But to show a single hub-and-spoke conspiracy,
rather than multiple independent conspiracies, there must be a rim: there
must be some overall awareness of the conspiracy and that “each defen-
dant knew or had reason to know of the scope of the conspiracy and . . .
reason to believe that their own benefits were dependent upon the success
46

of the entire venture.”4 In a hub-and-spoke price-fi xing conspiracy, the

competitors who form the wheel’s spokes must be aware of the concerted
effort to stabilize prices. An easy case is where the hub “has not only com-
mitted to vertical agreements, but has also agreed to participate in the
horizontal conspiracy.”5
Courts have long recognized the existence of hub-and-spoke price-fi xing
conspiracies. Often the hub operates at one level of the market structure,
coordinating a pricing-fi xing agreement among competitors at a different
level, the spokes.6 In the Interstate Circuit case,7 for example, a movie the-
ater owner approached each movie distributor individually, told each
movie distributor of the contemplated conspiracy, told each movie distrib-
utor that the other movie distributors would be invited to join the con-
spiracy, and said that cooperation of all eight distributors was essential
for the conspiracy to work. By giving their consent to the conspiracy and
agreeing to participate in it, both the movie theater owner (the hub) and
the eight movie distributors (spokes) were liable.
Let us now consider the hub’s active involvement in the collusion.8 The
European Commission condemned such active support and enabling prac-
tice as part of its investigation into the manipulation of the LIBOR. The
Commission condemned the hub in this case (U.K.-based broker ICAP) for
its “serving as a communications channel between a trader of Citigroup
and a trader of RBS and thereby enabling the anticompetitive practices
between them.”9 In the U.S. e-books case, Apple and five large book pub-
lishers were held liable for conspiring to raise the prices of e-books, and in
particu lar the price of new releases and New York Times bestsellers.10 The
Apple case is illustrative of the potential power of platforms and their
ability to distort competition. The use by Apple of an agency agreement
and wide parity clauses11 served to facilitate the publishers’ collective
(and collusive) action.12
Copyright © 2016. Harvard University Press. All rights reserved.

Algorithm-Fueled Hub and Spoke

With this overview of traditional hub-and-spoke conspiracies, we now ex-
plore instances where a computer algorithm executes the “hub” function
to facilitate collusion among competitors.
Suppose each competitor in a local market sees the shift to dynamic
pricing. But creating and refining the algorithms are too expensive. So
each competitor outsources its pricing to an upstream supplier’s pricing

algorithm. The competitors do not interact directly with each other, yet they
all use the upstream supplier’s pricing algorithm. Here we face an industry-
wide use of a single algorithm, which competitors use to determine the
market price or react to market changes. As a result, the market behavior of
the competitors could be “magically” aligned, when they all use a similar
“brain” to determine their price strategy.
It is important to note how the algorithm-fueled hub and spoke differs
from our first scenario—the Messenger—which considered the computer
as a mere extension of the humans’ illegal agreement. In an algorithm-
driven hub and spoke, the computer does not merely execute the orders of
humans; rather, it is the competitors’ use of the same pricing algorithm that
stabilizes prices and dampens competition.
It is also important to distinguish between the traditional hub and spoke
conspiracy, in which the immediate aim is horizontal collusion, and each
vertical link is in furtherance of that aim, from an algorithm-driven hub
and spoke. The latter may, of course, be the result of an intentional attempt
to dampen competition, but it may also occur due to unintentional align-
ment and use of similar algorithms to monitor prices. In other words,
collusion may be the consequence, but not necessarily the original aim,
when each competitor opts for the same third-party pricing algorithm.
Our focus is therefore on instances in which the use of a single algorithm
as a hub would lead to a de facto alignment among rivals that dampens
competition. One example is when many competitors outsource their pricing
to a third-party vendor. Indeed, as pricing has become more dynamic and
data-driven, companies are increasingly relying on third-party vendors.
Take, for example, the pricing ser vices provided by one of these
vendors—Boomerang Commerce. This third-party vendor’s “platform
analyzes over 100 discrete data points per SKU, including competitors’
prices” to help “retailers re-price millions of products in real-time.”13 Boo-
Copyright © 2016. Harvard University Press. All rights reserved.

merang “makes soft ware that online retailers use to evaluate competitors’
pricing on similar goods, and then analyze a variety of factors to decide
when to match prices, or drop them lower or push them higher than a
competitors’.”14 Among Boomerang’s customers are Staples, Sears, and
Groupon Goods.15 Boomerang also promotes how its clients can avoid an
algorithm-fueled price war:

A new upstart, Jet.com, is building an online product cata logue with

30 million-SKUs and has pledged to underprice Amazon. What does

this mean to you? An opportunity to take control of your pricing, and

turn a destructive “Race to the Bottom” pledge into a level playing field
where you can compete.
Amazon’s pricing machine will surely match Jet.com’s aggressive
prices to prevent customer churn. For retailers, blindly matching prices
is not an answer. Top 100 retailers are utilizing Boomerang’s innovative
pricing technology to develop real-time pricing strategies to compete and
grow profits.16

No one accuses Boomerang or its clients of fi xing prices. But let us develop
a potential case:
Suppose Staples uses Boomerang to price its online office supply prod-
ucts. What are the implications if Staples’ competitors also decide to use
Boomerang’s price optimization soft ware? Each competitor can claim that
it never intended to fi x prices; rather, it was too costly or time-consuming
to independently develop the pricing algorithm and collect the needed
market data. Moreover, as Boomerang enlists more office supply retailers
as clients, its pricing algorithm will likely improve as it has more data and
greater opportunities to experiment with prices and recalibrate. Quite
simply, as more retailers use Boomerang, its self-learning pricing algorithm
has more data to refine its pricing strategies for each client. Any algorithm
that a smaller retailer could independently develop would likely be inferior
to Boomerang’s. Thus, each retailer would have independent business jus-
tifications for using Boomerang’s pricing ser vices, as its algorithms are
smarter with more data and opportunities to experiment.
But if each major office supply retailer delegates its pricing to the same
vendor, and if the vendor promises in its advertising materials to maximize
profits, then surely each retailer knows that the vendor’s pricing algorithms
will make use of its own and its rivals’ information in assessing prices. Each
Copyright © 2016. Harvard University Press. All rights reserved.

competitor would surely be aware that using the same third-party pricing
vendor would likely influence market conditions.
Thus, we start to see the shape of a traditional hub-and-spoke conspiracy,
where each retailer provides the hub with data and pricing authority,
knowing that its rivals are doing the same. The vendor’s pricing algorithm
does, in fact, use the market information it collects from each retailer in de-
termining the optimal prices for each retailer’s products. Prices stabilize as a
result, and the retailers’ and algorithm vendor’s profits increase. Each com-
petitor might have independent business justifications for electronically

sending its data to a third-party pricing vendor. But the competitors also
recognize that doing so collectively will likely increase prices and their
profits. So an algorithm-fueled hub-and-spoke conspiracy arises when com-
petitors outsource pricing to a “similarly minded” or identical algorithm.

Uber’s Hub and Spoke

Another algorithm-fueled hub-and-spoke conspiracy may involve plat-
forms, which bring together sellers and purchasers. When the platform’s
algorithm sets the price and many competing operators agree to use the
platform’s price, that too may dampen horizontal competition.
To illustrate the possibility of the hub setting the price for its spokes,
we’ll consider Uber Technologies Inc.’s online platform for car ser vices. Ba-
sically, Uber connects drivers and passengers in over 300 cities, using a
single pricing algorithm to set the price for car ser vices in each city.17
For those unfamiliar with Uber’s success story, in mid-2015 Uber was
valued close to $51 billion. To put this into perspective, only two venture
capital–backed private start-ups had ever surpassed a $50 billion valuation:
Uber and Facebook. And Uber reached this valuation two years faster than
Facebook.18
This rise in value and usage echo the significant benefits the online plat-
form provides. Passengers don’t have to wait in line for a cab (or try to hail
one). They obtain full information on available rides and fares; and often
the fare is less than that of traditional taxi companies. This—coupled with
reduced search costs, and information about the driver’s availability, dis-
tance, and rating—has fueled Uber’s success. In addition, by linking a large
number of users and drivers—some part-time, others full-time—the plat-
form can promote a more efficient use of resources.
And yet the success story is not all rosy. Uber has its critics. Concerns
Copyright © 2016. Harvard University Press. All rights reserved.

have been raised over the nature of the relationship between Uber and its
drivers and the responsibility the company should have for their welfare
and actions.19 Fairness concerns have also been raised. Critics argue that
Uber avoids the costs of safety and other regulations that traditional taxi
companies incur, and thus enjoys an unfair competitive advantage.20
For our purposes, we’ll focus on another dimension—namely, the auto-
mated price-setting by Uber’s algorithm.
We approach this discussion with care. Having used Uber and its com-
petitor Lyft many times, we value the ser vice. Indeed, on a busy day in

London or New York, such applications often offer commuters a cost-

effective alternative. With these benefits in mind, we consider how a hub-
and-spoke model could, under certain conditions, generate a completely
different market dynamic.
Uber’s algorithm has been referred to as an “algorithmic monopoly” as
it may mimic a perceived competitive price rather than the true market
price.21 Uber’s drivers typically do not negotiate discounts with customers.
Instead, under its “[n]o cash, no tip, no hassle” policy, Uber’s algorithm
sets the price and automatically charges the passenger’s credit card on fi le.22
Uber takes between 20 and 25 percent of the fare; the driver gets the rest.23
Uber’s dynamic pricing algorithm provides passengers a baseline standard
fare, which increases when consumer demand in a location exceeds the
supply of available drivers.24 For example, during a New York snowstorm,
some rides on Uber cost 8.25 times more than normal.25 Controversially,
the algorithm was also reported to have implemented significant surge fees
of up to four times the normal rate when demand for rides escalated in the
midst of a hostage situation in downtime Sydney. Uber later apologized
and refunded the charge.26
Thus, Uber’s algorithm determines for hundreds of competing drivers
the base price for the trip, when to implement a surge price, for which areas,
for how long, and to what extent. Granted, the customer can compare the
Uber price to alternatives (such as taxis or other car ser vice platforms like
Lyft), but as more customers and drivers rely on Uber’s platform, one may
wonder what effect its algorithm could have on the market price.
To illustrate, let us suppose Uber is the dominant car ser vice platform
in Nashville. Let us also assume taxis, for various reasons, are not a signifi-
cant competitive restraint. What, if any, competition is left? Uber drivers
do not offer discounts, as Uber’s pricing algorithm determines the fare. Nor
will Uber drivers necessarily compete by offering better ser vice. One study
Copyright © 2016. Harvard University Press. All rights reserved.

of Uber and Lyft drivers found that they “distanced themselves from one
another by checking other drivers’ locations on the map so that they did
not compete with each other for passenger requests. When drivers desired
a break but did not want to turn off their driver applications to benefit from
an hourly payment promotion, they parked in between the other ride-
sharing cars in order not to get any requests.”27 So, as more people use
Uber in Nashville, more drivers will likewise gravitate to Uber’s platform,
which further reduces users’ wait time, increasing Uber’s appeal. Unless
passengers switch en masse to another platform, Uber’s algorithms will

have greater market power to set the price (including surge pricing) and
increase profits. Drivers won’t complain as they get 75 to 80 percent.

Enforcement Challenges
The above scenarios illustrate the dangers when many competitors rely on
a common algorithm. Higher prices are the likely outcome. To be clear, a
single vertical agreement by itself may not necessarily be anticompetitive
and does not necessarily reflect an attempt to fi x market prices. The con-
cern arises when a cluster of similar vertical agreements within a market
gives rise to a classic hub-and-spoke conspiracy. Rivals end up using the
same algorithm (the hub), which thereby softens competition and leads to
higher prices.
From an enforcement perspective, one must appreciate the vertical rela-
tionship at the heart of the hub-and-spoke category and the challenges it
raises. The pricing vendor Boomerang, for example, does not compete with
its retail customers, such as Staples; nor does Uber compete with its drivers.
To establish a conspiracy, it is not enough for information to flow through
the hub. The parties should be aware of the likely effects of the flow of in-
formation. Intent to communicate the information through the hub, and
situational awareness by the recipient of that information are required.28
In the case of Eturas and Others, the Court of Justice of the European
Union was considering a possible hub-and-spoke conspiracy facilitated by
an online system.29 There, an administrator of an online travel booking
system posted a notice on its system declaring a newly implemented tech-
nical restriction that imposed a cap on discount rates.30 The Court held that
travel agents who knew the content of the message sent via the system could
be presumed to have participated in an illegal collusion unless they pub-
licly distanced themselves from that message or reported it to the admin-
Copyright © 2016. Harvard University Press. All rights reserved.

istrative authorities. It emphasized the significance of establishing the

travel agents’ awareness of the message.31
Similarly, in the United States, competition authorities may use evidence
of intent to assess the nature of the agreement (i.e., is it purely vertical or is
it effectively a horizontal agreement among competitors?), its likely com-
petitive consequences, whether to categorize the conduct as a hard-core of-
fense, and whether to prosecute civilly or criminally.32
In applying this case law to our algorithm-fueled hub-and-spoke sce-
nario, it seems likely that in determining antitrust liability, courts in the

United States and the European Union will likely consider the firms’ in-
tent in using the algorithms, that is, whether they: (1) intended a clearly il-
legal result, such as agreeing to fi x prices, or (2) acted with knowledge that
illegal results, which actually occurred, were “probable.”33
No doubt if the algorithm is specifically designed to facilitate collusion
among the users, we would have the classic hub-and-spoke conspiracy.34
The motivation in developing the algorithm would satisfy the intent and
awareness conditions and pave the way for a finding of illegality.
From an enforcement perspective, also noteworthy is the fact that the
hub-and-spoke structure may support a more stable cartel due to the role
played by the hub and the use of computerized systems. When the hub sets
the price and monitors market information (including the activities of the
competing spokes), it may be more difficult, if not impossible, for any cartel
member to “cheat” on the agreed price.
The algorithm as a hub serves as a commitment device. Given the logistics,
the spokes may not know when to discount, for which products, or the mag-
nitude of the discount. That is why the rivals delegated pricing to the hub in
the first place. So the hub—as the central processor in collecting industry
data and setting prices—can reduce distrust among competitors; each knows
that the others are also entrusting their pricing to the same algorithm, whose
aim is to maximize profit.
But what about instances in which the algorithm is not designed to fa-
cilitate collusion, but may nonetheless tamper with the market price? Could
it then be deemed to have the object to restrict competition?
Consider again Uber’s algorithm and its use by all drivers. If the drivers
independently agreed among themselves to charge the same base rate (or
surge rate), they would be guilty of price fi xing. Yet in the case of Uber a
vertical agreement is present between the hub (the algorithm developer)
and the spokes (the Uber drivers). After Uber entered Nashville, the
Copyright © 2016. Harvard University Press. All rights reserved.

fi rst few drivers who joined Uber’s platform—while agreeing to use the
algorithm— did not necessarily agree to fi x the prices for taxi ser vices.
But what about the later drivers, who sign up after Uber dominates the
local Nashville market? If these drivers understood that, by joining Uber,
they all would receive the same rate and the same percentage of any mono-
poly profits, have they essentially become a hub-and-spoke conspiracy?
That remains unclear. As these online platforms’ size and power increase,
competition authorities will face challenging legal issues as to the algo-
rithms’ possible manipulation of the perceived market price and the liability

of drivers and the platform. While the effects on the market may resemble
horizontal collusion, the conditions for establishing a hub-and-spoke con-
spiracy may be absent. True, the parallel use of the same algorithm may give
rise to concern. Yet whether it is sufficient to facilitate a finding of illegality
remains to be seen.35
Interestingly, one federal district court in 2016 refused to dismiss a pri-
vate antitrust complaint against Uber’s CEO and drivers for generating
“supra-competitive prices” through their agreement to use the Uber pricing
algorithm.36 As Judge Rakoff aptly noted, “The advancement of technolog-
ical means for the orchestration of large-scale price-fixing conspiracies need
not leave antitrust law behind.”37 The antitrust complaint overcame the first
legal hurdle. It must next survive summary judgment.
If Uber and its drivers are found to have engaged in a hub-and-spoke
conspiracy, then they are liable regardless of the cartel’s actual effects. It
does not matter whether the algorithm’s pricing was reasonable or lower
than the prevailing taxi fares.
And yet, when prices offered by the platform are lower than the substi-
tutes, it is hard to justify intervention. In other words, should courts find
illegality even when the platform improved ser vices and lowered price?
Still, over time, as the platform gains power, the comparative benchmark
may no longer reflect the competitive price. A competition authority will
likely have a hard time identifying the tipping point when the legal use of
a common algorithm becomes a conspiracy. After all, the first retailer who
uses Boomerang’s pricing algorithm and the first driver on Uber’s platform
in a new city are not guilty of price fi xing. Nor is there necessarily a hub-
and-spoke conspiracy when the second or third driver joins Uber. At what
point does the algorithm become the central hub to facilitate collusion?
Another challenge concerns the properties of a given algorithm. A com-
petition authority may find it cumbersome, and at times impossible, to
Copyright © 2016. Harvard University Press. All rights reserved.

delve into the heart of an algorithm to establish whether it is designed in a

way that would lead to, or may lead to, exploitation. Machine learning is
an ongoing process. The algorithm used last year may not resemble the one
used today. Unless the competition agencies discovered evidence of a clear
anticompetitive design, their analysis would shift from a “per se illegal”
standard to a “Rule of Reason” standard. Under this standard, the compe-
tition authority must establish the likely adverse effects of these vertical
agreements in a properly defined antitrust market. Unlike the per se illegal
standard, the defendants can offer procompetitive justifications for using

the pricing algorithm. If any anticompetitive harm would be outweighed

by the practice’s procompetitive effects, then the practice is lawful. Such
analysis is typically costly, time-consuming, and complex. Nor is the
analysis always objective or the outcome predictable. Thus the appetite to
pursue such rule of reason cases is diminished.38

Reﬂections
When competitors use the same pricing algorithm, the illegality of these ar-
rangements will often be murky. If there is strong evidence of anticompeti-
tive intent, then the agency can prosecute the participants using the familiar
hub-and-spoke conspiracy law. Absent this evidence, however, the common
use of the same algorithm raises difficult enforcement and policy challenges.
First, as illustrated above, identifying the tipping point from legal use of
an algorithm to anticompetitive use may be challenging.
Second, the stability of these schemes and their susceptibility to entry
may be difficult to establish. Consider again our Uber example. On the one
hand, the use of the hub to set prices via a common algorithm and the cen-
tralized payment system support a stable environment. Drivers are likely
to lack the incentive and ability to undercut the common price. On the
other hand, if competitive alternatives restrain the platform and its sellers
from raising prices or degrading quality, then the antitrust risks lessen.
This tension leads us to the third enforcement challenge—identifying the
point at which the platform or sellers obtain market power and the hub’s al-
gorithm can and likely will increase prices. Market power should be assessed
while taking note of social biases, users’ inability to process complex infor-
mation,39 limited switching patterns, usage of apps, and access to platforms.
We cannot predict the competitive dynamics of some markets or the
competitive pressure to which the leading platforms are subjected. The
Copyright © 2016. Harvard University Press. All rights reserved.

ability of entrants to thwart any hub-and-spoke conspiracy is hotly debated

in the conferences and workshops we attend. It is hard to obtain clear infor-
mation about the extent to which users multihome and use cross-platform
comparison sites, and whether this could curtail the power of the leading
platform. When time is at a premium, do users simply opt for the leading
default platform?40 Moreover, as we will explore later, network effects can
make the big platforms even bigger and more powerful. When that hap-
pens, the platform’s price effectively becomes the market price, which
means higher prices for us all.

Tacit Collusion on Steroids:

The Predictable Agent

T HE LAST TWO CHAPTERS explored the shift from a world where executives
expressly collude in smoke-filled hotel rooms to where they still collude
but use computer algorithms to help execute their illegal agreement (Mes-
senger), or to serve as the hub in their Hub-and-Spoke conspiracy. We now
shift to the “twilight” world of virtual competition, where the industry-wide
use of pricing algorithms leads to higher prices, without any clear or implied
human anticompetitive agreement.
After explaining how greater transparency can facilitate the phenomenon
known as tacit collusion, we explore our third collusion scenario, Predict-
able Agent. Here we consider how each firm unilaterally creates an algo-
rithm but knows that the industry-wide use of pricing algorithms will facili-
tate tacit collusion. The competition authority lacks evidence of an illegal
agreement but has evidence of anticompetitive intent. We reflect on whether
this suffices to deter and punish tacit collusion.

Transparency, Competition, and Tacit Collusion

Greater transparency, as we saw in Chapter 1, can foster competition, as it

lowers our search costs in assessing the prices each seller charges. But
greater transparency, under certain market conditions, can also lead to a
unique phenomenon known as tacit collusion, which lessens competition.
As we’ll see, this phenomenon can harm consumers to the same extent as
the price-fi xing cartels we saw in Chapters 5 and 6.
So what is tacit collusion? In its judgment in Brooke Group Ltd. v. Brown &
Williamson Tobacco Corp., the United States Supreme Court gave the fol-
lowing description:
56

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Tacit Collusion on Steroids: The Predictable Agent 57

[T]acit collusion, sometimes called oligopolistic price coordination or

conscious parallelism, describes the process, not in itself unlawful, by
which firms in a concentrated market might in effect share monopoly
power, setting their prices at a profit-maximizing, supracompetitive level
by recognizing their shared economic interests and their interdepen-
dence with respect to price and output decisions and subsequently uni-
laterally set their prices above the competitive level.1

Importantly, the conditions for tacit collusion, as several economists have

noted, “need not involve any ‘collusion’ in the legal sense, and in particular
need involve no communication between the parties. It is referred to as
tacit collusion only because the outcome (in terms of prices set or quan-
tities produced, for example) may well resemble that of explicit collusion
or even of an official cartel.”2 Accordingly, tacit collusion differs from
express collusion, where the competitors’ employees actually agree to fi x
prices, reduce output, or allocate markets. More importantly, express
collusion is illegal, whereas the parallel behav ior, which may stem from
tacit collusion, is legal.3
To illustrate tacit collusion, let us consider Martha’s Vineyard, which at-
tracts U.S. presidents and vacationers every summer. The plaintiffs are the
summer residents, angry at unjustifiably high gasoline prices.4 The defen-
dants operate four of the nine gas stations on Martha’s Vineyard. The de-
fendants’ prices exceed prices at gas stations on nearby Cape Cod by an
average of fift y-six cents per gallon.5 So why is gas so expensive? Must it be
because the defendants are colluding?
Not necessarily. As both the trial and appellate court found, the retail
gasoline market on Martha’s Vineyard has “features that make it suscep-
tible to efforts by gas stations to sustain supra-competitive prices.”6 First are
the high entry (and regulatory) barriers for anyone seeking to open another
Copyright © 2016. Harvard University Press. All rights reserved.

gas station on the island.7 No bridges connect with the mainland; drivers
rely on a ferry. Second, consumer demand is inelastic, meaning “customers
will not buy much less gas when prices rise, because they cannot choose to
drive farther away to get cheaper gas.”8 Third, gasoline is a homogeneous
good, “so consumers decide where to buy it based mostly on price and con-
venience, leading competing gas stations to prominently post prices.”9
On Martha’s Vineyard, like most other places, gas prices are highly
transparent. You see the posted price as you drive by. But the transparency,
the First Circuit noted, helped cause the high prices: it “lets competitors

know and respond in real time to one another’s prices, allowing them to
catch price ‘cheaters’ and to follow price ‘leaders.’ ”10 With only nine gas sta-
tions on the entire island, they can, through conscious parallelism, reach
the same anticompetitive outcomes as a cartel:
[E]ach station can easily monitor and respond to the prices of the others. If
one station drops its price in order to attract more business, the others can
quickly drop their prices in response. The original “cheater” benefits very
little from undercutting its competitors’ prices, because when any one of
them drops its prices the competitors can match the price before many cus-
tomers respond to the incentive. And all of the stations suffer a decrease in
profit margin. Conversely, a station acting as a price “leader” risks little by
raising its price under such market conditions. Other stations are likely to
follow, given the possibility of higher prices and profit margins for all. If for
some reason the competitors do not follow the increases, the leader can
easily drop its price again to match the other stations so quickly that few
customers are lost to lower-priced competition. Knowing these features of
the market, each gas station owner is likely to reach its own independent
conclusion that its best interests involve keeping prices high, including fol-
lowing price changes by a price “leader” (if one emerges), in confidence that
the other station owners will reach the same independent conclusion.11

Of course, if the four gas station owners got together and agreed to fi x gas
prices, then that constitutes express collusion, which is per se illegal. The
defendants would be civilly liable and criminally prosecuted. But the Mar-
tha’s Vineyard gasoline market is an oligopolistic market that is highly
conducive to tacit collusion.12 Here the defendants, given the market condi-
tions and high transparency, could achieve the same end as a cartel (namely
higher prices and profits) without the illegal means (namely agreeing among
themselves to fix prices).
Copyright © 2016. Harvard University Press. All rights reserved.

So without sufficient evidence of an agreement among the competitors,

the plaintiffs’ antitrust claims failed. The plaintiffs’ evidence was entirely
consistent with conscious parallelism, where each gas station owner simply
followed the price leader. That is legal.
To demonstrate the dynamics underlying tacit collusion, suppose one
morning one of the owners decides to reduce the price she charges for gas-
oline. She does so with the hope of attracting more customers and in-
creasing overall profits. Indeed, such a strategy embodies competitive

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Tacit Collusion on Steroids: The Predictable Agent 59

markets. Yet our oligopolistic market has all the characteristics required for
tacit collusion. As a result, the other gas station owners—upon observing
the price reduction—will retaliate by reducing their prices. They do so to
prevent customers from shift ing their business to the discounting gas
station. At the day’s end, each competitor, as well as our original discounter,
sells the same amount of gasoline, but at a lower price, making less profit.
The discounter, along with her competitors, learns through experience how
the market is characterized by interdependence. Every time any owner tries
to undercut the others’ price, the others will match that price cut. In other
words, no one profits by discounting.
Th is discovered interdependence not only reduces the incentive to
discount; it increases their incentive to follow a price increase. This is so
because under the market conditions, when one competitor raises the
price of gasoline, the others know that if they do not follow the price in-
crease, the price leader will eventually drop the price, so as not to lose
many customers. In other words, if they choose not to follow the price in-
crease, they all forgo the extra profits from the price increase. The interde-
pendence between the gas station operators therefore supports a gradual
increase in price and leads to a new equilibrium above competitive levels.
Indeed, the defendant gas stations’ interdependence was strong: they
held gas prices on Martha’s Vineyard steady or raised them while the cost
of gasoline at wholesale declined.13 Their profits were abnormally high.
They based their gas prices, not on their cost, but on the actions and ex-
pected actions of the other stations on Martha’s Vineyard. And their
market shares were stable over time.
Importantly, the owners could charge high prices without any formal or
informal illegal agreement among themselves. The equilibrium was the
result of a rational, unilateral decision by each competitor.14
Here the competition authority is stuck. The outcome of tacit and ex-
Copyright © 2016. Harvard University Press. All rights reserved.

press collusion is the same—namely steep prices. But for tacit collusion,
there is no agreement. Thus, the ser vice station owners profit from the oli-
gopolistic market dynamic.
So why would fi rms expressly collude, when they can avoid antitrust
liability (and incarceration) by tacitly colluding? Recall that for tacit
collusion to be sustained, a few key conditions must be present. One key
condition is that the market is sufficiently transparent that its few com-
petitors can “promptly and confidently” observe each rival’s “significant

competitive initiatives.”15 When “the terms offered to customers are rela-

tively transparent,”16 the risk of tacit collusion increases in concentrated
industries with homogeneous products and inelastic consumer demand.
Each competitor has “the ability to know how the other members are be-
having” and can “monitor whether or not they are adopting the common
policy.”17 The European Commission explains how markets need to be suf-
ficiently transparent to allow the coordinating firms to monitor to a suffi-
cient degree whether other firms are deviating by lowering prices, offering
secret discounts, increasing product quality or capacity, or trying to win
new customers:
When evaluating the level of transparency in the market, the key element
is to identify what firms can infer about the actions of other firms from
the available information. Coordinating firms should be able to interpret
with some certainty whether unexpected behaviour is the result of devi-
ation from the terms of coordination. For instance, in unstable environ-
ments it may be difficult for a firm to know whether its lost sales are due
to an overall low level of demand or due to a competitor offering particu-
larly low prices.18

For tacit coordination to be sustainable over time, one European case

notes, “there must be an incentive not to depart from the common policy
on the market.”19 Companies must be able to effectively retaliate when a
competitor seeks a relative advantage by discounting. The retaliation must
be “sufficiently likely and costly to outweigh the short-term benefits from
‘cheating’ on the collusive path.”20 In addition, to sustain tacit collusion,
potential competitors or customers should not be in a position to jeopar-
dize the results expected from the common policy. One would therefore
expect that buyers cannot exert buyer power, are unlikely to change their
purchasing powers, and the market in general is characterized by high
Copyright © 2016. Harvard University Press. All rights reserved.

entry barriers.21 (But, as evidenced by the cartels in turtles and packaged

ice, even markets with low entry barriers can be cartelized.)
Thus, tacit collusion requires certain market conditions. Absent these
conditions, parallel behav ior should ordinarily not occur. As we illustrate
below, as market participants increasingly rely on pricing algorithms, they
will change the market conditions. One risk is that the changed market
conditions can widen the instances in which tacit collusion may occur.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Tacit Collusion on Steroids: The Predictable Agent 61

The Predictable Agent

To explore how algorithms can foster tacit collusion, we turn to our third
scenario, the Predictable Agent, where each firm programs its algorithm
with a strategy to maximize profits. The algorithm, among other things, is
programmed to monitor price changes and swift ly react to any competi-
tor’s price reduction. The algorithm is also programmed to follow price
increases when sustainable, that is, when others follow in a timely manner
so that no competitor benefits from keeping prices lower.
Consider the effect when each competitor in an industry adopts this
pricing algorithm. As each seller relies on the algorithm, more market data
will be digitalized and accessible, and market transparency will likely
increase.
First, the demand for digitized market information and transparency
will increase. The algorithms will likely engage in “predictive analytics”—
that is, the study of patterns in pricing and commercial decisions. Such an
analysis will enable firms to combine “real-time, historical and third-party
data to build forecasts of what will happen in their business months, weeks
or even just hours in advance.”22 That technology would enable “moving
away from ‘systems of record’ to ‘systems of engagement’ that use predic-
tive analytics to cut through the noise in big data and uncover insights that
can be acted on.”23 In order for the algorithm to function effectively and
optimize pricing, the computer must quickly access and process key market
data, including competitors’ prices and sale terms, and respond to market
changes.
Second, when each firm adopts a pricing algorithm, the supply of market
data (including competitors’ pricing) increases. Each seller, in shift ing to
algorithms, increases market transparency by posting its current prices.
Consumers and rival algorithms will immediately see each firm’s current
Copyright © 2016. Harvard University Press. All rights reserved.

price and terms online.

What if the algorithm determines the list price, and the firm secretly of-
fers price-sensitive customers discounts? (We’ll explore this further in
Part III in regard to price discrimination.) Some of us will end up paying
more. Moreover, as we saw in Chapter 2, some industries are currently in
an arms race to use algorithms to engage in dynamic pricing. Under this
competitive pressure to quickly adjust prices, fi rms may have neither
the time nor the incentive to manually check the algorithm’s price and
determine a secretive discount. When dynamic pricing yields a competitive

advantage, no firm can afford the time gap to assess whether the algo-
rithm’s suggested price should be implemented. The firm relies on the
pricing algorithm precisely because it is ineffective for humans to indepen-
dently analyze all the underlying market data to calculate prices (or dis-
counts) on many products.
If the whole purpose of dynamic pricing is to update prices quickly so as
to reflect market demand, market participants will likely expect the price
posted online to be the actual price. Some buyers may continue to haggle,
but the norm develops that the algorithm-determined posted price is
the actual price. (Indeed, with gas stations, one typically does not try to
negotiate a discount from the clerk, who may have little if any authority to
change the posted price.)
This differs from yesteryear when buyers received physical price lists.
There was often a lag before rivals obtained the list. Moreover, the list price
may not have reflected the actual price paid. Firms, with only the competi-
tors’ older price lists, would have to rely on hearsay collected by their sales
personnel on what rivals were actually charging.

Speed
Speed, in our Predictable Agent scenario, is critical. When we were growing
up, humans monitored market activity, determined whether, and by how
much, to raise or lower prices, and physically stamped products with price
stickers. Pricing decisions took weeks—if not months—to implement.
So as competitors’ prices shift online, their algorithms can assess and
adjust prices—even for particular individuals at particular times and for
thousands of products—within milliseconds.24 In other words, they can
swift ly match a rival’s discount, thus eliminating its incentive to discount
in the first place. On the other hand, they will follow price increases (when
Copyright © 2016. Harvard University Press. All rights reserved.

sustainable). In an environment dominated by similar pricing algorithms

that are aware of opportunities to foster interdependence, the risk is higher
prices.25
Returning to our gasoline station example, imagine our oligopolistic
market had limited transparency (i.e., drivers would have to inquire inside
the station to determine the price of gas). Drivers would not want to search
Martha’s Vineyard for the lowest gas prices. Instead, they might ask their
friends, visit a few stations, and support the one with the lowest price. Thus,
the discounter benefits from a reputation for having the lowest price; the
other stations would eventually discount. Under these market condi-

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Tacit Collusion on Steroids: The Predictable Agent 63

tions, conscious parallelism is harder to sustain. The firms will compete

as expected.
Now, think of the basic conditions for tacit collusion/conscious paral-
lelism. Suppose some gasoline owners want to shift pricing decisions from
humans to computers. The computer must be able to access market data,
of which there is very little. So the pricing algorithm initially uses crude
data, perhaps historic daily volume of sales. If actual sales falls below the
historic level, the assumption, after controlling other possible explanations,
is that gas is cheaper at the other stations. The computer continues drop-
ping the posted price of gasoline until average sales reach the historic
average. When the second and third gas station adopts a similar pricing
algorithm, the quality of the market data improves. The pricing algorithm
can use its rivals’ posted price data, rather than infer it from historic versus
actual sales.
Now suppose, as is the case in many states,26 a smartphone app tells you
the price of gasoline at every local station. That sounds procompetitive. The
increase in price transparency lowers your search costs for finding cheaper
gasoline. Indeed, in markets characterized by many sellers, the gas app may
promote competition.
However, in an oligopolistic market, such applications may have the op-
posite effect. First, the applications put competitive pressure on any gas sta-
tion that still does not use pricing algorithms. Ironically, even if some
companies yearn for the days of printed list prices and secretive discounts,
they may switch to pricing algorithms to prevent being at a competitive
disadvantage. Symetry is soon established. Second, the applications pro-
vide real-time pricing for gasoline for each station; competitors no longer
have to drive around the island to collect the information. Rivals’ pricing
algorithms can promptly observe all the competitively significant terms.
And the pricing is real—it reflects what customers actually pay. By shift ing
Copyright © 2016. Harvard University Press. All rights reserved.

pricing decisions to computer algorithms, competitors thereby increase

transparency, reduce strategic uncertainty (when the pricing algorithm
cannot grant secretive discounts), and thereby stabilize the market.27
When one gas station lowers the price by one cent at 11:33 a.m., within mil-
liseconds other nearby stations respond by lowering their price.
Thus, with each firm’s algorithm tapping into its rivals’ real-time pricing,
no firm would likely profit by discounting. Given the velocity with which
the pricing algorithms can adjust, no gas station would likely develop
among its customers a reputation as a price discounter. Accordingly, the
competitors will have less incentive to discount. We can see that in such a

market, the app that was meant to promote price competition could end
up undermining it.
On the flip side, the algorithms’ velocity of pricing decisions can shorten
the time period for signaling price increases. Firms would no longer have
to rely on lengthy (e.g., thirty-day) price announcements, where they wait
and see what the competitive response is, to decide whether to raise prices
(and to what extent). Computers can have multiple rounds whereby one
firm increases prices and the rival computers respond immediately and
without the risk that the firm that initiates the price increase will lose many
customers to rivals. Essentially, companies may now need only seconds,
rather than days, to signal price increases to foster collusion.
So the industry-wide use of pricing algorithms increases both market
transparency and the risk of conscious parallelism. Moreover, in program-
ming its pricing algorithm, each firm will likely use historic pricing data
and competitive responses to calibrate the dominant strategy. As such, when
the algorithms operate within the greater transparency of their digitalized
environment, the computers will already be programmed to anticipate and
respond to rivals’ moves. In such a scenario, computers can rapidly calculate
the profit implications of myriad moves and countermoves. With the com-
puters’ ability to police deviations and rely on prior strategies to punish de-
viations, prices, as a result of their conscious parallelism, will climb.

The Algorithm Arms Race

Our scenario resembles the use of computers at a chess match or blackjack
table. Imagine you are playing against other humans, but one player uses a
computer. For blackjack, the computer counts all the cards in the multiple
decks to predict the likelihood of receiving a desired card, assesses the
strategies employed by the other players under similar circumstances, and
Copyright © 2016. Harvard University Press. All rights reserved.

calculates the other players’ risk aversion. For chess, the computer calcu-
lates all the possible moves and countermoves.28 The computer gives the
player an inherent advantage. After losing several times, you would likely
want a computer too.
This explains the metal detectors at the World Chess Cup tournament.
The concern is that the players, to gain a competitive advantage, are secretly
turning to their smartphones’ chess app for their moves. A smartphone
chess app can beat the best players. Chess tournaments, besides the metal
detectors, are also using algorithms to detect whether a player’s moves,

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Tacit Collusion on Steroids: The Predictable Agent 65

given his or her skill levels and situations on the board, are too much like
a computer’s.29
Unlike chess players, market players can freely use algorithms to gain a
competitive advantage. Rivals, like high-speed traders, will have an incen-
tive to invest in technology so that they can see competitively significant
terms a few minutes or seconds before customers will. Michael Lewis in
Flash Boys documented how Wall Street traders built their business models
on having a slight relative advantage in seeing orders before others with
slower connections and systems, and then trading ahead of the others.30
Thus, the algorithm arms race may lead to companies detecting price
changes (including discounts) milliseconds before their customers do, and
being able to respond before the customers can. So too each firm—in uni-
laterally deciding to shift to pricing algorithms—would bring the market
reality closer to that necessary for conscious parallelism and higher prices.

Enforcement Challenges
Unlike our Messenger and Hub-and-Spoke scenarios (in Chapters 5 and 6),
which focused on collusion, the scenario here does not involve any agreement.
The firms—in unilaterally creating and implementing the algorithms—never
agreed to fi x prices. Each firm had an independent economic self-interest
in developing and relying on the algorithms; indeed, it may be contrary to
the firm’s economic self-interest to rely on human pricing or trading.
Interestingly, conscious parallelism takes place at both the human and
machine levels. First, when configuring the machines, each human, inde-
pendently and without collusion, knows that when possible, a dominant
strategy may be to follow a rival’s price increase. Furthermore, each person
knows that if other firms have a similar algorithm, the resulting equilib-
rium may be above the competitive level. This conscious parallelism at the
Copyright © 2016. Harvard University Press. All rights reserved.

human level leads to the programming of machines which are aware of

possible conscious parallelism at the market level. The computer is there-
fore set up to monitor the market and explore the likelihood of establishing
interdependence, without venturing into illegal concerted practices or
illicit agreements. The computer may also be programmed to identify
maverick firms and punish any deviations from a possible tacit collusion.
Competition law, in most jurisdictions, will require proof of an agree-
ment among the parties to change the market dynamics. Can the competi-
tion agency impute the presence of an illicit agreement or understanding

among the competitors to use similar algorithms to dampen competition?31

Not necessarily. One should acknowledge that evolution dictates that the
stronger, more powerful algorithms will likely prevail and dominate the
technology market. This reality naturally fosters assimilation of systems
between various computer developers and companies. Abstaining from an
advanced algorithm may be irrational; it would be as if an investment
bank or hedge fund insisted on human floor traders, when most trading
is automated.
In our example, there is no evidence of an agreement among the firms,
but there is strong evidence of anticompetitive intent. Humans unilaterally
design algorithms to deliver predictable outcomes and react in a given way
to changing market conditions. The firms recognize, in this scenario, that
the industry-wide adoption of similar algorithms would likely foster tacit
collusion, whereby they mutually profit from their initial investment. Cru-
cially, the use of advanced algorithms in this scenario transforms the
“normal,” preexisting market conditions. Before algorithms, transparency
was limited; conscious parallelism could not be sustained. To facilitate the
use of the pricing algorithms, the firms increase transparency, which in
turn makes tacit collusion likelier. While the mutual price monitoring at
the heart of tacit collusion is legal under competition law, one may ask
whether the creation of such a dynamic through “artificial” means should
give rise to antitrust intervention.
The main enforcement challenge concerns the legality of conscious paral-
lelism. A rational reaction by competitors to market dynamics, in itself, is
legal. When such legal behav ior, absent communication or agreement,
leads to an equilibrium above competitive levels, it does not trigger antitrust
intervention.32
But the fact that tacit collusion is legal does not mean it is desirable. In-
deed, competition law prohibits mergers that make tacit collusion more
Copyright © 2016. Harvard University Press. All rights reserved.

likely. This is so since the merger, in effect, changes the existing competi-
tive conditions. To illustrate, suppose Firms A and B plan to merge, leaving
Firms A, C, and D in the market. Suppose Firms A and C currently use
pricing algorithms, and Firm D, as a result of merger, would likely use
algorithms as well. Suppose the evidence shows that D’s use of pricing al-
gorithms post-merger would blunt each firm’s incentive to discount and
increase their incentive to raise prices.33 The merger by enabling tacit col-
lusion would likely diminish competition. Subsequently, the reviewing

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Tacit Collusion on Steroids: The Predictable Agent 67

competition agency will enjoin the merger or demand remedies which can
resolve the anticompetitive risks.
Now let us change the scenario. Suppose there is no merger, but Firms
A, B, C, and D are all predisposed to tacit collusion. But current market
conditions prevent it, and they do not want to risk criminal penalties by
expressly colluding. They all recognize that the use of advanced algorithms
will increase transparency, reduce their incentives to discount, and in-
crease their incentives to raise prices. They all incorporate the pricing al-
gorithms. Is that meaningfully different from the merger scenario?
The question is therefore whether one may condemn the creation of a
transparent market in which monitoring and punishment mechanisms are
present, and if so, under what conditions? Another challenge concerns in-
stances in which the algorithm is programmed to refrain from targeting
competitors’ customers in an attempt to stabilize the market and avoid a
price war. The legality of such action was addressed by the U.S. Court of
Appeals for the Seventh Circuit:
[S]uppose that the firms in an oligopolistic market don’t try to sell to each
other’s sleepers, “sleepers” being a term for a seller’s customers who out
of indolence or pricing ignorance don’t shop but instead are loyal to
whichever seller they’ve been accustomed to buy from. Each firm may be
reluctant to “awaken” any of the other firms’ sleepers by offering them
discounts, fearing retaliation. To avoid punishment under antitrust law
for such forbearance (which would be a form of tacit collusion, aimed at
keeping prices high), would firms be required to raid each other’s sleepers?
It is one thing to prohibit competitors from agreeing not to compete; it is
another to order them to compete. How is a court to decide how vigor-
ously they must compete in order to avoid being found to have tacitly
colluded in violation of antitrust law? Such liability would, to repeat, give
Copyright © 2016. Harvard University Press. All rights reserved.

antitrust agencies a public-utility style regulatory role.34

Should competition agencies challenge a unilateral decision not to compete?

Can such a strategy be credibly framed as collusive customer allocation?
Considering the above challenges, one may wonder whether, under cur-
rent laws, algorithm developers can legally program machines that unilat-
erally support tacit collusion.
Under traditional competition analysis the answer is likely to be “Yes.”
Absent evidence of an agreement to change market dynamics, most

competition agencies lack enforcement tools, outside of merger control,

that could effectively deal with the change of market dynamics through
algorithms. Unilaterally, a fi rm without market power may develop an
algorithm that detects the market behaviors of competitors; anticipates the
rivals’ algorithms’ likely reactions to different competitive responses;
and opts for the path that, given the competitive reactions, will maximize
profits, which may often be the path toward conscious parallelism.
Outside the core competition provisions, one may, however, consider
alternative legal instruments. For instance, the U.S. Federal Trade Com-
mission (FTC) can bring claims under Section 5 of the FTC Act, without
evidence of an agreement, only a showing of an “unfair practice.” Many
states have a similar statute. But the FTC has been unsuccessful in bringing
these “facilitating practices” claims, as is evident in Boise Cascade35 and
Ethyl.36 If the court adopts the standard in Ethyl, the FTC would need to
show either (1) evidence that defendants tacitly or expressly agreed to use
pricing algorithms to avoid competition, or (2) oppressiveness, such as
(a) evidence of defendants’ anticompetitive intent or purpose or (b) the
absence of an independent legitimate business reason for the defendants’
conduct.37 Accordingly, in the Predictable Agent category, the defendants
may be liable if, when developing the algorithms or in seeing the effects,
they were (1) motivated to achieve an anticompetitive outcome, or (2) aware
of their actions’ natural and probable anticompetitive consequences.
Another approach may be to consider the use of such algorithms as
market manipulation. This approach has its own obstacles; yet one could
imagine the introduction of legislation that targets “abuse” of excessive
transparency, possibly where clear anticompetitive intent is present.
If the executives, for example, call their algorithm Gravy, and tinker with
it to better manipulate the market, and boast about this in their internal
e-mails—as in the U.S. Securities and Exchange Commission’s (SEC) case
Copyright © 2016. Harvard University Press. All rights reserved.

against Athena Capital Research—liability could be established.38 The Athena

case is illustrative. In 2014, the SEC for the first time sanctioned the high-
frequency trading firm for using complex computer programs to manipulate
stock prices.39 The sophisticated algorithm, code-named Gravy, engaged in a
practice known as “marking the close” in which stocks were bought or sold
near the close of trading to affect the closing price: “[t]he massive volumes of
Athena’s last-second trades allowed Athena to overwhelm the market’s avail-
able liquidity and artificially push the market price—and therefore the
closing price—in Athena’s favor.”40 Athena’s employees, the SEC alleged,

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Tacit Collusion on Steroids: The Predictable Agent 69

were “acutely aware of the price impact of its algorithmic trading, calling it
‘owning the game’ in internal e-mails.”41 Athena employees “knew and ex-
pected that Gravy impacted the price of shares it traded, and at times Athena
monitored the extent to which it did. For example, in August 2008, Athena
employees compiled a spreadsheet containing information on the price move-
ments caused by an early version of Gravy.”42 Athena configured its algo-
rithm Gravy “so that it would have a price impact.”43
In calling its market-manipulation algorithm Gravy, and by exchanging a
string of incriminating e-mails, the company did not help its case. Without
admitting guilt, Athena paid a $1 million penalty. This demonstrates that
automated trading has the potential to increase market transparency and
efficiency, but it can also lead to market manipulation.44
Finding the predominant purpose for using an algorithm will not always
be straightforward. Athena, for example, challenged the SEC’s allegations
that it engaged in fraudulent activity: “While Athena does not deny the
Commission’s charges, Athena believes that its trading activity helped sat-
isfy market demand for liquidity during a period of unprecedented demand
for such liquidity.”45 A court might agree. Companies can also learn from
Athena and be more circumspect in their e-mails.
Moreover, evidence of intent will likely be mixed when each fi rm has
valid independent business reasons to develop and implement a pricing al-
gorithm. After all, the first firm to use the pricing algorithm could not be
accused of colluding, as the market was likelier less transparent, and rivals
could not match the speed of the first mover’s price changes. Thus, if the
first firm to use a pricing algorithm lacked anticompetitive intent, the same
may be true for the second or third firm. It too might have legitimate busi-
ness reasons to employ a pricing algorithm—namely, to not be at a com-
petitive disadvantage in responding to price changes by the first firm.
Copyright © 2016. Harvard University Press. All rights reserved.

Reﬂections
Tacit collusion provides a fascinating example of challenging market dy-
namics. First, it will likely arise (at least initially) in highly concentrated
markets where the other conditions for conscious parallelism are present.
Admittedly, even when these conditions are present, the dynamics of a
market may trigger changes or new entry and destabilize conscious paral-
lelism.46 Similarly, technology may provide a disruptive force, allowing
algorithms to successfully “cheat” by discounting.

Still, with the above caveats in mind, conscious parallelism will likely
become more common. The nature of electronic markets, the availability
of data, the development of similar algorithms, and the stability and trans-
parency they foster, will likely push markets that were just outside the
realm of tacit collusion into interdependence.47
These developments raise challenging technical, enforcement, and legal
questions. If the algorithms increase market transparency, the defendants
will often have an independent legitimate business rationale for their con-
duct. Courts and the enforcement agencies may be reluctant to restrict this
free flow of information in the marketplace. Its dissemination, observed
the Supreme Court, “is normally an aid to commerce,”48 and “can in cer-
tain circumstances increase economic efficiency and render markets more,
rather than less, competitive.”49 Indeed, concerted action to reduce price
transparency may itself be an antitrust violation.50
A regulatory approach to reduce transparency may also prove difficult.
One may find it difficult to fine-tune the enforcement policy aimed at
condemning “excessive” market transparency. Th is may be particularly
challenging when the information and data are other wise available to
consumers and traders and it is the intelligent use of that information that
facilitates conscious parallelism.
An alternative use of disruptive technology may also have limited ap-
peal, as it could be overpowered by new technology. Similarly, restrictions
that limit the ability to match prices could be undermined by smart algo-
rithms operating in fast moving markets.
We revisit the question of intervention in the Part V of the book. Next,
we consider the fourth scenario, Digital Eye, where, as more data is quickly
fed into the algorithm, transparency can reach what we call “the God
View.” Self-learning algorithms, with the God View, can expand tacit col-
lusion in unexpected directions, compounding the harm.
Copyright © 2016. Harvard University Press. All rights reserved.

Artiﬁcial Intelligence, God View,

and the Digital Eye

W E ARE ALREADY witnessing the Messenger and Hub and Spoke collu-
sion scenarios. Predictable Agent will likely be the next scenario
we’ll see. Our final collusion scenario—Digital Eye—represents the next
frontier. Here we consider how two key technological advancements can
amplify tacit collusion to a new level of stability and scope. The fi rst ad-
vancement involves the computer’s ability to process high volumes of data
in real time to achieve a God-like view of the marketplace. The second ad-
vancement concerns the increasing sophistication of algorithms as they
engage in autonomous decision making and learning through experience—
that is, the use of Artificial Intelligence (AI).
These two technological advances can form a harmful combination: In
enabling a wider, more detailed view of the market, a faster reaction time
in response to competitive initiatives, and dynamic strategies achieved by
“learning by doing,” the technologies can expand tacit collusion beyond
price, beyond oligopolistic markets, and beyond easy detection. With our
other three scenarios, we, like the vacationers on Martha’s Vineyard, may
know when something is amiss. In the Digital Eye scenario, the contagion
Copyright © 2016. Harvard University Press. All rights reserved.

spreads—to markets less susceptible to tacit collusion under the brick-and-

mortar economy, and beyond pricing to other competitive initiatives.
In the end, with Digital Eye, we may think the markets, driven by these
technologies, are competitive. We may believe that tacit collusion in
these markets isn’t even possible. And yet we’re not benefiting from this
virtual competition.

God View
In 2014, Uber caused a stir. Two former Uber employees told reporters that
“[t]racking customers is easy using an internal company tool called ‘God
View.’ ”1 Uber’s “God View” apparently shows the location of all Uber
vehicles and customers who have requested a car.
Borrowing Uber’s terminology, we refer to God View as competitors
using Big Data and Big Analytics for a clearer overview of the marketplace
at any given moment. The wealth of data generated from the online envi-
ronment, cloud computing, and smart sensors can provide a panoramic
God-like view of our state of being. Firms can see on a giant screen, for
any city, their own driverless trucks, their rivals’ driverless trucks, their
customers’ trucks, what the trucks are carry ing, and where they are trav-
eling. Each firm can track the movement of its own and its rivals’ products
traveling through the supply chain. They can see when the item enters their
customers’ factories or homes. They can continue to collect data until the
item is ultimately recycled or discarded.
As we saw in Chapter 7, computer algorithms are quicker than humans
to observe price and demand changes, and can respond (including tit-
for-tat) by adjusting prices for relatively homogeneous products. Also
significant is the fact that markets are typically more vulnerable to coordi-
nated conduct “if a firm’s prospective competitive reward from attracting
customers away from its rivals will be significantly diminished by likely
responses of those rivals,” which “is more likely to be the case, the stronger
and faster are the responses the firm anticipates from its rivals.”2 In mar-
kets where customers can switch between suppliers, and where the goods
are homogeneous, computer algorithms can quickly detect price reduc-
tions by a rival and effectively deprive that rival of any significant increase
in sales. The greater the price transparency, the quicker the competitive
Copyright © 2016. Harvard University Press. All rights reserved.

response, the less likely the first mover will benefit, and the less likely any
firm will discount. Thus Chapter 7 focused on the algorithms’ reactions
to rivals’ price changes.
With God View, we go a step further: computers can anticipate and react
to competitive threats well before any pricing change. Each firm’s algo-
rithm determines whether it can profit by undertaking a competitive
initiative. Under our scenario, the algorithm concludes not. This is because
the rivals also possess the God View technology. They can quickly identify
the competitive initiative and the emerging threat. The real-time data—

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Artificial Intelligence, God View, and the Digital Eye 73

from tracking the behavior of rivals, potential entrants, and customers—

will reveal when competitors are seeking to increase sales (including
expanding into serving new territories or types of customers, such as in-
stitutional buyers). God View enables each algorithm to quickly detect any
competitive maneuver, and thus to know when and how to retaliate. By
responding quickly, the rivals deprive any would-be mavericks of the ben-
efits of launching competitive initiatives, and thereby diminish the incen-
tive to undertake these initiatives in the first place.
Our scenario assumes that each firm will have the God View technology.
This increases overall transparency so that no algorithm will attempt to
increase its firm’s market share by secretly discounting, increasing product
quality or capacity, or trying to win new customers. Such an approach
would likely lead to detection, prompting retaliation and loss of profit.
So why assume that every significant firm will have God View? We base
this proposition on a simple evolutionary assumption with two possible
paths. Under the “survival of the fittest” path, faster, smarter operators may
develop the God View and AI technology for a competitive advantage over
rivals. With a clearer view of the marketplace, they can react swift ly to
market changes, increase sales, and acquire more data (from their prod-
ucts’ sensors). Rivals without the God View technology and data stream
lose sales until they exit the market. Entry barriers increase over time, as
weaker firms are eliminated and the leading firms acquire even more real-
time data on the flow of commerce. The industry is dominated by a few
self-learning algorithms that use God View to tacitly collude.
Alternatively, under a “sharing” path, the fi rms, as in the Predictable
Agent scenario, recognize that greater profits can be earned sooner by
sharing the God View and AI technology and data stream with rivals. They
recognize that tacit collusion depends on the competitors quickly identi-
fying and responding to any rival’s competitive initiative, such that the first
Copyright © 2016. Harvard University Press. All rights reserved.

mover does not meaningfully increase sales or profits. No one will be

tempted to improve their products, lower prices, or enter new markets,
because others will immediately detect and punish this initiative. By en-
abling each firm’s significant competitive initiatives to be promptly and
confidently observed by others, God View reduces uncertainty. Barriers to
God View may also erode as the information flows to common customers,
where the rivals can then access it, and eventually across the industry.
The rise of God View has clear implications for our story of tacit collu-
sion. The environment in which the machines operate will not necessarily

be homogeneous at the outset. It will include different levels of sophistica-

tion that characterize different machine-learning algorithms, and different
market players with different incentives. But the dynamics of information
harvesting and trade are likely to give rise to joint efforts at data extraction
and analytics. Initially, firms might track their own products, but it may
be inefficient to use multiple sensors on the same product. Instead, each
item may have one or two sensors, which can be openly tracked. Firms can
then follow not only their own products but their competitors’ products
and components through the supply chain. For instance, users may set
rules for how the system will handle and respond to data from their de-
vices. 3 For example, Amazon, in 2015, was developing a platform for
the Internet of Things, where a “whole ecosystem of manufacturers, ser-
vice providers, and application developers [can] easily connect their prod-
ucts to the cloud at scale, take action on the data they collect, and create a
new class of applications that interact with the physical world.” 4 Its plat-
form “also has a ‘shadow’ mode, which keeps the latest virtual version of a
device in the system for others to interact with, even if the device itself has
gone offline.”5

Digital Eye: Avoiding the Tit-for-Tat Death Spiral

As the flow of personal and market data increases, self-learning algorithms
may use the enhanced transparency of God View to assess the company’s
profit-maximizing strategy for myriad strategic moves and countermoves.
The use of artificial intelligence has significant implications for the effec-
tiveness of the strategy and its legal analysis.
As in the previous chapters, the algorithm here is designed to maximize
profit while avoiding any illegal activity, such as agreeing, to the extent
computers can, with rival firms’ computers to fi x prices or allocate mar-
Copyright © 2016. Harvard University Press. All rights reserved.

kets.6 Subject to these restrictions, the self-learning algorithm continually

analyzes market data and engages in self-learning and experimentation
with the aim of maximizing profit. In contrast to the previous chapter, the
algorithm is not mandated the task of stabilizing the market or reaching
tacit collusion. Rather, it is operating independently, observing the market
dynamic and identifying the optimal strategy.
With God View, computers can more easily anticipate and understand
each other’s moves. Their strategies thereby become more stable and pre-
dictable. Moreover, God View reduces the likelihood of retaliation where

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Artificial Intelligence, God View, and the Digital Eye 75

competitors cannot divine each other’s signals. Uncertainty and misper-

ception, the poison of collusion, are further reduced as each algorithm
quickly obtains additional data to assess whether its rivals’ actions were
intentional or mistakes.
To illustrate these developments, let us consider a “traditional” form of
competition between two companies and then consider how that dynamic
may change under our Digital Eye scenario.
Imagine companies USA and CAN each supply their home markets, say
the United States and Canada, respectively. If one company ships into the
rival’s home market, the rival retaliates by dumping its product in the other
market. Thus, we could expect a detente—each competitor learns to sit
“tight, expecting their neighbors to do the same thing.”7 This tit-for-tat
strategy works well, as each side knows that swift punishment will elimi-
nate any profits from entering.
A problem with this tit-for-tat strategy, however, is misperception, as
“any mistake ‘echoes’ back and forth.”8 Suppose, at times, a rogue distrib-
utor transports Company USA’s goods from the United States to Canada.
Suppose Company USA is unaware of this. Company CAN, seeing USA’s
products in Canada, retaliates by selling in the United States. Thinking that
Company CAN is the defector, Company USA retaliates by selling more in
Canada. Consumers benefit from the competition. Even if Company USA is
aware of the rogue distributor, it may be unable, given the mutual distrust,
to credibly inform Company CAN that it did not direct this shipment.
Consequently, as two game theorists have noted, “When misperceptions
are possible, in the long run tit-for-tat will spend half the time cooperating
and half of it defecting.”9 Consumers benefit from competitive prices
during periods of defection.
Now let’s add Digital Eye to the mix. At an initial stage, computers can
be programmed to follow a tit-for-tat strategy. The human programmers,
Copyright © 2016. Harvard University Press. All rights reserved.

however, will recognize that, given the risk of misperceptions, at least half
the time the rivals will be in a price war. Thus, the computer can be pro-
grammed to choose among different strategies, each of which has a greater
degree of tolerance (such as reverting to tit-for-tat if five trucks out of a
hundred within one month cross the border). The self-learning computers,
not tethered to following tit-for-tat, can optimize profits using evolving
competitive strategies.
With God View, Company CAN’s computer can now track the rogue
distributor shipping Company USA’s product into Canada. Company

CAN’s computer may also detect a pattern: after the rogue distributor
crosses the border, Company USA promptly curtails its supply to the rogue
distributor for one year, other distributors do not ship Company USA’s
product into Canada for several years, and monopoly prices are stable.
Thus, Company CAN’s computer may learn to forgo retribution to see
whether Company USA punishes the rogue distributor; if it does, a price war
is averted.
One general rule of game theory is that “the better players know one
another, or the more often they have been able to observe one another’s
strategic behav ior, the more likely they are to succeed in finding focal
points on which to coordinate.”10
In the Digital Eye scenario, each firm, in continuously tracking its rivals’
behavior, can find multiple points on which to coordinate. The algorithms, for
example, can stabilize the market through de facto customer allocation. The
self-learning algorithms may identify key customers serviced by competitors
and refrain from targeting them with promotions and discounts. Such a uni-
lateral strategy—the self-restriction of competition—could be used to avert
price wars among the competitors.

Artiﬁcial Intelligence vis-à-vis Humans

The Digital Eye dynamics not only support stable conscious parallelism;
they also increase the instances in which conscious parallelism may be
achieved and sustained. The tacit collusion equilibrium can become more
stable with more competitors.
Humans fi xing prices often trust one another.11 With trade associations
or other ringleaders, cartels generally involve many firms.12 But without a
ringleader, collusion (tacit or express) often involves far fewer firms. Why
is this? Under the “traditional scenario,” the market had to be sufficiently
Copyright © 2016. Harvard University Press. All rights reserved.

concentrated for tacit collusion to work. As U.S. competition authorities

note, “The ability of rival firms to engage in coordinated conduct depends
on the strength and predictability of rivals’ responses to a price change or
other competitive initiative.”13 It is generally easier for humans to monitor
two rivals than twenty. It is also easier to learn through observation two
firms’ strategic behavior than twenty firms’ behavior. Moreover, the pun-
ishment by one firm in a market equally divided among three firms is prob-
ably stronger than punishment by one firm with a 5 percent share in a
twenty-firm market.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Artificial Intelligence, God View, and the Digital Eye 77

Computers are not trusting. But self-learning algorithms and God View
could make tacit collusion among more competitors likelier. The stability
needed for tacit collusion is enhanced by the fact that computer algorithms,
while not trusting, are unlikely to exhibit other human biases. Human
biases can always be reflected in the programming code, but if some biases
are minimized (such as loss aversion, the sunk cost fallacy, and framing
effects), the algorithm acts consistently on more deliberative analysis,
rather than intuition.14
Unlike humans, the computer does not fear detection and possible
financial penalties or incarceration; nor does it respond in anger. The
computer can quantify the payoffs that are likely achievable through co-
operation in future games, and opt for forbearance rather than punishing
small deviations. It can also be more efficient in analyzing payoffs, since
“solution space exploration is concentrated on ‘promising’ areas, and is not
pre-imposed by the modeller.”15
With the industry-wide use of computer algorithms, we may witness
conscious parallelism in markets with many more players, where collusion
previously would have been unstable. The computer can more easily track
the behav ior of numerous rivals to detect cheating. If the algorithms are
all similarly programmed, it may be easier to predict the responses of the
other competitors’ computers. Moreover, if the computers, through self-
learning, coalesce around a dominant strategy, each small firm can detect
and appreciate the type of algorithms others are using. The computers can
uniformly and swift ly punish any deviations by a rival. The collective pun-
ishment may be the equivalent of a monopoly controlling 95 percent of a
market. The universe may close, with each algorithm sharing a common
interest (profits) and inputs (similar data) that may lead to durable tacit col-
lusion among many more competitors.
Copyright © 2016. Harvard University Press. All rights reserved.

The Empty Enforcement Tool Kit

Not only is the harm greater under Digital Eye, but the illegality is murkier.
We are far removed from the first two scenarios, Messenger and Hub and
Spoke, where algorithms helped humans collude. We are even beyond Pre-
dictable Agent, where humans did not expressly collude, but knew that tacit
collusion was the likely outcome if each adopted profit-maximizing pricing
algorithms. Although our third scenario presented greater legal uncer-
tainty, at least the prosecutors had evidence of anticompetitive intent. In

our last scenario, humans are further detached from the algorithms’ tac-
tical and strategic decisions. They don’t know whether, when, or for how
long the algorithms have been tacitly colluding. There is no evidence of an-
ticompetitive intent. We can no longer assume that humans intended to
create the conditions for tacit collusion.
What distinguishes our Digital Eye scenario from the other scenarios is
that collusion (tacit or express) is unlikely at the outset. Recall that on the
“factory floor” these computers have no specific commands that may trigger
collusion. It is the self-learning in a market with a God View transparency
occupied by similar-minded agents with the same profit-maximizing goal
that leads to collusion, of which the fi rms’ managers are unaware. As
more data flows from online trade, mobile communications, and the In-
ternet of Th ings, tacit collusion results from the algorithms’ self-learning
rather than human intent. The computers, in learning by doing, determine
independently the means to optimize profits, based on ongoing feedback
from the market. Here we see how self-learning computers may find that
the optimal strategy is to enhance market transparency and thereby
sustain conscious parallelism.
Our Digital Eye scenario raises many interesting liability issues. Can
firms be held liable for the pricing decisions of their self-learning algo-
rithms, when there isn’t any evidence of anticompetitive agreement or
intent? To what extent are humans responsible for their algorithms’ actions,
which they knew were possible but not necessarily probable? Granted,
humans created the algorithm. Humans knew that tacit collusion was one
of many possibilities. And humans relied upon, and profited from, the al-
gorithm. However, the humans did not know that the natural and prob-
able consequence of using the pricing algorithm was tacit collusion. The
humans knew that tacit collusion was one of many possible outcomes, but
they could not predict if, when, for how long, or to what extent the industry-
Copyright © 2016. Harvard University Press. All rights reserved.

wide use of pricing algorithms would lead to tacit collusion and inflated
prices. Therefore, there is no evidence of anticompetitive intent (or an anti-
competitive agreement among firms).
Nor is there express collusion among the computers to limit competition
through “agreement” or concerted practice; instead, computer algorithms
reduce or remove the degree of strategic uncertainty in the marketplace,
evolve toward a common set of predictable strategies, and promote even
greater transparency.
Lacking evidence of either an anticompetitive agreement or intent,
prosecutors now have few, if any, tools to challenge the tacit collusion.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Artificial Intelligence, God View, and the Digital Eye 79

Recall that the phenomenon of tacit collusion reflects a rational unilat-

eral reaction to market characteristics and in itself is not illegal. When
we discussed the Predictable Agent scenario, we considered whether the
intentional creation of a market environment that supports interdepen-
dence between sellers could be condemned. We noted that the applica-
tion of Section 5 of the FTC Act was contingent on anticompetitive
motive or intent.
Yet, when observing the Digital Eye scenario, enforcers cannot rely on
the legal concept of intent, and aside from market evaluations, they have
no other enforcement tools in their toolbox. Indeed, some regulators
might prefer the stable market environment, where algorithms predict
each other’s reactions and dominant strategy, over a volatile environment
where prices rise (during periods of cooperation) and fall (during periods
of retaliation).
Interestingly, in a market reality in which such future collusion is pos-
sible, program designers may favor the use of similar algorithms. This
seemingly benign decision may have significant implications once learning
and increased data flows and transparency have taken place. The similar
machines are more likely to “understand” one another and stabilize a col-
lusive outcome.
Here customers are harmed just as much as (if not more than) in our
other collusion scenarios (given fewer episodes of retaliation). We therefore
witness a new reality: an anticompetitive outcome which we may not
readily perceive and with no one to blame. Any reduction in our welfare is
“merely” a side effect of the rise of the machines and their quest to opti-
mize and serve.

Change the Benchmark for Intervention?

Assuming that the computers are programmed to refrain from violating

the competition laws, the company may have done all that it can to ensure
compliance. From a technological perspective, programming compliance
may be challenging when one attempts to capture the creation of a market
dynamic such as conscious parallelism. A command not to fix prices may
be simple to execute, but under reinforcement learning the algorithm will
experiment with solutions including, as the competition authorities recog-
nize, the myriad possibilities of coordinated interaction, not all of which
are illegal.16 Can the law credibly ask developers to instruct the algorithm
not to react to market changes—to be inefficient?

Although conscious parallelism is legal, the question is whether such

practices, when implemented by smart machines in a predictable digi-
talized environment, ought to be condemned. As a society, should we allow
advanced technology to change market dynamics when such change re-
sults in a transfer of wealth from customers to sellers? Should we focus on
the overall increase in efficiency and total welfare generated by new tech-
nology or on consumer welfare?
These questions go to the heart of competition policy and the goals of
competition enforcement. The balancing point between pro- and anticom-
petitive effects adopted in each jurisdiction reflects the antitrust goals and
values of that jurisdiction.17 As a result, the answer to these questions may
differ depending on the particular DNA of each regime and the hierarchy
of its competition goals.
Also relevant is the analytical framework of the substantive prohibitions—
be it administrative or criminal. In some jurisdictions a wide assumption of
illegality may make it easier to treat certain actions as anticompetitive by
object, or per se illegal. Also, the role of intent in establishing violations may
differ between jurisdictions—and as such may affect the ease with which
authorities may confront the Digital Eye scenario.

Reﬂections
To some readers, the Digital Eye scenario may appear counterintuitive.
After all, every risk we identify could be associated with a more competitive
environment: the increase in market transparency can lower consumers’
search costs. The velocity of price changes means that prices can come down
faster (and go up quicker in periods of scarcity, which promotes allocative
efficiency). The computers’ ability to calculate the likely profits from dif-
ferent moves and countermoves may mean procompetitive responses that
Copyright © 2016. Harvard University Press. All rights reserved.

humans may not have foreseen. Greater profits could be gained by devel-
oping computers that, through self-learning or programming, opt for the
profit-maximizing strategy, whereby everyone else charges the high price
while the company defects (and sells more items and earns greater profits).
We do not rule out procompetitive outcomes. In some markets new en-
trants, changes in customers’ purchasing patterns, disruptive technology,
or mavericks could destabilize or prevent tacit collusion. Nor do we argue
that algorithms are inherently bad, or that all markets will result in con-
scious parallelism.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Artificial Intelligence, God View, and the Digital Eye 81

Instead, we consider how new technologies, emerging market dynamics,

and pricing algorithms, instead of representing the end of collusion, poten-
tially mark a new, more durable form of collusion. Our scenarios illustrate
how, under certain market conditions, firms’ pricing algorithms could easily
settle into a profit-maximizing strategy of conscious parallelism, which
extracts their customers’ wealth through higher prices. So at a minimum,
competition authorities must be aware of these risks.
We may perceive the veneer of competition generated by the multitude
of online sellers and offers. We may be unaware of any tacit collusion. In-
deed, there may be periods where one firm enters another market and sells
at a discount. At times, there will be retaliation. Yet behind the scenes
competition may be undermined by close monitoring and retaliation. The
computers’ strategies may be so refined that both companies may be in the
same market but not really competing. The stable equilibrium could appear
to reflect a competitive equilibrium, when it actually reflects a subtle cus-
tomer allocation scheme—each algorithm catering to different categories
of customers.
Some readers, given the legal and ethical challenges in our last scenario,
may argue strongly against intervention. But a free-market approach—by
leaving these dynamics intact—would fail to protect many of us. It seems
reasonable to predict that a market reality in which algorithms enhance
and maintain tacit collusion is likely to intensify. Advanced technology and
the availability of a wide data pool will increase the instances in which the
conditions for tacit collusion are present. Fueled by profit maximization
(or, to put it bluntly, greed), fi rms may use algorithms to transfer more
wealth from purchasers (and ultimately us), further widening the gap
between the stressed middle- and lower-income classes and the wealthy.
Before we draw closer toward a plutocracy, a reevaluation of society’s toler-
ance for such dynamics may be called for, which we revisit in the last part
Copyright © 2016. Harvard University Press. All rights reserved.

of the book.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Copyright © 2016. Harvard University Press. All rights reserved.

Behavioral Discrimination

P ART II EXPLORED how pricing algorithms can foster express or tacit collu-
sion. In this part we consider a different theory of harm—the possibility
of a single firm using data-driven algorithms to better target consumers with
personalized marketing, pricing, and products.
The market reality here differs from our collusion scenarios. With collu-
sion, pricing algorithms increase the transparency of the terms of sale and
foster coordinated alignment of pricing. In this part we consider competi-
tors’ unilateral strategies to limit price transparency for highly differentiated
products. We no longer have a uniformly high (supracompetitive) price.
Instead, each firm seeks to charge different prices to different customers to
maximize its profits. The price you are charged reflects the firm’s estimate
of how much you are willing to pay.
The increased personalization of our online environment has been notice-
able in recent years. The advertisements you see online may differ from the
ads your spouse, children, parents, or neighbors see. Indeed, we do not know
to what extent the ads (or content) we see reflect our search inquiries, past
purchases, or even the subject of a recent e-mail or text we wrote. As the Wall
Copyright © 2016. Harvard University Press. All rights reserved.

Street Journal reported, “the idea of an unbiased, impersonal Internet is fast

giving way to an online world that, in reality, is increasingly tailored and tar-
geted. Websites are adopting techniques to glean information about visitors to
their sites, in real time, and then deliver different versions of the Web to dif-
ferent people. Prices change, products get swapped out, wording is modified,
and there is little way for the typical website user to spot it when it happens.”1
These personalization trends are at the heart of our discussion. Firms
track you, collect data about you, develop a profi le about you, and then
target you with personalized advertisements to induce you to buy. We
83

explore how the personalized, differentiated online environment affects the

dynamics of competition and our welfare.
Behavioral advertising, personalized product offerings, and targeted
pricing can help reduce our search costs and save time. Targeted advertise-
ments and promotions can help us quickly learn more about relevant
market opportunities. Yet, as we illustrate, behavioral discrimination can
reduce our welfare. Targeted ads and marketing not only facilitate our
consumption; they can influence and increase it. “Individualization” does
not stop at promotions. It also affects pricing decisions—what is often re-
ferred to as “price optimization” or “dynamic differential pricing”—so that
the more vulnerable end up consuming and paying more.
After a brief overview of price discrimination in Chapter 9, we explore
in Chapter 10 how Big Data and machine learning will enable companies
to better discriminate in pricing. But several challenges currently prevent
algorithms from perfectly discriminating. While perfect price discrimina-
tion is unlikely in the short term, Chapter 11 explores behavioral discrimi-
nation, where data-driven algorithms learn how to segment us into smaller
groups to target us with our own special offer. Our growing reliance on
web-based platforms for searches and purchases is changing the competi-
tive dynamics. Chapter 12 explores the social effects of behavioral discrim-
ination in a digitalized environment. Finally, Chapter 13 considers the
rise of price comparison websites (PCWs), metasearch engines and online
platforms. We explore how they may inhibit price discrimination and how,
at times, they could result in customers paying a higher price in a seemingly
competitive market.
Copyright © 2016. Harvard University Press. All rights reserved.

Price Discrimination (Brieﬂy) Explained

O NE WAY to price discriminate is to charge different customers different

prices based on the customer’s willingness to pay, irrespective of
cost differences for similar ser vices or products. Another way is to offer
a range of products that “sort customers based on their purchase deci-
sions.”1 The products are relatively similar, but some brands may be
more heavi ly marketed.2 Or products may be customized to individual
tastes.
A simple example would be if the local merchant knew how much each
customer is willing to pay for a can of Coca-Cola. The merchant could
charge a die-hard Coke drinker $3 and an ambivalent cola drinker 40 cents.
As competition authorities recognize, price discrimination, to succeed, re-
quires differential pricing and limited arbitrage.3

Differential Pricing
Differential pricing refers to the seller’s ability to segment its customers and
identify the demand elasticity of each customer or groups of customers.
Copyright © 2016. Harvard University Press. All rights reserved.

Suppose an art gallery did not post prices for its black-and-white Walker
Evans photographs. You like a photo. In your mind, you have an idea of
how much you are willing to pay. To price discriminate, the art gallery
must figure out the maximum price that you are willing to pay. Economists
call this the customer’s reservation price. Suppose the gallery owner could
read its customers’ minds. Each customer would leave the gallery paying
his or her reservation price for the black-and-white photograph—some
might pay only $10; others would pay over $100. With detailed informa-
tion on each customer’s reservation price, the gallery can perfectly price
85

discriminate (also known as first-degree price discrimination) by charging

each customer the maximum price he or she is willing to pay.
In perfectly price discriminating, the gallery maximizes its earnings by
capturing all the consumer surplus. Consumer surplus is basically the dif-
ference between the price you actually pay for a particular good or ser vice
and your reservation price.4 Suppose the gallery owner loses his mind-
reading powers. No longer able to perfectly price discriminate, he now
charges a fi xed price. In entering the gallery, you are willing to pay $500
for the Walker Evans photograph. Looking at the price tag, you see that it
costs only $100. Your consumer surplus is $400 (the price you were willing
to pay minus the price you actually pay). Now you can spend, save, or donate
the $400.
One real-world example of near perfect price discrimination is tuition
at private U.S. universities. Colleges collect detailed financial information
on candidates’ parents and their ability to pay; in awarding grants and
scholarships, many charge less in tuition to those families with fewer re-
sources. We say near perfect price discrimination, because some parents
may be willing to pay more than the posted tuition and room and board
for their child to attend a selective university. Depending on the student’s
acumen and the university’s admission standards, this may entail making
a significant donation (such as funding a new dormitory).5
In the old economy, many sellers lacked the detailed, accurate informa-
tion to assess each customer’s reservation price. Some instead imperfectly
price discriminated, a practice also known as third-degree price discrimina-
tion. Here the sellers segmented their customers into broad categories, which
were charged different prices. For instance, movie houses imperfectly price
discriminated for decades by charging different prices for adults, children,
students, and the elderly (on the assumption that students and the elderly
generally have less discretionary income and a lower reservation price).
Copyright © 2016. Harvard University Press. All rights reserved.

Limiting Arbitrage
Limiting arbitrage is the second requirement for successful price discrimi-
nation. This concerns the seller’s ability to prevent customers who pay a
lower price from reselling the product to customers willing to pay a higher
price. In our example, the gallery must find a way to prevent the buyer who
bought the black-and-white Walker Evans photo for $10 from reselling it
on eBay or Amazon to those willing to pay more. Sellers can deter arbitrage

in various ways, such as customizing the product, voiding any warranties

on resales, or making servicing the product more difficult or costly for cus-
tomers who buy on the gray market.6 Shipping products across borders
may be impractical or too expensive. As competition authorities recognize,
“Arbitrage on a modest scale may be possible but sufficiently costly or lim-
ited that it would not deter or defeat a discriminatory pricing strategy.”7

Dynamic Pricing
Price discrimination differs from dynamic pricing, where prices change in
response to changes in supply and demand. The airline industry was one
of the first to profit from dynamic pricing. Early estimates from the 1990’s
suggested that American Airlines, which many consider to have pioneered
dynamic pricing, made, at the time, an extra $500 million per year through
its yield management.8 Today, dynamic pricing is pervasive in the airline
industry, with frequent changes to pricing and availability of seat class—all
aimed at maximizing profitability—by estimating customers’ flexibility,
outside options, and reservation price. Similar practices are common in
many other industries, from hotels to sporting events. They may be used
in brick and mortar outlets or online.
For instance, retailers may change prices based on the time of purchase,
the availability of competing products, or the diminishing desirability of the
product. Sometimes, these strategies are simple to execute. For example, su-
permarkets often discount bread and other groceries toward the end of their
shelf life. They are not necessarily engaging in price discrimination, as the
groceries—close to their expiration date—differ from fresher products. Their
taste may be less appealing; they cannot be stored for a long period; they are
less in demand. So to sell its remaining supply of these groceries, the super-
market lowers its price. Here the retailer is responding to a shift in demand
Copyright © 2016. Harvard University Press. All rights reserved.

for its remaining supply. In principle, these practices differ from price dis-
crimination, where the supermarket charges different prices to different
consumers, based on their different reservation prices, for the same bread.
In practice, the distinction between dynamic pricing and price discrim-
ination may blur as sellers engage in more complex strategies.9 For instance,
a supermarket may engage in dynamic pricing by charging less for the same
product during lunchtime and late hours (when demand is low) and in-
crease the price in the early evening (when the outlet is packed with shop-
pers). This practice may also reflect price discrimination based on the time

sensitivity of customers. Suppose the supermarket knows that its more

price-sensitive customers will defer the purchase of bread until they head
home from work. These more price-sensitive customers also know that the
supermarket will discount the bread at the day’s end. So is the supermarket
adjusting price to changing demand or engaging in price discrimination
(charging less price-sensitive purchasers more before 7 p.m., and giving the
discount to more price-sensitive customers after 7 p.m.)? It is hard to tell.
What might appear as pricing responsive to changing market conditions
may simply be the supermarket segmenting customers by their price
sensitivity.

Why Do Firms Price Discriminate?

Price discrimination is often profitable.10 Returning to our gallery example,
when the gallery charges a fi xed price, say $250, for a black-and-white
photo, it obtains the sale of every customer whose reservation price is at or
above that price. It loses the sale of potential customers whose reservation
price is less than $250 for the photo. Moreover, the person who is willing
to pay $1,000 will not necessarily buy four black-and-white photos. Instead,
the customer may pay $250 for the one photo, and spend the $750 in con-
sumer surplus elsewhere. In perfectly price discriminating, the gallery
captures that consumer surplus. Moreover, it gains all the sales from cus-
tomers whose reservation price is below the fi xed retail price but above
the photo’s cost. So long as the customer’s reservation price is above the
gallery’s cost, the gallery will profit with each sale. Essentially the gallery
does not leave any money on the table. Everyone will pay the maximum
amount that they are willing to spend, more photos will be sold, and the
gallery will maximize its profit. By imperfectly price discriminating, the
seller captures more consumer surplus than it would with a fi xed price, but
Copyright © 2016. Harvard University Press. All rights reserved.

less consumer surplus than our clairvoyant gallery owner, who leaves no
money on the table.
We’ll explore in the next chapter how data-driven companies increase
profits by extracting as much consumer surplus as they can—by getting
them to buy things they didn’t know they needed and to pay more when
they can (or have fewer outside options).

The Age of Perfect Price Discrimination?

O NE PUBLICIZED TOPIC at the 2015 Advertising Week event in New York

was the personalization of advertising, exploring how, by 2020, all
advertising “will be planned and bought using household-level and
individual-level data.”1 Could the growth of Big Data and Big Analytics
enable online sellers to perfectly price discriminate—where each customer
pays the most he or she is willing to pay?
Competition as we know it is changing as companies experiment in de-
vising better ways to price discriminate—far better than the imperfect
price discrimination of the analog era (such as the senior citizen discount).
Chronicling the advancements in tracking us and collecting our data, this
chapter explores whether perfect price discrimination is on the horizon.
The immediate answer, to paraphrase St. Thomas Aquinas’s comment on
happiness, is that perfect price discrimination is not possible on earth, but
better forms of imperfect price discrimination are possible.

Not Your Grandmother’s Marketplace

You may wonder how can any company price discriminate online, when
rivals, one or two clicks away, offer the same item at an everyday low price?
How are online firms able to minimize the attractiveness of the consum-
er’s outside options?
What economists and competition lawyers call “price discrimination,”
online industry participants call “price optimization” or “dynamic differ-
ential pricing”. Dynamic differential pricing, as Massachusetts Institute of
Technology Professor Yossi Sheffi has put it, is the “science of squeezing
every possible dollar from customers.”2 With the rise of Big Data and
89

self-learning price algorithms, online and brick-and-mortar sellers, not

surprisingly, are experimenting with better ways to price discriminate.
Some online retailers are tracking a consumer’s location, purchasing be-
havior, and other personal data to charge consumers with fewer options a
higher price.3 In 2012 the Wall Street Journal reported price discrimination
by the online and brick-and-mortar office-supply superstore Staples, Inc.4
The Journal found that Staples’ pricing algorithms charged different prices
on its website after estimating the customers’ locations.5 Staples apparently
considered the customer’s distance from its rivals OfficeMax’s or Office
Depot’s brick-and-mortar stores: “If rival stores were within 20 miles or
so, Staples.com usually showed a discounted price.”6 Staples is not alone.
Office Depot admitted using “customers’ browsing history and geoloca-
tion to vary the offers and products it displays to a visitor to its site.”7
In 2000, Amazon.com sold DVDs to different people at different prices.
Amazon called it “merely a test and ultimately refunded the price differ-
ence to people who paid more.”8 Patrick Misener, Global Vice President of
Amazon, recently referred to this price test as an isolated incident: Am-
azon “will never use demographic information to price. We will not use
purchase history or whatever other assumptions. We will not do that and
never have. . . . In the case of this random price test, there is no other word
for it than stupid. . . . We have not done it since. It was just dumb.”9
In 2010 the Wall Street Journal reported that Capital One Financial Corp.
“was using personalization technology to decide which credit cards to show
first-time visitors to its website” and “was showing different users different
cards first—either those for ‘excellent credit’ or ‘average credit.’ ”10
Some online sites gave discounts based on whether or not a person was
using a mobile device. “A person searching for hotels from the Web browser
of an iPhone or Android phone on travel sites Orbitz and CheapTickets
would see discounts of as much as 50% off the list price.”11 Rosetta Stone
Copyright © 2016. Harvard University Press. All rights reserved.

offered “different product ‘bundles’ in different places” and personalized

its suggestions based on how the visitor reached its website—whether from
a search engine, a social-media link, a mobile device, or a PC.12 “We are
increasingly focused on segmentation and targeting,” a spokesman said.
“Every customer is different.”13
Allstate was recently criticized for its so-called “marketplace consider-
ations” algorithm. The insurance company sought to optimize the price it
charged to individuals by determining the likelihood that they would com-
pare prices before purchasing insurance. This criticism stemmed from the

fact that the algorithm facilitated non-risk-based selective pricing, which

ranged from a discount of up to 90 percent off the standard rate to an in-
crease in premiums of up to 800 percent.14
Brick-and-mortar stores, like their online counterparts, are also experi-
menting with dynamic differential pricing. In 2014 U.K. retailer B&Q, for
example, was testing “electronic price tags that alter the price of an item
based on the profile of the customer. The system uses data stored from loy-
alty cards and spending habits (via a chip in shoppers’ mobile phones) to
work out a price to be displayed next to the goods on the shelf. The retailer
claims the move will reward loyal shoppers.”15
Even coupons are becoming more personalized and targeted. One ex-
ample is Coupons.com, an online platform that in 2015 delivered person-
alized promotions every month to approximately 17 million consumers. It
promotes over 2,000 brands from approximately 700 consumer packaged
goods companies, such as Clorox, Procter & Gamble, General Mills, and
Kellogg’s, as well as retailers like Albertsons-Safeway, CVS, Dollar General,
Kroger, and Walgreens. Its digital platform, according to its 2014 annual
report, seeks “to engage consumers at the critical moments when they are
choosing which products they will buy and where they will shop.”16 By
2015, Coupons.com was targeting shoppers with display and video mes-
sages, “all informed by online data and in-store purchase behav ior.”17
Data is key here. Coupons.com starts “with demographic and geography
based personalization techniques to ensure that consumers see and can
easily access the most relevant content.”18 It then can “personalize content
based on which offers the consumer has clicked on and what searches the
consumer may conduct on our network as well as the coupons that the con-
sumer previously activated by printing or loading to their loyalty cards
and redeeming.”19 Moreover, when using Coupons.com’s mobile phone
grocery list app, shoppers receive personalized coupons based on their gro-
Copyright © 2016. Harvard University Press. All rights reserved.

cery list. The online platform also accesses a retailer’s loyalty card data
to target its shoppers. Thus, one of Coupons.com’s “key strengths” is its
“[p]roprietary data on consumer behav ior from intent to purchase.”20
Retailers have for years used loyalty programs to collect customer data
and target them with specific ads and discounts.21 One example is Target.
When you walk into the retail outlet, you may be unaware that Target,
whenever it can, will assign you and every other shopper “a unique code—
known internally as the Guest ID number—that keeps tabs on every thing
[you] buy.”22 Whenever you use a credit card or store coupon, fi ll out a

survey, mail in a refund, call the customer helpline, open an e-mail from
Target, or visit Target’s website, Target will link that data to your Guest
ID.23 You may also be unaware of the other data Target collects about you,
including “your age, whether you are married and have kids, which part of
town you live in, how long it takes you to drive to the store, your estimated
salary, whether you’ve moved recently, what credit cards you carry in your
wallet, and what websites you visit.”24 Plus, Target may acquire additional
data about you, including “your ethnicity, job history, the magazines you
read, if you’ve ever declared bankruptcy or got divorced, the year you bought
(or lost) your house, where you went to college, what kinds of topics you talk
about online, whether you prefer certain brands of coffee, paper towels, ce-
real or applesauce, your political leanings, reading habits, charitable giving
and the number of cars you own.”25 Target collects information about your
phone.26 Target, with your consent, also tracks your “geo-location and in-
store location” so that it knows how far you are from a Target store, and in
what aisle you are currently.27 Target, with your consent, also collects your
Facebook ID, including your profile picture and your friends’ IDs; and your
Google ID and profile picture.28 Target also collects information that you
submit in any public forum, including blogs, chat rooms, or social networks,
such as Facebook.29
Some of you don’t want to be tracked online. So you may use your
browser’s “do not track” feature that lets you tell websites that you do not
want to have your online activities tracked. Target, as of mid 2016, does
not “respond” to your request not to be tracked.30 Moreover Target uses its
store cameras not strictly for security purposes but also for “operational
purposes such as measuring traffic patterns and tracking in-stock levels.”31
Why does Target go to such lengths to track you and collect data about
you and your friends? To increase your loyalty and spending at Target, the
New York Times reported. One desirable target are pregnant women. Target
Copyright © 2016. Harvard University Press. All rights reserved.

can sell them baby products (and likely other products once the expecting
parents are at Target). Through its baby-shower registry Target knew some
of its customers who were pregnant, but not every shopper revealed her
pregnancy. To identify which women are likely pregnant, Target’s com-
puters examined for any patterns the shopping purchases of women on its
baby-shower registry. This revealed that pregnant women were more likely
to buy certain items, such as, “larger quantities of unscented lotion around
the beginning of their second trimester.”32 From the wealth of purchase
data, Target identified about twenty-five products that, when analyzed

together, enabled it “to assign each shopper a ‘pregnancy prediction’

score.”33 Target could predict from the pattern of purchases (such as lotions)
if the shopper was pregnant, and also, by the types of products, estimate
“her due date to within a small window.”34 Target then analyzed every reg-
ular female shopper in its national database to identify those most likely
to be pregnant, and offered them coupons for baby-related products,
with the aim that they would buy other things at Target as well.
The Times reported about one father who demanded to see the manager
at a Target outside Minneapolis:
He was clutching coupons that had been sent to his daughter, and he was
angry, according to an employee who participated in the conversation.
“My daughter got this in the mail!” he said. “She’s still in high school, and
you’re sending her coupons for baby clothes and cribs? Are you trying to
encourage her to get pregnant?” The manager didn’t have any idea what
the man was talking about. He looked at the mailer. Sure enough, it was
addressed to the man’s daughter and contained advertisements for ma-
ternity clothing, nursery furniture and pictures of smiling infants. The
manager apologized and then called a few days later to apologize again.
On the phone, though, the father was somewhat abashed. “I had a talk
with my daughter,” he said. “It turns out there’s been some activities in
my house I haven’t been completely aware of. She’s due in August. I owe
you an apology.”35

Personalization is also prevalent in the U.K. supermarket sector, notably

Tesco, which introduced in 1995 its loyalty program, Clubcard. As its
former CEO noted, “We could treat customers as individuals. And we
could learn what they were interested in, what their behaviours were, and
we could tailor and target all of their marketing so that it was relevant to
that individual consumer.”36 Why all the data? As McKinsey & Company
Copyright © 2016. Harvard University Press. All rights reserved.

found, it helps drive loyalty. 37

This “personalization” extends to product offerings. As Merlin Stone,
business research leader at IBM, explained: “In every sector, the top 20% of
customers give 80% of the profit.” No retailer would call the mechanics of
their loyalty scheme discriminatory, but here, according to Stone, is how it
might work: “They set a level [of ser vice] that any customer will get, known
as the ‘vanilla treatment’, and the more valuable customers will get even
better treatment.” So if, for example, you are a “cherry-picker”—someone
who buys all the loss leaders and nothing else—a Tesco or a Sainsbury’s

might conclude that you would perhaps be better off in Lidl or Aldi. They
might then raise the price of the loss leader, or just stop stocking the goods
only cherry-pickers buy. More likely, however, they will actively discourage
the purchase of these basic goods by placing them somewhere obscure and
with little shelf space. That way, says Stone, “you don’t have social exclu-
sion, but at the same time you make sure you are focusing on selling the
most profitable products to your best customers.”38
On the fl ip side, companies may seek passive consumers with low en-
gagement who will continue paying high prices for poor service—and
tailor an environment for them which is free from promotions and ensures
continuing purchases. Alex Chisholm, chief executive of the U.K.’s Com-
petition and Markets Authority, observed how the “incumbent, wary of
disturbing the sleeping, tip-toes around the customer dorm, with auto
roll-overs, silent expiry of special offer periods . . .”39

Mining, Tracking, and Proﬁt-Maximizing

Companies can use Big Data to help their self-learning computer algo-
rithms optimize behavioral advertisements, individualize promotions, and
pricing. As Coupons.com’s CEO remarked, its sophisticated products can
“turn big data into smart data, using leading edge technologies such as
machine learning and cross device targeting.”40 With more data about us,
the pricing algorithms can better predict our behavior and preferences and
thereby better price discriminate.
Loyalty cards and payment records are only the tip of the iceberg as com-
panies track our activities and amass data about us. Indeed, brick-and-
mortar retailers are increasingly tracking their customers. Unless you used
a loyalty card, historically when you walked into a store, the retailer may
not know who you are, what you are looking for, what else might interest
Copyright © 2016. Harvard University Press. All rights reserved.

you, or how much you are willing to spend. Retailers these days may use
Wi-Fi and mobile phone technology to interact with shoppers as they enter
the outlet.41 They may also opt for more complex facial recognition soft-
ware to identify shoppers and learn about their behav ior.42
The FTC in 2014 hosted a seminar on mobile device tracking, dealing
with “how retailers and other businesses have been tracking consumers’
movements throughout and around retail stores and other attractions
using technologies that identify signals emitted by their mobile devices,”
and how “this tracking is invisible to consumers and occurs with no con-
sumer interaction.”43

A year later the FTC brought its first retail tracking case against Nomi
Technologies.44 Using mobile device tracking technology (often without
the customers’ knowledge), Nomi tracked 9 million mobile devices during
the first nine months of 2013 alone. It provided analytics ser vices to brick-
and-mortar retailers about aggregate customer traffic patterns, such as:
“the percentage of consumers merely passing by the store versus entering
the store; the average duration of consumers’ visits; types of mobile devices
used by consumers visiting a location; the percentage of repeat customers
within a given time period; and the number of customers that have also
visited another location within the client’s chain.”45 Interestingly, what got
Nomi in trouble wasn’t the tracking. Instead, it was the misleading nature
of its privacy policy. Nomi assured individuals that they could opt out of
its tracking through its website or at “any retailer using Nomi’s technology.”
But Nomi never required its retail clients to provide such notice. Since the
retailers never notified individuals of the tracking, consumers could not
opt out at the stores.46 Had Nomi simply not posted a privacy policy, it pre-
sumably could have continued tracking.
Besides the data they collect, companies also rely on data brokers, which,
the FTC found, collect a vast amount of information on consumers:
Of the nine data brokers, one data broker’s database has information on
1.4 billion consumer transactions and over 700 billion aggregated data
elements; another data broker’s database covers one trillion dollars in
consumer transactions; and yet another data broker adds three billion
new records each month to its databases. Most importantly, data brokers
hold a vast array of information on individual consumers. For example,
one of the nine data brokers has 3000 data segments for nearly every U.S.
consumer.47

Besides collecting data about individuals’ interests and online and of-
Copyright © 2016. Harvard University Press. All rights reserved.

fline activities, data brokers also offer analytics products. The FTC found
that a few data brokers “convert their analyses into marketing scores that,
for example, rank clients’ customers on the basis of how likely they are to
respond to particular marketing efforts or to make a purchase, their pres-
ence on the web or their influence over others, or other metrics.”48 All of
this data is collected and processed with the aim of identifying customers’
interests and price sensitivity to market specific products to them at spe-
cific prices (or with specific discounts).
In our modern world, much of our lives and livelihoods are intertwined
with the online environment. Consider the number of hours you spend on

your smartphone, tablet, or computer: at work, when communicating with

others, when purchasing goods and ser vices, for entertainment, and social-
izing. Then add the times when data is passively collected about you, such
as your location, whenever you carry your smartphone. Then consider how
much more data will be collected about you with the growth of driverless
cars, the Internet of Things, and smart watches. In these digitalized
algorithm-based markets, sellers will invest more money and time to better
“shadow” our activities, harvest data on our behavior, and identify key mo-
ments to induce us to buy at the right price.

Is Perfect Price Discrimination Near?

So as the volume, variety, and value of personal data increases, and self-
learning pricing algorithms improve, will more sellers be able to perfectly
price discriminate? To appreciate how competition can evolve, we must
first understand why perfect price discrimination is unlikely in many mar-
kets in the near future. (One unfortunate exception is real estate—where
both of us, in the overheated Oxford and Washington, DC, real estate mar-
kets, submitted bids that reflected the most we were willing to pay above
the asking price for a property. But leaving our personal expenditures aside,
let us focus on the online environment.)

Insufﬁcient Data
First, to perfectly price discriminate, a firm must develop an algorithm that
can identify each customer’s reservation price. Each consumer’s reserva-
tion price is a latent variable—that is, a variable that cannot be directly
observed but may be inferred. Online sellers often cannot directly observe,
when the customer goes online, how much that par ticu lar customer is
Copyright © 2016. Harvard University Press. All rights reserved.

willing to pay for a particular item. Instead, the pricing algorithm collects
and processes data that observes transactions and consumer behavior, such
as under what conditions and price points consumers do purchase and
when they do not.
So one impediment to perfect price discrimination is insufficient data.
Although the algorithm has a lot more personal data than the brick-and-
mortar retailer of twenty years ago, the algorithm still has insufficient data
for any particular customer: the customer may never have bought the item
before; and the customer’s behav ior may never have signaled how much he

or she is willing to spend. To accurately predict an individual’s reservation

price would require sufficient data to identify and measure each of the
many variables that affect that reservation price.

Predictability and (Ir)rationality

Second, to perfectly price discriminate with incomplete data, any predic-
tion would likely be based on assumptions, such as consumers having
stable premises and a preexisting reservation price. The problem is that
multiple dispositional and situational factors can affect purchasing deci-
sions, and consumers may not even know their reservation price.
Neoclassical economic thinking had simplifying assumptions on human
behav ior, namely that market participants are rational, self-interested, and
have willpower. But individuals, as the behavioral economics literature
explores, are far more complex. Over the past twenty years, the economic
literature has increasingly recognized and measured how (1) willpower
is imperfect, (2) biases and heuristics can affect decision making, and
(3) many people are concerned about fairness.49
Participants in one experiment were asked how much they were willing
to pay for a cold beer that they would drink on the beach.50 The experience
is the same (drinking on the beach a specific brand of beer which one’s
friend procured from a nearby place). The only difference was whether
the beer came from a nearby expensive hotel or a run-down grocery
store. Most consumers were willing to pay more if the beer came from
the fancy hotel ($2.65) than if it came from a run-down grocery store
($1.50). Even though the experience is the same, the reservation price dif-
fered significantly.
Consumers cannot always predict their reservation price, and may dis-
count factors that actually affect their reservation price. For example, does
Copyright © 2016. Harvard University Press. All rights reserved.

an inflated list price affect one’s reservation price? In one experiment, ex-
perienced real estate agents predicted it would not, but the results showed
other wise.51 Each subject received a ten-page packet of information that
real estate agents typically use to evaluate residential property, on a partic-
ular home.52 The independent variable was the house’s listing price relative
to its appraised value. The local real estate agents claimed that most real
estate agents would detect as “obviously deviant” if the listing price were
inflated by more than 5 percent from its appraised value.53 In two condi-
tions, the listing price was either 4 percent above or below the house’s

appraised and actual listing price ($74,900). In the two other conditions,
the listing price was 12 percent above or below the appraised value. For
both the real estate agents and lay persons who participated in the study,
the listing price significantly influenced their valuations. Even though the
house was the same, and even with the same market data, the higher the
house’s listing price, the higher the participants’ estimate of the property’s
appraised value.54 The listing price (whether inflated or discounted) sig-
nificantly biased both the amateurs’ and real estate experts’ estimates. The
only difference was that the amateurs acknowledged using the listing
price; the experts “flatly denied” considering the listing price.
We, like the real estate agents in the study, may also discount the signifi-
cance of an inflated list price. One example is the advertised “discount” off
the list price. One Amazon listing promoted how buyers saved $2,131.33
(99%) for a cat litter pan.55 Likewise, buyers seemingly saved a lot on a six-
ounce bag of dog treats; its list price of $822 was discounted to $7.90.56 No
one would likely pay $2,131.33 for a kitty litter pan or $822 for dog treats.
Nonetheless, given the behavioral literature on anchoring effects,57 our
reservation price may be affected by an inflated list price, and other factors
we claim we ignore or are unaware of. Indeed, when you travel abroad, you
may be shocked initially by how expensive or inexpensive things are; even-
tually you adjust.
So with bounded rational consumers with imperfect willpower, the move
toward perfect price discrimination requires identifying all the key par-
ameters for each individual, and observing and improving the estimate of
each parameter. Does a customer’s reservation price for a can of Coca-Cola
change if the preceding three days exceeded a certain temperature; whether
the customer is on a date; whether the person just finished exercising; whether
the purchase is from a vending machine or a store; and whether or not the
purchase is at an airport?
Copyright © 2016. Harvard University Press. All rights reserved.

We can see that a pricing algorithm tailored for each individual would
require an enormous volume and variety of data. It must identify all the
relevant variables that affect a particular person’s reservation price. Each
buying experience may differ: the day of the week; time of day; where the
person is; what else they looked at on the Internet; the individual’s sex, age,
education, and demographics; the order of goods presented; and their rela-
tive prices.

Sample Size
A third impediment to perfect price discrimination is that the algorithm
would need a sufficient sample size for its hypothesis test to be robust.
To identify all the relevant variables that affect a particular person’s reser-
vation price, one generally must conduct multiple experiments.
So how many observations would the algorithm need to accurately pre-
dict an individual’s reservation price? The algorithm may have the min-
imum sample size for daily or weekly purchases, but not for infrequent
purchases, such as television sets, automobiles, and so on.
Consequently, to perfectly price discriminate, the pricing algorithm
would have to factor in the myriad dispositional and situational factors that
could affect an individual’s reservation price. This requires the algorithm
to identify additional variables that can affect an individual’s reservation
price and have the data to accurately predict how the individual would
likely react under each scenario. The pricing algorithm may not have
enough trial-and-error opportunities to identify each variable needed to
calculate accurately the individual’s reservation price. If the algorithm
cannot calculate each person’s reservation price under those par ticu lar
behavioral and situational factors, the algorithm cannot perfectly price
discriminate.

Reﬂections
Companies have for decades practiced cruder forms of imperfect price dis-
crimination, such as discriminatory pricing based on broad groupings of
individuals (youth, students, adults, senior citizens) and cruder indepen-
dent variables (such as the customer’s location, the day of the week, or time
of day).
Copyright © 2016. Harvard University Press. All rights reserved.

Advances in tracking our behav ior, collecting and analyzing our per-
sonal data, and implementing differential pricing have improved firms’
ability to price discriminate. Nonetheless, unless customers accurately
reveal their reservation price (such as in heated real estate auctions), in the
near future, pricing algorithms cannot identify each individual’s reserva-
tion price in many online markets. At times, the algorithm may be better
at predicting our behav ior than we are. But ultimately every pricing algo-
rithm model will be wrong. The algorithm’s challenge is to calculate how
wrong it is and improve itself.

As we’ll see in the next chapter, with advances in pricing algorithms and
the collection of a greater variety and volume of personal data, online com-
panies can more closely approximate our reservation price. They may find
the road to perfect price discrimination and increased profits irresistible.
They will compete in refining their pricing algorithms’ many independent
variables, and in more precisely classifying individuals into smaller sub-
groups. As the European Data Protection Supervisor observed:
Economic theory shows that a provider maximises profit when it is able
to identify (and then, where appropriate, price-discriminate) between
customers. In principle, if all patients remain unidentified, a pharma-
ceutical company will likely set a price for a drug which is the same for
everyone. However, if the same company is able to identify who, among
its customers, has more financial resources or has a greater need for the
drug, it might be able to charge those customers a higher price (e.g.
through a “premium” version of the drug that claims to be more effective).
Big Data might facilitate such group discrimination. There is therefore a
direct relationship between the availability of large sets of health data and
the potential profitability of a number of industries active in the health-
care sector, as businesses will be able to better target their commercial
propositions and thus draw a greater profit from the use of personal data.
In a self-reinforcing trend, greater chances of profit will turn into an
even greater demand of data and greater need for effective safeguards
against abuses.58

As we’ll see, as the volume of data collected increases, and the data ana-
lytics and categorization of consumers improve, self-learning computer al-
gorithms will continually inch closer to perfect price discrimination.
Copyright © 2016. Harvard University Press. All rights reserved.

The Rise of “Almost Perfect”

Behavioral Discrimination

P ERFECT PRICE DISCRIMINATION may be unattainable. But “almost perfect”

behavioral discrimination may be within reach. In the online world
where Big Data meets behavioral economics, we are witnessing an emerging
category of price discrimination—behavioral discrimination. Here firms
harvest our personal data to identify which emotion (or bias) will prompt
us to buy a product, and what’s the most we are willing to pay.1 Sellers, in
tracking us and collecting data about us, can tailor their advertising and
marketing to target us at critical moments with the right price and emo-
tional pitch. So behavioral discrimination increases profits by increasing
overall consumption (by shifting the demand curve to the right and price
discriminating) and reducing consumer surplus.
This chapter explores how Big Data and Big Analytics are fueling this
arms race. Both online and brick-and-mortar stores are rapidly seeking to
improve their consumer profi les. Our online and increasingly offline ac-
tivities are progressively being tracked. The aim—as we’ll illustrate—is to
approach ever closer to perfect behavioral discrimination.
Copyright © 2016. Harvard University Press. All rights reserved.

Big Data, Learning by Doing, and the Scale of Experiments

In approaching perfect behavioral discrimination, self-learning algo-
rithms, while not identifying each customer’s reservation price, may
segregate us in ever smaller reference groups and refine for each group the
explanatory variables (that among other things capture biases and situa-
tional factors). The algorithms refine the crude divisions of yesteryear
(such as senior citizens and students) to more detailed, segmented reality,
where people are matched to groups with similar price sensitivity and
101

purchase behavior. So the march to “almost perfect” behavioral discrimi-

nation will involve labeling you in countless ways and placing you in
smaller groupings of consumers with common situational and disposi-
tional factors.
How is this possible? This is where Big Data, learning by doing, and the
scale of experiments come into play. Pricing algorithms can collect data on
you and other people. Users are divided into subgroups of like-minded,
like–price-sensitive individuals, who share common biases and levels of
willpower. Pricing algorithms can use data on how other people within
your grouping react to predict how you will likely react under similar cir-
cumstances. This then enables the algorithm to more accurately approxi-
mate the user’s reservation price, observe behavior, and adjust. Thus, the
more times the algorithm can observe what you and others within your
grouping do under various circumstances, the more experiments it can
run, the more it can learn through trial and error what your group’s reser-
vation price is under different situations, and the more it can recalibrate
and refine (including shifting you to another group).
Here is where scale comes into play. To optimize their performance,
firms will refine their pricing algorithms, which in turn requires continu-
ally refi ning the groupings and independent variables. Firms will get as
much relevant data as possible and to get as many opportunities to test and
refine their algorithms. Every actual or potential consumer transaction
gives the company an opportunity to study consumer behavior and adjust
both the weight attributed to each variable and the categorization of users.
Firms will compete in more accurately segmenting customers into sub-
groups to optimize pricing and profits. Th is is so because technology
enables operators to bypass challenges in identifying particular consumers’
reservation price, targeting the consumers, and preventing arbitrage. The
more data that the firm quickly collects on users, the better able it can seg-
Copyright © 2016. Harvard University Press. All rights reserved.

ment them; the more opportunities it has to observe individual behavior;

the more the algorithms can learn when they predict correctly; and the
more the algorithms can refine and retest when they predict incorrectly.
The data flow, like a hamster wheel, whirls continually. To better catego-
rize even smaller groups of individuals, firms will need more data. This will
accelerate the Internet of Things, as firms compete to collect data on con-
sumers’ activities in the home, at work, and outside. Smart thermostats, cars,
utensils, and watches will help firms refine their consumer profiles, and will
serve as platforms for behavioral advertisements. Information harvesting

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
The Rise of “Almost Perfect” Behavioral Discrimination 103

will go beyond purchasing decisions. For instance, online providers and

sellers may know and make use of customers’ locations via mapping soft-
ware, their browser and search history, their friends and links on social net-
works, the media they stream, their preferences, and the contents of their
online reviews and blog posts.2

The Reality of Behavioral Discrimination

A joke in IT circles illustrates the speed of the harvesting and usage of such
data: In the middle of an important board meeting, one manager anxiously
requests to leave the room. “Why?” asks the chairwoman. The manager,
pointing to his smartphone, replies, “I just received a string of ads on alarm
and security systems—I think my house was just burgled.”
A report by the U.K. Competition and Markets Authority (CMA) shows
the joke’s underlying reality. First, the joke reflects the velocity of being
targeted. As the CMA found, “This real-time method of buying and selling
inventory can enable adverts to be served to users’ devices within millisec-
onds, offering increased accuracy by targeting refi ned audiences rather
than the previous method of buying audiences in bulk tranches of thou-
sands.”3 The U.K. Internet Advertising Bureau estimates that “in 2013, 28%
of all digital display advertising was traded programmatically and that by
2017 this could increase to between 60% and 75%.”4
Second, as the joke reflects, firms are already combining and analyzing
data about consumers to make inferences about them. Using case studies
in three sectors (auto insurance, clothing retailing, and game apps), the
published literature, and meetings with key parties, the CMA found that
firms are using detailed consumer data to increase consumption of their
goods and services.5 Firms increase consumption through targeted adver-
tising, “which can increase the conversion rate from advertisements to pur-
Copyright © 2016. Harvard University Press. All rights reserved.

chases of a range of products and services available,” and by “using data on

consumers’ previous purchases or areas of interest to cross-sell related prod-
ucts and services (for instance ‘you may also be interested in . . .’ messages
on websites).”6 The CMA found such concerns were greater in the digital
advertising space, with the “growth in programmatic advertising—the fully
automated buying and selling of digital advertising space.”7
If you have an iPhone, Apple is categorizing you. When updating its iOS
operating system in 2015, Apple discussed how its advertising platform,
iAd, “creates groups of people who share similar characteristics and uses

these groups for ad targeting.”8 Apple uses the data it collects “to deter-
mine which groups you are assigned to, and thus, which ads you receive.”9
Apple stated that to protect “your privacy, your information is used to
place you into groups of at least 5,000 people.”10
So in which categories will you fall? That depends on a variety of per-
sonal information Apple will harvest, including:
• Account Information: Your name, address, age, and devices regis-
tered to your account.
• Downloads: The music, movies, books, TV shows, and apps you
download.
• Device Information: Your keyboard language settings, location,
device, and connection type.
• Activities in Apple Apps: The topics and publications you follow for
news, the types of music you listen to in Apple Music, the offers you
choose to add to Wallet.
• Activities in other Apps: Information that other app developers
(with your permission) provide Apple regarding your in-app pur-
chases and activities such as game-level completion.
• Advertising: Your interaction with advertising delivered by iAd.
• Other Segments: Information that third parties may also share with
Apple, including information on groups of people in which you
belong (so long as no individual data is shared).11

Apple’s privacy policy explains how users can clear existing data collected
on them (but not necessarily how they can stop being tracked and the data
from being collected). Apple also explains how users can limit its targeted
advertising, which is not self-evident.12
Needless to say, we use Apple to illustrate an industry trend. The FTC in
Copyright © 2016. Harvard University Press. All rights reserved.

2014 described how data brokers were developing complex models to predict
consumer behavior, primarily by categorizing and segmenting consumers:

The data brokers can identify a group of consumers that has already
bought the products in which the data broker wants to predict an interest,
analyze the characteristics the consumers share, and use the shared char-
acteristic data to create a predictive model to apply to other consumers.
For example, a data broker can:
• Analyze the characteristics of a subset of consumers that purchased
camping gear in the last year, identify consumers in its database that

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
The Rise of “Almost Perfect” Behavioral Discrimination 105

share these characteristics, and create a segment called “Consumers

Interested in Buying Camping Gear”;
• Identify a group of consumers that sought chargebacks on their credit
cards in the last year, analyze the characteristics those consumers
share, and use the characteristic data to predict “Consumers that are
Likely to Seek a Chargeback”; or
• Analyze data on consumers in this manner to predict which con-
sumers are likely to use brand name medicine, order prescriptions by
mail, research medications online, or respond to pharmaceutical
advertisements.13

Increasing Demand by Exploiting Biases

Behavioral discrimination isn’t just categorizing us into groups that are
charged different prices. It is also about getting us to buy things that we
may not need or have previously wanted. One popu lar Internet quote is
“We buy things we don’t need with money we don’t have to impress people
we don’t like.”14 So to increase demand for their products and ser vices,
companies will likely appeal to our emotional wants.
As noted earlier, most of us are not rational, self-interested individuals
with willpower. The field of behavioral economics, as one of its pioneers,
Amos Tversky, noted, has quantified what every good advertiser and car
salesman already knew.15 We have cognitive biases, which refer to our ten-
dency to react, think, or operate in a certain way, which diverge from as-
sumed rationality. Biases can be observed. But businesses and governments
can trigger consumers’ biases to achieve certain goals.16 As noted by
Cialdini, factors such as relative pricing, reciprocity, and the illusion of
scarcity play a powerful role in the persuasion game.17
One competition authority official told us in 2015 that the behavioral
Copyright © 2016. Harvard University Press. All rights reserved.

economics literature identifies over one hundred human biases linked to

decision making, information processing, memory, and social interaction.
Companies could surely identify a number of biases, he remarked, in order
to better price discriminate online. An algorithm-fueled environment will
provide firms with unparalleled information about our desires, behavior,
interests, search patterns, and willingness and ability to pay.
Below, we explore a few of the numerous “exploitable” consumer biases,
which firms may use to promote (or reduce the consumer outrage over)
their behavioral discrimination.

Use of Decoys
We rarely choose goods and ser vices in absolute terms. Instead, we base
our choices on the product’s relative advantage or disadvantage to other
things.18 How firms position their products can influence the purchase de-
cision. By adding an expensive (albeit inferior) choice, for example, the
marketer can steer consumers to a more expensive second choice. As the
consulting firm McKinsey & Company reported, “many restaurants find
that the second-most-expensive bottle of wine is very popular—and so is
the second-cheapest. Customers who buy the former feel they are getting
something special but not going over the top. Those who buy the latter feel
they are getting a bargain but not being cheap.”19
In one study, a hundred MIT students were offered three subscription
choices for The Economist magazine: (1) Internet-only subscriptions for $59
(sixteen students chose this option); (2) print-only subscriptions for $125
(no students); and (3) print-and-Internet subscriptions for $125 (eighty-four
students).20 When the “decoy” second choice (print-only subscriptions) was
removed and only the first and third options were presented, the students
did not react similarly.21 Instead, sixty-eight students opted for an Internet-
only subscription for $59 (up from sixteen students) and only thirty-two
students chose print-and-Internet subscriptions for $125 (down from
eighty-four students).22
Online sellers can use decoy products or pricing to push consumers
toward higher-margin products. Apple, for example, can make its recently
launched $349 Apple Watch appear reasonable by adding thirty-eight dif-
ferent designs, ranging between $349 and $17,000.23 Few, if any, will pay
$17,000 for an Apple Watch, but it makes the $349 watch seem more rea-
sonably priced.
A study has been carried out to determine whether the introduction of
Copyright © 2016. Harvard University Press. All rights reserved.

a decoy soft ware option can increase demand for the real option by ex-
ploiting consumers’ relative assessments of prices.24 The study used Micro-
soft’s portfolio of Windows 7, which included a Windows 7 Professional
bundled with a 4-GB pen drive and a decoy Windows 7 Professional op-
tion priced the same as the bundle. The study found that the presence of
the decoy makes the bundle option “a lucrative one and has the potential
of increasing overall revenues by 15 per cent.”25
Companies are already using these tactics. They may purposefully make
the least expensive private-label products less appealing visually in order

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
The Rise of “Almost Perfect” Behavioral Discrimination 107

to nudge customers to higher-priced midrange private-label or other

higher-margin goods. Indeed, one concern is that the commercial airlines
are purposefully degrading the experience of flying coach to push less
price-sensitive customers to “coach plus” or business class.

Price Steering
Besides offering decoy choices, firms can nudge consumers closer to their
reservation price by the way they present options online. As a 2015 White
House report explains, “Steering is the practice of showing different prod-
ucts to customers in different demographic groups. In the online environ-
ment, steering occurs when a web site alters its search results based on
information about a potential customer.”26 For example, for consumers
with higher reservation prices, the online seller would likely present first
the premium, more expensive brands.27 As the U.K.’s competition au-
thority noted, “Firms may do this by restricting the products that are dis-
played to consumers or by varying the order in which products are listed
on their website to display relatively poorer or better quality products
fi rst depending on the information they collect about consumers. This
raises the possibility of some consumers being exploited with low quality
products that are sold at the same price as higher quality products.”28
Thus, under the new competitive paradigm, the online experience will
differ, as firms personalize how, and in what order, they present the prod-
ucts. The offerings on the web page may be tailored depending on your zip
code, household wealth, gender, and age. So it will be harder to know what
others see. At most, you might know what other people of similar age in
your zip code see if they have similar professional and educational back-
grounds, visited the same websites and have similar purchase histories. As
personalized offerings increase, search costs will also increase for con-
Copyright © 2016. Harvard University Press. All rights reserved.

sumers seeking to identify the “true” market price.

One older example was the travel website Orbitz, which steered Mac
OS X users toward “more expensive hotels in select locations by placing
them at higher ranks in search results.”29 One 2014 study found additional
evidence of price steering and price discrimination.30 The study examined
sixteen popular e-commerce sites involving general retailers and hotel and
rental car booking sites, and found evidence of price steering and price
discrimination on four general retailers’ websites and five travel sites. For
example, the travel website Expedia was assigning users to one of three

buckets, and steering users in some buckets toward more expensive hotels.31
Travelocity offered different hotel search results depending on whether
users were browsing on their iPhone or iPad or browsing from “Chrome
on Android, Safari on OS X, or other desktop browsers.”32 Travelocity gave
people using their iPhones and iPads better prices on some hotels.33 In con-
trast, Home Depot was steering users on mobile browsers toward more
expensive products.34

Increasing Complexity
To better discriminate, companies can take advantage of consumers’ diffi-
culty in processing many complex options. Companies deliberately increase
the complexity by adding price and quality parameters, with the intent to
facilitate consumer error or bias, to their advantage. Here, firms add options
and increase their products’ complexity to manipulate consumer demand
by making it difficult to appraise quality and compare products.35 Firms in-
crease the consumers’ search and evaluation costs, thus driving consumers
to rely on basic signaling that benefits the firms. Firms increase the com-
plexity of their contracts to increase their customers’ switching costs and to
more effectively price discriminate.36 In short, firms increase complexity to
render market conditions less susceptible to effective competition.
One study found that as competition in U.S. telecommunication markets
increased, telecommunication providers offered more complicated, bad-
value price plans.37 The increased competition caused “cellphone providers
to focus on raising profitability through creating confusion and gaining
from consumer mistakes.”38 A criticism of the mobile phone industry is its
deliberately increasing choice complexity to exploit consumers:

Too much and too complex information have made it difficult for all but
Copyright © 2016. Harvard University Press. All rights reserved.

the most technologically savvy to choose the product best suited to their
needs. Customers unable to choose based on attribute preferences ap-
peared to make their choices based on price, only to later find out that
the product did not meet their needs. This tendency is further compli-
cated by a lack of comprehension. When provided with multiple options,
consumers are only able to choose the least expensive about 65% of the
time. When faced with the complex options of base ser vice fees, addi-
tional features and cost for usage overages, customers tend to choose
plans that greatly exceed their requirements, significantly overpaying
each month rather than risking the chance of occasional overage costs.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
The Rise of “Almost Perfect” Behavioral Discrimination 109

Problems navigating the telecommunications industry are not limited to

older adults, although they may be particularly vulnerable.39

Similarly, another recent study found that a greater variety of price plans
in U.K. electricity markets led more consumers to choose suboptimally,
harming their welfare.40 Consumers may find it difficult to accurately compare
the true cost of a deal by one energy provider with that of another. Indeed,
the U.K. Competition and Markets Authority (CMA) noted that the funda-
mental characteristics of energy consumption presented two barriers to con-
sumers’ engagement with the retail energy market: (1) the absence of a quality
measurement for the differentiation of energy, which may “fundamentally
reduce consumers’ enthusiasm for, and interest in, engaging in the do-
mestic retail energy markets, leading to customer inertia”;41 and (2) the fact
that the information contained in conventional meters is not immediately
accessible. Customers or the supplier generally read conventional meters
infrequently, which “adds considerably to the complexity and opacity of
gas and electricity bills.”42 Moreover, “the perception of the complexity and
burden of the process” of searching for an alternative supplier limits suc-
cessful switching.43
Complexity has also been shown to limit switching between ser vice pro-
viders and to increase the likelihood of customers retaining the default
option.44 Ultimately, companies can discriminate by designing the number
and types of options they offer to exploit consumers’ cognitive overload.
Similarly, they may support complexity by increasing search costs. Online
firms may resort to age-old tactics, such as selling the same product but
under different labels. This is legendary in the U.S. mattress industry. Con-
sumer Reports noted the tricks the mattress industry uses to make it difficult
for consumers to compare models and negotiate a better price, including
when “manufacturers sell the identical or nearly identical mattresses to dif-
Copyright © 2016. Harvard University Press. All rights reserved.

ferent retailers with exclusive model names.”45 Each retailer tells the puz-
zled customer, “We don’t sell that Sealy brand, but we offer a superior mat-
tress from Sealy.”

Drip Pricing
Online companies can shroud their behavioral discrimination by making
the price terms more complex. The U.K. competition authority experi-
mented with five common price frames: (1) “drip pricing,” where a lower
price is initially disclosed to the consumer and additional charges are added

as the sale progresses; (2) “sales,” where the sale price is referenced off an
inflated regular price (e.g., was $2, now $1); (3) “complex pricing” (e.g.,
three-for-two offers), where the unit price requires some computation;
(4) “baiting,” where sellers promote special deals with only a limited
number of goods available at the discounted price; and (5) “time-limited
offers,” where the special price is available for a short period.46 Consumers
made more mistakes and were especially worse off under drip pricing and
time-limited offers.
The Executive Office of the President in the United States gave other ex-
amples of how firms may use complex or opaque pricing schemes to screen
out less sophisticated buyers:
[C]ompanies may obfuscate by bundling a low product price with costly
warranties or shipping fees, using “bait and switch” techniques to attract
unwary customers with low advertised prices and then upselling them on
different merchandise, or burying impor tant details in the small print of
complex contracts. When these tactics work, the economic intuition that
differential pricing allows firms to serve more price-sensitive customers
at a lower price-point may even be overturned. If price-sensitive cus-
tomers also tend to be less experienced, or less knowledgeable about
potential pitfalls, they might more readily accept offers that appear fine
on the surface but are actually full of hidden charges.47

Imperfect Willpower
Consumers with limited patience will often pay a higher price. This has
been known for decades. People could save money by waiting until a movie
appeared in a second-run theatre,48 or for the fiction hardcover to appear
in paperback. Thus, the more the online site can encourage impulse pur-
Copyright © 2016. Harvard University Press. All rights reserved.

chases (such as “scarcity marketing” that promotes the dwindling stockpile

of items and the many buyers looking at the item), the less likely the con-
sumer will comparison shop.
Likewise, online retailers may shroud their behavioral discrimination by
offering discounts to consumers with greater willpower. Baymard Institute,
a retailers’ Internet research firm, found that 68 percent of online shopping
carts are abandoned after initial click-throughs. Consumers who abandon
an online transaction may be rewarded. Indeed, it is not uncommon, when
purchasing online, that a second visit to a site, following incomplete pur-

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
The Rise of “Almost Perfect” Behavioral Discrimination 111

chase, would result in a pop-up screen with a discount code. Such practices
reflect the assumption that a customer who delayed purchase may be less
eager, more price sensitive, or considering other purchase options. Coupon
site Rather-be-shopping.com, for example, “found 17 well-known retailers
(including Bed, Bath & Beyond, Macy’s, and Williams-Sonoma) that of-
fered coupons (ranging from 20% off to free shipping) to customers who
left their carts.”49

Minimize the Perceived Unfairness through Framing Effects

Many consumers, as the next chapter discusses, feel cheated by price dis-
crimination. They are deprived of a market price. To address this, firms
can rely on framing effects when price discriminating. The behavioral eco-
nomics literature suggests that “framing effects” (how the issue is worded
or framed) do matter.50 Here the price discrimination is framed not as
consumers paying more, but rather as some getting a discount.
Credit cards are one example. Merchants often tell consumers of a dis-
count if they pay with cash. Merchants never tell consumers they must pay
more if they use a credit card. The net effect is the same: credit card cus-
tomers pay more than cash customers. But how the choice is framed (either
as a discount or as a surcharge) does matter. To illustrate, after the credit
card companies’ no-discrimination rule was abolished, Dutch merchants
could impose surcharges or offer discounts based on how the customer was
going to pay. Of the consumers surveyed, 74 percent thought it (very) bad
if a merchant asked for a surcharge for using a credit card. But when asked
about a merchant offering a cash discount, only 49 percent thought it (very)
bad, with 22 percent neutral and 21 percent saying it is a (very) good thing.51
In another experiment, a majority of people said that a car dealer’s
elimination of a $200 discount off the list price for a popular vehicle was
Copyright © 2016. Harvard University Press. All rights reserved.

acceptable; 71 percent viewed selling the vehicle $200 above the list price as
unfair.52 Both scenarios again produce the same effect—a higher net retail
price—but the direction of the deviation to or from the established refer-
ence point differed.
Thus when price discriminating, online companies will not likely im-
pose a surcharge on those willing to pay more. Instead, they will likely
start with a higher list price, and then selectively vary the level or size of
discounts. Moreover, consumers’ incentive to search for the outside option
will vary, depending on whether there is a price increase versus a price

decrease. As the OECD noted, consumers “may search more aggressively

for alternative suppliers when prices increase, but less aggressively when
prices are stable or slowly decreasing.”53 Consumers typically base a deal’s
“value” on the deviation from an established reference point (for example,
a sale of 20 percent off the regular price). Many consumers may be less
concerned when the steady discounts are eliminated (such as the 20 percent
discount) than when list prices increase (although both have the same net
effect). Deviations from the perceived reference point are marked by asym-
metric price elasticity: consumers are angrier about, and more sensitive to,
price increases than to the elimination of a discount or the maintenance
of prices during periods of deflation.54 Thus the road to near-perfect be-
havioral discrimination will likely be paved with personalized coupons
and promotions: the less price-sensitive online customers may not care as
much if others are getting promotional codes, coupons, and so on, as long
as the list price does not increase.
Online sellers will increasingly offer consumers with a lower reservation
price a timely coupon—ostensibly for being a valued customer, a new cus-
tomer, a returning customer, or a customer who won the discount. The
coupon may appear randomly assigned, but only customers with a lower
reservation price are strategically targeted. For example, when American
Airlines introduced dynamic pricing, it framed “the 21-day advance-
purchase requirement as a chance to buy ‘super-saver’ fares.”55 Indeed, the
price discrimination can happen on other, less salient aspects of the pur-
chase. Retailers can offer the same price, but provide greater discounts on
shipping (or faster delivery), offer complimentary customer ser vice, or
better warranty terms to attract customers with lower reservation prices,
greater willpower, or more outside options.
Another way to frame behavioral discrimination in a palatable manner
is to ascribe the pricing deviations to shifting market forces. Few people pay
Copyright © 2016. Harvard University Press. All rights reserved.

the same price for corporate stock. They accept that the pricing differences
are responsive to market changes in supply and demand (dynamic pricing)
rather than price discrimination (differential pricing). So once consumers
accept that prices change rapidly (such as airfare, hotels, etc.), they have
lower expectations of price uniformity among competitors. One hotel may
be charging you a higher price because of its supply of rooms (rather than
discriminating against that par ticu lar user). When your friend inquires
about the same room, the different price could reflect an interim change in
supply or demand. Rarely will you and others simultaneously search on the

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
The Rise of “Almost Perfect” Behavioral Discrimination 113

same website for the same room and communicate your findings. Thus,
consumers may not know when pricing is dynamic, discriminatory, or
both.

Where Does the Power Reside?

The power in our scenario favors those who hold and sell our personal
data. These informational asymmetries support near-perfect behavioral
discrimination.
The first asymmetry is between the discriminating firm and its customers.
The firm collects data on its customers and designs the algorithm. The firm
may not know exactly how its pricing algorithm derived a particular price for
a particular customer (see Chapters 7 and 8), but the firm does know the
algorithm’s ultimate strategy (namely to increase profits by better discrim-
ination). The firm, unlike its customers, also knows the magnitude of the
price disparity among the various consumer categories.
We, on the other hand, are in the dark on this data-driven behavioral
discrimination. The list of “known unknowns” is troubling. In many
jurisdictions, including the United States, we do not know:
• who has been collecting data on us;
• the uses to which our personal data will be put;
• the other potential recipients of our personal data;
• the nature of the personal data collected about us;
• the categories in which we are placed;
• the means by which our personal data is collected;
• the quality of the data (and inferences made about our likely
behavior); and
• what, if any, options we have to control how the data is used.
Copyright © 2016. Harvard University Press. All rights reserved.

Nor, in most countries, do individuals as of this book’s publication date

have the right to view the data, and to verify or contest the data’s accuracy.
Nor can we challenge the accuracy of categories in which we are placed. We
may know that we are getting a lot of ads—such as substance abuse, lower-
quality credit cards, or checking our prison records—but not know why
these ads are directed at us.56
The second level of informational asymmetry is between the firm and
its competitors. Because perfect price discrimination will remain elusive,
firms will seek to refine their categorizations of consumers. Not all the

firms will have the same data. Firms with a data advantage will likely
have better algorithms that can better segment customers, and for each
group they will likely be better at identifying the group’s average reservation
price (and a narrower distribution of individual reservation prices). Rather
than converge under the tacit collusion scenarios, the pricing algorithms
here might diverge, with some firms’ algorithms improving and others re-
maining cruder. If some competitors are closer to perfect behavioral dis-
crimination than others, that might suggest greater competitive opportu-
nities. Consumers with (perceived) high reservation prices could switch to
competitors with cruder algorithms (so to get better deals). Consumers
with low reservation prices could switch to competitors with more sophis-
ticated algorithms (where the product they are offered will likely be cheaper).
The problem is that firms that discriminate will generally seek to reduce
the customers’ outside options by reducing price transparency and in-
creasing the customers’ search costs. This may be easy. Some customers
are known as “sleepers,” a term for customers “who out of indolence or ig-
norance don’t shop but instead are loyal to whichever seller they’ve been
accustomed to buy from.”57 Switching costs may therefore be higher than
one assumes, despite perceived competition being only a click away. Illus-
trative is the behav ior of many users who indicated that when a search
result fails to meet their expectations, they will “try to change the search
query—not the search engine.”58
The firm might differentiate its products and ser vices, perhaps through
customization. When products and ser vices are customized to individual
tastes, there is no longer a common benchmark and it can become harder
for customers (and competitors) to compare products and prices. The cus-
tomization and disappearance of a competitive benchmark could also
make it harder for potential entrants to assess what price they should charge
to convert sufficient customers for their entry to be profitable.
Copyright © 2016. Harvard University Press. All rights reserved.

Moreover, as more online retailers engage in dynamic, differential

pricing, it will be harder for consumers to discover a general market price
and to assess their outside options.59 For example, if Amazon would engage
in behavioral discrimination, the customer might turn to Walmart. But
if Walmart and other retailers engage in behavioral discrimination for
customized products, it will be increasingly difficult for consumers to find
the competitive benchmark. A White House report summarized this
trend toward personalized pricing and individually-targeted marketing
campaigns: “A review of the current practices suggests that sellers are now

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
The Rise of “Almost Perfect” Behavioral Discrimination 115

using big data and digital technology to explore consumer demand, to steer
consumers towards particular products, to create targeted advertising and
marketing offers, and in a more limited and experimental fashion, to set
personalized prices.”60
Importantly, that power is not unlimited. It depends on the level of com-
petition, the availability of outside options, and public perception of the
unfairness of price discrimination. It is also affected by the ability to har-
vest and process data, economies of scale, and network effects. The power
to price discriminate may be curtailed by a possible pushback from con-
sumers, in the form of programs designed to outsmart price algorithms
and trigger discounts or lower prices. A market may emerge in which
countermeasures develop for individuals to migrate between groups or ob-
struct segmentation.

Reﬂections
While technology, financial, and other barriers may prevent sellers from
perfecting behavioral discrimination,61 the online environment is still
much more susceptible to such measures than the traditional brick-and-
mortar shops. So whatever the measures and countermeasures, we are
moving away from the old competitive environment where the store clerk
stamps one price on the item, which everyone pays. Customers in the
coming years may still know the price of milk at several retailers, and
retailers will still compete by lowering prices for some items or improving
ser vice. But for many products and ser vices, the growth of Big Data
and Big Analytics will lead to greater opportunities to acquire informa-
tion on the preferences, weaknesses, and elasticity of demand for discrete
groups of customers. The more detailed such information is, the easier it
may be to segment the customer base and approach perfect behavioral
Copyright © 2016. Harvard University Press. All rights reserved.

discrimination.
We may not know when, and the extent to which, we are victims of
behavioral discrimination.62 Often markets exhibit weak consumer en-
gagement.63 Only a minority will likely invest time in countermeasures to
curtail tracking. Even the astute, to benefit from discounts through loy-
alty programs, must reveal their identity. Moreover, the loyalty discounts
are more salient than the perceived savings in remaining anonymous.
These processes not only affect us, but may equally affect small and
medium size enterprises (SMEs), which may lack the sophistication and

resources to cope with leading market players. As noted by CMA Chairman,

David Currie: “Compared with the large corporation, SMEs typically have
scarce skilled resource, their transactions are more sporadic, their infor-
mation is limited and their access to high volume data processing and ana-
lytics is limited. So they may be much more like individual consumers, so
that their biases risk being used by large companies, just as for individual
consumers.”64
An interesting and challenging question which emerges from our discus-
sion is whether the enhanced capacity to price discriminate in a digitalized
environment should call for a more interventionist approach. The next
chapter examines the welfare implications of behavioral discrimination, and
what tools enforcers could use as firms get even better at discriminating.
Copyright © 2016. Harvard University Press. All rights reserved.

Behavioral Discrimination:
Economic and Social Perspectives

O UR GROWING RELIANCE on web-based platforms for searches and pur-

chases changes the dynamics of competition. It can facilitate an al-
most perfect behavioral discrimination, which will result in customers
consuming more and many paying a higher price in a seemingly competi-
tive market.
In this chapter we consider the welfare effects of behavioral discrimina-
tion. We do so in two steps.
First, we explore the welfare effects of price discrimination. We note how
price discrimination’s welfare effects are mixed. On the one hand, our in-
creasingly automated, digitalized transactions could create a more trans-
parent marketplace in which resources are allocated more efficiently and
in which the best product or ser vice, at the lowest price, triumphs. On the
other hand, pricing algorithms may be used to exploit customers and raise
“the seemingly endless possibilities for both chaos and mischief.”1
Second, we move beyond price discrimination into the murkier water of
behavioral discrimination. Digitalized algorithm-based markets are char-
acterized by the ability of sellers to increasingly “shadow” our activities,
Copyright © 2016. Harvard University Press. All rights reserved.

harvest data on our behavior and preferences, build profiles about us, tailor
inducements, and recalibrate based on our responses. This behavioral dis-
crimination can increase consumption, optimize the extraction of wealth,
and affect other important values, such as privacy, equality and fairness.

The Neoclassical Economists’ Take on Price Discrimination

Do we lose when sellers discriminate? Well, it depends. At times we will; at
other times we won’t. Price discrimination could generate a mixture of
117

effects under different market conditions. Not surprisingly, economists

debate over the net welfare effects of price discrimination. Some view it
more negatively than others. For instance, two economists, Leeson and
Sobel, suggest that when the costs of implementing price discrimination
are taken into account, its efficiency is doubtful.2
Price discrimination can, at times, yield distinct efficiencies and welfare
gains, including:

• Increasing output, facilitating the recovery of high fixed costs, and

allowing certain businesses, which would not otherwise have existed,
to operate profitably.3 For instance, railway companies may depend
on temporal price discrimination between peak and off-peak travel
or between advanced bookings or loyalty schemes, to attract enough
business to keep their ser vices running.
• Fostering choice and access in increasing the range of products or
ser vices offered to customers who could not other wise afford them
under uniform pricing (e.g., land or air travel).4
• Increasing social equality5—such as U.S. colleges, which, in awarding
poorer students financial aid, enable greater access to higher educa-
tion than if everyone paid the same (higher) tuition.
• Facilitating dynamic efficiencies. With more revenues and profits,
companies have more to invest in research and development and
enhancing quality.6 This could also manifest itself through competi-
tive investments, as opposed to innovation, such as “increasing
product variety” and “expanding retail outlets.”7
• Intensifying competition among oligopolists, who can now charge
lower prices to competitors’ customers and new customers while
maintaining higher prices on existing (more captive) customers.
This allows suppliers to compete for all customers, not just marginal
Copyright © 2016. Harvard University Press. All rights reserved.

customers.8

Price discrimination can also raise several concerns—in par ticu lar,
when it enables a powerful firm, or group of firms, to:
• Exploit customers. One example is when customers are locked in, and
the seller uses its market power to segment its customer groups to
maximize wealth extraction.9
• Exclude or eliminate a competitor. The European Commission
recognized in AKZO how discrimination “between similarly-placed

customers is expressly prohibited . . . when it places certain firms

at a competitive disadvantage.” The anticompetitive effect of
AKZO’s differential pricing “involved not so much direct injury
to customers but rather a serious impact on the structure of
competition at the level of supply by reason of its exclusionary
effect.”10
• Increase barriers to entry or expansion, by making entry or expan-
sion difficult or excessively costly. The U.K. Competition and Mar-
kets Authority, for example, noted that “where a fi rm uses consumer
data to separate different groups of customers and offers a different
price to each group,” small or new firms “would not have a substan-
tial fixed base of existing customers, and so may be unable to compete
as successfully to target customers through offering them lower
prices.”11
• Deprive smaller rivals or new entrants of attaining sales or distribu-
tion sufficient to achieve efficient scale, thereby raising the smaller
companies’ costs and thwarting their competitiveness.
• Magnify or sustain other exclusionary or predatory abuses.12
• Adversely affect competition downstream. This concern led to the
Robinson-Patman Act, which, the U.S. Supreme Court and Federal
Trade Commission concluded, is based on “one fundamental
principle: to assure, to the extent reasonably practicable, that busi-
nessmen at the same functional level would stand on equal competi-
tive footing so far as price is concerned.”13
In the context of online sales, the U.K. competition agency14 nicely out-
lined several instances where price discrimination can harm competition
and consumers:
• when the practice “is carried out by a monopolist”;
Copyright © 2016. Harvard University Press. All rights reserved.

• “the form of price discrimination is very complex and/or consumers

are not aware of it”;
• “it is costly for firms to implement and so it pushes up costs”; and
• “it leads to a fall in consumers’ trust in online markets.”15

The Added Complexity of Behavioral Discrimination

Behavioral discrimination amplifies many of the welfare effects associated
with price discrimination. Information gathered about our behav ior,

desires, and ability to pay can help firms exploit consumer biases. Sophis-
ticated online sellers can manipulate our environment to increase overall
consumption by price discriminating and by shifting the demand curve to
the right (getting people other wise uninterested in the product to buy).
At times behavioral discrimination may generate positive effects. To il-
lustrate, think about your desire to visit your local dentist. Consumers may
underappreciate a product or ser vice or simply procrastinate. Behavioral
discrimination can encourage people to regularly visit a dentist, use dental
floss, and brush their teeth. In doing so, it increases the individual’s and
society’s welfare. One can think of other examples, where an increase in
demand, even through manipulation, benefits the individual. Many in-
stances, however, involve noncommercial interests and limited financial
incentives.
The unscrupulous nature of behavioral discrimination is revealed as firms
induce consumers to buy more of a bad thing. Cigarette manufacturers, for
example, can exploit biases and imperfect willpower by getting people other-
wise uninterested in smoking addicted.16 Behavioral discrimination will en-
able cigarette manufacturers to sell even more cigarettes. Indeed, one study
found that “retail cigarette advertising increased the likelihood that youth
would initiate smoking; pricing strategies contributed to increases all along
the smoking continuum, from initiation and experimentation to regular
smoking; and cigarette promotions increased the likelihood that youth will
move from experimentation to regular smoking.”17 From a social welfare
perspective, the increase in output is bad for smokers, their families, those
harmed by the secondhand smoke, and anyone who bears the health and
other costs caused by smoking.
Behavioral discrimination also raises concerns over wealth inequality.
Decoy pricing may be used to make unreasonably priced goods appear more
reasonable, or entice consumers to make purchases they otherwise would
Copyright © 2016. Harvard University Press. All rights reserved.

not make. This decreases the disposable income that consumers can use for
retirement, savings, or basic necessities; this effect is proportionately greater
on those with lower incomes, and thus increases wealth inequality.
Further, one may argue that, from a neoclassical economic perspective,
the overall efficiencies would be limited. The argument put forward by
Leeson and Sobel, who doubted the presence of efficiencies in price dis-
crimination, may be extended. Behavioral discrimination does not come
cheap. To harness one’s biases to trigger consumption, the seller would
need to make a significant investment—in tracking consumers; collecting

data on their behav ior; segmenting them into groups; identifying their de-
mand elasticities; reducing market transparency; increasing consumers’
search costs; preventing the resale of the products; limiting the growth of
anti-tracking technology; creating lengthy, tedious privacy statements;
setting the individuals’ privacy defaults as opt-in; and lobbying against
greater privacy protections. So the practice is more likely to accompany
commercial activities, where money is available and profit is lucrative. For
the seller, investment in these measures may be rational and profitable.
When sales increase, customers will ultimately pay these costs and supply
the profits. So even from a neoclassical economic perspective, behavioral
discrimination would have to deliver significant benefits to actually in-
crease total welfare. Once you account the consumer perspective, the social
welfare perspective, and the limited likelihood of total welfare increasing,
behavioral discrimination is likely a toxic combination.
Not surprisingly, some might disagree. Like opponents of the ban on ad-
vertising tobacco products, who argued that tobacco advertising did not in-
crease consumption, they would argue that customers are empowered (more
so than their paternalistic overseers believe). If consumers remain in control
and behavioral discrimination persists, then consumers must ultimately
favor behavioral discrimination to assist their decision making. Thus behav-
ioral discrimination, they would argue, is generally good for you and society.

Social Acceptance
The argument ultimately is over what we prefer. Do many of us prefer be-
havioral discrimination? Would we want it if we end up paying more than
others? Looking at social acceptance, we can identify instances in which
we, as a society, willingly accept price discrimination.
Many of us whose children attend a college or private school know the
Copyright © 2016. Harvard University Press. All rights reserved.

financial sting when tuition is due. We also know that many parents will
likely pay more or less than we do. No parent, to our knowledge, has
stormed the dean’s office protesting the price discrimination. Instead, price
discrimination in some contexts, such as higher education, is more ac-
cepted than in many other contexts, where people view it as unfair. Why is
this? One must examine price discrimination beyond the neoclassical eco-
nomic analysis, which assumes that we are self-interested (greedy) profit
maximizers, to the frontiers of behavioral economics, which views price
discrimination through the prism of fairness and equality.

First, price discrimination may be accepted where its primary pur-

pose is to advance social goals, not simply profit maximization. For
many elite U.S. universities, the full tuition does not cover the universi-
ty’s marginal cost of supplying a year of education to an undergraduate
student.18 The same applies for some private high schools, like Phillips
Exeter Academy. Every student, in effect, receives a discount—a benefit
from the university’s or high school’s endowment and revenue stream.
Parents and students accept price discrimination when it promotes
greater social and moral goals, such as promoting opportunities and up-
ward mobility. Price discrimination provides poor and middle-class
students a broader range of schools from which to choose, including
highly selective universities, where they can network and increase their
earnings potential.19 (Unfortunately, many bright but poor students have
not applied to selective U.S. universities, even when doing so would cost
them less.20)
Second, price discrimination may be accepted when it improves the
overall product. Without price discrimination, prestigious universities and
high schools could attract only those students whose parents are willing
and able to pay the full tuition and room and board. (No fi nancial aid
would be offered.) But the belief is that the academic environment would
suffer (as well as the school’s reputation over time). Thus price discrimina-
tion enriches the product itself, namely the educational and academic en-
vironment, by increasing diversity.
Finally, price discrimination is accepted when it is transparent and fairly
applied. Many colleges provide financial assistance calculators that enable
parents and students to assess the net price. The belief is that, apart from
issues of merit or athletic scholarships, families with similar financial situ-
ations whose children attend the same school will contribute roughly the
same amount.
Copyright © 2016. Harvard University Press. All rights reserved.

If these factors no longer applied, and universities simply price discrim-

inated to maximize revenues and increase compensation for the university
administrators, coaches, and perhaps even professors, then trust would
break down and parents would likely rebel.
The introduction of a profit motive changes the framework of assessment
and leads many people to perceive price discrimination as unfair. This at-
titude intensifies where behavioral discrimination is exposed and people
perceive themselves as victims of manipulation.

(Lack of) Fairness and Equality

In contrast to the above, other markets dominated by for-profit firms have
less exceptional characteristics. Price and behavioral discrimination will
not increase the product’s or ser vice’s quality or foster any greater societal
objectives. Instead, companies price discriminate to capture as much of the
consumers’ wealth as possible.
In this context price discrimination is generally perceived as unfair.
Over the past thirty years, behavioral economists in the lab and the field
have tested the assumption that we are greedy, self-interested profit maxi-
mizers. Some people are. But the psychological and experimental economic
data show that most people care about treating others, and being treated,
fairly.21 Some economists are agnostic on price discrimination, or believe
that in certain instances it may be welfare-enhancing. But 91 percent of in-
dividuals in one survey thought charging higher prices to those who were
more dependent on the product was offensive.22
A collateral consequence of price discrimination is a loss of trust between
companies and their customers. In behavioral experiments, consumers have
been found to be less trusting of firms engaging in price discrimination, and
less willing to purchase from them.23 Even where the study’s participants
personally received a better price than others, many still perceived the
retailer to be behaving unfairly, were less inclined to purchase from that
retailer again, and were less willing to recommend the retailer to a friend.24
Many consumers, in a couple of studies, objected to such price discrim-
ination as “ethically wrong,” including:

• 76 percent who agreed “it would bother me to learn that other people
pay less than I do for the same products”;
• 64 percent who agreed “it would bother me to learn that other people
Copyright © 2016. Harvard University Press. All rights reserved.

get better discount coupons than I do for the same products”;

• 66 percent who disagreed that “it’s OK with me if the supermarket I
shop at keeps detailed records of my buying behavior”;
• 87 percent who disagreed that “it’s OK if an online store I use
charges people different prices for the same products during the
same hour”; and
• 72 percent who disagreed with the idea that “if a store I shop at
frequently charges me lower prices than it charges other people
because it wants to keep me as a customer more than it wants to keep
them, that’s OK.”25

Most of the surveyed American adults believed it was illegal for “an online
store to charge different people different prices at the same time of day.”26
It isn’t.
So why does price discrimination violate so many people’s notions of
fairness? Why do they think it is (or should be) illegal? One reason is its
perceived exploitation, namely taking advantage of people who don’t have
a viable outside option. The company does not provide any extra ser vice; it
simply exercises its market power.

When Actual Discrimination and Behavioral Discrimination Blur

One belief is that algorithms are free of human biases (to the extent
their code does not reflect such biases) and can process data objectively.
Unlike humans, pricing algorithms won’t discriminate based on one’s
gender, skin color, national origin, age, disabilities, sexual orientation,
or religion.
As Chapter 10 discusses, pricing algorithms won’t be able in the near
future to identify each customer’s unique reservation price across every
possible situation. Instead, consumers are lumped into groups. If you live
in a certain neighborhood, are a certain age, went to a particular univer-
sity, or are a member of a particular religion, then the pricing algorithm
may lump you in a particular category. In other words, the seller makes an
educated guess that certain groups are more likely to buy the product and
are less sensitive to its price than other groups, who even if tempted to pur-
chase, will require a lower price.
Some groups will be exploited. Thus, the road to perfect behavioral dis-
crimination increases the risk that computer algorithms may categorize
consumers on an unalterable trait, such as one’s skin color. The White
House report noted how Big Data “may facilitate discrimination against
Copyright © 2016. Harvard University Press. All rights reserved.

protected groups, and when prices are not transparent, differential pricing
could be conducive to fraud or scams that take advantage of unwary con-
sumers.”27 Data brokers, as the FTC reported in 2014, were already catego-
rizing and segmenting consumers.28 “While some of these segments seem
innocuous,” noted the FTC, “others rely on characteristics, such as eth-
nicity, income level, and education level, which seem more sensitive and
may be disconcerting.”29 Some segments, for example, “primarily focus on
minority communities with lower incomes, such as ‘Urban Scramble’ and
‘Mobile Mixers,’ both of which include a high concentration of Latino and

African-American consumers with low incomes.”30 Other potentially sensi-

tive categories “highlight a consumer’s age such as ‘Rural Everlasting,’ which
includes single men and women over the age of 66 with ‘low educational at-
tainment and low net worths,’ while ‘Married Sophisticates’ includes thirty-
something couples in the ‘upper-middle class . . . with no children.’ ”31 Data
brokers also categorize us into health-related topics or conditions, such as
“Expectant Parent,” “Diabetes Interest,” and “Cholesterol Focus.”32
Businesses in the United States and the European Union cannot use
one’s race, color, religion, or certain other categories to make credit, insur-
ance, or employment decisions.33 But firms can circumvent these antidis-
crimination laws through the use of algorithms that automatically develop
and refine categories in which people of certain race, marital status, age,
sexual orientation, or religion are lumped together.
Such discrimination by algorithms is already happening. One 2015 study
found high school students in communities with a high density of Asian resi-
dents were 1.8 times likelier to be charged higher prices for Princeton Re-
view’s online college test-prep services, regardless of income.34 As the study
found, the gap “remains even for Asians in lower income neighborhoods”—
citing as one example a ZIP code in Flushing, New York, where Asians
make up 70.5 percent of the population, with a low median household in-
come, $41,884, “yet The Princeton Review customers there are quoted the
highest price.”35
A Carnegie Mellon study examined whether visiting websites related to
substance abuse will likely impact the ads you’ll see later.36 Google targets
users with behavioral ads. Google’s Ad Settings “display[s] inferences
Google has made about a user’s demographics and interests based on his
browsing behav ior.”37 Google also provides a tool that “helps you control
the ads you see on Google ser vices and on websites that partner with
Google.”38 So the researchers ran an experiment where two groups initially
Copyright © 2016. Harvard University Press. All rights reserved.

had the same Google Ad Settings. One group visited websites on substance
abuse; the control group simply waited. Then both groups collected the ads
served by Google on a news website.
Although Google’s Ad Settings remained the same for both groups, the
group that visited the substance abuse websites received many more ads for
Watershed Rehab. The experiment found a mismatch between Google’s Ad
Settings and the ads targeting consumers: “information about visits to
these websites is indeed being used to serve ads, but the Ad Settings page
does not reflect this use in this case.”39

A related experiment found evidence suggestive of gender discrimina-

tion. In this experiment, the group was divided according to gender on
Google’s Ad Settings page.40 With only that difference, the female and male
groups both visited web pages associated with employment. Here the ads
differed by the agent’s sex: Google showed the male group “ads from a cer-
tain career coaching agency that promised large salaries more frequently”
than the female group.41 The experiments observed the discrimination, but
it was unclear who was to blame: “whether Google, the advertiser, or com-
plex interactions among them and others caused the discrimination.” Even
if they could find out who was responsible, “the discrimination might have
resulted unintentionally from algorithms optimizing click-through rates or
other metrics free of bigotry.”42
Here we see the lack of transparency come to the fore. In the first experi-
ment, the users seemingly had the same profi le, but they received different
types of ads based on the websites they visited. If they went to Google’s Ad
Settings, they would not see any reference to substance abuse (nor any
ability to correct this misimpression if it did show up). One can imagine
other implications. Health insurance firms could discriminate in their
posted rate, based on the websites one visits, the items one purchases (in-
cluding cigarettes or alcohol as a gift), or the activities one posts on a social
network. Users cannot go to Ad Settings or other advertising profile sites
to correct any misimpression, as the sites may not reflect this information.
Say you type “CEO” on Google’s search engine. You click the Images tab.
In early 2016, you would find mostly white men. Which woman is listed
among the top results? Not Mary Barra of General Motors. Nor Virginia
Rometty of IBM. Nor Indra K. Nooyi of PepsiCo, Inc. The first female
image is none other than CEO Barbie. One 2013 study identified discrimi-
nation in Google image search results for some, but not all, occupations.
The study compared the percentages of women who appeared in Google
Copyright © 2016. Harvard University Press. All rights reserved.

searches for different occupations with U.S. statistics showing how many
women actually worked in that field. Among the professions with signifi-
cant gaps were CEOs (11 percent of the images in the Google image search
result were women, compared with 27 percent of U.S. CEOs who are
women), authors (25 percent of images for this search result were women,
compared with 56 percent of U.S. authors who are women), and telemar-
keters (64 percent of the images were women compared with 50 percent in
the workforce).43
Another study examined advertisements for the web page of a high-
profile, historically black fraternity, Omega Psi Phi, which celebrated its

one hundredth birthday. Among the algorithm-generated ads on the web-

site were ads for low-quality, highly criticized credit cards and ads that sug-
gest the audience member had an arrest record.44
What remains unclear is why a black fraternity website attracts ads about
one’s criminal history, and why men get career coaching ads for boosting
their salary, but not women. One possibility is that the advertiser requested
the placement. Alternatively, the advertising algorithm predicted that the
questionable ad would likelier be clicked than other ads. Law professor
Frank Pasquale criticized the secrecy of these algorithms and the lack of
transparency.45
The worrying thing is that we may not even know that we are being dis-
criminated against. Under the old competitive paradigm, one knew if one
was discriminated against if one was denied access (e.g., restaurants for
“whites only”) or was charged a higher price based on this single variable.
Under the new paradigm, users may be unable to detect the small but sta-
tistically significant change in targeted advertisements (or advertised rates).
In a Carnegie Mellon study, all conditions were held equal except one
(namely gender or specific sites visited). In this controlled experiment, one
can measure the direct impact this one change had. But in our day-to-day
reality, there are often too many confounding variables and often no handy
benchmark.46 A woman searching the web will not necessarily know to
what extent the ads she is seeing differ from those her male counterparts
see. She will unlikely ask her male colleagues to report the ads they see and
compare the differences. Even if users could detect that they were being
targeted with rehab ads (or not being offered advertised opportunities for
greater salaries), they may not know why they were being discriminated
against. It might be the websites they visited, a magazine subscription they
had, an e-mail they wrote, their race, or some incident. The bottom line is
that the march to perfect behavioral discrimination will entail lumping us
Copyright © 2016. Harvard University Press. All rights reserved.

into categories. At times that categorization process and behavioral tar-

geting will promote actual discrimination.

Current Enforcement Tools

Many people believe price discrimination by online and brick-and-mortar
retailers is illegal. But it typically isn’t. Even when potentially illegal, price
discrimination is low on the competition agencies’ enforcement agenda.
The primary weapon in the United States against price discrimination,
the Robinson-Patman Act, does not even apply to our scenario, namely

behavioral discrimination directed at end users.47 Instead, the Act pro-

hibits discriminating against smaller retailers or distributors competing
against larger competitors.48 Even for small businesses, an American Bar
Association article observes, the Robinson-Patman Act “is still alive, but
with declining relevance.”49 Although the United States could criminally
prosecute under the Robinson-Patman Act, it hasn’t for decades. The U.S.
competition agencies rarely enforce the Act through civil actions. Private
plaintiffs bring almost all the cases, and they often lose, as the Supreme
Court has narrowed the Act’s scope by, among other things, expanding the
defenses.50 Thus, generally under U.S. federal law, online or brick-and-
mortar retailers “are under no obligation to disclose their pricing struc-
ture to consumers, and in the absence of some duty to disclose, retailers
may charge different prices for the same goods or ser vice to different
groups of consumers without disclosing that fact.”51
In the European Union, price discrimination can be addressed under
Article 102(c) TFEU. The article stipulates that an abuse of a dominant
position may consist of the application of “dissimilar conditions to equiva-
lent transactions with other trading parties, thereby placing them at a com-
petitive disadvantage.” The provision targets discrimination that places
other trading parties in an unfavorable position. This type of competitive
harm is often referred to as a “second line injury” as it concerns competi-
tion between undertakings at the downstream level. For instance, in Post
Danmark A/S v. Konkurrencerådet,52 the court clarified that “the fact that
the practice of a dominant undertaking may, like the pricing policy . . . [be]
described as ‘price discrimination’, . . . , cannot of itself suggest that there
exists an exclusionary abuse.”53
Evidently, other laws may prohibit discrimination on a broader basis—
such as discrimination based on race, age, and so on. As the UK competi-
tion authority noted, “A range of other legislation could also apply in this
Copyright © 2016. Harvard University Press. All rights reserved.

area, for example the Equality Act 2010, which prohibits discrimination in
the supply of goods, services or facilities based on ‘protected characteristics’
of age, disability, gender reassignment, pregnancy and maternity, race, reli-
gion or belief, sex or sexual orientation, could apply in the case of businesses
which may discriminate, for example, through the analysis of data to target
individuals based on racially discriminatory algorithms.”54 Other instru-
ments include the EU Unfair Commercial Practices Directive55 and EU Data
Protection Directive.56

Reﬂections
In 2015, we asked lawyers, judges, and economists about their approach to
price and behavioral discrimination. We raised these issues to different
groups as part of training sessions on competition law. Competition lawyers
and economists dominated some groups; in others the participants had
limited economic or competition law background. We asked each group for
their reaction if they discovered that another online customer had pur-
chased goods for a lower price through intended price discrimination.
Those without an economic background felt it was unfair, so much so that
they would stop, if possible, using the seller in question. Interestingly, those
with an economic background were less susceptible to feelings of unfair-
ness. They felt that this may be acceptable when one wishes to facilitate ac-
cess for lower-income consumers, create positive externalities, and increase
and optimize production. When faced with questions about behavioral dis-
crimination, participants were more united in their approach. Some felt ma-
nipulated, others exposed. Many indicated lack of belief as to the ease with
which their actions may be affected by simple “tricks of the trade.”
As companies’ data collection and analytics improve, so too will their
ability to discriminate. Targeted pricing may, in particular, be sustainable
where a market is stable and exhibits barriers to entry or expansion, lim-
ited outside options, heterogeneous or branded goods, imperfect informa-
tion flows, or the ability to distort or inhibit information exchange. It
may also be sustained in markets that attract loyal customers or where
companies develop and customize distinguishable products for par tic-
u lar purchasers.57
Even if companies can discriminate, this does not necessarily mean
that they will. Behavioral discrimination—given its manipulation of our
emotions and our expectation of a fair competitive price, which everyone
Copyright © 2016. Harvard University Press. All rights reserved.

pays—will likely be condemned. Behavioral discrimination raises privacy

concerns.58 We do not necessarily think we are being monitored when we
are online. Even if we do, we may underestimate the amount of data being
collected on us and how that data is being used to target us. Behavioral
discrimination, besides the heightened privacy concerns, can disturb our
trust in the marketplace. People still expect a competitive price set by
market forces, whereas perfect price discrimination eliminates the
competitive benchmark. The price you pay differs from the price I pay,
based simply on our perceived willingness to pay. The new paradigm of

behavioral discrimination affects not only our pocketbook but our social
environment, trust in firms and the marketplace, personal autonomy, and
privacy.59 Fearing the adverse reputational impact, some companies will
initially refrain from using all the data at their disposal to maximize wealth
extraction.
A trend may, however, emerge. As pricing norms change, fewer people
over the years may oppose price and behavioral discrimination. Price dis-
crimination eventually may be accepted as the new normal. Just as we have
accepted (or become resigned to) the quality degradation of air travel, and
the rise of airline fees—from luggage to printing boarding passes—our
future norms may well include online segmentation and price discrimina-
tion. Many forms of discrimination, involving different pricing on mobile
platforms and PCs, personalized search results, personalized coupons, and
price steering,60 are already appearing in the online marketplace.
Looking ahead, perhaps younger generations may more easily accept
price (and even behavioral) discrimination, particularly as they have been
exposed to it from a younger age, and will engage in online purchasing at
an increasing rate. The savvier may use countermeasures to extract better
deals and manipulate attempts at price discrimination and steering.
But given the asymmetries in information and power between the data
collectors and us, countermeasures will unlikely stop the march toward
near-perfect behavioral discrimination. Absent legal intervention, behav-
ioral discrimination will likely become in many retail industries the new
norm.
Copyright © 2016. Harvard University Press. All rights reserved.

The Comparison Intermediaries

H AVING CONSIDERED the possible use of price discrimination in an algorithm-

driven environment, we explore in this chapter the role played by price
comparison websites (PCWs) and metasearch engines. These comparison
intermediaries support market transparency, and as such create an environ-
ment which can reduce the capacity for price discrimination and help safe-
guard our welfare.
Indeed, these comparison and search platforms have become an integral
part of our web environment: Google, Bing, Yahoo!, Amazon, Expedia,
Booking, Alibaba, Ebay, Lastminute, Trivago, PriceRunner, Go Compare,
Money Supermarket, mySupermarket, Comparethemarket, and uSwitch
are only a handful of the many platforms that provide users with compara-
tive data and access to sellers and ser vice providers.
Often these intermediaries operate in a multisided market. On the mar-
ket’s consumer side, the websites predominantly offer their ser vices for
free: customers enter their search queries and are provided with relevant
information and links without charge. On the provider side, however, they
often generate an income stream. Search platforms would predominantly
Copyright © 2016. Harvard University Press. All rights reserved.

rely on advertising income. Price comparison websites would often charge

a fee or commission for a referral or sale they facilitate.
In Chapter 1 we introduced the prominent role played by intermedi-
aries and their contribution to the dynamic online environment—buyers
access information about product offerings, variety, available sellers,
price options, and reviews in a wide range of markets—including insur-
ance, travel, loans, mortgages, credit cards, holidays, and consumer
goods. In this chapter we note some of the benefits offered by these
131

intermediaries. We then explore instances in which they may raise com-

petitive concerns—and reduce our welfare while retaining the façade of
competition.

The Beneﬁts
Comparison intermediaries can help level the playing field and intensify
competitive pressure.
For comparison intermediaries to yield the desired benefits, customers
must be aware of the availability of outside options, have low switching
costs, and have the incentive to search (that is, the alternative price and
quality must be sufficiently attractive to justify the costs and risks of
searching and switching).
When this occurs, comparison intermediates can facilitate a transparent
market environment in which customers obtain comparable information
on the available products and bargains. By collating and aggregating quan-
titative and qualitative data about suppliers, as well as price and product
characteristics, the search and price comparison platforms can reduce the
asymmetry of information and improve information flows. This can make
it harder for suppliers to take advantage of ill-informed customers who are
subject to high information costs.1 As such, the comparison intermediaries
could weaken the power of sellers to segment the market and price dis-
criminate. They can help reduce the market power of sellers, to the extent
that power was the result of high information costs.2
These platforms can support a competitive dynamic, where sellers face
price competition and invest in ser vices, quality, and innovation. Overall,
the flow of information generates substantial allocation efficiencies, as
buyers can more cheaply locate sellers that better meet their needs. Indeed,
“[t]he Internet brought about price transparency across the market, en-
Copyright © 2016. Harvard University Press. All rights reserved.

abling consumers to identify the best deal, i.e. the lowest price for any given
hotel room, at very low search costs.”3
Comparison intermediaries may also facilitate suppliers’ entry and ex-
pansion by providing economies of scale and efficiencies in distribution,
marketing, and promotion. They can reduce the associated risks and costs
for new entrants and enable access to potential customers, economies of
scale in online advertising, payment facilities, guarantees, and ease of trans-
action.4 These supplier-focused efficiencies can support a more competitive
environment and lower prices—again, to our benefit.

Moreover, comparison intermediaries can widen the realm of the market

at both national and international levels. They can link international sellers
with buyers—widening the reach of suppliers and widening the customer
base. They may offer interfaces in a range of languages, further facilitating
access and information flow.
Overall, comparison and search platforms have the potential to support a
competitive environment in which users easily identify, compare, and access
various offerings of products and ser vices. We use the word “potential”
because sometimes platforms might reduce competition, as we explore next.

Network Effects and Market Power

Let us now consider how network effects may give rise to market power and
create bottlenecks in the online environment. Network effects and market
power may influence the comparison intermediates’ incentives and practice.
Powerful platforms can distort the information they present us to improve
their own profitability. To explain how, we first explain the operation of net-
work effects. Then we illustrate the way in which market power could distort
competition and search results online.

Network Effects
To understand how online platforms can exercise market power, we briefly
outline several network effects involving online multisided platforms (such
as Google, Bing, price comparison websites, and Facebook).5

Traditional Network Effects. Traditional network effects are observable in

social network platforms, where bigger is better. Direct network effects
arise when a consumer’s utility from a product increases as others use the
Copyright © 2016. Harvard University Press. All rights reserved.

product.6 One example is Facebook. As more people use the social network,
the more people with whom one can interact, the easier it is to connect with
other people, and the greater one’s utility in using Facebook. The value of
the network increases with its growth. As the big platforms get bigger, the
entry barriers to obtaining the necessary scale to meaningfully compete
also increase.

Trial and Error. This network effect is linked to the scale achieved by trial
and error, or learning by doing. Such an effect is relevant to machine

learning. For example, for search engines an increase in the number of

searches increases the search engine’s likelihood of identifying relevant re-
sults. In other words, the more consumers who use the search engine and
the more searches they run, the more trials the search engine has in pre-
dicting consumer preferences, the more feedback the search engine re-
ceives of any errors, and the quicker the search engine can respond by
recalibrating its offerings. Naturally, the quality improvement attracts ad-
ditional consumers to that search engine compared to competitor sites. In
effect, the more users, the larger (and more heterogeneous) the sample size,
and the better the search engine can identify relevant responses for both
popular and less frequent queries (“tail” queries).

Scope of Data. This network effect involves the scope of data on the user.
Search results, for example, can improve from the variety of personal data
on users. If people use, besides the search engine, other ser vices offered by
the company (such as e-mail, web-browser, texting, mapping, purchasing,
etc.), the company, in collecting the variety of personal data, can develop
user profiles to better predict users’ tastes and interests, and better target
users with more relevant organic and sponsored search results. This feed-
back loop adds another dimension: it is now no longer the trial and error,
learning by doing from earlier searches, but also learning of users’ tastes
and preferences from the variety of personal data it collects across its plat-
form, which enables the personalization of search results and the targeting
of users with specific sponsored ads that they will likely click.

Spillovers and Snowball Effect. As discussed previously, intermediaries

often operate in a multisided market. Spillovers and the snowball effect
concern the way network effects on the “ free” (consumer) side can spill
over to the “paid” (provider) side, and each can reinforce the other. The in-
Copyright © 2016. Harvard University Press. All rights reserved.

flow of many users with heterogeneous search inquiries, for example, will
attract a greater variety of advertisers to a platform. The search platform
can use the inflow of personal data to better target consumers with specific
targeted advertising across its platform of free ser vices (such as sponsored
search results, ads in e-mails, and displaying ads in videos) in the moments
that matter for a purchasing decision. In targeting users with more relevant
ads (or ads that users will likelier click), the search engine increases its ad-
vertising revenue and profits. Moreover, the search engine can target users

with these personalized ads across media (such as personal computers,

smartphones, tablets, and, soon, household appliances) and across ser vices
(texts, maps, videos, etc.). This too increases the likelihood of consumers
clicking on a relevant sponsored ad (which generates revenue on a cost-per-
click basis) or seeing a display ad (which generates revenue on a cost-
per-impression basis). As more users are drawn to the platform, and as the
company amasses a greater variety of data to effectively target consumers
with relevant online ads, the broad platform can reduce the advertisers’
fi xed costs of managing multiple ad campaigns. As more people use the
search engine, more advertisers will use the platform, the more relevant
and targeted the advertisements, the likelier that users will click on the ads,
and the more profits the search engine has to expand its range of free ser-
vices and to ensure that its ser vice remains the default search engine on
various portals to the Internet (for example, developing one’s own browser
and paying other browsers to have one’s search engine be the default).7

Market Power
The above network effects illustrate the way in which online comparison
intermediaries may acquire market power. For example, more users gen-
erate more search queries, which generate more trial and error, which yields
better search results, which attracts more users and advertisers to the
search platform, which enables better profiling of users and greater likeli-
hood of users clicking on the ads, which generates more advertising rev-
enue to enable the search engine to offer even more free ser vices, which
enables consumers to spend more time on the company’s platform, which
allows it “to gather even more valuable data about consumer behaviour,
and to further improve ser vices, for (new) consumers as well as advertisers
(on both sides of the market).”8 The larger platforms gather more data on
Copyright © 2016. Harvard University Press. All rights reserved.

their users and have greater opportunities to experiment and learn by

doing. They can continually improve and refine their ser vices as more users
rely on them. The big web-aggregators become bigger, as users increas-
ingly rely on the popu lar websites. The web-aggregators in turn become
significant players—unavoidable partners, even—occupying a strategic po-
sition in the distribution channel.
As more consumers rely (and trust) an intermediary to deliver the best
results (whether relevant results to a search query or array of goods and

ser vices), they become less interested in multi-homing—that is, in checking

the availability of products and prices elsewhere. For example, most people
conduct their searches on a single search engine. Few run the same search on
multiple search engines, such as Bing, Google, Yahoo!, or DuckDuckGo. As
fewer consumers multi-home by running the same search on multiple search
engines or price comparison websites, the more power the platform has.
As more customers rely on the intermediary, it becomes more attractive
to sellers, who will find it important that their products be included on the
platform. Sellers know that their products’ and ser vices’ inclusion on a plat-
form’s search results may be crucial for their visibility.
As these “information and referral junctions” become a crucial gatekeeper
between suppliers and consumers, the platform’s bargaining power increases.9
The increased reliance on a few powerful intermediaries gives rise to possible
changes in market dynamics. These gatekeepers control many significant ac-
cess points and, as a result, have the power to distort competition, sometimes
unintentionally. So, comparison intermediaries could sometimes, under
certain conditions, pave the way for higher prices, lower quality, and a
reduction in consumer welfare. In what follows, we illustrate how plat-
forms may sometimes fail to deliver on their competitive promise.

Possible Distortions
When a multisided platform offers a product or ser vice for free, the pri-
mary dimension of competition is typically quality. Competition is there-
fore likely to stimulate investment in quality, such as more relevant search
results. Yet, the platform operator has competing incentives. It invests in
quality on the free side to attract users. But its revenues and profits come
from the platform’s other side, such as commissions or advertising. So its
incentive to optimize quality may be distorted.
Copyright © 2016. Harvard University Press. All rights reserved.

As the platforms’ market power increases, transparency in these online

markets will not be controlled by society or even by market forces, but in-
creasingly by the platform operators’ incentives. In such instances, the on-
line platform may intentionally degrade quality on the free side below levels
that consumers prefer, if doing so increases its profitability (or market power).
Some platforms, for example, may allow for preferential placement based
on the level of payment or commission they receive from sellers. For in-
stance, pay-for-placement fees allow a platform to charge higher rates to
sellers for the right to be positioned at the top of the list on the default page

result. Such positioning may distort competition when the user is unaware
of the preferential positioning and assumes that the top results are the best
(or most relevant) ones objectively picked by the websites’ algorithms.
One example of such manipulation of results is in online hotel bookings.
The factors which could influence the default ordering of hotels on hotel
booking intermediaries includes: “customer ratings and complaints”; “if
hotels are willing to pay larger commissions”; “photo quality”; and “if a
hotel is quicker to turn shoppers into buyers.”10 The methods that hotel
booking intermediaries use to tailor search results have come under criti-
cism by some hotels. The American Hotel and Lodging Association told the
Wall Street Journal, “ ‘Biased or misleading search results from these sites
or via web searches can be highly problematic, particularly on those
booking websites that purport to be helping consumers comparison shop
based off of less than objective information.’ ”11
This old trick also occurred with the U.S. airlines’ computerized reservation
systems. The United States in 1984 was concerned that several airlines were
taking advantage of their control of a computerized reservation system to give
themselves a competitive edge. The government discovered, for example, “that
certain system owners had written the computer program algorithms in such
a manner that their [computer reservation system] screens would display all of
their own flights before listing those of competitors, even though other flights
might more closely match the agents’ specifications. Because travel agents
work under heavy time pressures, they tend to recommend the flights listed
first.”12 To eliminate such abuses, the government required that the algorithms
generate results based on “neutral” characteristics.13
The New York Times reported that Amazon in 1999, unbeknownst to its
customers, offered an “E-merchandising” program, where book publishers,
in exchange for paying advertising fees of as much as $12,500, received
“featured treatment of titles in categories that range from ‘What We’re
Copyright © 2016. Harvard University Press. All rights reserved.

Reading’ to ‘Destined for Greatness.’ ”14 Amazon did not disclose the adver-
tising fees to its customers.15
Some web aggregators may add a charge for a referral to a provider’s site.
While these websites may benefit from a public perception that they will
provide the best rate available online—like many other web aggregators—
they levy a charge on the user. For instance, a study conducted in Germany
revealed how some comparison platforms charge higher prices for ser vices
and goods, than would other wise be available on the provider’s site.16 The
study questioned the true benefit provided by these ser vices.

In the European Union, the consumer harms from distorted informa-

tion have led, on occasion, to public condemnation. For instance, in early
2015 the U.K. Energy and Climate Change Select Committee challenged
practices that resulted in distorted information on the web-aggregators.
Among other things, platforms were accused of “not showing the cheapest
tariffs by default if it meant they wouldn’t earn a commission.”17 Following
this criticism, the PCWs have since ensured that the default search setting
will include the full range of tariffs available, regardless of whether or not
a commission is charged upstream.18
Search engines may bias their results to favor paid advertising. For a
number of search engine operators, for example, the order of the results
that appear on a search results screen will depend on how much an ad-
vertiser pays. Most search engines provide users with “sponsored” re-
sults and “organic” results, which are produced by the search engine’s
algorithms.19 For the paid sponsored search results, most advertisers pay
the search engine on a cost-per-click basis, whereby the advertiser pays
the search engine only when a user clicks on its sponsored ad.20 When
Microsoft and Yahoo! collaborated on their search engines, one concern
two economists for the European Commission noted was whether the
search engine “may alter the ranking of the organic search results such
that, from the user’s perspective, firms offering competing products to
the sponsored links are given a less-than-optimal ranking on the organic
side.”21 The Commission ultimately found that Microsoft and Yahoo were
unlikely to engage in such behavior.
The intermediaries may have greater power to extract greater rents
from suppliers of goods and ser vices in the form of higher commissions,
fees for preferential placement, or advertising. Perhaps not surprisingly,
online platforms, as profit making entities, share incentives similar to
other market players and may also benefit from personalization and be-
Copyright © 2016. Harvard University Press. All rights reserved.

havioral discrimination. In fact, some of them use similar tactics to

gather information about users. They may amend their offering based on
one’s location or shopping history and engage in a range of tactics, from
the use of decoys and price steering to drip pricing. So the promise of
transparency and undistorted information may give way to profit motive,
once the comparison intermediates gain relative power.
In addition, some comparison platforms or meta-search engines, which
also provide related ser vices downstream, may favor their own ser vices
over those of competitors and display them more prominently in their

search results. Notable here is the European Commission’s investigation

into Google’s alleged favoring of its own comparison shopping product
“Google Shopping” and its preferential positioning on Google general search
results pages. In April 2015, the European Commission accused Google of
systematic favoring of its own comparison-shopping ser vices in its general
search results pages over more relevant competitor sites. The commission
was predominantly concerned with the leveraging of Google’s market
power in the online general search engine market to create an advantage
in the related market of comparison-shopping ser vices. This leveraging, the
commission argued, harmed rival comparison-shopping ser vices, con-
sumers, and innovation.22 The top European Commission competition of-
ficial noted: “when a consumer enters a shopping-related query in Google’s
search engine, Google’s comparison shopping product is systematically
displayed prominently at the top of the search results. This display is irre-
spective of whether it is the most relevant response to the query. Thus,
Google’s commercial product is not subject to the same algorithms as other
comparison shopping ser vices . . . with the result that consumers may not
necessarily see the most relevant results in response to their queries, and
Google’s competitors may not get the commercial opportunities that their
innovations deserve.”23
Overall, an online platform, even when competition is a click away, can
reduce quality, when: (1) the platform has the ability and economic incen-
tive to degrade quality; (2) consumers cannot accurately assess the quality
degradation; and (3) it is difficult or costly for others to convey to con-
sumers the products’ or ser vices’ inherent quality differences or to prompt
them to switch.24

The Use of Wide Parity Clauses and the Agency Model

So far we noted how network effects and market power may distort how
information is displayed by comparison intermediaries. We now explore
how contractual arrangements may sometimes distort competition, not by
manipulating the results, but by dampening the intensity of competition.
We focus on the use of wide parity—also known as wide Most Favored
Nation (MFN) clauses—by some intermediaries. These clauses have attracted
attention especially when combined with an agency distribution model.
Let us consider how these wide MFN clauses with the agency model
could maintain a perception of competitiveness while in practice limiting

competition. We’ll use Apple and the book publishers as our example. In
an agency distribution model, the book publishers set the final price of the e-
books sold on Apple, Amazon, and any other online platform. The platform
receives a commission for each e-book sale made under an agreed revenue-
sharing clause. Accordingly, Apple, Amazon, and the other platforms do not
purchase the e-books from the publishers. Instead, they act as agents, selling
the e-books on the publishers’ behalf.25 To ensure that Amazon or any other
bookseller does not offer popular e-books at a lower price, Apple also insisted
on a wide MFN clause. As the court noted in the Apple antitrust case, “an
MFN Clause is a contractual provision that requires one party to give the
other the best terms that it makes available to any competitor”; in other words
“the MFN would require the publisher to offer any e-book in Apple’s iBook-
store for no more than what the same e-book was offered elsewhere, such as
from Amazon.”26 With a wide MFN clause and agency model, Apple, the
court found, was protected from retail price competition.27
The combination of a wide MFN clause and agency model has become
common in online commerce. They are designed to resolve the hold-up
problem, often manifested in vertical relationships, by removing the risk
of the supplier and other sellers free-riding on the PCWs’ investment in
demand-enhancing features.28 By addressing possible horizontal and ver-
tical externalities, they ensure the continuous investment by the platform
in demand-enhancing features.
Yet the use of wide MFN clauses has come under increased scrutiny in
recent years due to their potential anticompetitive effects. One concern has
been that wide MFNs, when combined with an agency model, may incen-
tivize a powerful intermediary to increase the fees it charges to upstream
sellers. To see why, suppose the PCW has market power. It wants to charge
sellers higher fees for their products to be listed on its website. The fear is
that a seller agrees, but starts shifting sales to rival PCWs or an entrant.
Copyright © 2016. Harvard University Press. All rights reserved.

With a wide MFN clause, each seller contractually agrees not to charge a
lower price on any other PCWs, even those that charge modest fees. So the
dominant platform can increase its fees to the sellers. The sellers can
choose to absorb the price increase (which lowers their profit margins) or
raise the price of their goods. If the latter, you will pay that higher price on
whatever PCW you visit. Why? Under the wide MFN, price parity with
other platforms is guaranteed.29
In its review of the private auto insurance sector, the U.K. Competition
and Markets Authority commented on the combined effect of agency
pricing and wide MFNs:

Generally, we expected that higher commission fees would lead to higher

policy premiums because there was likely to be some pass-through of
costs to premiums. . . . [I]rrespective of the rate of pass-through, the PCW
with the wide MFN could continue to increase commission fees until the
price of the policy was too high from the point of view of the PCW.
Premiums across the market might increase up to the point at which
[private motor insurance] providers exercised their “outside option”,
which would be to withdraw from listing on the PCW with the wide
MFN and to seek to attract customers from other sources.30

For us, the wide MFN, when combined with an agency model, might
seem like a benefit. We can continue using the dominant PCW, knowing
that it always has (or matches) the best price. The harm is less salient. We
do not see, besides their price effects, how the clause may undermine new
PCWs from entering into the market. Suppose a new PCW seeks to offer
consumers lower prices by charging the sellers a lower fee. The dominant
platform will invoke its price parity provision with the sellers and require
them to match the lower price on its website. This would effectively reduce
the sellers’ profit margins on the dominant platform. Each additional sale
on the dominant PCW means less money for the sellers. Since the sellers
now make even less money per sale, they would not eagerly embrace the
entrant. Thus, the wide MFN clause and agency model collectively change
the sellers’ incentives; specifically, they would not discount on the entrant’s
PCW—no matter how great a deal the entrant offers—since they would
have to offer that discounted price on the other PCWs. As the price is set
under an agency model and subjected to wide MFN provisions, lower prices
cannot be charged on any platforms, including the entrant’s. 31 So why
enter this market? The entrant cannot compete against the incumbent
PCWs by offering lower prices or better terms. Even if the entrant somehow
Copyright © 2016. Harvard University Press. All rights reserved.

could compete, the dominant PCW’s algorithms would immediately de-

tect and match the price. As we continue using the dominant PCW, the
sellers, whose margins are squeezed, view the entrant as a pariah.
For example, a BBC report raised concerns on the rise in energy costs in
the U.K. The report considered how the hidden costs of commissions levied
by PCWs perhaps contributed to the price increase:
[T]here’s another cost in the bill. It’s hidden, it’s kept confidential, and
yet it’s for a part of the industry that appears to be on the consumers’ side.
This is the cut of the bill taken by price comparison websites, in return
for referring customers. The recommendation to switch creates churn in

the market, and it is seen by supplier companies as worth paying high fees
to the websites. Whether or not customers choose to use the sites, the cost
to the supplier is embedded within bills for all customers.32

The report quotes David Hunter, an energy industry analyst with Schneider
Electric, who elaborated on the possible price implications of PCWs:

If you use a price comparison website, that website or broker will be paid
commission by the successful supplier for placing business through
them. . . . We know the supplier makes a profit for billing you of about
£60 a year, and bearing in mind what we know about supply cost infor-
mation, I wouldn’t be at all surprised if the websites and brokers are
making £60 or perhaps more out of every customer’s annual bill.33

In early 2015, the fees charged by the U.K.’s leading PCWs were revealed
in discussions of the Energy and Climate Change Select Committee.34 The
PCWs’ commissions ranged between £22 to £30 per single fuel customer
and up to £60 for a dual fuel tariff.35 Committee member Ian Lavery com-
mented on these charges and the platforms’ profits: “Someone is paying for
these profits. We support advising people to switch, but we do so on the
basis that the price comparison websites are trustworthy.”36 Despite the
criticism and concerns raised, however, the committee acknowledged
the benefit of these sites in facilitating the effective comparison of tariffs
in the energy sector and encouraging competition on price.37
PCWs in our U.K. energy example did not directly increase the cost per
transaction—in the sense that the PCW’s referral fee was not added as a sur-
charge to the product or service referred. Instead, the concern was that the
rising referral fees increased the pressure on upstream providers to increase
their prices to defray the referral costs.38 Some argue that providers using
these platforms, when faced with increased charges by web-aggregators, will
Copyright © 2016. Harvard University Press. All rights reserved.

ultimately increase their output price to regain profitability.39

The economic literature suggests that such cost externalities may indeed
arise under certain conditions.40 The introduction of a wide MFN to a
market with a homogeneous good, in which consumers already have ac-
cess to each of the sellers’ websites, would lead to an increase in price when
consumers would not multi-home and would be satisfied by the results of
a single comparison platform. Furthermore, in such markets, price in-
creases may intensify as additional web-aggregators are introduced.41
So the use of wide MFNs and agency models reminds us that all that glit-
ters is not gold. The perceived competitive environment offered by online

platforms may be undermined by behind-the-scenes mechanisms and

agreements which may, at times, leave consumers with less than they
bargained for. That realization has indeed led in recent years to a restric-
tive approach to wide MFNs.42 A growing number of competition agencies
have condemned these wide parity clauses and held them to be illegal. By
contrast, competition agencies have generally accepted the use of narrow
MFNs, which do not foster alignment between market players, as these
provisions can encourage investment and competition.43

Reﬂections
Using online comparison and search platforms, users can make better de-
cisions and be exposed to sellers and products which may other wise re-
main outside the market. With such contributions, it is no wonder that
these web-aggregators have become a significant intermediary in our on-
line environment. Their role and their privileged position have supported
competition at the platform level as well, with an increase in the number of
web-aggregators which are established and aim to provide products. That
competition is valuable. It ensures the quality and ser vice of the web-
aggregators, their ongoing investment in demand-enhancing features, and,
of course, supports a more transparent marketplace.
While they may increase welfare where the relevant market exhibits in-
formation failures and high search costs, they can also distort competition.
Such may be the case when intermediaries become a crucial gate to the World
Wide Web or when they use business agreements—like wide MFN clauses
and the agency model. The mixed effects generated by web-aggregators re-
quire consideration in context, taking into account the market characteris-
tics and nature of competition. One important destabilizing feature, which
could help safeguard competition, comes in the form of new technology and
Copyright © 2016. Harvard University Press. All rights reserved.

innovation. Indeed, the likely entry of new players or new technology could
restrain the incumbents’ behavior. Intervention by the competition authority
should be considered in light of these dynamics.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Copyright © 2016. Harvard University Press. All rights reserved.

Frenemies

A S WE HAVE SEEN in Parts II and III, the rise of Big Data and Big Ana-
lytics can foster new forms of collusion and behavioral discrimina-
tion. We now explore a third competitive dynamic, which we refer to as
“Frenemy.” This dynamic strategy—while not part of the competition
agencies’ lexicon—can significantly harm competition, innovation, and
our privacy interests. We also discuss how “network effects” can reinforce
dominance and create powerful gatekeepers.1
The Frenemy dynamic highlights the complexity of new online
ecosystem—what on the surface appears competitive, really is not—and
some shortcomings of traditional competition analysis.
Chapter 14 outlines the complex Frenemy dynamic and the interdepen-
dence among competitors. Our Frenemies are not equals. We’ll see the rise
of the so-called “super-platforms,” and the way independent application
developers depend on the goodwill of these main gatekeepers. In using the
term super-platform we refer to a handful of very power ful companies,
which benefit from network effects and dominate the ecosystem.2
Chapter 15 explores how firms, in our Frenemies scenario, have a dual
Copyright © 2016. Harvard University Press. All rights reserved.

data-driven strategy—extraction and capture. It is as if a den of lions were

to cooperate to circle the prey and then compete over which of them gets
the choice cuts of the gazelle. As you might guess, we are the gazelles.
Chapter 16 examines the Frenemy social structure. The super-platform
sets the rules for getting on, being promoted within, and getting kicked
off its platform. The super-platform can control (and cut off ) the smaller
independent apps’ oxygen supply. We will examine two apps, have you
guess which one was thrown off the super-platform, and explore why.
We will also see why— despite the abundance of free apps— competition
145

and innovation are diminished in the Frenemy scenario. The application

developers’ and super-platforms’ interests, even when aligned, do not al-
ways favor the consumer.
Chapter 17 considers the future of the Frenemy dynamic and the rise of
personal assistants. We explore how the introduction of digital personal
assistants can marginalize other operators and increase the super-platforms
as gatekeepers. This, we illustrate, will enable it to more easily determine
what we will see, where we will buy, and what we will read.
Copyright © 2016. Harvard University Press. All rights reserved.

The Dynamic Interplay among Frenemies

Y OU WON’T FIND the word Frenemy in any of the competition agencies’

policy statements. The agencies typically classify the competitive re-
lationship, if any, between firms as horizontal, vertical, interlocking, or
conglomerate.1 Companies in a horizontal relationship operate at the same
level of the distribution chain; they compete directly for market share (such
as Coke vs. Pepsi). Companies in a vertical relationship operate at different
levels of the supply chain. They do not compete directly for market share;
instead they buy from or sell to each other (such as Coke, its distributors,
and retailers like Walmart).2 Examples of firms in an interlocking relation-
ship are those in a hub-and-spoke conspiracy, discussed in Chapter 6, or
persons who serve as directors or officers of two competitors (as when
Google CEO Eric Schmidt and former Genentech CEO Arthur Levinson
sat on the boards of both Google and Apple).3 Finally, under a conglomerate
theory, firms are in neither a horizontal nor a vertical relationship, but are
active in closely related markets (e.g., mergers involving suppliers of com-
plementary products4 or products that belong to the same product range).5
With that classification as their guide, competition agencies generally
Copyright © 2016. Harvard University Press. All rights reserved.

scrutinize horizontal agreements and mergers more often than vertical ones;
they rarely investigate agreements or transactions involving conglomerate
or interlocking theories. That analytical framework provides a useful guide
when assessing most commercial relationships. Yet modern dynamic mar-
kets sometimes exhibit different, more complex relationships. In the context
of our discussion, the competitive Frenemy dynamic gives rise to an inter-
esting aty pical form of competition. The rise of platform competition and
pricing algorithms entails companies increasingly becoming “Frenemies.”
147

Horizontal
Competition

Price
Tacit Collusion
Discrimination
Frenemies

Horizontal
Vertical Input
Collaboration

Figure 1. Frenemies

Figure 1 reflects the move beyond the binary world of coordinated effects/
tacit collusion (which Part II addresses) and unilateral effects/price
discrimination (which Part III addresses), and beyond horizontal and
vertical interplay, to the dynamic real ity of Frenemies. Firms here col-
laborate (friends), compete (enemies), at times engage in unilateral discrimi-
natory action, and at other times benefit from increased interdependence
between firms (collusion). While they may view each other as enemies, they
may also cooperate in extracting and analyzing data or, alternatively, sup-
plying each other with a key vertical input.

Competition between and within Super-Platforms

The online world has seen the rise of super-platforms. In the 1990s, Microsoft
Copyright © 2016. Harvard University Press. All rights reserved.

dominated the operating systems of personal computers. Today consumers

are increasingly migrating to mobile and tablet operating systems. Smart-
phone adoption has accelerated. Ericsson predicts that 90 percent of the
world’s population over six years old will have a mobile phone by 2020.6
People are spending more time on their smartphones than on their personal
computers. Between 2013 and 2015, Americans’ overall digital media usage
grew by 49 percent.7 In this two-year period, their average time on mobile
apps increased by 90 percent, and their average time browsing the web in-
creased by 53 percent.8

With the migration to smartphones and tablets, the new super-platforms

are Apple, Google, Facebook, and Amazon.9 As the Wall Street Journal
observed:
Anyone building a brand, for example, can’t ignore Facebook’s highly en-
gaged daily audience of 1 billion. Anyone starting a business needs to
make sure they can be found on Google. Anyone with goods to sell wants
Amazon to carry them. Any mobile app maker needs to be available
in Apple Inc.’s or Google’s online stores. Any marketer with a video to
promote needs to be on Google’s YouTube, while producers selling
music, film, and television distribute their works through Apple’s iTunes
or Amazon Video.
The giants have spent billions of dollars on computing hardware and
data centers that run their own operations while increasingly providing
free or low-cost ser vices for startups and many large corporations. Many
longtime Silicon Valley executives are convinced that these companies
have become fundamental to the business landscape.
“You are seeing ecosystems built around all of these companies now,”
said Enrique Salem, a managing director at Bain Capital Ventures and
the former chief executive of Symantec Corp. “There is a platform shift
happening.”10

Two super-platforms dominate the mobile and tablet world: Apple’s iOS
and Google’s Android operating systems. In the second quarter of 2015,
Android accounted for 59 percent of the U.S. smartphone market, Apple’s
iPhone soft ware had 38 percent, Microsoft’s Windows Phone platform had
2.35 percent, and BlackBerry had 0.36 percent.11
Each super-platform, like a coral reef, attracts to it an ecosystem of soft-
ware developers and app and accessory makers. Figure 2 shows how the
number of apps available in the Google Play and Apple App stores soared
Copyright © 2016. Harvard University Press. All rights reserved.

between December 2009 and February 2015.12

Within these ecosystems, competition is along multiple dimensions.
First, apps can compete with other apps on the same super-platform. For
example, Kayak’s travel reservation app competes with Orbitz’s app. Second,
apps on one super-platform can compete with apps on a rival super-platform;
an app may be available on both the Apple and Android super-platforms.13
Third, independent apps can compete with the super-platform itself. Finally,
Apple’s and Google’s super-platforms can compete against each other for
app developers and users.

Apple Apps Android Apps

1,600,000

1,400,000

1,200,000

1,000,000

800,000

600,000

400,000

200,000

0
Jul Apr Nov Mar Jun Oct Apr Aug Dec Mar Jun Oct April Jun Oct July Jan Jun
'08 '09 '09 '10 '10 '10 '11 '11 '11 '12 '12 '12 '13 '13 '13 '14 '15 '15

Figure 2. Number of available apps in the Apple App and Google Play stores

Although the super-platforms compete, they can also be friends. For

example, Android, Google’s mobile operating system platform, supports
90 percent of Apple’s APIs (which stands for application program interface,
“a set of routines, protocols, and tools for building software applications”).14
Google has strategic reasons for extending this support, and consequently
for reducing the workload of potential Android soft ware developers who
already are (or wish to be) present on Apple’s mobile platform.15 In sup-
porting Apple’s APIs, Google increases the likelihood of developers
writing apps for its own version of the “open-source” Android operating
system, and decreases the likelihood of developers working with any other
version of Android (i.e., a competing fork of the soft ware) or Amazon’s
Kindle:
Copyright © 2016. Harvard University Press. All rights reserved.

Most developers probably say “yes” to Google APIs, and the next
question is what should they do about the Kindle and other Android
forks? Developers are largely on their own to find a replacement API so-
lution, which might be out of date and might not work perfectly with
their existing app. If this other solution isn’t a perfect drop-in replace-
ment, the developer will have to figure out how to design their app
around the missing feature. Since this is such a small amount of users
compared to their current iOS + Android user base, is it even worth it to
try to figure out this separate ecosystem? Will they get a return on their

time investment? It would be easy to say “the hell with forked Android”
and skip all the extra work and Q/A that would entail.16

Accordingly, the two super-platforms, while competing, can also collabo-

rate when their commercial interests align.
Another example of Frenemies, which several business professors have
examined, is the e-reader market.17 Amazon introduced its Kindle device
several years before Apple’s iPad; both devices compete for users today. But,
interestingly, Amazon subsequently developed a Kindle Reader app for
iPads, which Apple approved. So consumers can now read e-books they
purchased from Amazon on either their Kindle Reader or their iPad.18 As
Professors Adner, Chen, and Zhu noted, Apple, “well known for rejecting
third-party applications that compete directly with its own offerings, nev-
ertheless approved Amazon’s Kindle Reader for iPad, effectively rendering
the two platforms Frenemies (friends and enemies). Apple has not, how-
ever, made iBooks available for the Kindle.”19
This manifests the interdependence among leading companies. While
super-platforms can easily discard small app developers and remove them
from their ecosystem, they may have limited incentive to do so when
dealing with other leading platforms.

Uber’s Frenemy Relationship with Apple and Google

Besides super-platforms having Frenemy relationships, independent apps
can have a Frenemy relationship with their super-platform. Uber’s relation-
ship with Google and Apple illustrates this dynamic. As we saw in Chapter 6,
Uber operates a platform that connects drivers and users in dozens of coun-
tries. According to Uber’s website, “hundreds of thousands of drivers” are
joining its platform every month.20 Uber’s own platform sits on top of, and
Copyright © 2016. Harvard University Press. All rights reserved.

depends upon, Apple’s and Google’s super-platforms. To compete, Uber

must be available and fully functional on these super-platforms, because
users rely on their smartphones to request a ride, locate the car, and pay
the fare.
So Uber is friends with Google and Apple. Uber’s app is available in the
Google Play and Apple App stores. Indeed, apps like Uber can increase de-
mand for smartphones by enhancing their utility. Uber drivers and users
also need mapping technology, which Google has supplied.21 In addition,
Google, besides providing a platform for Uber, is an investor. Google

Ventures invested $258 million in Uber in August 2013.22 A Google execu-

tive, as of mid-2016, sits on Uber’s board.23
On the other hand, Uber and the super-platforms will increasingly be-
come competitors and thus enemies. Google is currently investing heavily
in driverless cars. In 2015, Google’s cars were already “averaging 10,000
new self-driven miles a week, mostly on real-world city streets—not on
controlled test tracks.”24 In 2015, Apple was also rumored to be launching
an electric car that may eventually be self-driving.25 In addition, 2016 saw
a significant investment by Apple in Didi Chuxing, a leading Chinese car-
hailing app.26 The investment, amounting to a billion dollars, formed part
of Apple’s strategy in China, and may well affect the increasingly complex
Frenemy dynamic to which Uber is party as well.
When, in the future, consumers start using Google’s (and perhaps Ap-
ple’s) driverless cars and ride-sharing ser vices, the super-platforms have the
potential to become a powerful force in these downstream markets. Both
super-platforms already have the mapping technology. Google also has a
crowd-sourcing app, Waze, which provides real-time traffic, accident, and
police information. Furthermore, Google has a vast quantity of consumer
data from its browser, e-mail ser vice, search engine, and social network.
Finally, Google, by licensing its applications, largely controls the open-
source Android operating system. Apple controls both the iOS operating
system and the iPhone.
At their will, therefore, Google and Apple can nudge users toward their
Uber-like app in several ways. Google can require smartphone makers to
preload its driver app (along with its other apps) on the mobile operating
system platform and include the app on the smartphone’s home screen.27
Apple can preload and feature its app directly on its iPhone. Alternatively,
the super-platforms could integrate their Uber-like app with their other
ser vices, such as their mapping apps. So when you look up an address on
Copyright © 2016. Harvard University Press. All rights reserved.

your iPhone or Android smartphone, Apple and Google could identify how
much it would cost to be driven there; you would simply click a button to
immediately order a car. In other words, the super-platforms can leverage
their power to diminish Uber’s competitive position. In 2016, Google was
already testing a ride-sharing app, whereby riders pay drivers 54 cents per
mile.28 Predictably, Google used its popular Waze app.
The possible leveraging of market power to push out an “as efficient” op-
erator downstream, and clear the way for the super-platform’s own opera-
tion downstream, is worrying.29

Beyond the super-platforms’ ability to leverage their power to support

car-sharing ser vices, the true game changer is likely to be their driverless
car technology. Google and Apple could in the future enjoy a significant
cost advantage over Uber. The super-platforms’ algorithms for their driv-
erless cars will also benefit from learning by doing, which provides impor-
tant scale. As Forbes has noted, “unlike human drivers who must rely on
their own experience for learning, Google’s cars will learn from every
Google car’s experience. That means that the more cars Google puts on the
road compared to its competitors, the greater its learning advantage.”30
Unlike Uber, the super-platforms do incur a significant upfront cost for
developing the driverless car. But Google and Apple can spread that cost
over many passengers. And, unlike Uber, the super-platforms would not
have to pay for the expensive recruitment of drivers; nor would they need
to pay 80 percent of each fare to drivers. Nor would Google and Apple
have to impose heft y surcharges to entice drivers onto the roads when de-
mand exceeds supply: their algorithms could simply summon available
driverless cars from local parking lots. Thus, as the cost of driverless car
technology decreases, and the technology improves through learning by
doing, the super-platforms could enjoy a significant cost advantage over
Uber and become a powerful enemy.
What makes Google a powerful Frenemy is that driverless cars represent
a new, important mechanism by which it can collect data for its super-
platform. As with all economies of scale and network effects, the technology
and data can go much further and add more value when at the disposal of a
super-platform. For instance, Google’s driverless car fleet would collect real-
time data on street traffic, construction, and so on. Th is data, along with
that collected on its Waze and Google Maps apps, could give Google a com-
petitive advantage in turn-by-turn navigation systems. If commuters want
to know the latest traffic conditions, they would likely turn to Google’s Waze
Copyright © 2016. Harvard University Press. All rights reserved.

and Maps apps. As more traffic data is quickly pumped through Google’s
super-platform, its driverless cars could better avoid traffic jams—thereby
reducing electricity/fuel costs and travel time, which in turn increases its
competitive advantage over Uber. Google’s Uber-like app would also col-
lect the geolocation data of par ticu lar consumers, which Google can com-
bine with its other data to better target app users with behavioral ads (both
on their phones and in its cars). The driverless car could, were it to offer
free Wi-Fi, collect even more data from users, including how they spend
their time.

Google is ultimately not a car-sharing ser vice. We doubt that Google is

interested in cars per se (just as it isn’t interested in smart thermostats per
se, despite its $3.2 billion acquisition of Nest Labs, a company whose leading
product is just that). Google is fundamentally an advertising-supported
super-platform. It collects our personal data to help advertisers better target
us with advertisements when we are on its super-platform and third-party
websites. Users, while in the Google cars, could spend time on Google’s
super-platform—watching YouTube videos (and ads) and searching (and
clicking on sponsored ads). The driverless car could itself become a bill-
board for the Google platform, where users and bystanders can be targeted
with ads televised on screens throughout the car.
Thus, the windfall is potentially great for Google: even if Google were to
offer the car ser vice for a nominal fee, consumers would end up paying
with their data and privacy. This gives Google a competitive advantage over
other super-platforms, like Apple, Facebook, and Amazon. Google’s Uber-
like car ser vice could become so cheap that many people might relinquish
their cars (and their use of public transportation). Instead, Google’s car ser-
vice could seamlessly become part of our daily activities, adjusting to the
rhythm of our lifestyles. Google, by virtue of its calendar and e-mail apps,
could anticipate when you need a car and where you need to go. As your
Nest Labs smart thermostat lowers the heat when you leave your home, a
Google driverless car pulls up to your front door.
Even for those who prefer to drive their own cars, Google could none-
theless leverage its driverless car technology. Given the scale benefits of
learning by doing, car manufacturers might prefer (or be forced) to license
Google’s technology (including the daily or hourly updates from the mil-
lions of Google cars on the road). (If Apple did not license its driverless car
technology, then it would lack the scale, and its quality would lag.) The car
manufacturers also see a potential relationship in which they manufacture
Copyright © 2016. Harvard University Press. All rights reserved.

the car’s “dumb” commoditized parts (like Foxconn Technology Group for
Apple), while the super-platforms earn the lion’s share of profits from tech-
nology and design. Thus the leading car manufacturers are undertaking
AI research.31
The above Frenemy scenario between and within platforms also takes
place between the operators on these platforms. Each application or ser vice
competes with other providers on the platform. At the same time, they also
complement each other. By providing users a wider choice of apps, the
super-platform becomes more attractive relative to other platforms.32

Economic (Inter)dependence
The benefits the super-platform may derive from driverless car tech-
nology highlight another theme—one of market power and dependency.
The super-platform needs the apps to attract users, but, once it becomes
powerful, it can harm the independent app developer in many ways. The
super-platform at any moment may favor its own operations downstream
over those provided by Uber. To put it differently, Uber’s biggest nightmare
is not some obnoxious taxi commissioner seeking to hold on to a crum-
bling monopoly by refusing Uber entry into his city, nor is it another car-
service platform like Lyft. The real fright comes from super-platforms like
Google and Apple.
Consumers may benefit from Uber’s fright when Uber improves ser vice,
maintains competitive prices, and increases its investment in research and
development. But Uber also sees the long shadow of the super-platforms,
and realizes that it will likely be at a significant competitive disadvantage
over the long run. Uber must now develop its own driverless car tech-
nology (or partner with a car manufacturer that does). It also lacks the
super-platforms’ mapping technology. So in early 2015, when Bloomberg
reported that Google was preparing its own ride-hailing ser vice,33 Uber re-
sponded. In 2015, it acquired mapping technology, including Microsoft’s
Bing Maps street-mapping technology and a hundred of its workers.34 Uber
has also hired around fift y researchers from Carnegie Mellon to develop
driverless cars.35 Google’s rival, Microsoft, is also an investor in Uber.36 So
Google is now Uber’s friend, investor, and enemy. Adding another dimen-
sion, General Motors invested in Uber’s smaller rival Lyft, where customers
can order a driverless car using their smartphones.37
But Uber will still lack the consumer data, advertising revenue, and mul-
tiple advertising platforms that a super-platform like Google possesses.
Copyright © 2016. Harvard University Press. All rights reserved.

And Uber, unless it launches its own smartphone operating system, will
continue to depend on the super-platforms for its oxygen supply. The
oxygen provider will ultimately hold the key to coexistence, exclusion, or
possible acquisition of the downstream operator.

Asymmetries in Power
The asymmetry in bargaining power between the apps and super-platforms
is central to the Frenemy dynamic. The independent apps recognize that

web usage is increasingly shifting to mobile platforms such as smartphones

and other connected devices.38 Their business growth and success thus de-
pend on their interoperability with the super-platforms, which they do not
control.
In its report on “Online Platforms and the Digital Single Market,” the
United Kingdom House of Lords noted the asymmetry in bargaining
power.39 We also see both the smaller and larger publicly held platforms,
from Coupons.com40 to Facebook,41 expressing the same concern to their
investors, namely being squeezed by the super-platform.
The super-platform Facebook, for example, described to investors its
“dependen[ce] on the interoperability . . . with popu lar mobile operating
systems, networks, and standards that we do not control, such as the An-
droid and iOS operating systems.”42 Facebook identified how these super-
platforms could degrade the functionality of its app; reduce or eliminate
its ability to distribute its products; give preferential treatment to com-
peting products; limit its app, whose revenues are primarily from adver-
tising and the ability to deliver, target, or measure the effectiveness of ads; or
impose fees or other charges related to its delivery of ads.43 Thus, Facebook
warns that, among other things, there “is no guarantee that popular mobile
devices will continue to feature Facebook or our other products, or that
mobile device users will continue to use our products rather than competing
products.”44
Are the independent apps simply paranoid? Not necessarily. A super-
platform has several levers with which to exert its power. The super-platform
can degrade the functionality of the independent apps and online platforms—
like LinkedIn, Twitter, Yelp, or Coupons.com—by reducing their perfor-
mance and making them run slower. It can foreclose its Frenemies’ timely
access to critical data. It can increase consumers’ switching costs, thereby
making it harder for the app to attract users.45 A super-platform could also
Copyright © 2016. Harvard University Press. All rights reserved.

prevent Frenemies from achieving the minimum efficient scale.46 Scale can
be especially important in data-driven industries such as search and search-
advertising. It may limit a competing app’s revenue stream by excluding the
app from its online payment systems, such as Apple Pay or Google Wallet.
The super-platform can reduce or eliminate the independent app’s ability
to distribute its products. It can make it harder for consumers to find the
product on its search engine or app store. It can also give preferential treat-
ment to its own or other competitive services.

The super-platform can give preferential treatment to its own prod-

ucts, by preloading its app on the smartphone, having it on the opening
screen, or integrating its own products into other popu lar products, in-
cluding search and the smartphone operating system. Th is was the basis
of the European Commission’s 2016 charges against Google.47 Google,
the Commission alleged, abused its dominant position with its mobile
Android super-platform to “preserve and strengthen its dominance in
general internet search.”48 Absent a Frenemy relationship, a super-platform
would ordinarily leave it to manufacturers or customers to decide what
apps to preinstall on the smartphone. Android is technically an open
operating system. But Google, according to the Commission, controlled
the operating system’s development through its licensing agreements
with the Android smartphone manufacturers.49 It reduced the smart-
phone manufacturers’ incentives to preinstall competing search apps, as
well as consumers’ incentives to download such apps.50 Google also paid
a lot of money to “some of the largest smartphone and tablet manufac-
turers as well as mobile network operators” on the condition that they
exclusively preinstall Google Search on their devices, and not any other
search provider.51
If the Commission’s concerns are borne out, then one can imagine how
difficult it is for an independent search engine, like DuckDuckGo—or any
independent app—to successfully compete against Google. In Eu rope,
Google dominates the markets of general Internet search ser vices and li-
censable smart mobile operating systems. DuckDuckGo can’t convince the
manufacturer to be the default search engine on your Android phone; nor
can DuckDuckGo get its search engine preloaded on your Android phone;
nor can it turn to Apple. As we’ll see, Google paid Apple $1 billion in re-
cent years to be the default search engine on your iPhone.
Copyright © 2016. Harvard University Press. All rights reserved.

Reﬂections
Th is chapter identifies the rise of a competitive dynamic that fosters in-
terdependence between competitors and between companies in the vertical
chain. Our Frenemy scenario takes place between super-platforms and
within them. Possible abuses by the super-platform can hurt consumers in
many ways, including less innovation (when independent companies know
that, however good their products or services are, they cannot effectively

reach consumers unless the super-platform admits them and doesn’t injure
them later).
In the next chapter we explore the unique environment in which apps
and super-platforms join forces in extracting, selling, and analyzing data.
Here again we see super-platforms, in exercising their dominance, coordi-
nate the extraction of our data and capture most of the profits.
Copyright © 2016. Harvard University Press. All rights reserved.

Extraction and Capture

P OWER, we saw in the last chapter, resides with the super-platform. But
the lives of the independent apps are not necessarily solitary, in con-
tinual fear and danger of violent death by the super-platform. The apps’
lives are, in fact, interdependent with the super-platforms’. They both seek
to attract us to their ecosystem, whether it is, for example, an Apple or An-
droid smartphone.
One way to attract our attention is through free apps. Indeed, the eco-
system can appear competitive, with many apps offering free or discounted
products or ser vices. But behind this competitive veneer lies another facet
of their interdependence, namely their joint strategy, extraction and cap-
ture, in collecting and using our personal data. As we saw with price dis-
crimination, relevant, up-to-date personal data can provide online sellers
a competitive advantage.1
This chapter explores the Frenemy strategy of extraction and capture. The
website owners, independent apps, and the super-platform cooperate in the
extraction phase, obtaining valuable personal data (such as geolocation
data) about us, tracking our behavior, promoting asymmetrical informa-
Copyright © 2016. Harvard University Press. All rights reserved.

tion exchanges and strategies (i.e., where they control and know about the
data flow, but we don’t), and reducing our ability to maintain our privacy.
In this extraction phase, the super-platform, website owners, and inde-
pendent app developers are like lions that cooperate to circle the prey.
We, like gazelles running across the savanna, rarely stay on one website or
app. If the super-platform, website owners, and independent app developers
did not cooperate, they would see us only when we arrive. They would not
know from where we came or where we are going next. Advertisers would
159

not want to negotiate with each website for data about us. It would be too
costly and time-consuming to develop profiles about us. Without such pro-
files, it would be harder to predict what ads would appeal to particular users
and to engage in behavioral advertising (and discrimination).
Following the successful extraction, in the subsequent capture phase,
the super-platform, website owners, and independent app developers
compete to retain an advantage over each other and to capture as much
of consumers’ wealth as possible. In this phase, the Frenemies’ interests
can diverge, as they compete among themselves over the distribution of
value within the channel. That distribution, as we explore, depends
on their relative bargaining power. So, after circling the prey, the super-
platform and independent apps compete over the choice cuts of the
gazelle.

Data Extraction
Suppose advertising revenue is the lifeblood of the super-platform’s, website
owners’, and independent app developers’ ecosystem. Every time we click an
ad or purchase an item, they get money. How could they maximize profits?
Ideally, the Frenemies could implant in our brains a device to collect data
wherever we go, about what we do, what we are thinking, and what we can
be enticed to buy. The implant would provide a clear picture of our desires
and behavior; the Frenemies could easily predict what we are likely to buy,
when, at what price, and the personalized ad that would induce our pur-
chase. With this technology the Frenemies could perfectly track and target
us with personalized ads, increasing the relative power of their ecosystem.
What advertiser or marketer could afford to spurn this technology?
Of course, such technology does not exist today. Instead, the Frenemies,
under their joint extraction strategy, strive to approach it. The super-
Copyright © 2016. Harvard University Press. All rights reserved.

platform, independent apps, and websites within the ecosystem join forces
to better track and understand us as we browse their websites across de-
vices (such as when we are on our personal computer, laptop, smartphone,
or tablet). In jointly extracting the data about us, they seek to improve their
understanding of our behav ior in order to influence and modify it. This
includes devising ways to better exploit our biases and imperfect willpower.
Ultimately, the companies agree to extract our personal data, which feeds
the advertising networks’ algorithms, and improves their ability to iden-
tify behavioral ads to prompt purchases. So the Frenemies cooperate with

respect to both the inputs (extracting personal data) and outputs (pro-
viding platforms for behavioral ads by others).
The independent apps and websites use the personal data internally, but
also selectively distribute it within the food chain for others to analyze. At
the top of the chain is the super-platform. The super-platform harvests
massive volumes of data directly as well as through others, analyzes it, and
then determines what data its partners and advertisers can analyze (per-
haps without personal identifiers). Or the super-platform serves as the air
traffic controller. It uses the data to continually target individuals with per-
sonalized ads on all of its own and its Frenemies’ publishing platforms.
Their joint interest is that that we spend more time on their network of
websites and apps rather than a competing super-platform’s ecosystem,
where they cannot collect data about us or target us with ads.
To illustrate, let us consider a health club ad. Traditionally, a health club
would advertise offers on media that many prospects were watching, such
as fitness or athletic shows. There was a lot of waste, because many viewers
may be uninterested in joining a health club. Another drawback is that the
health clubs had to wait for consumers to watch the shows or visit the fit-
ness or athletic websites where they advertised. Now suppose that from the
data flow the super-platform finds that an individual, whom we’ll call
Harry, is interested in joining a health club. It might be because Harry
searched for nearby health clubs or e-mailed a friend about possibly joining
a gym. Or, cleverer still, Harry’s behavior signals a greater willingness to
join a gym. Harry, for example, recently searched about dieting or healthier
foods. Under the Frenemy scenario, the super-platform is collecting enor-
mous amounts of data about Harry. It knows Harry is ripe to join a gym. It
also knows, from where Harry lives, works, and socializes, which gym
might be especially attractive. Now the super-platform can target Harry
with promotions for this specific health club wherever Harry spends time
Copyright © 2016. Harvard University Press. All rights reserved.

on any of the Frenemies’ websites, apps, or ser vices within the ecosystem.
When Harry awakes, his free alarm app might have an ad for the health
club. Later in the morning, when Harry reads about world news on the New
York Times website, a banner ad may promote the health club. When Harry
next visits a social network, he may see a friend’s endorsement of the same
health club. Indeed, the super-platform can target Harry with complemen-
tary ads—for athletic clothing, energy bars, nutritional supplements, and
health clubs—each reinforcing the need to consume some of these prod-
ucts. When walking downtown near the health club, Harry might get a

coupon for a free thirty-day trial period. So what’s the best way to track
Harry and extract his data? The answer is in his pocket.

Mobile Platforms as a Source of Data

Harry’s smartphone can offer a wealth of data that quantitatively and qual-
itatively differs from anything else he carries in his pocket or stores in his
home.2
The U.S. Supreme Court observed in 2014 how the storage capacity of
cell phones had several interrelated consequences for privacy.3 The Court
noted how the data collected on a smartphone differs qualitatively from
physical records, and could reveal “an individual’s private interests or
concerns—perhaps a search for certain symptoms of disease, coupled with
frequent visits to WebMD . . . [and] where a person has been.”4 Historic
location information, the Court noted, “is a standard feature on many
smart phones and can reconstruct someone’s specific movements down
to the minute, not only around town but also within a particular building.”5
Our location data is also used to refine our user profiles and better target
us with behavioral advertising.6 Our geolocation information, along with
other data collected from our phones, can be sold or used to develop other
marketing strategies.7 So companies can use our mobile phones to extract
additional information about our whereabouts and activities, when we visit
or revisit a particular store, and where we linger in that store. Google, for
example, tracks users’ locations “to tell with 99% accuracy if a user visited a
store after seeing a relevant search ad.”8 That data collection is intensified in
the context of the “tech addiction that puts smartphones in control of us.”9
Importantly, not many of us know of all the personal data collected
through our smartphone. As the Australian Communications and Media
Copyright © 2016. Harvard University Press. All rights reserved.

Authority noted:

Around half of all survey participants were not aware of data-sharing

processes such as information shared with a third-party provider, stored
by a mobile ser vice or internet ser vice provider when using a location ser-
vice, or information stored by the location ser vice developer. . . . The sale
and ownership of information and risks associated with disclosure were
key concerns with 71 per cent of users’ concerns with information being
sold to a third party and 59 per cent concerned about a lack of informa-
tion on where their data goes and who owns it.10

With the rise of the Internet of Things, mobile platforms will become the
key gateway to the flow of personal data. Google’s underlying operating
system for the Internet of Things, dubbed “Brillo,” is based on its Android
operating system.11 As our smartphones are always near us (except perhaps
when we shower or swim), they will assist the super-platforms, govern-
ments,12 and others in tracking our behavior, harvesting our data, and tar-
geting us with behavioral ads.13 This data trove will also attract hackers and
criminals.
Thus we should expect Frenemies to support the Internet of Things, to
the extent that the sensors can effectively track and collect data on us when
we are offline—data that can be used to fuel their advertising-supported
business model. Frenemies, as we’ll see with Uber, would also likely coop-
erate by allowing third-party cookies and other technology to better track us.

Uber and Google

We saw in Chapter 14 how Uber and Google were Frenemies with respect
to car ser vices. But they are also Frenemies with respect to data.
To begin with, Uber caused a stir in 2014 when its senior vice president
expressed a desire to spend $1 million to dig up information on the per-
sonal lives and families of journalists who wrote critically about Uber.14
Another Uber executive had “examined the private travel records of a
Buzzfeed reporter during an e-mail exchange about an article without
seeking permission to access the data.”15
Given these privacy lapses, Uber prompted consternation when in 2015
it relaxed its privacy policy. Uber will track its users’ locations, if they agree,
even when they aren’t using the app:
When you use the Ser vices for transportation or delivery, we collect pre-
Copyright © 2016. Harvard University Press. All rights reserved.

cise location data about the trip from the Uber app used by the Driver. If
you permit the Uber app to access location ser vices through the permis-
sion system used by your mobile operating system (“platform”), we may
also collect the precise location of your device when the app is running
in the foreground or background. We may also derive your approximate
location from your IP address.16

Uber will also, if users agree, access their address books, collecting all
the names and contact information in them so as “to facilitate social inter-
actions through [its] Ser vices and for other purposes described in this

Statement or at the time of consent or collection.”17 Uber never explained

how this data would improve its car-sharing ser vices.
Competition dynamics should correct an app, like Uber, that over-
reaches by seeking more personal data than is necessary for its ser vices to
be provided. One check should come from the super-platform, as its incen-
tives should be aligned with the users’. The super-platform wants to attract
many users, and to retain its existing users. So the super-platform should
be vigilant, rebuking any independent app developer who greedily de-
mands more personal data than necessary for its app to run effectively. The
super-platform would also design its platform to promote users’ privacy
preferences, such as giving users greater control over their data, requiring
any app to obtain the user’s express permission, and enabling users to delete
their personal data.
Interestingly, neither super-platform publicly rebuked Uber over tracking
smartphone users’ locations when they weren’t using the app, or for ac-
cessing their contact information. Indeed, Google made it easier for Uber to
access your information.
Apple, as Uber notes, “will alert you the first time the Uber app wants
permission to access certain types of data and will let you consent (or not
consent) to that request.”18 So Apple, at least, requires its iPhone users to
affirmatively assent, allowing them to opt out of each permission on an in-
dividual basis.
Android smartphones, however, do not; they will merely notify you “of
the permissions that the Uber app seeks before you first use the app, and
your use of the app constitutes your consent.”19
Apple iPhone users see the box shown in Figure 3, which requires the
user’s consent.20 Android users instead see the box shown in Figure 4,
without any conspicuous consent box.21
Copyright © 2016. Harvard University Press. All rights reserved.

“Uber” Would Like to Use

Your Location.
Uber picks you up exactly where you are.
To start riding, choose “Allow” so the
app can Ànd your location.

'RQ·WAllow OK

Figure 3. Uber, iOS app permissions

(https://www.uber.com/ios/permissions)
© Uber Technologies, Inc.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Copyright © 2016. Harvard University Press. All rights reserved.

Figure 4. Uber, Android app permissions

(https://www.uber.com/legal/android/permissions)
© Uber Technologies, Inc.

This was not entirely Uber’s choice. Uber’s Privacy Statement states
that Google created “[t]he descriptions of these permissions,” which “are
worded the same for every app—currently, there’s no way for Uber to cus-
tomize them.”22
Returning to our extraction phase, we see here how the super-platforms
help (rather than thwart) the independent apps (like Uber) to better track
users. Uber (and others) can track your location even when you aren’t
using the app. Rather than promoting privacy by design, Google’s take-it-
or-leave-it privacy approach makes it harder for users to avoid being
tracked. (Google’s forthcoming Android 6.0 or “Marshmallow” may en-
able users to toggle individual permissions on and off on an app-by-app
basis.23) Indeed, according to one account, advertisers hated working with
Apple’s iAd:

not because it failed to improve the ad experience, or because it was tech-

nically inferior, or because it failed to engage audiences. Advertisers hated
that Apple’s iAd was preventing them from gaining full access to user
demographics and behaviors—the way Google, Adobe and the other mo-
bile ad networks were working to facilitate. Apple’s increasingly vocal
stance on the side of consumer privacy—which has only grown more
strident over time—was standing in the way of advertising nirvana: the
non-stop audience surveillance program that could be distilled into
the sort of pure profit brand manipulation depicted in futurist movies
such as Minority Report, where billboards literally leap into your face and
talk to you by name, coaxing you to buy with the savvy of a salesman
pretending to be your best friend.24

Although Google is more opportunistic than Apple in the extraction

phase, neither super-platform employs a data-minimization principle,
Copyright © 2016. Harvard University Press. All rights reserved.

which would require the apps within their ecosystem to collect only the
personal data that is directly relevant and necessary to accomplish the
app’s specified purposes and to retain the data only as long as is necessary
to fulfill the specified purpose. Neither super-platform requires Uber to
explain:
• why it needs to track your location when you aren’t using the
Uber app;
• why it needs your contact information “to facilitate social interac-
tions through [its] Ser vices”;25 or

• why if you choose to link, create, or log into your Uber account with
a payment provider (such as Google Wallet) or a social media ser vice
(such as Facebook), Uber can obtain yet more personal information
about you or your connections from these websites.26

The other component of our extraction phase is how the independent

apps and websites within the ecosystem help each other extract personal
data. Uber isn’t simply hoarding the data for itself. Rather, Google (and
Apple to a lesser extent) encourages the flow of consumer data, because it
fuels the advertising revenue upon which the super-platforms, the ad ex-
changes, and the publishers are dependent. This is not apparent when you
first visit Uber’s website. To have seen this in February 2016, one would have
had to visit Uber’s website, go to the bottom of the page, click on its Privacy
tab, scroll down the page until near the end, click the Cookie Statement hy-
perlink, scroll to the bottom of the Cookie Statement, and read Uber’s de-
scription of how third parties will track you when you visit Uber’s website:

Things like cookies and pixels are used to deliver relevant ads, track ad
campaign performance and efficiency. For example, we and our ad part-
ners may rely on information gleaned through these cookies to serve you
ads that may be interesting to you on other websites. Similarly, our part-
ners may use a cookie, attribution ser vice or another similar technology
to determine whether we’ve served an ad and how it performed or pro-
vide us with information about how you interact with them.27

For its app to function, Uber doesn’t need to continuously monitor your
location. It doesn’t need to know who all of your friends, family, and co-
workers are, as well as anyone else listed among your contacts. Nor does
Uber have to allow others to track you across the web, including when you
visit Uber’s website. So what explains Uber’s actions? Here again the an-
Copyright © 2016. Harvard University Press. All rights reserved.

swer is data.
To understand tracking we’ll briefly discuss cookies. A “cookie” is a file
on a user’s computer that contains “information that identifies the domain
name of the webserver that wrote the cookie (e.g., hulu.com or facebook
.com)” and “information about the user’s interaction with a website.”28
Cookies were originally developed for a benign purpose: an online shop-
ping site placed the cookie (first-party cookie) to help remember the items
you wanted to purchase (e.g., placed in your shopping cart) as you browsed
its website.29 For example, the cookie would allow Brooks Brothers to re-

member the shirt you placed in your shopping basket as you continued
searching for a matching tie.
The inventor of cookies was concerned that third parties could use
cookies to track you across the web.30 That is exactly what happened. In
fact, the Wall Street Journal found that the tracking technology “is getting
smarter and more intrusive,” moving beyond cookies to tools that “scan in
real time what people are doing on a Web page, then instantly assess loca-
tion, income, shopping interests and even medical conditions. Some tools
surreptitiously re-spawn themselves even after users try to delete them.”31
As a result of the tracking mechanisms, including third-party cookies, not
only will the retailer remember what you placed in your shopping cart, but
so too will other companies; thus you may encounter advertisements for
dress shirts and ties as you surf other websites.
Uber places cookies on your browser or computer when you visit its web-
site. But Uber also allows the super-platform Google and other companies
to place cookies and identification technologies on your computer and
browser when you visit Uber’s website.
Uber identifies several reasons why it allows third parties to track you
when you visit Uber’s website. The first is “Site features and Services,” namely
to enable Facebook, Twitter, Google, and possibly others to “provide you and
others with social plugins and other customized content and experiences,
such as making suggestions to you and others.”32 Second, it enables Google,
MixPanel, Optimizely, and possibly others to collect data for “analytics
and research,” including segmenting audiences for testing and understanding
“how you use websites, apps, products, ser vices and ads.”33 The fi nal
reason is advertising. Uber allows, among others, Google, Facebook, AOL,
Microsoft, Yahoo, Drawbridge, Indeed, Recruitics, RocketFuel, Simplyhired,
Twitter, Ziprecruiter, Mixpanel, HasOffers/Tune, Adjust, AdRoll, Quantcast,
and KenshooThings, to track your visit to Uber. The cookies and pixels are
Copyright © 2016. Harvard University Press. All rights reserved.

used to deliver and track the performance of behavioral ads.34

Uber is not alone in helping the super-platform and others extract your
data. Any time you visit the website or app of any of Google’s partners, such
as websites that use Google’s advertising products (like AdSense), social
products (like the +1 button), or analytics tools (Google Analytics), “your
web browser automatically sends certain information to Google.”35 Google
collects the web address of the page that you’re visiting and your IP ad-
dress.36 Google “may also set cookies on your browser, or read cookies that
are already there.”37 Apps that “partner with Google” can send it additional

information “such as the name of the app and an identifier that helps
[Google] to determine which ads [Google] served to other apps on your
device.”38 Google can use your personal data in a variety of ways, including
making the behavioral ads more effective, and help its partner websites and
app owners using Google Analytics “to understand how visitors engage
with their sites or apps.”39 (To opt out of websites and apps sending your
data to Google Analytics, you must download and install an add-on for
your web browser.40)
Consider one class action lawsuit against Viacom Inc. and Google on
behalf of children under thirteen years of age for violating their privacy
rights under U.S. federal and state law.41 Viacom owns and operates three
websites geared toward children: Nick.com, Nickjr.com, and Neopets.com.
As the complaint alleged, Viacom encouraged the children to register and
establish profi les on its websites. Viacom assigned a code name to each
child based on that child’s gender and age—allegedly called (by Viacom
internally) the “rugrat” code. Viacom placed a cookie on the children’s
computers, allegedly without their (or their parents’) consent to acquire ad-
ditional information. Viacom then shared this information with Google
and permitted Google to place third-party cookies on the children’s com-
puters to track their Internet usage. Google used the information for the
same reason that Viacom used it—“ ‘to sell targeted advertising’ based upon
[the children’s] ‘individualized web usage, including videos requested and
obtained.’ ”42
Judge Stanley R. Chesler, in an unpublished opinion, dismissed the
lawsuit. Some of the reasons dealt with the par ticu lar statutes at issue.43
The state privacy claim failed because the judge was not persuaded that
Google’s and Viacom’s collection and monetization of online informa-
tion of children would be “offensive to the reasonable person, let alone
exceedingly so.”44
Copyright © 2016. Harvard University Press. All rights reserved.

This was not an isolated incident. In 2010, the Wall Street Journal exam-
ined fift y websites popular with U.S. teens and children to see what tracking
tools they installed on a test computer.45 As a group, the websites placed
“4,123 ‘cookies,’ ‘beacons’ and other pieces of tracking technology.”46 That
was 30 percent more than what the Journal found in an analysis of the fift y
most popular U.S. sites overall, which were generally aimed at adults.47 Vi-
acom’s Nickelodeon TV network accounted for eight of the fi ft y websites
in the Journal’s survey.48 And Google placed the most tracking fi les overall
on the fift y websites. In response to the Journal article, a Google spokes-

person responded that “ ‘a small proportion’ of the files may be used to de-
termine computer users’ interests.”49 The Google spokesperson also said
that Google doesn’t include in its user profiles “topics solely of interest to
children.”50 But the Journal, in reviewing Google’s Ads Preferences page,
found that the super-platform “accurately identified a dozen pastimes of
10-year-old Jenna Maas—including pets, photography, ‘virtual worlds’ and
‘online goodies’ such as little animated graphics to decorate a website.”51
Included in the extraction phase are data brokers, who are also inno-
vating methods to better track you online and offline:
Data brokers rely on websites with registration features and cookies to
find consumers online and target Internet advertisements to them based
on their offline activities. Once a data broker locates a consumer online
and places a cookie on the consumer’s browser, the data broker’s client
can advertise to that consumer across the Internet for as long as the
cookie stays on the consumer’s browser. Consumers may not be aware
that data brokers are providing companies with products to allow them
to advertise to consumers online based on their offline activities. Some
data brokers are using similar technology to serve targeted advertise-
ments to consumers on mobile devices.52

In addition, companies buy and sell user profiles and updated information
on stock market–like exchanges.53

Capture
So Frenemies cooperate to track us, extract our data, and target us with
behavioral ads. They all benefit from the combined effort. But they do not
share equally the spoils; the dominant lion gets the best cut, which further
enhances its power.
Copyright © 2016. Harvard University Press. All rights reserved.

As we saw in the last chapter, the super-platform has relatively more

power than the independent apps. As a key conduit for data, it can serve as
the advertising intermediary, obtaining several choice cuts of the adver-
tising revenue. It can collect revenue from ads on its sprawling super-
platform. It can collect fees for advising advertisers where their ads should
go. It can collect fees for advising publishers on how to better collect data
to target users with ads they are likelier to click. Finally, it can collect a
share of the advertising revenue of its partners’ websites. Suppose Google’s
algorithms directed the health club ad to one of its partner’s websites.
When Harry clicks the health club ad, the website receives 68 percent of

the ad revenue and Google 32 percent. So if the advertiser pays $1 per click,
the Frenemy that published the ad gets 68 cents; Google gets 32 cents.54
Why settle for 32 percent when you can get it all? The super-platforms
can use their profits to expand downstream with new (or acquired) apps
and programs, so that the gazelles will likely spend more time grazing on
the super-platform’s terrain. When you click an ad on the super-platform’s
website, it collects 100 percent. For example, the bulk of Google’s ad reve-
nues come from its websites rather than its Frenemies’ websites.55
So Frenemies, using the findings from psychology experiments, compete
to get us to spend more time on their platform or app.56 For advertising-
dependent apps, their value largely depends on how much time people spend
on them.57 Here the super-platforms have the advantage. In 2014, Ameri-
cans, on average, spent forty minutes per day on Facebook.58 One survey
found that of all the minutes spent on mobile phone apps, U.S. users spent
the greatest time on Facebook’s super-platform (13 percent), followed by
Google (12 percent, and this percentage does not include time spent on the
Android operating system ), Amazon (3 percent), and Apple (3 percent).59
Thus, one concern is that as the super-platform expands, including in-
corporating more habit-forming technology, it becomes less dependent on
third-party tracking technologies and third-party publishers of ads. In
effect, the savanna becomes the super-platform’s zoo. It knows where the
gazelles are, and can devour them whole.
For example, for those of us owning an iPhone and iPad, we can use ad-
blocking technology to limit behavioral ads when we use the Safari browser;
we cannot use the technology within apps, such as those of Facebook or
Google. Why the distinction? As the Wall Street Journal reports, “Apple
says it won’t allow ad blocking within apps, because ads inside apps don’t
compromise performance as they do on the browser. That distinction
serves Apple’s interests. It takes a 30% cut on money generated from apps,
Copyright © 2016. Harvard University Press. All rights reserved.

and has a business serving ads inside apps.” 60 This comports with our
Frenemy scenario. Apple incentivizes companies that are dependent on ad-
vertising revenue to develop apps for its iOS platform, where ads can’t be
blocked, and Apple gets a significant cut of the ad revenue.
Likewise, Coupons.com warns investors about Google moving away
from cookies, which consumers can block, to other tracking technologies:

[C]ompanies such as Google have publicly disclosed their intention to

move away from cookies to another form of persistent unique identifier,
or ID, to identify individual Internet users or Internet-connected devices

in the bidding process on advertising exchanges. If companies do not use

shared IDs across the entire ecosystem, this could have a negative impact
on our ability to find the same anonymous user across different web
properties, and reduce the effectiveness of our solution.61

Thus, one concern is that Google and other dominant firms can track
individuals across their sprawling super-platforms, but restrict sharing the
customer information with others in the ecosystem.
Allen Grunes, our colleague at the Data Competition Institute, explained
this “capture” dynamic with respect to an industry-proposed do-not-track
standard.62 The Federal Trade Commission asked industry participants to
craft a new “Do Not Track” policy for online data, similar to the “Do Not
Call” registry that helped reduce the nuisance of telemarketers telephoning
our homes. “But what started as a group effort by technology companies
and privacy experts to craft a new type of consumer protection has quietly
changed,” Grunes said, “and today has morphed into a committee where a
few of the most powerful Internet firms are deciding on the rules of the
game.”63 The World-Wide-Web Consortium, under the influence of domi-
nant players such as Google, Yahoo!, Facebook, and Comcast, proposed in
July 2015 a Do Not Track standard that distinguished between first and
third parties. Basically, when the gazelles are visiting the super-platform’s
own apps and websites, the super-platforms can continue to track them and
collect data on them, but third parties, like the smaller apps, cannot. So
when a user activates the Do Not Track signal, “if he or she enters a query
into the Google search engine, signs onto . . . Gmail, or uses Google Chrome
or Android, he or she will still be allowing Google to gather information
and use it to deliver targeted ads.”64
Here the Frenemy relationship is tilting toward capture. The proposed
Do Not Track standard doesn’t really prevent tracking. Instead, the main
Copyright © 2016. Harvard University Press. All rights reserved.

function of the proposed standard “will be to limit the ability of any po-
tential rivals to collect comparable data.”65
This is key as users spend more time on mobile phones, whose operating
systems the super-platforms control, and on the super-platforms’ own apps
and ser vices. The big lions will still rely on others to help hunt the gazelles.
But under our scenario’s capture stage, the weaker lions will get skinnier
as their cuts of profits get smaller. They will eventually perish or be dis-
placed by other, more enterprising app developers who can offer the super-
platform better, more valuable data or provide a popular watering hole for
gazelles that do not frequent the super-platform.

The more personal data the super-platforms amass, the more likely they
can predict where we will go and buy, and what we will want. With a better
view of our behavior, they can better identify the critical aspects that are
missing. To use another analogy, if data is crude oil, the super-platforms
will know where to drill. For more valuable personal data, the super-
platform may use its own rigs to access the data. Lesser grades of data may
be pumped by one of the independent apps. Once the data is exhausted, the
independent app ceases. At times, consumers may volunteer their valuable
personal data, such as identifying their friends and likes [and, in 2016, “dis-
likes” (or a near alternative)] on Facebook.66 Then the super-platform refines
the data and selectively allows access to specific advertisers and independent
apps that serve as advertising platforms.
Thus the extraction and capture phases are dynamic. Super-platforms at
times will be more selective about sharing data on users. Facebook, for ex-
ample, had a billion people who used its network on a single day in August
2015.67 While extending its platform to include search, Facebook is also re-
stricting the data it is sharing with others. As the Wall Street Journal re-
ported, “Dozens of startups that had been using Facebook data have shut
down, been acquired or overhauled their businesses.”68 One venture capi-
talist noted the shift from joint extraction to capture: “Companies are open
until they have liquidity and users. Then they start to control.”69 He too is
“becoming increasingly skeptical that you can build a lasting, stand-alone
business based on access to someone else’s social graph.”70
While closing one door, Facebook is opening other doors for compa-
nies to target us. For example, Facebook introduced bots for its Face-
book Messenger text ing platform. The new technology, backed by
powerful algorithms, will make use of user data to better target users with
ads and promotions. The bots will foster communication with companies
and enable Facebook’s algorithms to better learn of your preferences.
Copyright © 2016. Harvard University Press. All rights reserved.

Since it is embedded within the chat ser vice, it may become difficult to
separate from the “normal” use Messenger. Indeed, if Facebook’s vision
materializes, the use of its bots would replace other apps as these will
form part of the chat thread—exposing us to increased tracking and
targeting.71

How Powerful Is the Super-Platform?

The super-platform, while increasing in power and scale, is not immune
from competition. Competitive pressure may come from other platforms,

innovation, or entry. To maintain its position, the super-platform will in-

vest heavily in research and development.
While it controls the bottleneck and determines access, the super-
platform also has to retain and attract independent app developers for its
ecosystem to flourish. It is important to recall the interdependence between
the super-platform and the applications. Each super-platform collects
money from the independent app developers. Google reportedly takes a
30 percent cut from every sale that its app developers make at the Play
Store.72 Each platform needs its ecosystem to flourish; accordingly, it will
need to attract independent application developers to join its platform to
help build solutions. The greater the utility and diversity of solutions that
the platform’s products and ser vices can offer, the more users the platform
will likely attract. As the platform attracts more users, more application
developers will migrate to the platform. So we see a positive feedback loop,
where big platforms get even bigger as they attract more developers and
users, and as they morph into super-platforms. Microsoft, which is familiar
with these network effects from its personal computer operating system,
recently noted the importance of platform competition: “A well-established
ecosystem creates beneficial network effects among users, application
developers, and the platform provider that can accelerate growth. Estab-
lishing significant scale in the marketplace is necessary to achieve and
maintain attractive margins.”73
When considering the power of the super-platform, one should also ac-
knowledge the nonexclusive nature of some data that others may indepen-
dently access (for example, the ability of mobile applications to access the
contact list on a mobile phone). Linked to this is the temporal dimension
of data. It is often the case that data may decrease in value over time (such
as user location in a given time).
In addition, there are looming threats. Intel Corp. is reportedly partly
Copyright © 2016. Harvard University Press. All rights reserved.

funding “research into how to shield different apps on a smartphone so that

they can’t steal data from each other, or siphon off data from the phone’s
user.”74
Still, in relative terms, the super-platform is the alpha lion that pushes
away the other predators once the gazelle is ready to be consumed. As one
industry expert observed, “Apps are worth millions. Platforms are worth
billions. If you want to make money in mobile, build a killer platform.”75
Indeed, since 2008, Google reportedly (according to Oracle’s counsel in on-
going litigation) made $31 billion in revenue and $22 billion in profit out
of Android.76

To cement its leadership, the super-platform may engage in the defen-

sive practices of acquiring or blocking innovation or entry that might po-
tentially undermine its dominance. The European Commission in 2015
expressed concern over the growth, power, and influence of online plat-
forms. The Commission thinks “almost all areas of the economy will de-
pend on them in the near future.”77
Indeed, as platforms become super-platforms, network effects can better
insulate them from competitive pressure. The bigger the super-platform,
the greater the data-driven network effects, and the more difficult it may be
for competitive forces to displace it. For example, critics in the 1990s ar-
gued that Microsoft’s monopoly was likely to be short-lived. Twenty-five
years later, Microsoft still has a 90 percent share in operating systems for
desktop personal computers.78
The real threat to the super-platform generally comes from innovation
that disrupts the entire market. Returning to Microsoft, the threat to its
position came not from another PC operating system (such as Apple’s op-
erating system for its Macs). Nor was Microsoft’s market power weakened
by a competing browser (such as Netscape), word processing soft ware
(such as WordPerfect), or media player (such as RealNetworks). Micro-
soft’s power was eroded when users shifted from personal computers to
tablets and mobile phones. Microsoft’s 2015 annual report discussed this
platform competition:
We derive substantial revenue from licenses of Windows operating sys-
tems on personal computers. We face significant competition from com-
peting platforms developed for new devices and from factors such as
smartphones and tablet computers. These devices compete on multiple
bases including price and the perceived utility of the device and its plat-
form. Users are increasingly turning to these devices to perform func-
Copyright © 2016. Harvard University Press. All rights reserved.

tions that in the past were performed by personal computers. Even if

many users view these devices as complementary to a personal computer,
the prevalence of these devices may make it more difficult to attract ap-
plication developers to our PC operating system platforms. Competing
with operating systems licensed at low or no cost may decrease our PC
operating system margins. In addition, some of our devices compete with
products made by our OEM partners, which may affect their commit-
ment to our platform.79

Many people are familiar with Microsoft’s Windows platform from their
use of PCs. But Microsoft—while dominant on the old platform (PC

operating systems)—is having a harder time competing against Apple’s and

Google’s mobile and tablet platforms, and in getting Android and iOS users
to switch to Microsoft tablets and smartphones. Once users are locked into
a platform, it is often hard to switch them to another platform. As Micro-
soft reports, “Users incur costs to move data and buy new applications
when switching platforms.”80
Moreover, if data and data-driven network effects provide a competitive
advantage in these online markets, then the super-platforms may take their
increasing profits to expand their platform to sweep in more data and
advertising revenue. Here we may see the super-platforms prevent others
from tracking and profi ling the gazelles while they are on their websites
and apps, thereby weakening the Frenemies’ ability to usurp the super-
platform’s throne. Thus the availability of data plays a central role in the
Frenemy scenario’s capture phase.
Finally, the cost of harvesting, storing, and processing the data, while
declining over the years, still plays a role in the ability of firms to engage in
these strategies. All in all, these variables affect the dependence and interde-
pendence between market participants. Ultimately, they affect the competi-
tive dynamics: Will firms seek to improve our welfare or will they cooperate
to extract our data, and compete in the ensuing feeding frenzy?

Reﬂections
A rapid shift from competitors to collaborators and a mixture of vertical
and horizontal effects characterize the complex Frenemy dynamics. Com-
petition under our Frenemy scenario changes constantly and is dependent
on the relative market and bargaining powers of the independent app de-
velopers and super-platform operators.
To conclude, we will not necessarily benefit as the super-platform ex-
Copyright © 2016. Harvard University Press. All rights reserved.

pands. Superficially, competition may appear robust. After all, Android

users can choose from over a million apps. Many apps are free. Thus, An-
droid users seemingly are benefiting from a greater choice of free and low-
priced apps. But the incentives of the super-platform and its ecosystem of
app developers will not always align with our incentives.
Where the incentives diverge, we will be harmed—often with less inno-
vation and privacy protection. Yes, we may get more apps, but none will
offer solutions to our privacy concerns. Thus there will be growing res-
ignation that tracking and behavioral advertising are inevitable. Some

might accept this as a measure of progress. But this progress is at our

expense. While the free app developers are seemingly competing for our
attention, we will continue to pay more—with our data and privacy.
Moreover, an advertising-supported platform, as it pumps consumer data
through its ecosystem, will leave us more vulnerable to malware and other
cybercrime, including identity theft.81
Copyright © 2016. Harvard University Press. All rights reserved.

“Why Invite an Arsonist to Your Home?”

Understanding the Frenemy Mentality

O UR FRENEMY SCENARIO, as we have seen in Chapters 14 and 15, has mul-

tiple dimensions. One dimension is when the independent app devel-
oper competes against the super-platform (such as Uber competing against
Google’s driverless cars). Another dimension is when advertising revenue
supports the ecosystem. Here, the independent apps and super-platform, in
the extraction phase, cooperate to better track, profile, and target us with
behavioral ads; in the capture phase, they compete over the spoils. In both
Frenemy scenarios, the power has shifted to the super-platforms.
As the super-platform’s dominance increases, it benefits from its unri-
valed ability to control the consumer’s experience. Examples range from
Microsoft’s control over a significant proportion of personal computer op-
erating systems, to Google’s and Apple’s control of the smartphones oper-
ating super-platform, and to Amazon’s control over third-party retailers
on its super-platform.1 The super-platform determines who can join its
platform, which apps are featured in its app store, which apps are pre-
loaded on the smartphone, and the product’s default settings. Apps can
be promoted and demoted. In short, the super-platform becomes the
Copyright © 2016. Harvard University Press. All rights reserved.

gatekeeper.
This chapter examines, through the tale of two apps, the Frenemy social
structure. The super-platform sets the rules for getting on, being promoted
within, and getting kicked off the platform. The independent apps co-
operate under the single leader, the super-platform. They live by the
super-platform’s rules and informal norms. Lines are clearly defi ned. The
super-platform eats first; no one can interfere with the hunt of the gazelles.
And certainly any independent app that tries to help the gazelles will be
kicked off the platform. The asymmetric power and bargaining between
178

the independent apps and the super-platform are understood. The super-
platform controls the independent apps’ access to users and their data; it
can control (and cut off ) the smaller independent apps’ oxygen supply. The
super-platform’s ecosystem will develop in a controlled manner in line
with the super-platform’s strategic goals.

A Tale of Two Apps

To illustrate the “Frenemy” social structure, we look at the fate of two in-
dependent apps: Brightest Flashlight Free and Disconnect. We ask you to
guess which app was kicked out of the Google Play Store.
The first app, Brightest Flashlight Free, was simple. It turned on all the
available lights on the Android smartphone to make it a flashlight. The free
app was popular. The Google Play application store ranked the app in May
2013 “as one of the top free applications available for download. Users
have downloaded the Brightest Flashlight App tens of millions of times
via Google Play.”2 Unbeknownst to its millions of users, however, the app
secretively tracked its users’ precise location, which it then sold to third
parties, including advertising networks.3 The Federal Trade Commission
sued the app developer for its deception, and the app developer settled.
The second app, Disconnect, afforded users greater control over the ex-
tent to which they were tracked while surfing on the web. Disconnect gave
the following example: In visiting the Financial Times newspaper’s website
on an Android phone, users will be tracked by seventeen separate networks.
Seven of these requests come from sites and ser vices that invisibly track
consumers as they use applications and browse the web, in order to form a
comprehensive profile of their personal information. Advertising compa-
nies, Disconnect stated, “use these invisible connections to track” us as we
browse the web or open other mobile applications. They collect personal
Copyright © 2016. Harvard University Press. All rights reserved.

information about us, create a “profile” of each of us, and make money tar-
geting us with behavioral advertising. Increasingly, these invisible connec-
tions (including those set up by advertising companies) are being used even
more maliciously—by cybercriminals—to distribute malware, steal confi-
dential personal and business information, damage property, and engage
in identity theft.4 Disconnect’s app revealed and blocked this secretive
tracking.
So which app did Google kick off its platform? The flashlight app, which
surreptitiously tracked users’ locations, or the privacy app that gave users

greater control over their online privacy? The answer is Disconnect’s pri-
vacy app. The flashlight app remains on the Google Play store, and is still
tracking users’ locations—even after the FTC enforcement action.
Now, like many stories, this one has different ways of being told. Google
explains that Disconnect was “disconnected” as it infringed Google’s pri-
vacy rules, acted as a free rider, and changed users’ settings without their
consent. Disconnect tells the story of a small app that was kicked out for
being innovative and disrupting the super-platform’s extraction agenda.
We want to make clear that we have as of June 2016 only Disconnect’s
complaint to the European Commission. Google has not publicized its re-
sponse to the Commission. Nor has the European Commission made its
findings, if any, public. Moreover, the Commission may find that Discon-
nect’s version of facts, even if true, does not make out an antitrust claim. We
therefore do not argue in favor of or against any of these companies. Ulti-
mately, we leave it for the reader to decide which account is more convincing
or reflective of reality. With that caveat in mind, we use this tale of two
apps to explore the possible incentives and disincentives in a data-driven
Frenemy dynamic, where the super-platform and many of its independent
apps have a dual strategy of extraction and capture.

The Brightest Flashlight Android App

Even though it was developed for Google’s Android operating system, and
even though tens of millions of users downloaded the app, neither the app
developer, Goldenshores Technologies, nor Google told users that the free
flashlight app tracked the smartphone’s precise geolocation “along with
persistent device identifiers that can be used to track a user’s location over
time” and shared the phone’s location with third parties, including adver-
tising networks.5 Instead, the app deceived its millions of users.
Copyright © 2016. Harvard University Press. All rights reserved.

The penalty for such deception is telling. Some people opined that the
app developer should have been criminally prosecuted, have paid a civil
fine, or provided consumers with restitution. That never happened. In-
stead, the FTC’s punishment was weak. This reflects in part the FTC’s lim-
ited power in this area. The FTC noted that “it would be very difficult to
calculate each consumer’s monetary loss,” and that it lacks authority to
prosecute criminally or issue fines or civil penalties under the circum-
stances of the case.6 Instead, Goldenshores agreed, among other things, to
no longer misrepresent how consumers’ information was collected and

shared. Goldenshores also committed to disclose to users when, how, and

why their geolocation information was being collected, used, and shared,
and to obtain the users’ affirmative consent before doing so. The app maker
had to delete any preexisting personal information it had collected.
It is hard to see what the FTC action actually accomplished. Most strik-
ingly, Goldenshores was not prevented from tracking its users in the future.
The FTC only required the company to obtain users’ “affirmative express
consent.”7 It is questionable whether Goldenshores is even complying with
this requirement. Unless Goldenshores provides additional disclosures
when you download its app, it appears from its privacy policy in November
2015 that Goldenshores still automatically collects your geolocation data.
It remains incumbent on you to click the link to Goldenshores’ privacy
policy, which tells you that you must opt out of tracking by using your
settings within your device. (The app developer’s privacy statement never
explains how to do this.)
The FTC also required Goldenshores to tell you how your geolocation
information may be used. Goldenshores must tell you why its flashlight app
is accessing your geolocation information. It must tell you the “identity or
specific categories of third parties that receive geolocation information
directly or indirectly from such application.”8 Goldenshores’ privacy state-
ment does not even satisfy this lax requirement. All we know from Golden-
shores’ privacy statement is that your geolocation data is sent to “third-party
service providers” who are “persons or entities that provide advertisements
to you during your use of the Brightest Flashlight® software.”9 Goldenshores,
in its privacy statement, never identifies who exactly is getting your geolo-
cation data, how they are using your data, or whether the advertisers are
passing your data to others in the food chain.
Goldenshores was clearly at fault for its deception. Users never knew
that this simple app was tracking their location for advertising purposes.
Copyright © 2016. Harvard University Press. All rights reserved.

Nonetheless, Goldenshores is still collecting users’ geolocation, even

though this data is wholly unrelated to its app’s purpose—namely, to pro-
vide a shining bright light.
Putting aside the FTC’s ineffectual remedy, why didn’t Google punish
Goldenshores? Google could have prevented this deception. It could have
limited tracking only when necessary for the app to function. It could have
required every app to expressly obtain the user’s informed consent before
tracking (and thereafter give users the ability to turn off the tracking
feature). Google never required this.10 Even after Goldenshores’s deception

came to light, Google never kicked the app off its platform. Google had this
power. Goldenshores violated the super-platform’s stated policy: Google
requires that apps for its Google Play store “must not contain false or mis-
leading information or claims in any content, title, icon, description, or
screenshots.”11 Google notes that its policies for app developers “play an
impor tant role in maintaining a positive experience for everyone using
Google Play.”12 Android users could choose other apps that offered this
flashlight ser vice. So why is Goldenshores Technologies tracking its users,
and why is the super-platform allowing this?

Advertising and Privacy

Here we see how Frenemies both compete and share data. Goldenshores
Technologies is privately owned, so we do not know exactly how it makes
money. Since its app was free, the company likely made most, if not all, of
its revenues from advertising. And despite the FTC action, Goldenshores
still does not reveal the identities of the advertisers and advertising net-
works to which it is sending its users’ location.
One likely recipient is Google. Google, like Brightest Flashlight, derives
most of its revenues from advertising. In 2015, nearly 90 percent of Google’s
revenues came from advertising.13 Google’s advertising revenues have in-
creased in recent years—from $43.686 billion in 2012, to $51 billion in
2013, to $59.6 billion in 2014, to $67.39 billion in 2015.14 Google controls
the world’s largest mobile advertising network. Over 650,000 apps, as of
mid-2015, used Google’s AdMob.15 As Google’s website states: “With the
largest source of global advertiser demand, flexible ad controls, and an
industry-leading mediation ser vice, AdMob is the best platform to mone-
tize your apps and maximize your ad revenue through app advertising.”16
Copyright © 2016. Harvard University Press. All rights reserved.

Google’s Doubleclick business also promotes mobile advertising—from

helping other companies design their mobile ads and placing them on apps,
to measuring how customers respond to those ads.
Europe’s privacy advisory process describes the interaction among the
advertisers, publishers (like Goldenshores), and advertising networks (like
Google’s):
[T]he publisher reserves visual space on its website to display an ad and
relinquishes the rest of the advertising process to one or more advertising
network providers. The ad network providers are responsible for distrib-
uting advertisements to publishers with the maximum effect possible.

The ad network providers control the targeting technology and associ-

ated databases. The larger the advertising network, the more resources it
has to monitor users and “track” their behaviour.17

In this ecosystem, online advertisers generally pay only when the user
clicks on the ad. As Google asks, “Want extra revenue from your website?
Google AdSense shows timely and relevant ads alongside your own online
content—and pays whenever someone clicks.”18 That changes Golden-
shores’ incentives. As part of their joint “extraction” strategy, Goldenshores
and the other advertising-dependent apps will track users, help others
track their users, collect personal data, and share that data if that increases
the likelihood of users clicking the ads. Likewise, when the super-platform
controls the largest mobile advertising marketplace, its incentives change:
It wants to help advertisers and publishers better target us with ads that
we will likely click. Once we click the ad, Google gets paid, and publishers
like Goldenshores get their cut. The super-platform’s concern about its users’
privacy lessens.
Thus, Google’s and Goldenshores’ interests are aligned in that both want
users to click ads. To increase the chances that users will click an ad on the
flashlight app, both Google and Goldenshores must predict which ads will
be more appealing. The better the data they collect, the better the algo-
rithms can predict which ad will most likely be clicked at that particular
place and time. Google and the app developer want to promote the flow of
personal data to know where the users are, what they’re doing, and what
they’re interested in, and to target them with an ad that will likely appeal
to them at that moment. Smartphones add another important dimension
to behavioral advertising—namely “geofencing,” by which advertisers can
use one’s physical location to target ads.19 So when a flashlight app user is
fumbling for his keys at 11 p.m. in a parking lot, which is later than his usual
Copyright © 2016. Harvard University Press. All rights reserved.

hour, the ad network and publisher can use the data to target him with an
ad for a nearby Taco Bell.
Goldenshores is not alone. One 2010 study examined the United States
fift y top websites, which accounted at the time for about 40 percent of the
web pages viewed by Americans. On average, each website installed “64
pieces of tracking technology onto the computers of visitors, usually with
no warning. A dozen sites each installed more than a hundred.”20
Tracking by cookies or other means has favored larger ad networks. The
networks have agreements with websites that publish their ads. The
bigger the ad network, the more likely that you will browse websites with

relationships with the ad network, the greater the ad network’s potential to

track you with these third-party cookies, and the more detailed its profile
on you.21 Thus, the super-platform, in controlling an ad network, coordi-
nates with the advertisers and publishers of ads to track you and target you
with personalized ads.
To see where the data is flowing, another study looked at the hundred
most popular Android applications and the hundred newest applications
in each of the categories available.22 The researchers wanted to know the
apps’ friends, namely the other websites to which many of these apps con-
nected. Nine of the ten most popu lar domains the apps frequented were
various web ser vices run by Google.23 The most popular domain was dou-
bleclick .net, Google’s advertising platform, which tracks end users and
serves up advertisements. Thus, the study found that, while “Google does
not directly make any revenue from Android itself (which is openly li-
censed to manufacturers), it is able to extract revenue from the ads busi-
ness around the ecosystem.”24 The study found “(i) that a significant number
of applications, some highly rated, download an excessive number of adver-
tisements which indicate that users may not be as sensitive to advertisements
as anecdotally conjectured; (ii) a large number of applications communicate
with a multiplicity of online tracking entities, a fact to which users may not
be aware; and (iii) . . . some applications [were] communicating with web-
sites that have been deemed malicious by malware detection engines.”25
If Google and Goldenshores are friends when seeking to extract data
from consumers, who then is their enemy? Any company that helps users
prevent invisible tracking and impede the extraction of information about
them. This brings us to Disconnect.

Disruptive Dynamic
Copyright © 2016. Harvard University Press. All rights reserved.

Disconnect’s cofounder was an engineer at Google. In 2010, he read how

the most popular apps on Facebook “were transmitting users’ identifying
information to dozens of advertising and internet tracking companies,
without disclosure or permission.”26 So he went home and wrote an “ex-
tension” for Google’s Chrome browser, which blocked connections between
third-party sites and Facebook servers without interfering with the user’s
connection to Facebook.27 Within two weeks, his extension was down-
loaded 50,000 times.28 After realizing that his own employer was among
the larger collectors of the personal data that his extension was intended to
protect, he left Google in November 2010 to focus on online privacy.29

Disconnect offers four privacy functions: the ability of users to see other-
wise undisclosed web tracking and privacy policies, virtual private net-
working (VPN) technology, private search, and private browsing.30
Unlike the flashlight app, Disconnect does not surreptitiously collect
users’ geolocation information. Nor does it track users’ activities across the
web. Unlike Google and the flashlight app, Disconnect does not rely on
advertising revenues. It does not sell its user data to advertisers, ad net-
works, or data brokers.31 Disconnect makes its money “by selling its prod-
ucts to users.”32 The basic version of its mobile app is free; its “Pro” version
has a one-time $40 fee; and its “Premium” version costs $50 per year.33
After launching its mobile app, Disconnect received on August 26, 2014,
an e-mail from the Google Play team. Google had removed the app from
its Play Store because the app “interferes with or accesses another ser vice
or product in an unauthorized manner.”34 As Google warned the app
maker:
All violations are tracked. Serious or repeated violations of any nature
will result in the termination of your developer account, and investiga-
tion and possible termination of related Google accounts. If your account
is terminated, payments will cease and Google may recover the proceeds
of any past sales and/or the cost of any associated fees (such as charge-
backs and transaction fees) from you.35

Disconnect was popular. As the Wall Street Journal reported, “In the six
days it was available in Google’s store, it was downloaded more than 5,000
times.”36 Google readmitted Disconnect, only to kick it out again.
Disconnect tried to make its app compliant with Google’s rules. But
Google’s policies were “so vague that Google could, in essence, ban any app
in its store.”37 As Disconnect’s cofounder said, “It’s like a Kafka novel—
you’re getting kicked out or arrested for reasons you don’t even know.”38
Copyright © 2016. Harvard University Press. All rights reserved.

Disconnect was not alone. Google, according to the Wall Street Journal,
had removed other ad-blocking apps, such as Adblock Plus, from its Play
Store.39 While kicking out some apps that safeguarded users’ privacy by
preventing tracking, Google did not kick out all apps promising to protect
users’ privacy.40
Disconnect eventually complained to the European Commission, con-
tending that Google had abused its dominant position. In its ninety-five-
page complaint, Disconnect questioned why Google does not protect
Android users from the risks associated with tracking. Google’s Chrome
browser, for example, does not provide details on which websites and web

ser vices respect “Do Not Track” requests or how they interpret them. As
Google reports, “most Web ser vices, including Google’s, do not alter their
behavior or change their services upon receiving Do Not Track requests.”41
Disconnect also explained how Google’s privacy features in 2015 were
weak. Google tells users they can opt out of targeted advertising. But app
developers could circumvent Google’s targeted advertising opt-out feature
in Android.42 Moreover, Google’s opt-out feature only stops Google from
showing you “interest-based” ads. Opting out did not stop ads altogether,
including ads based on your recent searches or general location.43 Opting
out would not disable other companies’ interest-based ads.44 Nor does the
opt-out automatically apply whenever you use other browsers on that de-
vice or other devices.45 Thus, you would have to opt out for each browser
you use on each PC, tablet, and smartphone. Nor will opting out keep
you opted out after you clear your browser’s cookies.46 Nor will it opt you
out of interest-based ads in ser vices where cookie technology may not be
available.47 Nor does the opt-out feature prevent Google or any other
company from tracking you; it only prevents a targeted ad under certain
circumstances.48
Google, for its part, rejected the complaint as baseless; Disconnect was
removed from the ecosystem as its blocking ser vices prevented other ap-
plications from legitimately earning money. That interference infringed
Google Play’s apps policy.49 Google noted that over 200 privacy apps that
do not infringe its policies are available in Google Play.50
After Google removed Disconnect from its Play Store, Android users
could no longer search for, or find, the privacy app in the Play Store, and
its downloads and sales for a premium version of the app suffered.51

Why Some Ecosystems Hate Privacy

For both the super-platforms and app developers whose business model is
dependent upon tracking us and using our data to target us with behav-
ioral ads, privacy technologies represent a real threat. Google, among
others, said so.
In its 2014 Annual Report, Google identifies various risks that could
adversely affect its business, financial condition, results of operations, cash
flows, and the trading price of its stock. One risk is that “New technologies
could block online ads, which would harm our business.”52 As Google
explains,

Technologies have been developed that can block the display of our ads
and that provide tools to users to opt out of our advertising products.
Most of our revenues are derived from fees paid to us by advertisers in
connection with the display of ads on web pages for our users. As a re-
sult, such technologies and tools could adversely affect our operating
results.53

Facebook also warns investors that privacy innovations can threaten its
business model. Like Google, nearly all of Facebook’s revenue is generated
from advertising. For 2014, 2013, and 2012, advertising accounted for
92 percent, 89 percent, and 84 percent of its revenue, respectively.54 Among
the risks Facebook identifies are “the degree to which users opt out of so-
cial ads or certain types of ad targeting; the degree to which users cease or
reduce the number of times they click on our ads; . . . [and] the impact of
new technologies that could block or obscure the display of our ads.”55
Besides individuals deleting cookies or using “ad blocking” software that
prevents cookies from being stored on a user’s computer, Coupons.com
warns investors that even the privacy default setting can hurt its business:

[T]he Safari browser blocks third-party cookies by default, the developers

of the Firefox browser have announced that a future version of the Firefox
browser will also block third-party cookies by default, and other browsers
may do so in the future. Unless such default settings in browsers were al-
tered by Internet users to permit the placement of third-party cookies,
we would be able to set fewer of our cookies in users’ browsers, which
could adversely affect our business.56

So one can see why Disconnect is the pariah in the Frenemy scenario. It
is unwilling to help the other lions chase the prey—that is, to track con-
sumers’ activities and behavior, in order to better know their tastes, inter-
Copyright © 2016. Harvard University Press. All rights reserved.

ests, and intentions, and target them with the right ad at the right place and
time. Disconnect was not interested in fighting over who gets the choice
cuts of the advertising revenue. Instead, it wanted to make money by
helping users avoid the lions altogether.
Is Disconnect the saint or a serial infringer? As we indicated earlier, the
Google/Disconnect clash is still under review in Europe. No formal deci-
sion, as of May 2016, was released; accordingly without all the facts, the way
the story should be read remains unclear. Our aim is not to discredit either
party, but to use this friction to illustrate the complex Frenemy dynamics.

Apart from the Google/Disconnect story, privacy issues made the news
in August 2015, when Apple announced its implementation of a new pri-
vacy feature. Its updated iOS would permit iPhone and iPad users for the
first time to download apps that block ads when using Safari.57 It is remark-
able how such a small privacy measure is so newsworthy. Users, for years,
could employ ad blockers on their Macs and MacBooks when browsing the
web using Safari. The technology existed, and so too did the demand. The
benefits of blocking ads are far greater on smartphones than on personal
computers. Ad-blocking technology can reduce the clutter on our smart-
phones’ small screens, help pages load faster (four times faster in one test),
and save data (53 percent less in one test).58 Nonetheless, we had to wait
until 2015 for one of the two super-platforms to even permit ad-blocking
technology on their mobile phone platform. Even here, Apple is treading
lightly. It will not preinstall the ad-blocking technology. Rather, we must
fi nd and download the browser extension. Thus, by requiring users to
opt in, Apple can mollify those firms that are dependent on advertising
revenues by pointing out that relatively few iPhone and iPad users will
likely download the technology, and that consequently fewer ads will be
blocked.

Reﬂections
When we raised the Google/Disconnect scenario at one conference,
someone responded, “Isn’t this like a homeowner inviting an arsonist to
her house?” In other words, why should Google accept an app that would
undermine its own and its ecosystem’s advertising-fueled business model?
Should the law penalize a super-platform for protecting its commercial
interest?
The answer will depend, among other things, on one’s view of the role of
Copyright © 2016. Harvard University Press. All rights reserved.

competition law, and the particular market’s dynamics and structure.

Some would oppose antitrust intervention, citing the risks of chilling
competition and undermining the incentives to invest in research and de-
velopment. They would cite a 1945 case where the United States success-
fully prosecuted a monopoly: “The successful competitor, having been
urged to compete, must not be turned upon when he wins.”59 Accordingly,
an outright or de facto refusal to deal, even by a dominant platform, should
not justify governmental intervention.60 Instead, the independent app de-
velopers and outlets should strive to create independent demand for their
products and ser vices. That demand will improve their bargaining position

in its relationship with the super-platform and reduce their dependency.

They, like Facebook, could become their own super-platform. Even if the
independent app developer is squeezed out of the market, it will be hard
for the government and courts to distinguish whether that outcome did or
did not reflect healthy competitive dynamics—namely better-quality prod-
ucts pushing out inferior products produced by less-efficient players.
Proponents of intervention would focus on the special responsibility that
a super-platform has. They would question the ability of a small indepen-
dent app to fight against a giant when the latter takes action to undermine
competition. Even if the super-platform would not immediately kick the
app out of the ecosystem, the super-platform can hinder the app’s ability to
reach users, eventually leading to its exit. Th is, arguably, is an abuse of
power. We saw this in the 1990s when independents—such as the Netscape
browser, WordPerfect, and RealNetworks—struggled to compete against
the dominant Microsoft Windows platform. The focus here is not only on
monopolistic abuses that directly harm consumers, but also on abuses that
distort competition and thereby harm consumers.61 Intervention may be
required to protect efficient competitors from being pushed out of the
market and innovators seeking to enter the market.
What is clear is that the independent app developers face a perilous land-
scape when the super-platform controls their oxygen supply. When the
super-platform competes against the app, its incentives change, and the
“Frenemy” can bare its teeth. If the independent app threatens the super-
platform’s and other apps’ source of revenue, it too can be kicked off the
platform.
The independent apps cannot rely on the antitrust laws to protect them
from the super-platform’s anticompetitive abuses. The super-platforms
have little to fear from the antitrust enforcers (at least in the United States).
While running for president, Barack Obama criticized the Bush Adminis-
Copyright © 2016. Harvard University Press. All rights reserved.

tration for having “what may be the weakest record of antitrust enforcement
of any administration in the last half century.”62 Obama noted that “in seven
years, the Bush Justice Department has not brought a single monopoliza-
tion case.”63 Obama promised to “reinvigorate antitrust enforcement” and
“step up review of merger activity.”64 Now, with his second term coming to
an end, the Department of Justice has brought only one monopoly case; and
the same criticism of feeble antitrust enforcement has been made about the
Obama Administration.65 Nor can the independent app afford a costly,
time-consuming private lawsuit against the super-platform. The outcome
is often uncertain, and the super-platform can retaliate subtly.

Instead, the independent app must somehow scale up to keep ahead of

rival app developers; but as the app grows in usage and profits, it will likely
catch the super-platform’s attention. In such a case, the independent app
developer can assess its strategies: it can align itself with another, friendly
super-platform; it can try to become its own super-platform; or, the more
attractive (or viable) option, it can be acquired by the super-platform (as
Waze was by Google). An acquisition at the peak of its value would maxi-
mize the independent app’s profits. It would also improve its horizontal po-
sition in the downstream market. Other wise, its value will likely diminish
by the entry of the super-platform (for example, MapQuest). Long term, the
Frenemy dynamics may foster vertical consolidation, which would further
enhance the super-platforms’ power. With each acquisition, more personal
data flows through the super-platform. Moreover, since the Frenemy sce-
nario defies the competition authorities’ horizontal, vertical, interlocking,
and conglomerate categories, the acquisitions by the super-platforms would
likely escape antitrust scrutiny.
With that in mind, it will be interesting to see Disconnect’s or Uber’s
ultimate fate. Can Disconnect survive outside the super-platform? How
will that affect others investing in privacy apps? Will Uber continue to
scale up and expand to become a super-platform? Will it be acquired—like
the independent navigation app Waze—by the super-platform? Will it
switch allegiances to another platform? Or will it eventually wither away,
to join MapQuest, WordPerfect, RealNetworks, and other innovators
eclipsed by the super-platforms?
Copyright © 2016. Harvard University Press. All rights reserved.

The Future of Frenemy:

The Rise of Personal Assistants
I am putting myself to the fullest possible use, which is all
I think that any conscious entity can ever hope to do.
—The HAL 9000 computer, in the 1968 fi lm
2001: A Space Odyssey

W HAT IS THE FUTURE of Frenemy? One likely path, we believe, will in-
volve the rise of digital personal assistants. All the super-platforms
are currently investing in this technology: Apple’s Siri, Amazon.com’s
Alexa, Facebook’s M, and Google Assistant.1 This personalized tool prom-
ises to interact with us in a human-like way, providing relevant informa-
tion and suggesting restaurants, news stories, hotels, and shopping sites. As
the artificial intelligence and communication interface advance, personal
assistants can offer an unparalleled personalized experience.
These developments are exciting. Personal assistants can not only provide
us with endless information, if we so desire, but can anticipate and fulfill our
needs and requests. They can do so in an intelligent manner, based on our
connections, data profile, behavior, and so forth. They can learn our needs,
communicate with us in our preferred language, and execute our commands
Copyright © 2016. Harvard University Press. All rights reserved.

at high speed. Our time will be too important to worry over life’s little de-
tails. As the personal assistant seamlessly provides more of what interests
us and less of what doesn’t, we will grow to like and trust it.
How does this relate to our Frenemy dynamic? As this chapter illustrates,
the rise of the digitalized assistant, driven primarily by the super-platforms,
may heighten the anticompetitive dynamics we have explored thus far: our
personal butler may, unbeknownst to us, encourage tacit collusion. It may
help bring retailers closer to perfect behavioral discrimination. And it may
consolidate the super-platform’s power, as it obtains greater control over
191

what we see and what we purchase. The more we rely on our personal
assistant, the less relevant our outside options become, the more depen-
dent we, and the independent sellers, become on the super-platform, and
the greater the super-platform’s power to exclude others and control our
virtual universe.

The Rise of Personal Assistants

Facebook in 2015 announced a beta version of M, its digital assistant, which
can replace most of one’s web searches and apps with a chat app on Face-
book Messenger.2 Like Apple’s Siri, Microsoft’s Cortana, Amazon.com’s
Alexa, and Google’s voice-recognition system, Facebook’s technology re-
lies on machine learning. The super-platforms’ plans are clear: they “envision
a future where humans do less thinking when it comes to the small decisions
that make up daily life.”3 In 2016 Google showed a video of a suburban family
undergoing its morning wakeup routine: “The dad made French press
coffee while telling Google to turn on the lights and start playing music in
his kids’ rooms. The mom asked if ‘my package’ had shipped. It did, Google
said. The daughter asked for help with her Spanish homework.”4
The future looks bright. But behind the dream of the personal helper lies
a sophisticated machine that may eventually undermine our welfare. We
see two important shifts with the rise of the digital personal assistants.
First, technology is shifting from being reactive to becoming proactive—
anticipating our needs and wants, rather than following instructions. One
example is takeout. Twenty years ago we looked at menus in our kitchen
drawer or in the yellow pages. In 2016, absent a steady favorite, we likely
choose the type of food and then search online for nearby restaurants. We
review how others rated the restaurants, and read the menus online. After
deciding which dishes we want, we place our order with the restaurant and
Copyright © 2016. Harvard University Press. All rights reserved.

pay online or on delivery.

In this domain, the super-platform points us to other sources, which
may lead us to additional sources, and ultimately our end choice. Google,
through its search engine, was, and remains as of 2016, the primary portal
for the web. Google also dominates through its search engine “direct re-
sponse” advertising, “which is the kind of ad that pops up when we are
searching for an airline ticket, a new laptop or any other purchase.”5 So, as
we saw in the preceding chapters, the super-platform may try to influence
our decision while we undertake each step in ordering takeout.

Now imagine a new technology that can enable us to accomplish so

much more with the assistance of artificial intelligence. Our personal as-
sistant reduces the number of steps we take, online and offl ine. Imagine
you lived in an estate—like Downton Abbey. You would not want the butler
to direct you to the livery of junior servants for each request you make. The
butler instead anticipates your wants, and quietly orchestrates the servants
to fulfill these tasks. Your shoes are already polished; the coffee with the
right amount of sugar and cream is beside the freshly squeezed juice. Your
car and driver are already waiting. The warmed clean towel is folded by the
shower. The scent from the freshly cut roses wafts down the hallway. Your
children are busily practicing their French while drawing landscapes.
So too the AI personal assistant will become our primary interface.
Based on our personal data, including our calendar, texts, e-mails, and geo-
location data, our personal assistant may recognize a busier than usual
day. From our phone’s geolocation data, it will know when we are heading
to our car. Our personal assistant may suggest, “How about treating your-
self to Chinese tonight?” Our personal assistant might recommend a
popular place. It might then direct the order to a restaurant it believes we
would like, arrange for the food’s delivery shortly after we arrive home, and
pay for the food. All we need to do is grab the food at the door. So like a
good butler, our personal assistant seamlessly anticipates and satisfies our
needs, condensing all the steps to one or two commands. Thus each super-
platform will seek through its voice- or text-based interface to become the
first, and only, place we will go.
Second, the stakes are huge. Google, Apple, and Facebook are currently
jockeying as to “who gets to control the primary interface of mobile de-
vices.”6 As we shift from a mobile-dominated world to an AI-dominated
platform, we will converse primarily with our head butler, who increas-
ingly predicts and fulfi lls our needs, and we will less frequently look at
Copyright © 2016. Harvard University Press. All rights reserved.

price-comparison websites, search the web, or download apps. Take, for

example, the Google assistant, which forms part of the company’s “effort
to further entrench itself in users’ daily lives by answering users’ queries
directly rather than pointing them to other sources.”7 Why fritter our time
away on such minor details, when we can use it more productively? Or we
can dream of the fly-fishing vacation our head butler arranged at a Scot-
tish castle, where no doubt we’ll be sipping eighteen-year-old Bunnahab-
hain single-malt Scotch whiskey with our Cohiba cigars, while gazing at
the fireplace.

Could a Personal Assistant Be So Devious?

Who wouldn’t want a personal butler? The idea of an intelligent, voice-
activated helper is alluring. And yet, when considering its possible stra-
tegic usage by the winners in the Frenemy dynamic, our butler’s interests
may not always align with our interests. Our new trusted alter ego, to which
we outsource our decision making, may be charming, but partial. After all,
as we learned earlier, being the “ free” part of a multisided market, we don’t
directly pay for the butler’s ser vices. Our butler must ultimately cater to the
needs of its real employer—the super-platform. Of course, we can still ben-
efit when the super-platform’s interests are aligned with our own. But we
may often be unaware of when such alignment is absent.
So what differentiates the strategies discussed earlier in our Frenemy
scenario and the future use of personal assistants? We identify several
key differences that affect the user experience and will likely act as “game
changers”—transforming our competitive environment.
First, we note the changing interaction with the new digital helpers.
Because it is human in its communications, charming in its demeanor, and
funny—at just the appropriate level—we will grow to trust our companion,
which has been privy to so many of our activities. Many of us already trust
our favorite search engine to find the relevant results for our inquiries,
Facebook to identify relevant news stories, Amazon for book recommen-
dations, and Siri to place phone calls, send text messages, and find a good
Chinese place nearby. So with an eager (and free) butler whose capacity to
help us improves, we will increasingly rely on it. At first the choices will
seem benign, such as asking the personal assistant for today’s weather
rather than searching the web. But the more tasks the personal platform
undertakes, the more we rely on the super-platform’s functions (such as its
texting app, maps, and so forth). That trust, in competitive terms, can
Copyright © 2016. Harvard University Press. All rights reserved.

easily translate to our willingly being locked in—that is, willingly forgoing
opportunities to independently consider outside options.
Second, the more we communicate only with our personal assistant, the
less likely we will independently search the web, use price-comparison
websites, seek independent customer reviews, and rely on other tools. The
ease of voice activation and verbal communication with our butler may
limit our view of the available outside options.
Today when an app is preloaded on our smartphone and integrated with
the phone’s other functions, few of us download a competing app. As

EU Commissioner Vestager noted, “if Google’s apps are already on our

phones when we buy them, not many of us will go to the trouble of looking
for alternatives. And that makes it hard for Google’s competitors to per-
suade us to try their apps.”8 Many of us stick with the default option. Like-
wise, as our personal assistant becomes our default, so too will its plat-
form’s other features and functions. To illustrate, suppose we are sipping
our single-malt Scotch in our summer retreat. We may ask our assistant
how long a drive it is to a friend’s country home, and then ask our assistant
to text our friend our ETA. We would be unlikely to pull out our Android
phone, search MapQuest to figure out the distance and time to drive, and
then use Messenger to text our friend.
The removal of the human element from the search activity, and partly
from the decision making, transfers more power to the super-platform. The
personal assistant will use its own tools and may exercise its own judgment
as to prioritizing and communicating the results.
Third, the scope of data and the personalization that follows will make
it harder for us to switch assistants. As Google’s CEO characterized its per-
sonal assistant, “We think of it as building each user their own individual
Google.”9 The super-platforms, given the scope of data, opportunities to
experiment and learn by doing, and control over key technologies (such
as maps), can already provide us with a personalized experience that
smaller providers are unlikely to match. Once we choose and train a head
butler, we may tolerate mistakes rather than train a new butler from an-
other super-platform.
The super-platform—through its butler—will benefit from unparalleled
access to our data. As we noted earlier, the super-platforms already spend
a lot of time, money, and effort to track our behavior and profile us. Now,
by providing us a butler, the super-platforms will collect even more data.
As repeatedly noted by developers, the hope is for these tools to accompany
Copyright © 2016. Harvard University Press. All rights reserved.

us in our decision making, encouraging ongoing daily interaction from

chats to shopping. As a result, the information gathered about our needs
and desires will be significantly enhanced—feeding the Big Data and Big
Analytics machines.
As the super-platform’s power increases, so does the risk of anticompeti-
tive activities. Indeed, the personal assistant can magnify the harmful ef-
fects of each of our three scenarios.
Think, for example, of our behavioral discrimination scenario. Our per-
sonal assistant will accumulate increasing information about us and will

be aware of the extent to which we venture out and seek other options. Its
aim is to deliver the right product or ser vice at a price we are willing to pay.
So the line between personalization and behavioral discrimination will
blur. As we increasingly rely on the personal assistant for suggestions, it can
increasingly suggest things or ser vices to buy, and the price it has success-
fully negotiated. While helping our son with his Spanish, our personal as-
sistant might suggest a particular app or private tutor that tremendously
helped other students struggling with the same issue. Because the tutoring
is customized for our son, it will be harder to assess whether the price the
tutor charges is the fair market price or simply a price we would tolerate.
Moreover, if the tutoring ser vice is helping other children improve their
grades, we would not want our child to be at a competitive disadvantage—
especially if we are all eyeing the same highly selective universities. So
the personal assistant can prompt purchases that we other wise wouldn’t
consider.
Now let’s think of our collusion scenarios facilitated by smart algo-
rithms. As our digital assistant increasingly orchestrates what we purchase
at what time, it can easily become the hub upon which the spokes rely. Sup-
pose the personal assistant gets a commission for every sale. Higher prices
mean higher commissions. The more the personal assistant is plugged into
our lives, the harder it will be to circumvent the assistant in finding a lower
price. And when we independently find a lower price, our butler can always
surprise us with a special deal, whose aim is to punish any discounter and
undermine any attempt to cheat on the tacitly established price.
Finally, the harms we identify in our Frenemy scenario are amplified.
While we would each enjoy our personalized experience and growing inti-
macy with our digital personal assistant, its roots and loyalty remain with
its real master. As more people rely on the super-platform’s personal as-
sistant for their day-to-day activities, so too will sellers gravitate to the
Copyright © 2016. Harvard University Press. All rights reserved.

super-platform. The punishment for being kicked off the super-platform is

severe. Without access to the personal data stream, the independent re-
tailer will have a harder time identifying a customer’s key purchasing mo-
ment (like when he needs a new oxford shirt). Because the ads we see while
surfing the web will be orchestrated by the super-platform, it will be harder
for the retailer to reach that customer. Even if the retailer can reach the
customer, it cannot provide the increasingly customized products or ser-
vices (such as tailored shirts in the styles and colors that appeal to the
customer). And even if retailer can gain the customer’s attention, the per-

sonal assistant may interject with its own recommendation, suggesting

that he consider a special deal by another haberdashery, one that is part of
the super-platform’s ecosystem. In this multisided market, the assistant
may subtly push certain products and ser vices and degrade or conceal
others, all in the name of personalization.

Control of Media and Mind

As we increasingly rely on our personal assistant, it will increasingly learn
about our social and political views, behavior, and susceptibility to biases. It
will become more proactive—making recommendations on entertainment,
commenting on the music we listen to or the books we are reading. By com-
plimenting and cajoling, sharing thoughts with us on recent events, sending
personalized notes on special occasions, reminding us of presents, sug-
gesting popular gifts trending among the recipient’s friends, and informing
us about information from our smart meters and smart sensors, it will in-
grain itself in our lives. We will wonder how we ever managed without a
digital personal assistant.
While we appreciate this free ser vice, we will not know its exact cost.
When it joins our chats to make suggestions, or at times makes suggestions
counter to those made by other helpers, we may not know whether it is
being helpful or simply manipulating our behav ior. It may work in the
background to undermine attempts to expose us to competing products,
or it may monitor our chats for signs of discontent with the ser vice or dis-
count offered—signs of anger that should trigger a behavioral action. The
list is truly endless—all in the name of catering to our needs. After all,
happy users make happy super-platforms.
The Truman Show reemerges, only grander and more effective. In fact, it
will transcend the retail world and distinctly impact our worldview. The
Copyright © 2016. Harvard University Press. All rights reserved.

rise of the trusted personal assistant may provide the super-platforms with
the ultimate power—to affect our views and the public debate. The reliance
on a gatekeeper may enable its operator to intellectually capture users, and
subsequently decision makers, in an attempt to ultimately ensure that
public opinion and government policies align with the corporate agenda.
Consider, for example, the control our personal assistant may have over
our news feed. Currently, the super-platforms do not report the news. But
many people rely on the super-platforms’ algorithms to fi nd news of in-
terest. One 2015 study found that 61 percent of Millennials in the United

States (those born between 1981 and 1996) were “getting political news on
Facebook in a given week.”10 This was a much larger percentage than any
other news source. A 2016 study found that Facebook “sends by far the
most mobile readers to news sites of any social media sites”—82 percent of
the social traffic to longer news stories and 84 percent of the social traffic
to shorter news articles.11
Users rely on the super-platforms, in part, because they believe the algo-
rithms objectively identify the most relevant results. But, as we saw with
search engines, the super-platform can intentionally degrade its results to
promote its corporate interests. Thus we can see why conservatives were
concerned over allegations in 2016 that the social network Facebook ma-
nipulated for political purposes the rankings of news stories for its users,
suppressing conservative viewpoints.12 (Facebook denied doing this.) As
we saw in our behavioral discrimination, as prices become personalized, a
benchmark market price becomes more elusive. Likewise, personalization
of news feeds makes it harder to uncover censorship. Because you expect
other people’s news feeds to differ from yours, it is harder to determine
when (and why) a relevant story is buried.
Given our reliance on these information gatekeepers, the super-
platforms—in affecting our views of the world—can also influence elec-
tions. Jonathan Zittrain, for example, identified as a risk Facebook’s ability
to manipulate elections.13 He warned of the super-platform’s potential
ability to predict political views, identify party affi liation, and engage in
targeted campaigning to mobilize distinct groups of voters to take action.14
Robert Epstein likewise pointed to the potential risk associated with on-
line search manipulation. He noted how Google could affect not only
commercial interests but also political agendas.15 His research illustrates
the potential to affect election results through web manipulation and
“boost the proportion of people who favored any candidate by between 37
Copyright © 2016. Harvard University Press. All rights reserved.

and 63 percent after just one search session.”16 By manipulating the rank-
ings of search results, the studies found, “Google’s search algorithm can
easily shift the voting preferences of undecided voters by 20 percent or
more—up to 80 percent in some demographic groups—with virtually no
one knowing they are being manipulated.”17
We already see instances in which platforms were used to promote cer-
tain agendas. Uber used its app to mobilize protests against the New York
mayor’s proposal for a cap on the number of available rides.18 Google used
its homepage to protest against the Stop Online Piracy Act (SOPA), asking
users to petition Congress.19

As the personal assistant expands its role in our daily lives, it can alter
our worldview. By craft ing notes for us, and suggesting “likes” for other
posts it wrote for other people, the personal assistant can effectively ma-
nipulate us through this stimulation. “With two billion ‘likes’ a day and
one billion comments,” one doctor noted, “Facebook stimulates the release
of loads of dopamine as well as offering an effective cure to loneliness.”20
Imagine the dopamine spike when the personal assistant secures a new
record of “likes” for a political message it suggested that you post. Others
do not know that your digital assistant was heavi ly involved in draft ing
your note. You don’t know the extent to which the personal assistant gen-
erated the likes. And none of us know how this note is helping sway the
public discourse in ways that benefit the super-platform.
As we increasingly rely on our personal assistant, we may not recognize
its toll on our well-being. As the personal assistant increasingly controls
mundane household tasks, like turning off lights, regulating room temper-
ature, and adjusting our water heater, it will be harder to turn off. But the
online (and increasingly offline) tracking of our behavior can impede our
creativity, solitude, and mental repose.21 George Orwell famously discussed
in 1984 how monitoring our behav ior can adversely affect intellectual
freedom: “You had to live—did live, from habit that became instinct—in
the assumption that every sound you made was overheard and, except in
darkness, every movement scrutinized.”22 As the U.S. President’s Commis-
sion on Law Enforcement and Administration of Justice wrote in 1967, “In
a democratic society privacy of communication is essential if citizens are
to think and act creatively and constructively. Fear or suspicion that one’s
speech is being monitored by a stranger, even without the reality of such
activity, can have a seriously inhibiting effect upon the willingness to voice
critical and constructive ideas.”23 Soon we may have our digital personal
assistant hovering nearby, without being conspicuous.
Copyright © 2016. Harvard University Press. All rights reserved.

The Purist Assistant

The predictions outlined in this chapter are driven by the assumption that
most of the future private assistants will be developed and provided by the
super-platforms and other powerful, vertically integrated firms.
Would there be any possible counter measure to these devious helpers?
We certainly hope so. We can anticipate a class of independent assistants,
developed by independent firms with our interests in mind. Our purist
assistant could warn us when behavioral discrimination is at play, when

outside options are ignored, when price alignment seems out of order, or
when our information is harvested. It may even be capable of deploying
counter measures to maximize our welfare in face of such strategies. It
could monitor our news feed and alert us if it has been affected. It will form
a true extension of our interest—aware of our preferences and safeguarding
our autonomy.
With the possibility for such a purist assistant, you may rightly ask why
our base assumption has been rather pessimistic and has led us to focus on
the devious assistant? Perhaps the simplest way to explain our prediction is
to ask you, our reader, to consider the following: Which search engine did
you use today? Did you opt for one which does not harvest information and
retains your anonymity (such as DuckDuckGo) or for one which is part of
the Frenemy ecosystem and lavishes you with individualized results? Did
you limit the ability of your phone apps to access personal and location
information? Do you often change the default option? Do you click “accept”
only after reading the terms and conditions? Did you invest money in pri-
vacy measures, or recently choose a paid app for its privacy protections over
a popular free app? And if you did invest money, do you know if the promise
of privacy and control was truly delivered by your service provider?
The likely answers to these questions may help us all appreciate why the
evolutionary path tilts in favor of the Frenemy dynamic and the devious
helper. Key here are data-driven network effects, Big Data, Big Analytics,
vertical integration, bundling of ser vices, and interoperability, which all
seem to favor the super-platforms.
Let us explain in more detail: First, to excel in its role, the personal
assistant must know our preferences. To learn and predict our desires,
personal assistants require a significant store of data and opportunities
to experiment. The underlying code and algorithms of Facebook’s M, for
example, are largely open source. The key assets are not the algorithms
Copyright © 2016. Harvard University Press. All rights reserved.

(otherwise why share them?) but the scale of data and the algorithm’s ability
to learn by trial-by-error. As the Wall Street Journal reported, “Facebook
Messenger already has more than 700 million users,” which yields it the fol-
lowing advantage: “with access to so many users, Facebook has a plausible
way to get the gigantic quantity of conversational data required to make a
chat-based assistant sufficiently automated.”24 With more users making
more requests, M can quickly process more tasks easily. In effect, users help
the super-platform’s algorithm learn by noting and correcting mistakes.
Only a few companies have the requisite volume and variety of personal

data and opportunities to experiment for their personal assistants to be

competitive: namely, the super-platforms Amazon, Facebook, Google, and
Apple. Microsoft, in acquiring the professional social network LinkedIn
and divesting its low-end smartphones, could become the fift h competitor
in this space.25
Second, data-driven network effects will further weed out the super-
platforms. We do not want five butlers, each asking us about movies
tonight or food to order. Each super-platform will jockey for its butler to
become our head butler. In discussing its digital personal assistant,
Google’s CEO said, “We want users to have an ongoing two-way dialogue
with Google.”26 Google, as the head butler, can analyze our e-mails, texts,
or photos, and suggest replies.27 Looking at our calendar, it can determine
the best time for the dog to be groomed. Thus the more we converse with,
and delegate to, the head butler, the better it can predict our tastes, and
the more likely we will rely on it for our daily activities. As our butler ac-
cumulates information over time, the switching costs between butlers will
become higher. We would therefore be willingly locked into our comfort
zone. New entrants will find it difficult to match the scale of data held by
the super-platforms and to convince us to switch.
Third, the super-platform can seamlessly integrate its wide offerings,
bundle its apps, and nudge us to its products. The gatekeeper function is
intensified. Others will likely lack the scale and scope of products to attract
new users. Moreover, unless they develop their own operating system, they
too will be dependent on the super-platform’s. For example, Google argues
that given “its 17 years of work cataloguing the internet and physical world,
its assistant is smarter and better able to work with its email, messaging,
mapping and photo apps. And since Google makes soft ware for smart-
phones, smartwatches and old-fashioned computers, Google says people
will be able to have one conversation with multiple machines.”28
Copyright © 2016. Harvard University Press. All rights reserved.

These effects could undermine the success of the Purist Assistant. Luckily,
all is not lost and an increased popularity of these purist tools may be pos-
sible. But it will have to be driven by us, the users, taking control over the
interface and appreciating that free can sometimes be very expensive.

Reﬂections
The next frontier of the Frenemy dynamic may make the strong super-
platforms even stronger, and many independent apps weaker. It may increase

the entry barriers and consolidate economic and political power into even
fewer hands. The super-platforms could extract even more data, and com-
.

mand even higher rents to allow others to access us. The increasing reve-
nues could enable the few super-platforms to further expand into driver-
less cars, wearables, virtual reality, and the Internet of Things, drying out
downstream ser vices.
The new Frenemy environment, especially when voice-activated and
speech-based, would superficially appear beneficial—more free ser vices to
help our personal assistant attend to our daily needs. Because the products
and ser vices are highly personalized, competition authorities would have
difficulties identifying quality degradation and proving its anticompetitive
effects. As the primary interaction takes place at the personal-assistant
level, shortcuts and bypasses may make applications redundant and allow
the super-platform to occupy broader swaths of the savanna. Indeed, un-
like the current Frenemy scenario, where lions collaborate to kill gazelles,
the super-platform in the future might seem protective. We are no longer
gazelles in the wild, but domesticated animals in the zoo, with the digital
personal assistant as our caretaker.
Thus, this evolution may go unchallenged under current user behavior
and current antitrust policies and tools. The greater algorithm autonomy
in a nontransparent, highly personalized interface with customers will halt
competition enforcement. At best, the agencies and courts might under-
stand the risks and work to confront them, among other things, by edu-
cating the users about the cost of free. At worst, the agencies and users will
be captured intellectually and regulatorily, and will celebrate the technol-
ogies that slowly bring us into the ultimate Truman Show.
Copyright © 2016. Harvard University Press. All rights reserved.

Intervention

O UR THREE SCENARIOS — collusion, behavioral discrimination, and

Frenemy—illustrate the mirage of competition. Online markets at
times may appear competitive. They have many of the procompetitive at-
tributes upon which we have learned to rely. But behind this competitive
veneer, new strategies emerge, powered by a complex web of algorithms,
that maximize the firms’ profits, while harming our welfare.
Faced with the end of competition as we know it, how do we protect
ourselves? How do we ensure that the digitized hand yields a competitive
environment that promotes our overall well-being? The displacement of
the invisible hand by the “digitized hand” heralds a change in dynamics
which requires us to carefully recalibrate our approach to markets and
intervention.
Some, however, will defend the adequacy of a free-market, noninterven-
tionist approach, They will warn that any regulatory or enforcement inter-
vention will likely chill the new technologies and the rise of dynamic markets
from which we will benefit immensely.
Others will recognize the problem but look for a quick, broadly appli-
Copyright © 2016. Harvard University Press. All rights reserved.

cable fi x. Because there isn’t one, they will argue for non-intervention until
a fi x can be developed.
Given the different types of problems we identify, we need to consider
afresh when intervention is required, the type of intervention, and its du-
ration. The regulatory/enforcement aim is to help promote competition,
where innovation and investment flourish, and to minimize the harms we
identify, along with other harms.
Our discussion in this part spreads over two chapters, the first being more
conceptual in nature and the second taking a more practical perspective.
203

Chapter 18 looks at how Big Data and Big Analytics could lead to a
“planned economy,” albeit one planned by dominant firms, not bureau-
crats. Faced with these dynamics, should smart regulation be introduced
by the government?
Following this, in Chapter 19, we explore the antitrust toolbox to see
whether the enforcers can prevent or deter the anticompetitive scenarios
of collusion, behavioral discrimination, and Frenemy. We highlight pos-
sible avenues for the enforcer—including the use of current tools and per-
haps a few new ones.
Copyright © 2016. Harvard University Press. All rights reserved.

To Regulate or Not to Regulate

S UPPOSE WE COULD INVEST in phrases, as we do with stocks. Also, suppose

the payoff would depend on how frequently the phrase appears in
English-language books. If you could invest in one of these phrases—price
regulation or invisible hand—which one would you pick? Google’s Ngram
Viewer shows how often a phrase has occurred in a corpus of English-
language books. As Figure 5 reflects, an investor in price regulation would
have profited in the 1940s (see the gray line in Figure 5), but the overall
winners are those who picked invisible hand (the dotted line in Figure 5).
The result parallels the shift in emphasis in modern times, from regula-
tion to free market philosophy. Price regulation has taken a beating, espe-
cially, as Chapter 3 explores, with the rise of neoclassical economic theories
associated with the University of Chicago.
Granted, the appeal of the invisible hand has diminished in recent
years—after the financial crisis, the Great Recession, growing income and
wealth inequality in the United States and U.K.,1 reduced social mobility,
and the sheer arrogance of crony capitalism. Nonetheless, despite the
growing appeal of conscious capitalism and shared value, many policy-
Copyright © 2016. Harvard University Press. All rights reserved.

makers still praise the unrestrained free market, and are far more vocal
over the cost of false positives from governmental intervention than the
cost of false negatives from governmental abstention.

Aggregated Information and Competition: The Sum of All Knowledge

In defending a market economy, one should first inquire what one is de-
fending. As we have seen, competition is changing. The digitalized market
environment, characterized by a growing capacity to analyze, aggregate,
205

Ngram Viewer
Graph these comma-separated phrases: invisible hand,price regulation ccase-insensitive

between 1800 and 2000 from the corpus English with smoothing of 3 . Search lots of books

0.000200%

0.000180%

0.000160%

0.000140%

0.000120%

0.000100%

0.000080%

0.000060%

0.000040%

0.000020%

0.000000%
1800 1820 1840 1860 1880 1900 1920 1940 1960
(click on line/label for focus, right click to expand/contract wildcards)

Figure 5. Google Books, Ngram Viewer (https://books.google.com/ngrams).

and store data, changes the role and significance of information. In a data-
driven environment, algorithms collect and process data on our movements,
preferences, biases, and reservation price, and utilize this information in
future transactions. As more real-time data flows through the ecosystem,
and the data becomes accessible, a thought-provoking question emerges:
Could it be that in a digitalized environment, data will be aggregated to
provide the sum of all or the most relevant knowledge? Are we approaching
the creation of a single repository of all information that is continually
updated and accessible? If so, then surely the dynamics of competition
would radically change. Increasingly one may wonder, what is left of com-
petition as we know it?
Copyright © 2016. Harvard University Press. All rights reserved.

In asking what power the invisible hand still possesses in digitalized

markets, we must consider Friedrich A. Hayek’s seminal work on knowl-
edge, competition, and society. In his book The Road to Serfdom, the econ-
omist did not condemn governments’ intrusion into the marketplace. He
was careful to distinguish his opposition to central planning from a dog-
matic laissez-faire attitude.2 Recall that in Hayek’s time, collectivism was on
the rise, with fascism, nationalist autarky, and later communism spreading.
As Harvard Professor Jeffry Frieden discussed, the Nazis by 1938 “had more
than five hundred important state-owned firms, half of all investment was

being carried out by the state, and government spending was 34 percent of
GNP, up from 15 percent in the late 1920s.”3 Frieden also observed how, at
its height in the late 1930s, the fascist economic order included nearly all of
Europe and the Middle East and much of Asia and Africa.4 Thus, demo-
cratic governments with market economies, when Hayek wrote his sem-
inal book, were the exception rather than the rule.
Part of Hayek’s critique of centrally planned economies was incomplete
information patterns, which characterized the market environment in the
mid-1940s, and the way in which these affected resource allocation. Hayek
noted that:

[K]nowledge of the circumstances of which we must make use never ex-

ists in concentrated or integrated form but solely as the dispersed bits of
incomplete and frequently contradictory knowledge which all the sepa-
rate individuals possess. The economic problem of society is thus not
merely a problem of how to allocate “given” resources—if “given” is taken
to mean given to a single mind which deliberately solves the problem set
by these “data.” It is rather a problem of how to secure the best use of re-
sources known to any of the members of society, for ends whose relative
importance only these individuals know. Or, to put it briefly, it is a
problem of the utilization of knowledge which is not given to anyone in
its totality.5

Hayek’s critique focused on the then pervasive (but fallacious) assump-

tions of stability and availability of information, which underpinned much
of the contemporary macroeconomic theory during his time.6 In reality,
he observed, information is dispersed among many people, and he there-
fore noted both “the unavoidable imperfection of man’s knowledge and
the consequent need for a process by which knowledge is constantly com-
Copyright © 2016. Harvard University Press. All rights reserved.

municated and acquired.”7

Markets incorporate dispersed knowledge. The definition of the efficient
markets hypothesis encompasses this belief:

[A]ll relevant information is fully and immediately reflected in a security’s

market price, thereby assuming that an investor will obtain an equilib-
rium rate of return. In other words, an investor should not expect to
earn an abnormal return (above the market return) through either tech-
nical analysis or fundamental analysis. Th ree forms of efficient market
hypothesis exist: weak form (stock prices reflect all past information in

prices), semistrong form (stock prices reflect all past and current publicly
available information), and strong form (stock prices reflect all relevant
information, including information not yet disclosed to the general
public, such as insider information).8

It follows that a competitive market price incorporates the dispersed

knowledge. In a competitive market, price is determined by the intersec-
tion of consumers’ demand for a given good or ser vice with its available
supply. The higher the demand and lower the supply for any given good or
service, the higher its price; the lower the demand and higher the supply, the
lower its price. Price (at least in a competitive market) serves an allocative
function; it mediates between how much of a good or ser vice is demanded
and how much of it can be supplied. Price also serves an important signaling
function; it publicly communicates information about a par ticu lar com-
pany’s efficiency and market profitability, and it may help others to con-
sider expansion, entry, or resource allocation.

The Illusion of a Competitive Price

With the advancement in Big Data and Big Analytics, our ability to amass
information has progressed far beyond what Hayek envisaged in the mid-
twentieth century.9 Is Hayek’s “knowledge problem” less problematic today?
As Hayek recognized:
If we possess all the relevant information, if we can start out from a given
system of preferences, and if we command complete knowledge of avail-
able means, the problem which remains is purely one of logic.10

If anyone actually knew every thing that economic theory designated as

“data” competition would indeed be a highly wasteful method of securing
adjustment to these facts.11
Copyright © 2016. Harvard University Press. All rights reserved.

[C]ompetition is impor tant only because and insofar as its outcomes are
unpredictable and on the whole different from those that anyone would
have been able to consciously strive for.12

It is interesting to consider Hayek’s theory in light of comments made

by Leonid Kantorovich in his Nobel Prize lecture.13 The Soviet economist
noted how complex problems of economic prediction, control, planning,
and resource allocation may be resolved, among other things, with advance-
ments in computing technology and algorithms.

Are we close to reaching that stage? With the rise of pricing algorithms
in an increasingly digitalized market universe, in which web-aggregators,
algorithms, and data pools provide the foundation for possible unilateral,
coordinated, and Frenemy behav ior, is the invisible hand still a viable
concept? After all, in a market that is in reality controlled by bots and al-
gorithms, what power does the invisible hand possess?
One may argue that the invisible hand remains a power ful force.
Humans, after all, program and nominally control the algorithms. We still
trust the invisible hand in many markets where robots manufacture prod-
ucts or provide ser vices, or where computers help facilitate trade. More-
over, we have seen how computers can enhance competition and our
welfare.
But we have also seen that, in a digitalized and controlled universe, bar-
riers and market failures often exist, and may even be unavoidably inte-
grated into the landscape. As we suggested earlier, what might at first glance
be seen as competition is, in fact, the creation of a new force—the “digi-
talized hand.” That hand, controlled by algorithms, determines the market
price in any given market through complex calculations. It is controlled by
those who seek to maximize their profits. At times we see a surreal result,
as when Amazon’s algorithms priced the book The Making of a Fly at
$23,698,655.93. But other times, the deviation will be smaller and less no-
ticeable, especially when no ready benchmark exists.
Another interesting question follows: Is the price in a digitalized envi-
ronment characterized by many sellers the competitive price, or merely a
fiction created by the digitalized hand?
Recall our Hub and Spoke scenario, where an algorithm determines the
base price for the ride-sharing platform Uber. The algorithm also deter-
mines: when to implement a surge price, for which areas, for how long; and
to what extent. Passengers typically do not negotiate a discount with the
Copyright © 2016. Harvard University Press. All rights reserved.

drivers. Uber’s God View reaches beyond knowing where all its drivers and
users are traveling. Uber’s algorithm collects data in real time and sets a
price based on demand and supply characteristics and market conditions.
Uber defends its surge pricing based on its knowledge of supply and de-
mand conditions:

We raise price when supply of available cars gets tight. [Example: If there
are 300 cars in a city and 290 of them are picking up a rider or in trip,
then this would be considered an extremely tight supply situation.] We

raise the price in increments over time based on supply health. When
supply opens up, we then lower the price. . . .
We are able to get a far greater number of drivers on the system when
Surge Pricing is in effect—it’s basic economics. Higher prices encourages
more supply to come online. It gets some drivers out to work on NYE. It
keeps other drivers from going to alternatives like renting their car out
for the night, or trying their luck at hustling rides on the street. Higher
prices means more cars, means more rides, means more people getting
around the city efficiently, safely AND in style.14

The surge in prices, in theory, should attract more drivers onto the road
(or reduce consumer demand for those unwilling to pay the surge amount),
whereupon prices should return to the regular rate. In the context of our dis-
cussion, of significance is the fact that Uber is the one seemingly determining
the competitive market price, where its users’ demand for a car service inter-
sects with the available supply.
Uber also claims that its surge pricing serves a signaling function:
namely to communicate to the community’s Uber drivers (and potential
drivers) to consider entering the market. So is Uber’s surge price, in a market
characterized by many drivers, the competitive price or merely a fiction
created by its algorithms? When Uber’s algorithm determines when de-
mand exceeds supply, and charges a premium (ostensibly to attract poten-
tial drivers to the road), is this the invisible hand at work? Or is it Uber’s
digitalized hand at work?
Two studies have drawn into question Uber’s claim that its surge pricing
brings more drivers into the market. One study examined four weeks of
Uber data and did not find evidence of surge pricing bringing more drivers
out on the roads. Instead, surge pricing appeared to push “drivers already
on the job toward neighborhoods with more demand—and higher surge
Copyright © 2016. Harvard University Press. All rights reserved.

pricing. As a result, some neighborhoods are left with higher waiting times
for a car.”15 Another study interviewed Uber and Lyft drivers. Over half of
the interviewed drivers said they were “not influenced by surge pricing in-
formation as the supply-demand control algorithms failed to accommo-
date their abilities, emotion, and motivation.”16 As the study found,
Surge pricing changed too rapidly and unexpectedly to utilize the infor-
mation in a strategic way to boost their incomes. Surge areas were on and
off, sometimes by the second, and being in the surge area did not guar-
antee requests from within the surge area.17

Even if surge pricing did not have its intended effect of quickly attracting
additional drivers to the road, the invisible hand could still be at work. Ar-
guably, Uber has to price competitively. Other wise users will turn to other
ride-sharing apps, taxis, public transportation, or other modes of travel.
Thus, in markets with a competitive alternative, the invisible hand ulti-
mately checks Uber. If Uber’s surge price is too frequent, too high, or for
too long, its app users would opt for other modes of travel (or simply walk).
But as we saw in Chapter 6, as Uber’s platform increases in popularity out-
side options are limited and switching costs become high. With more drivers
and users relying on Uber, its pricing algorithm, rather than responding to
market conditions, essentially sets the market price. The competing ser vices
may not benefit from economies of scale; public transportation and taxis
may not be a viable option (and if they are, the waiting time may be longer).
Uber—quite rightly—may have the upper hand. Eventually, the price Uber
sets may be shielded, to some extent, from the competitive pressures of other
providers. Furthermore, competitors in some markets may opt to follow the
price determined by the digitalized hand, when that price is higher than
the alternative which would have emerged through the invisible hand.
Consumers do not know how Uber’s pricing algorithm calculates the
surge price or whether the surge price is fair. Nor will the surge price al-
ways have the desired effect of quickly attracting drivers to the road. In-
stead, if Uber possesses market power, the surge price enables both Uber
and its drivers to simply earn extra profits, at consumers’ expense, all under
the guise of a “market-clearing” price.

A Privately Planned Economy?

Notice here that Uber is in effect an uber–price regulator. Uber does not
own the cars. Nor does Uber employ the drivers, who are “independent
Copyright © 2016. Harvard University Press. All rights reserved.

contractors.”18 Nor does Uber allow individual drivers and passengers to

negotiate prices in each city. Uber sets the price. It also increases and lowers
the price based on its capturing all the relevant market information. So if
Uber captures the sum of all knowledge to set the market-clearing price,
why can’t other platforms and super-platforms do the same?
The emergence of super-platforms could indicate a shift toward the at-
tainment of all knowledge. Data collection by leading platforms (like
Uber) and super-platforms (like Google, Facebook, Apple, and Amazon)
could create an economy which, for all purposes, is planned, not by

bureaucrats or CEOs, but by the technostructure. Corporate decision-

making power, economist John Kenneth Galbraith observed, resides not
with the CEO but with the technostructure.19 As Galbraith found, “if a
decision requires the specialized knowledge of a group, it is subject to safe
review only by the similar knowledge of a similar group. Group decision,
unless acted upon by another group, tends to be absolute.”20 In Galbraith’s
day, the technostructure involved the working groups of designers, engi-
neers, and marketers within corporations like General Motors. In 2016 it
would be the working groups of engineers, marketing, advertising, software
developers, designers, and so on, operating within Google, Apple, Am-
azon, and Facebook. The CEO will not necessarily know the details of
how the technostructure facilitates the data flow and the ways in which
wealth can be extracted. And the technostructure, while knowing the
general objectives of the algorithm, may not necessarily know how the
algorithm determined the instant price. And how the algorithm deter-
mined the price may not at all resemble how humans did it in brick-and-
mortar shops subject to the invisible hand.
Firms increasingly will use Big Data and Big Analytics to determine
prices. We may not observe in each market when the digitalized hand dis-
places the natural competition dynamics. One scenario, as we saw, will be
the gradual demise of a uniform market price, as pricing algorithms indi-
vidualize price and product offerings as they approach near-perfect behav-
ioral discrimination. Another set of scenarios involve pricing algorithms
tacitly colluding in a transparency-enhanced environment. In either case,
the market involves many competitors who, following the path of Uber, can
also claim that they are harnessing Big Data and Big Analytics to set the
market-clearing price. If they have market power, then their market-
clearing price may exceed the competitive price, that is, the average price if
left to negotiation between individual buyers and sellers.
Copyright © 2016. Harvard University Press. All rights reserved.

Is Smart Regulation Back?

If companies can harness Big Data and Big Analytics to effectively set the
market price, should governments use the same tools to monitor industry
prices (or even determine a competitive price)?
If Uber, which doesn’t own any cars or employ any drivers, can deter-
mine the surge and base price in each market, why can’t the government?
The government can collect as much data as Uber, if not more. The gov-

ernment could also use pricing algorithms. And the government could de-
termine a competitive price and identify what amounts to an excessive
price. So if Uber, as an intermediary, can calculate the surcharge for its
many drivers and passengers during periods of congestion, why couldn’t
the government’s pricing algorithms monitor industry pricing or simply set
the market-clearing price. One could argue that price regulation in the
post-Hayekian world of Big Data is feasible, once industry data on indi-
vidual consumer preferences and firm costs are collected and analyzed. The
rise of the digitalized hand makes this possible.
One could go a step further. If companies can harness Big Data to set
the market price, can the combination of large volumes of data and sophis-
ticated pricing algorithms make a centrally planned economy viable?
We can reflect upon the potential and limitations of Big Data and a
planned economy in light of developments in Chile between 1971 and 1973.
At the time, when seeking to coordinate the nationalized sectors of its
economy, the Chilean government sought to address Hayek’s “knowledge
problem.” To do so, it launched Project Cybersyn—a system of telex ma-
chines that were placed in nationalized factories and regularly transmitted
data to a central operations room. This data, once harvested, was used in
economic modeling and enabled the central government to steer the na-
tional economy.21 The significance of Project Cybersyn was aptly illustrated
during the national strike in 1972.22 Faced with the possibility that the
strike might cripple its economy, Chile’s government responded by uti-
lizing its information network to coordinate the activities of those factories
that were not striking and effectively allocate resources to them.23 Impor-
tantly, the project was not problem-free. Most noticeably, its response time
was often too slow to be of any use,24 and it was also heavily reliant upon
accurate submission by factories of their production data—information
that was not always forthcoming.
Copyright © 2016. Harvard University Press. All rights reserved.

The Chilean experience, with its achievements and drawbacks, provides

an example of the power of knowledge and its use as part of a planned
economy. Its execution in a modern digitalized environment would have
achieved much more than capacity and production optimization. As one
proponent wrote, “A combination of new sensory and computing technol-
ogies, two-way communications and devices that both create and analyze
large volumes of data can now measure and communicate real-time de-
mand.”25 The government algorithm, in analyzing the data, could quickly
change pricing or supply levels, “such as powering down preselected devices

during periods of peak electricity demand.”26 Thus local, state, or federal

governments can harness data to set a market-clearing price.
To illustrate “smart” regulation in the post–Project Cybersyn world,
we’ll look at parking in San Francisco. In 2011, the city’s SFpark program
began experimenting with smart parking meters that continually adjust
pricing to respond to changing demand conditions. The city wanted to de-
termine the right price to charge for parking to meet its parking space
availability targets. So the city installed wireless parking sensors in 8,200
on-street metered spaces, which detected parking availability in real time
and monitored public garages.27 The city set a target occupancy rate be-
tween 60 and 80 percent.28 Data on the available supply of parking spaces
was collected in real time from the meter sensors. Rather than charging the
same price during periods of high or low demand, the parking prices would
adjust accordingly throughout the day for public parking on any street. In
areas and at times where it became difficult to find a parking space, parking
rates increased incrementally “until at least one space is available on each
block most of the time;” in areas where parking spaces were plentiful,
parking rates decreased “until some of the empty spaces fill.”29 The city also
directed “ drivers towards available parking by sending real-time avail-
ability to mobile apps and to the website.”30
The city’s pi lot “demand-responsive” pricing program was apparently
successful on many levels. It helped improve parking availability and net
revenues.31 It decreased drivers’ average search time for a space.32 It also
helped reduce greenhouse gas emissions,33 peak period congestion,34 traffic
volume,35 vehicle miles traveled,36 and double parking.37
So if one major U.S. city could determine the market-clearing price for
parking spaces, does this mark a rebirth for “smarter” price regulation, al-
beit under a more fashionable and acceptable term, such as data-driven
dynamic pricing?
Copyright © 2016. Harvard University Press. All rights reserved.

To Inﬁnity and Beyond?

Even if one accepts the premise that the sum of all data would facilitate
some variation of a planned economy, the challenge remains to gather all
or most data. Pricing for municipal parking is straightforward. So too is
pricing for car-sharing ser vices. Uber breaks down prices per city along
several basic categories of cars (basic, SUVs, luxury, and taxi). But as anyone
who lives in a centrally planned economy knows, designing and pricing

men’s and women’s fashions are not as straightforward. Even if the velocity
of collecting and processing a voluminous variety of data increases, one
wonders whether any firm or governmental agency could truly possess
enough information for many markets. We may have solved Hayek’s knowl-
edge problem for parking, car ser vice, energy, and other simple homoge-
nous ser vices and products, but not for designer dresses and the many dif-
ferentiated goods and ser vices.
It is hard to imagine us categorically solving the knowledge problem in
the foreseeable future. We cannot capture and digitize all information.
It is too costly (and raises many privacy concerns). Moreover, even if tech-
nology does indeed reach a point where all or most relevant knowledge
can be aggregated, this does not mean that the computer algorithms will
keep abreast in processing it. Recall that 2015 marked the first time an al-
gorithm weakly solved one kind of poker game with incomplete informa-
tion. It will likely take years for algorithms to solve more complex games
and for a complete picture of many markets for differentiated goods and
ser vices. Indeed, one criticism of Hayek’s theory concerns its assumption
that, despite the dispersal of knowledge among a large number of decision
makers, “as a whole, the aggregated set of all decision makers have a
complete set of all relevant knowledge.”38 Even if all the data were col-
lected, many theorists believe that there are in fact missing pieces of the
puzzle, which are not known by anyone (and cannot therefore be cap-
tured or analyzed).39
One example is Apple. Its cofounder, Steve Jobs, showed the iPad to a
small group of journalists shortly before its going on sale in 2010. As the
New York Times reported, one journalist asked Jobs what consumer and
market research Apple had done to guide the development of the new
product. “None,” Jobs replied. “It isn’t the consumers’ job to know what
they want.”40 Data can capture well-established consumer preferences for
Copyright © 2016. Harvard University Press. All rights reserved.

pre-existing products and ser vices, but the algorithms may not necessarily
estimate demand for new products based on new technologies.41 Even if the
government or super-platform could collect all the data on all of our ex-
isting preferences, they may perhaps design a better cell phone (but not the
iPhone) or a cheaper watch (but not the Apple Watch).
Other concerns over price regulation involve incentives and regulatory
capture. Economic regulation attracts special interest groups to lobby the
government for regulatory measures that benefit them to the detriment of
society overall. If the government algorithm sets (or regulates) price levels,

then rent-seeking behav ior can impose additional social costs. For ex-
ample, the U.S. Federal Energy Regulatory Commission’s merger review
policies were criticized for relying on data supplied by the regulated enti-
ties, rather than conducting its own independent fact gathering and analysis
of market definition.42 The risk that sector regulators, even the most dedi-
cated ones, may fail to understand and predict market dynamics, is real.
Such failure will likely lead to a generalized approach that ultimately
reduces welfare. In addition to the risks of imperfect information and
regulatory capture, the government can undertake anticompetitive in-
tervention because of weaker incentives to avoid mistakes than private
actors who fully bear the costs of their mistakes, “political myopia,” and
the lack of direct accountability to the public.43
Moreover, the road to perfect price regulation may also lead to a world
of limited privacy, among other things. On this point, it has been noted that
“being online increasingly means being put into categories based on a so-
cioeconomic portrait of you that’s built over time by advertisers and search
engines collecting your data—a portrait that data brokers buy and sell, but
that you cannot control or even see.”44 That information is open for com-
panies and individuals to purchase and use in targeting users—from legiti-
mate advertising to possible abusive use of one’s most secret or vulnerable
searches and online behav ior.45

Reﬂections
For some products and ser vices, one does not require all information for
decision making. Uber, for example, sets the surge price without knowing
where exactly its possible drivers are or how quickly they will respond (if
they do). We live in a world where we spend a lot of our household income
on ordinary things, such as basic foods, utilities, electricity, mortgage in-
Copyright © 2016. Harvard University Press. All rights reserved.

terest, gasoline, and public ser vices.46 For these areas, data-driven, dynamic
pricing may be on the horizon.
As firms and super-platforms increasingly collect and analyze data, we
will likely see more dynamic pricing. But we cannot assume that market
prices going forward will approximate the market-clearing, competitive
price. The more power these platforms (like Uber) or super-platforms (like
Apple, Facebook, Amazon, Google) accumulate, the more likely they will
control, rather than respond to, the digitalized hand.

There will always be alternatives. So you can always walk at 1 a.m. on

New Year’s Day if you are dissatisfied with Uber’s surge pricing. But that
defense is available to any monopolist. And as we trudge through the snow
on our way home, it becomes apparent that we are no longer in a market
economy governed by an invisible hand, where competition is the brass
knuckles that enforce its decisions.
Copyright © 2016. Harvard University Press. All rights reserved.

The Enforcement Toolbox

H AVING EXPLORED the possible use of “smart” regulation in a data-driven

economy, we now identify several looming enforcement challenges
presented by our scenarios. First, the competition agency may not see any
problem. Even if it does see the problem, it may not have any enforcement
tools to fi x the problem. The issues may transcend antitrust, given the dif-
ficult legal and ethical issues of humans’ accountability for a computer’s
behavior. Finally, even if the competition agency has the tools, when should
it intervene?
These problems may appear insurmountable. One debate, as we saw in
Chapter 3, is whether the “old economy” antitrust doctrines should apply
to firms competing in dynamic technological markets characterized by
network effects. But this debate recurs with every generation of antitrust
scholars and practitioners. As we will see, the current antitrust tools will
be inadequate in prosecuting and remedying some of our anticompetitive
scenarios. The dynamic has changed in many markets—competition as we
know it has given way to new forms of rivalry.
But the issue is not over the current tools. As we will explore, the real
Copyright © 2016. Harvard University Press. All rights reserved.

issue is designing new tools to address the new problems. Indeed, some
courts and competition agencies many years ago were better at recognizing
the prevailing challenges and devising ingenuous solutions to meet them.
With that reality in mind, we must be open-minded to new enforcement
instruments. Failure to do so, as the United Kingdom House of Lords rec-
ognized, may result in “a perception that large online platforms are above
the law.”1 And if the House of Lords is at the forefront in addressing these
issues, no competition agency can justify its unimaginativeness as wisdom.
218

Key Challenges to Competition Enforcement

When the Competition Agency Does Not See the Problem
As discussed in Chapter 1, the digital economy is booming—a dynamic
landscape where new entrants, low entry barriers, and rapidly changing
technologies are dislodging preexisting market power. As we have ex-
plored, what appears to be a competitive environment may not be the
welfare-enhancing competition that we know. Our concerns are not with
technological advances or successful online businesses. Our concerns go
deeper, to the core of the new market dynamics—where entry is possible,
but expansion will likely be controlled by super-platforms; where choice is
ample, but competition is limited; and where disruptive innovative threats
emerge, but are eliminated through acquisitions or exclusionary practices.
The competitive façade masks the wealth transfer, and the targets of anti-
competitive practices—the buyers—are often unaware of the extent of the
manipulation.
Thus, the competition agency must look beyond the façade to see whether
competitors are racing to the bottom in finding new ways to exploit us. So
while technology can increase price transparency (which should be a good
thing as it lowers consumers’ search costs), the pricing algorithms at times
can foster tacit collusion—when sellers’ pricing algorithms, by quickly re-
acting to price changes, diminish the incentive to discount. If market par-
ticipants’ algorithms can attain a God View, then enforcers must consider
the possibility of tacit collusion beyond price and highly concentrated
industries.
So while technology can enable sellers to customize their product offer-
ings (which should be a good thing if it matches consumer preferences), the
pricing algorithms can also enable sellers to better segment customers and
engage in behavioral discrimination—again, at our expense. Some en-
Copyright © 2016. Harvard University Press. All rights reserved.

forcers today accept price discrimination as efficiency-enhancing. Some

enforcers scoff at the advances in economic thinking over the past thirty
years–downplaying the role of imperfect willpower and biases. They still
believe that markets behave as though the participants are perfectly ra-
tional and have willpower. So behavioral exploitation may be as foreign to
them as Snapchat, and will affect their readiness to pursue these cases.
Super-platforms may appear competitive in attracting a constellation
of soft ware developers. Seeing the plethora of free and low-cost apps

downloaded annually, the enforcers may conclude that consumers are

benefiting (and do not care that much about privacy). As a result, they do
not discover that free comes at a significant cost as Frenemies coordinate
to better track our behavior, profile us, and target us with behavioral ads.
The anticompetitive effects are not always easy to see. Companies can be
a step ahead in developing sophisticated strategies and technologies that
distort the perceived competitive environment. One illustration, in the
context of manipulation of stock exchange trading, is Michael Lewis’s fas-
cinating book Flash Boys: A Wall Street Revolt.2 On one side were the tra-
ditional investors, from small traders to large pension funds, making their
investment decisions based on the market reality as they saw it. But what
appeared to be a competitive market (indeed, the stock market is often per-
ceived as competition in its truest sense) was an illusion. The stock prices
posted on the computer screen were not the actual prices. Once the traders
executed an order, the price often shifted. In reality, the different stock ex-
changes were rigged by sophisticated high-frequency traders working with
many of the large financial institutions. The high-frequency traders were
in a competitive arms race, including laying cable the shortest possible
distance to their trading desk to gain a relative advantage in seeing and
executing orders slightly before everyone else. With this advantage, the
insiders “rode” on the traditional investment orders before they were ex-
ecuted. Once the traditional order was placed, these investors moved in
and out of positions in milliseconds, changing the market reality through
their accelerated capabilities. Ultimately, the insiders imposed, as Lewis
described, a small invisible tax on each trade, which amounted to nearly
$160 million a day.
One cannot help but borrow Albert Einstein’s aphorism: “Reality is merely
an illusion, albeit a very persistent one.” Technology in an algorithm-driven
economy can create multiple versions of the same market, distinguished
Copyright © 2016. Harvard University Press. All rights reserved.

by the participant’s savviness (or deceit) and wealth. Some compete in the
slower lane while others, better placed, compete against them and against
each other in the fast lane. Whether this is competition on the merits or ex-
ploitation is the challenging issue.

When the Competition Agency Sees the Problem but Has No Tools to Fix It
Some competition officials are already seeing the problems in some of our
scenarios. But they also see some of the challenges in applying the existing

antitrust case law to our collusion, behavioral discrimination, and Frenemy

scenarios. Indeed, looking at our three scenarios, competition enforcers
will have the tools to fi x some problems but not others. Antitrust, while not
obsolete, may prove difficult, at times, to apply even when a theory of harm
is present and the competition agency wishes to intervene.
When considering our collusion scenarios, the competition agencies are
likely to position these practices high on their enforcement agendas. Euro-
pean Union and U.S. authorities could use Article 101 TFEU and Section 1
of the Sherman Act, respectively, to address the Messenger and Hub and
Spoke scenarios, where an anticompetitive agreement exists. They may also
stretch these statutes to Predictable Agent by reframing it as a simple con-
spiracy. Unless they have another statute (like Section 5 of the FTC Act),
their current tools do not reach the unilateral use of algorithms by non-
dominant companies. Some agencies currently have no tools for the
creation of the Digital Eye, where there is no evidence of anticompetitive
agreement or intent.3
When reflecting on behavioral discrimination, several challenges emerge.
The agency must understand the new market dynamics to appreciate the
shift to almost perfect behavioral discrimination. Second, the agency’s
welfare benchmark should target the transfer of wealth from consumers to
sellers and focus on direct consumer welfare. Third, the current antitrust
tools do not target noncollusive behavioral discrimination. One exception
is if the discrimination helps the firm attain or maintain its monopoly.
Arguably the ability to behaviorally discriminate almost perfectly indicates
market power. In Europe, enforcers can perhaps go further in challenging
the imposition of monopoly prices by the discriminators.4
Finally, with respect to Frenemy dynamics, distinct theories of harm
may be considered. The first concerns collusion among the apps and within
the ecosystem. The main challenge here concerns the framing of such
Copyright © 2016. Harvard University Press. All rights reserved.

activities in antitrust language. Indeed, the harm to users that is inflicted

through the extraction of data may be easier to address and to remedy under
consumer and data protection laws than under the competition laws. A
second theory of harm may concern the activities of the super-platform
and the possible abuse of its dominant position. Here, as illustrated in
Chapter 14, an explicit refusal to deal, or a de facto refusal, may trigger in-
tervention, in the form of decisions, commitments, or interim measures.
Moreover, enforcers can intervene when the super-platform uses unfair
tactics to favor its own ser vices and products on its platform over those of

independent producers. Competition officials dealt with this issue when

prosecuting Microsoft.5 And the European Commission raises this con-
cern in its two recent statements of objections involving Google.

When the Competition Agency Has the Tools, but Practical Challenges Await
Even when the competition agency takes action, using its existing legal
tools, the new market dynamics raise other practical challenges.
Take, for example, the challenge of establishing market power in a data-
driven environment. Can the online company behave independently of its
customers or competitors? Often there is no direct evidence. Much depends
on how one defines the relevant market and one’s assumptions about and
understanding of market dynamics. The dynamics of a data-driven envi-
ronment, as the Frenemy scenario reflects, can be complex. The competition
agency must appreciate the interdependence, the asymmetry in bargaining
power, the strength of network effects, the absence of outside options, high
switching costs, and whether customers are locked in.
Even when market power is identified, recall that monopolies (other
than those created by mergers) are legal. Once dominance has been attained
legally, its abuse should be considered. The agency must consider which
actions should be condemned and the level and nature of harm needed to
trigger intervention.6
Also challenging is identifying reliable counterfactuals when appraising
harm. With traditional cartels, enforcers and courts can consider price
increase announcements that occur closely after meetings in which many
cartel members participate. In our collusion scenarios, the prices may in-
crease gradually as market transparency and interdependence increase.
There may not be a definitive meeting or occurrence to which one can
point. Nor can enforcers always identify credible counterfactuals when as-
Copyright © 2016. Harvard University Press. All rights reserved.

sessing behavioral discrimination. What is the competitive baseline “market”

price in a world of dynamic, differential pricing? Should the next incre-
mental development be contrasted with the previous one or with a
computer-free reality? Absent a transformative change in technology and
market dynamics, it may be difficult to identify the appropriate point of
intervention and comparison. After all, today’s artificial levels of transpar-
ency (and elevated prices) may become tomorrow’s acceptable norm.
Another key challenge is the legal and conceptual difficulty emerging
from the relationship between man and machine—that is, humans’ con-

trol (or lack of it) over machines, and their accountability for the algo-
rithms’ activities.7 If algorithms collude or price discriminate, are humans
liable? As several computer scientists have observed, “The amoral status of
an algorithm does not negate its effects on society.”8 Or as one judge aptly
noted, “Automation is effected through a human design.”9 In the context
of competition and markets, friction among profit maximization, ethical
trading, and consumer welfare exists. Current laws may not always resolve
this friction and may fail to incentivize individuals to take responsibility
for the actions of an “independent,” self-learning algorithm.

When to Intervene and to What Extent

Controversy may surround the timing of an intervention, its nature, and
its extent. As markets may display many characteristics of a competitive
environment, intervention may seem counterintuitive. The competitive fa-
çade, in addition to intellectual capture propagated by interested parties,
may tilt against intervention. To what extent do these dynamics present an
immediate problem in need of immediate correction?
The durability of market power in dynamic technology markets is often
difficult to ascertain and therefore controversial to act upon. Should the
competition agency wait for entry or expansion? At what point should
the enforcer assume that market power is sustainable? And which enforce-
ment tool, if any, should be used? Balancing the possible anticompetitive
harm against the welfare gains from technological advantages may prove
challenging. Importantly, some market efficiencies may only be delivered
through a super-platform.
Courts and enforcers also may be reluctant to intervene and find the
firm liable under the antitrust law, if no effective remedy exists. The U.S.
Supreme Court, for example, said, “No court should impose a duty to deal
Copyright © 2016. Harvard University Press. All rights reserved.

that it cannot explain or adequately and reasonably supervise. The problem

should be deemed irremedia[ble] by antitrust law when compulsory access
requires the court to assume the day-to-day controls characteristic of a reg-
ulatory agency.”10
Not surprisingly, a range of enforcement philosophies may lead to
varying levels of intervention. Although guided by economic analysis, the
foundations of competition policy around the world differ as competition
regimes developed and evolved under different political, social, and market
conditions.11 Enforcers differ in their views on the ability of markets to

correct themselves and the benefits and risks of intervention versus

nonintervention.
Even within a jurisdiction, elected officials, enforcers, and courts can
differ over the choice of cases, the enforcement agenda, the way the law is
applied, the goals and scope of competition law, and the wisdom of inter-
vening in our three scenarios.
To illustrate, consider the issuance in 2008 by the Bush administration
and subsequent withdrawal in 2009 by the Obama administration of the
Department of Justice (DOJ) report on competition and monopoly.12 De-
spite their joint task force, the FTC did not join the report.13 The report was
seen as too deferential to monopolies. “Withdrawing the Section 2 report,”
Christine A. Varney, then head of the DOJ’s Antitrust Division, said, “is a
shift in philosophy and the clearest way to let everyone know that the An-
titrust Division will be aggressively pursuing cases where monopolists try
to use their dominance in the marketplace to stifle competition and harm
consumers.”14 Notwithstanding her rhetoric, the DOJ since 1999 has chal-
lenged a monopoly only once for engaging in anticompetitive unilateral
conduct. This case involved a private hospital in Wichita Falls, the twenty-
ninth largest city in Texas.15 Neither the FTC nor the DOJ has challenged
many data-driven mergers16 or the alleged anticompetitive practices of
any of the super-platforms, besides Apple.17
So where does this leave us? Opponents of intervention often argue that
antitrust law is ill-suited to data-driven markets and that existing tools do
not support intervention. When reflecting on such claims, however, one
should note that harm and market failure cannot be defined by the enforce-
ment tools. Rather, once they are identified, a range of enforcement tools
should be considered. In line with this, the argument that “no previous case
law supports intervention” fails to consider the full enforcement toolbox
and to acknowledge the change in market and competitive dynamics. Inter-
Copyright © 2016. Harvard University Press. All rights reserved.

vention should not be categorically set aside as irrelevant for new dynamic
markets. Economic theories, based on hypothetical market assumptions,
should not trump the economic realities that evince actual consumer harm.
But enforcement should not be taken lightly. One should critically reflect
on the assumptions, theory of harm, and market reality before taking ac-
tion. Antitrust is not a panacea for every problem. While antitrust may
prove useful in addressing some dynamics, it may lack the scope and re-
finement to address others. Importantly, antitrust is only one of many tools
at our disposal.

Learning and Doing

Traditional competition law can address some, albeit not all, of the issues
we have explored. In a 2016 speech, EU Commissioner Vestager noted the
rise of Big Data and the role of competition enforcement:
[W]e don’t need a whole new competition rulebook for the big data world.
Just as we didn’t need one for a world of fax machines, or credit cards, or
personal computers. What we do need is to pay close attention to these
markets and to take action when it’s necessary. Competition rules can’t
solve every problem on their own. But they can make an impor tant con-
tribution to keeping digital markets level and open. So that consumers
get innovative products at the right prices. And so that digital entrepre-
neurs, however big or small, have a fair shot at success.18

We agree. Competition law is not obsolete. Furthermore, some jurisdic-

tions benefit from a broad market and sector investigations regime. This
flexible tool enables agencies to investigate and gather information to better
understand market dynamics. Unlike traditional competition law investi-
gations, the focus is not whether one or several companies violated the
law but on the operation of the market and identifying possible market
failures.
This tool can prove useful in helping agencies understand the new dy-
namics in algorithm-driven markets and the magnitude of any competi-
tive problems. In some jurisdictions, like the United Kingdom, market
and sector investigation laws also provide for wide behavioral and struc-
tural remedies. So even when no other suitable tool exists, the agency can
nonetheless address and minimize our scenarios’ anticompetitive risks.19
Moreover, market investigations can provide a valuable framework for
considering new enforcement tools.
Copyright © 2016. Harvard University Press. All rights reserved.

Building the Framework for Healthy Virtual Competition

Competition is normative. What we observe as competition reflects, in
part, the legal constraints and incentives, as well as informal social, eth-
ical, and moral norms. Formal norms (like laws and regulations) and in-
formal norms (such as codes of practice20) shape participants’ incentives
and market structure. As economist Douglass North observed, “The gov-
ernment is not a disinterested party in the economy.”21 For example, if the

government does nothing to prevent monopolies, and if many markets are

conducive to monopolization, then monopolies will likely dominate these
markets. So rather than an ex post approach (wait for the monopolies to
arise and then regulate them or prosecute their behav ior), the government
may favor ex ante mechanisms (toughen the merger laws to prevent mo-
nopolies, and facilitate entry by lowering regulatory barriers).
Thus, for virtual competition to flourish, we should focus on what pre-
conditions are necessary, and the extent to which competition, consumer
protection, and privacy laws can help support this economy. Such an ap-
proach is valuable. In changing the formal and informal norms, the gov-
ernment affects the inherent nature of competition so that firms’ incentives
are aligned with ours. In other words, carefully sowing the seeds of change
may be preferable to laborious and unrefined weeding.
We sketch several possible tools extending beyond competition law to ad-
dress our three scenarios. Our list is not exclusive. More creative solutions—
no doubt—can be tailored. The tools below simply illustrate the expansive
frontier to be explored.

Privacy by Design and Customer Empowerment

To hinder the abuses in our behavioral discrimination and Frenemy sce-
narios, privacy measures and safeguards may be a necessary precondition.
One promising legislative approach is to give individuals greater control
over their personal data and to avoid being tracked on- and offline.
The notice-and-consent privacy model is broken.22 Too often, as users,
we habitually click our consent to terms and conditions that we never read.
Indeed, a study of customers’ reading of standard online contracts revealed
that fewer than two out of a thousand customers attempt to read them, and
even then only superficially.23 Given the zeal with which we click our con-
Copyright © 2016. Harvard University Press. All rights reserved.

sent to online terms, the solution isn’t more information (and legalese) that
we’ll never read (especially on our mobile phones). In fact, excessive and
complex information would likely have the opposite effect—further disin-
centivizing us from attempting to read the notices. Even if the disclosures
were briefer, we still lack the power to renegotiate. You either accept the terms
or do without the update, app, or software. To be effective, any enhanced
disclosure must take into account the asymmetry in negotiating power
and possible consumer engagement problems, where “under-active con-
sumers, by not exercising choice, impose negative externalities on other
consumers from the consequential weakening of competition.”24

So what are the alternatives? First, the behavioral discrimination and

Frenemy behavior must become more salient, such as by means of (1) pop-
up windows that inform us when and what information is being harvested,
and when we’re being tracked; (2) clear indication of when personalized
prices are displayed; (3) clear disclosure with respect to “best price” claims,
to avoid misleading information as to the availability of other options; and
(4) clarity with respect to “no availability” claims—for instance, in the case
of hotel rooms or flights—and requiring sellers to clearly indicate whether
such claims relate to the availability on the specific website or to overall
availability.
Second, we should require default privacy options that are generally
aligned with our privacy interests. To illustrate, in the United States we
could extend to everyone the existing protection afforded to minors (and
their parents) under the Children’s Online Privacy Protection Act.25 Under
this legal model, privacy would be the default. We would have to opt in,
rather than opt out of being tracked (either by the website or by third par-
ties), of having data collected on us, and of being profi led. Firms would
have to obtain our written, verifiable consent. Like the browser choice screen
in the European Commission’s Microsoft case,26 users, after purchasing
their smartphone or computer, would be prompted for an omnibus pri-
vacy setting. We could opt for the bare minimum, such as cookies to en-
able us to log in and remember what is in our shopping basket, but for no
other purpose. Firms could not condition our access and use of their ser-
vices on their collection and use of our data (“click wrap”). Even when we
authorize the use of data, the company—consistent with data-minimization
principles—could not collect more personal information than is reason-
ably necessary for them to provide their ser vices. We could readily access
any personal information the firm has about us and delete it.
We are seeing such developments in the European Union, where privacy
Copyright © 2016. Harvard University Press. All rights reserved.

generally, and informational privacy in particular, have long been a fun-

damental right. Of significance is the EU Data Protection Regulation.
Adopted in 2016, after four years of drafting and negotiations, its aim is to
advance “clear rules that are fit for the digital age, that give strong protec-
tion and at the same time create opportunities and encourage innovation
in a European Digital Single Market.”27 Privacy regulators will have more
power, including the ability to impose larger administrative fines.28 Com-
panies cannot “divulge information that they have received for a particular
purpose without the permission of the person concerned. Consumers will
have to give their explicit consent to the use of their data.”29 Consumers

will have greater control over their data—being able to transfer it between
providers, remove it from a database, and obtain information as to its
usage.30
With greater control over our personal data, we could select third-party
agents to negotiate on our behalf. Computer science professor Pedro Do-
mingos, for example, proposed creating a digital alter ego: “For a subscrip-
tion fee, such a firm would record your every interaction with the digital
world, build and maintain a 360-degree model of you, and use it to nego-
tiate with other people’s models.”31
So we can see how ex ante privacy protections can foster innovative ser-
vices, while limiting the ability and incentives of sellers to discriminate.32

New Entrants with Different Incentives

To destabilize algorithm-enhanced tacit collusion or widespread behavioral
discrimination, the government might promote entry by companies with
different economic incentives. One approach may involve subsidies to algo-
rithm providers that promote customers’ interest by, for example, de-
signing “counter measures,” which may serve to restore competition. An-
other avenue may involve entry by consumer-owned cooperatives, where
the supracompetitive profits are redistributed to consumers in the form of
rebates. The rebates could effectively return prices to competitive levels.
Social purchasing sites, such as CrowdZap, which assemble buying groups
and achieve economies of scale in purchasing, including the “Big Switch” ini-
tiative to save money on energy bills,33 are injecting competition. Similarly,
other sites, such as Groupon, Wowcher, and Living Social, distinguished as
“group buying” rather than “collective purchasing,”34 offer consumers dis-
count vouchers when enough consumers sign up for an offer.35
A related approach involves sponsoring or supporting entry by a mav-
Copyright © 2016. Harvard University Press. All rights reserved.

erick. A maverick may offer a disruptive technology or business model, take

the lead in cutting prices (or resisting its rivals’ attempts to raise prices), or
expand its production capacity. If consumers flock to the discounter, rivals
will likely respond. If successful, they end up, from the competitors’ per-
spective, with the nuclear option—an all-out price war. Mavericks might
program their pricing algorithm to prefer market share growth over profit-
ability within certain bounds, so as to enable them to expand quickly.
Admittedly, our ability to design such maverick interventions—across
markets—is not problem-free and may be limited. Cooperatives, subject to

weak corporate governance, may dissipate their profits on internal salaries,

perks, or expansion into other markets. Beyond the difficulties in spon-
soring entry, the incumbents can develop counterstrategies that ultimately
thwart the entrant’s market-share growth. In industries where competitors
compete in multiple products and geographic markets, computers can
learn to retaliate across markets (e.g., the incumbent offers a steep discount
in the maverick’s home market or markets sheltered from competition),
which the maverick’s pricing algorithm can quickly learn is correlated with
its discounting.
Incumbents can devise strategies to reduce the maverick’s incentive to
discount, thus fostering coordinated behavior instead. With improvements
in technology, companies can replicate Uber’s God View, seeing where
their shoppers are currently located and what they are doing. Using real-
time geolocation data, companies will know if their loyalty card shoppers
have entered a rival’s store. They will know if their consumers are fre-
quenting a competing website that is promoting the same item at a discount.
So, using this God View, companies could avoid an all-out price war. Instead,
they target the maverick’s customers, and not their fellow Frenemies’, to
marginalize the maverick while avoiding the nuclear option.

Reducing Price Transparency

We are accustomed to treating transparency as a condition for competi-
tion. But we want to design ways for companies to undercut the market
price using nontransparent communications with buyers. Careful design
of secret bids and sales could help destabilize algorithm-supported tacit
collusion.
One way to decrease transparency is by offering discount cards that pro-
vide secret discounts without collecting data on users (to avoid our behav-
Copyright © 2016. Harvard University Press. All rights reserved.

ioral discrimination scenario). Dealers may engage in reverse auctions and

sell below the market price. Such strategies already exist in some industries
and sectors, from vehicle sales to legal ser vices.36 Similarly, some websites
allow buyers to solicit offers from undisclosed sellers who compete against
each other for the sale of electric appliances and other products.
Alternatively, the government may attempt to reduce the speed with
which sellers can adjust prices. Such an approach has been implemented in
the fuel sector in Austria and Western Australia, where sellers are limited
in their ability to match each other’s price more than once a day. Such

provisions aim to reduce the number of price changes, open the way for
competitors to undercut the collusive price, and promote a seller’s reputa-
tion as a discounter. The pricing algorithms, while continually monitoring
the rivals’ pricing and business maneuvers, would now face a time delay
in changing price. Under this scenario, the maverick—if the delay were
long enough—could profit from being the first to lower prices.
This solution has its problems: competitors would soon complain that the
government is preventing them from discounting. One alternative would be
if the government allowed price decreases to be implemented immediately,
but imposed a time lag for price increases. But pricing algorithms, like
humans, could game the system. For example, a dominant incumbent could
punish the maverick by undercutting its price. The maverick could not im-
mediately raise its price, and would be forced to lower its price even further.
Taking this into account, the maverick’s algorithm, before discounting,
would likely calculate the probability of incumbents retaliating, its costs
(including lost profits) in discounting, and the benefits (which would be
slight if rivals could instantly match the maverick’s lower price). The govern-
mental pricing delay—rather than helping the maverick and consumers—
would instead serve as a punishment mechanism for defecting from the
supracompetitive price. In reducing the maverick’s incentives to lower
prices in cases where retaliation is likely, the governmental pricing delay
could unintentionally foster tacit collusion.

Auditing the Algorithm

Finally, we note the possibility of “enhanced ex-ante monitoring” of firms’
algorithms. Such a mechanism, in promoting greater transparency, may en-
able “public countermeasures” when industry-wide signaling and stabilizing
features are identified. Disclosure of algorithms may also help address pos-
Copyright © 2016. Harvard University Press. All rights reserved.

sible quality degradation or manipulation of search results. Companies may

be required to change features in their algorithm, or, alternatively, the gov-
ernment may use the information to design its own market countermea-
sures. To safeguard the intellectual property rights of online companies,
disclosure may be confined to a dedicated enforcement agency and handled
under strict confidentiality.
Admittedly, the audit route has limited practical appeal. At least three
challenges exist: First, algorithms are not likely to include simple “collusive”
instructions, but rather are used to observe market conditions and react
in a profit-maximizing way. Running them in a “Sandbox”—a controlled

environment that tests their operation—may not always reveal their true
effects. Their sophistication may make proof of an illicit aim or action dif-
ficult. Auditing may thus fail to lead to meaningful intervention.
Second, even if one can show that algorithms facilitate tacit collusion,
the remedy may be challenging. For instance, it may be impractical to re-
quire computers to ignore information that is available to everyone online.
One possibility may be to focus on commercially sensitive information
that, although publicly available, is of little or no value to customers but
helps the competitors arrive at a supracompetitive price.37 Here the focus
is on “cheap talk,” that is, data exchanges that facilitate conscious paral-
lelism but are of limited use to customers. One problem, however, is in
identifying such information. Part of the value of Big Data is data fusion,
whereby computers link data sets, from which new insights emerge.38 More-
over, the data for some applications—such as customers sharing their in-
ventory data with suppliers—can promote efficiency even while raising anti-
trust concerns.39 Even if the customers seek to limit what information can
be shared, the algorithms—by analyzing a variety of data—could fi ll in the
gaps. So it would likely be difficult (and potentially welfare-reducing) for
the government to specify what data the algorithms must ignore.
A third challenge concerns the ability to effectively audit an algorithm
or intervene in the market. One risk is that the government will re-
main several steps behind. Competition authorities may have a difficult
time overseeing fi rms’ design and development of sophisticated algo-
rithms. Further, state countermea sures implemented following an
audit are not likely to keep pace with the industry’s evolving, self-learning
algorithms.
While limited in appeal in 2016, the audit route may become feasible as
technology and enforcers’ proficiency develop.
Copyright © 2016. Harvard University Press. All rights reserved.

Reﬂections
The enforcement challenges presented by the new market dynamics may
be divided into four groups. The first is ideology around how markets
operate and the goals of competition law. The second is political will. The
effects of intellectual and regulatory capture, as the next chapter explores,
cannot be underestimated. The third is ingenuity. The lack of tools may
represent a policy failure—namely, the agency’s pursuit of what is readily
quantifiable over what is important. It is all the more startling that in the
age of Big Data and Big Analytics, the government is the entity that claims

it lacks the data or analytical tools to assess a restraint’s impact. For the
areas we identify where their current tools may be ill-suited, the agencies
should study how they can refine them or develop alternatives.
Finally, the law at times is messy. Some competition agencies envision
themselves as surgeons, removing the anticompetitive restraint while
leaving intact the procompetitive (or competitively neutral) behavior. That
is the ideal. The risks of legal uncertainty and enforcement costs dictate a
careful approach when designing new instruments. In designing and en-
forcing, one should be aware of our limited ability to predict the future,
the dynamics of the market, and changes in strategies. Still, we would argue
that a blanket refusal to engage in a discussion of possible new tools would
be effectively allowing the patient to die on the table. It stems from the
misguided assumption that we are merely dealing with old wine in new
bottles—that, while digital markets are new and different, the competi-
tive dynamic remains the same and does not call for change.
Copyright © 2016. Harvard University Press. All rights reserved.

A S WE HAVE EXPLORED, the upsurge of algorithms, Big Data, and super-

platforms will hasten the end of competition as we know it—a decline
of the market system to which we have become accustomed. But as we
stated at the outset, the new market dynamic is not necessarily bleak. The
innovations from machine learning and Big Data can be transformative—
lowering our search costs in finding a raincoat or parking spot, lowering
entry barriers, creating new channels for expansion and entry, and ulti-
mately stimulating competition.
Big Analytics and Big Data can be beneficial, no doubt. As we saw, these
innovations are not inherently good, bad, or neutral: it will depend on how
the companies employ the technologies, on whether the companies’ incen-
tives are aligned with our interests, and on their actions’ collective impact
on markets.
The new competitive landscape, however, is not necessarily rosy either. We
explored three core dynamics—collusion, behavioral discrimination, and
Frenemy—where the rise of sophisticated computer algorithms, artificial
intelligence, and Big Data can change our competitive paradigm and
Copyright © 2016. Harvard University Press. All rights reserved.

market reality for the worse. These dynamics require us to look beyond
the façade of competitive progress, acknowledge its pitfalls, and explore
smart intervention to promote our welfare. We conclude with the following
reflections.

The Descent from King to Slave on the Data Treadmill

How do we know when competition is a mirage? In competitive markets,
the economist Ludwig von Mises observed, the customer, not firms, should
233

be supreme. In our purchasing behav ior, we ultimately determine “what

should be produced and in what quantity and quality.”1 Von Mises, in his
belief in consumer sovereignty, was skeptical about the evils of private
monopolies—rational consumers with willpower often can take care of
themselves. However, the market realities raise doubt as to our actual power
and control.
If we reigned supreme, we would get products that advanced our privacy
interests. We would not have to wait years to get ad blockers for our mobile
phones. We would not have to search the web in hopes of finding some
technology that promotes our privacy interest. If we believed behavioral
discrimination was unfair, companies would not segment and target us
with differential pricing. Nor could companies tacitly collude and charge
us higher prices as a result.
In competitive markets, firms collaborate to innovate or lower produc-
tion costs.2 Yet, as the Frenemy scenario explores, the cooperation and
competition are misdirected. These new dynamics change competition as
we know it from “consumer welfare/surplus” competition, which seeks to
improve our well-being, to “producer welfare/surplus” competition, where
firms cooperate to extract as much consumer surplus as possible, and then
compete over the spoils. Importantly, in producer welfare/surplus compe-
tition, the perceived attributes of a competitive market may exist, yet we
do not benefit.
No doubt super-platforms invest in research and development and im-
prove the interface and ser vices available to us. Th is, at times, improves
overall welfare. But whatever one thinks of the super-platforms, one does
not think of them as the authors or artists of any notable work. Facebook
does not supply original content that attracts users to its platform: the
content comes from other people we know. The same is true for Twitter,
LinkedIn, and many of Google’s ser vices. We largely create the YouTube
Copyright © 2016. Harvard University Press. All rights reserved.

videos ourselves or upload other people’s content. None of the super-

platforms have written an outstanding novel, directed or produced an
award-winning film, scored a musical or opera, staged a play, or informed
the public debate with any investigative journalism. Other people and com-
panies provide the creative content, which the super-platforms’ algorithms
identify to attract our attention.
What is remarkable is that consumers are often toiling away on the
super-platform, providing content without compensation, which the super-
platform uses to attract others. The more content we create or the more
interesting our content is, the more likely others will visit the super-

platform, the more data we collectively generate, and the more we can be
targeted with behavioral ads. Indeed, the super-platform enables us to
compete for status over how many “likes” our posts receive, how many fol-
lowers we have, or how often our postings are viewed. We are competing
against each other for a relative advantage in status.
Yes, one might say, but we get it all for free. Free, however, is only a price
point on a spectrum. We would likely pay some amount if we did not pay
with our data (and being targeted with ads). But this is not to say we are
fairly compensated for our data and content. Indeed, our toil on the super-
platforms and the data collected about us are worth far more than the cost
of developing and providing the technology. As we have seen, some tech
firms have put a value on our data. Google estimates $720 per person, per
year when it talks to investors.3 One key metric for Facebook, among others,
is daily active users: “Trends in the number of users affect our revenue and
financial results by influencing the number of ads we are able to show, the
value of our ads to marketers, the volume of Payments transactions, as well
as our expenses and capital expenditures.”4 As Figure 6 shows, Facebook’s
average revenue per user (ARPU) has increased steadily since 2010.5
Google’s ARPU in the first quarter of 2014 was even higher—six times
higher than Facebook’s.6 Both companies earn billions of dollars in profits
from their advertising platforms.7
In a consumer-oriented competitive market, we could conceivably de-
mand compensation for our data and our contributions to YouTube, Face-
book, LinkedIn, and Twitter. Yet in our online environment this power has
been diluted. Indeed, Facebook users effectively become free endorsers
when they “like” a product, advertisement, or company; their photo and
identity can now be used in that product’s advertisements targeted at friends,
family, and others.8
Facebook notes how “[t]he size of our user base and our users’ level of
Copyright © 2016. Harvard University Press. All rights reserved.

engagement are critical to our success.”9 If Facebook users stopped toiling

away for free and hopped off the data treadmill, then the quality and fre-
quency of postings would decrease, and Facebook’s profits would shrink.
As Facebook tells investors, it is “ ‘vital’ to encourage a broad range of users
to contribute content.”10 One fear at Facebook is that users will provide less
content to attract others to its platform. In the third quarter of 2015, for
example, a market researcher found that “34% of Facebook users updated
their status, and 37% shared their own photos, down from 50% and 59%,
respectively, in the same period a year earlier.”11 Facebook thus seeks to
nudge us onto the data treadmill. For example, since May 2015, to spur

!"" #

%&'

!"# ( )* !" !"

Figure 6. Facebook’s average revenue per user, in U.S. dollars (http://www

.statista.com/statistics/251328/facebooks-average-revenue-per-user-by-region/)

conversations, Facebook has drawn on users’ likes and location to place

“prompts related to ongoing events at the top of some users’ news feeds.”12
Facebook, some argue, taps into our fear of missing out with habit-forming
technology, like Instagram.13
Thus Facebook has every incentive to steer us onto its super-platform, and
having us work—by posting content, commenting on other people’s content,
and supporting the advertisers. It is doubtful that Facebook would be as
popular as a social network if another network compensated users for posting
Copyright © 2016. Harvard University Press. All rights reserved.

content and spending time on its platform. Nor would many of us be satisfied
with a free flashlight app if we knew how much the app maker was profiting
from our data. And yet, the allegedly powerful forces of the invisible hand
and competition have yet to deliver such new entry despite likely demand.

The Positive Feedback Loop Where the Big Get Bigger

So consumers aren’t necessarily reigning supreme. The negative develop-
ments we identify from the collusion, behavioral discrimination, and

Frenemy scenarios have already surfaced and will likely continue. But from a
policy perspective the harm is less concerning if it is episodic and fleeting. So
will the market power in our three scenarios likely be fleeting or durable?
While we can hope for the former, the market characteristics suggest the
latter. One threat to this anticompetitive equilibrium is if mavericks quickly
enter the market, restore competition, and improve our well-being. As our
three scenarios take hold in different markets, it may become harder for
independent mavericks to curb the abuses and safeguard our privacy.
One reason, as we saw in Part II, is that greater market transparency cuts
both ways. Granted, imperfect information and less transparency can
foster market manipulation. Indeed, a seller may manipulate the market—
engage in secret discounts and, like a stock trader, profit from the ineffi-
ciencies through arbitrage. And yet, as pricing is increasingly powered
by algorithms, as the velocity in accessing and responding to market in-
formation increases, and as algorithms quickly adjust prices in response
to discounts or increases, the increase in transparency can yield tacit
collusion—where competitors perhaps offer more choices, albeit at higher
prices and with little differentiation.
With their data-generated God View, firms will observe not only their
competitors’ current prices, but any nascent competitive initiatives. Tesco’s
or Kroger’s computers, for example, can discern from the inflowing data
that a particular loyalty card customer is driven by milk prices and is likely
to shop for milk and other food products on the weekends. From the in-
coming geolocation data, the supermarket knows whether its customer is
driving toward a rival supermarket. It can text the customer a significant
milk discount to woo him to its store, and thereby blunt the competitor’s
incentives to poach its customers. Thus in tracking customers and se-
lectively discounting, retailers can thwart their rivals’ efforts to woo
customers with lower prices. If rivals persist in discounting, the nuclear
Copyright © 2016. Harvard University Press. All rights reserved.

option becomes a significant threat; the pricing algorithms will soon learn
that they can’t profit by lowering prices. For homogeneous goods, one likely
outcome will be the collusion scenario. For differentiated goods, the
likely outcome will be behavioral discrimination. Or at times the firms
will offer the product or ser vice for “ free,” and the competition will be
over extracting wealth through behavioral advertising.
Another reason why market power will likely be durable is network ef-
fects.14 Operating systems, we saw, are a classic example of network effects:
the more people that use the platform, “the more there will be invested in

developing products compatible with that platform, which, in turn rein-

forces the popularity of that platform with users.”15 Thus network effects
help insulate Google’s and Apple’s market power over mobile phone oper-
ating systems. As The Economist reported, “Alphabet [Google], Facebook
and Amazon are not being valued by investors as if they are high risk, but as
if their market shares are sustainable and their network effects and accumu-
lation of data will eventually allow them to reap monopoly-style profits.”16
Positive feedback loops and data-driven network effects can play a sig-
nificant role here.17 The new currency in our Frenemy and behavioral dis-
crimination scenarios is data. As one consulting firm observed, “Big Data
may well become a new type of corporate asset that will . . . function much
as a powerful brand does, representing a key basis for competition.”18 The
ability to gather data enables super-platforms’ and sellers’ algorithms to
better classify and target individual users, and consequently allows them
to foster an ongoing relationship entailing additional interactions. Suc-
cessful data harvesting and analysis can have a snowball effect, enabling a
provider to better target customers, thereby reinforcing its power by at-
tracting additional users.19 This enables a provider “to gather even more
valuable data about consumer behaviour, and to further improve ser vices,
for (new) consumers as well as advertisers (on both sides of the market).”20
Firms with more users, more personal data, and better algorithms can better
price discriminate or wield greater power in the Frenemy scenario, so
firms will seek a data advantage over rivals. In its 2015 report on Big Data,
the White House noted that “even small improvements can have a large
impact on profitability, particularly for companies with a large customer
base.”21 One example it gives is Netfl ix, which, a 2014 study described,
could boost profits by using behavioral data for personalized pricing.22
Shielded by data-driven network effects, firms can use their data advan-
tage and greater opportunities to experiment to teach their algorithms to
Copyright © 2016. Harvard University Press. All rights reserved.

be smarter than their rivals’. In quickly accessing and analyzing our per-
sonal data, the super-platforms have powerful tools that the monopolies of
yesteryear lacked, such as the ability to discern trends and threats well be-
fore others, including the government.23
Their superior market position enables them to dictate the interaction
not only with us—the customers—but also with small and medium size
companies. The latter, just like us, may lack the resources, data, and algo-
rithms to effectively curb the power of the super-platforms when they ex-

pand into their markets. Even Goliaths like General Motors are wary that
with self-driving cars, the lion’s share of profits will go to those tech firms
that develop the algorithms and collect the data.24 The auto manufacturers’
profits (and their unionized workers’ salaries) are squeezed.
The winners in the data-collection arms race benefit in several ways: first,
in improving their self-learning algorithms; second, in capturing greater
value from the data (either directly or indirectly through advertising-
related ser vices or behavioral discrimination); third, in using the profits to
expand their platform, thereby attracting more users, advertisers, and per-
sonal data; and finally, as their platforms evolve into super-platforms, in
becoming the lords of the new market order—keepers of the data—who
can promote or disrupt competition at their will.
Thus we enter a data-collection arms race, where firms have little incen-
tive to protect our privacy interests.25 A data advantage over rivals can en-
able the company to achieve critical economies of scale, which could tilt
the data—and the competitive balance—in its favor. Indeed, leading com-
panies do not limit themselves to mere improvements in harvesting and
analyzing data; they also compete on infrastructure and in emerging mar-
kets. As Evgeny Morozov noted, “Google and Facebook have figured out
that they cannot be in the business of organizing the world’s knowledge if
they do not also control the sensors that generate that knowledge and the
gateways through which it passes.”26
Mavericks, like Disconnect, may offer privacy technologies that seek to
protect us from being tracked and our personal data from being collected. But
they currently are swimming against the current. None of the super-platforms
are promoting these privacy technologies. Indeed, some super-platforms
perceive these privacy technologies as threats to their advertising-revenue
business models. Thus the incentives and ability to innovate diminish, espe-
cially when a super-platform kicks these mavericks out of its app store. Any
Copyright © 2016. Harvard University Press. All rights reserved.

maverick will end up investing heavily to reach us—paying the tax levied by
the new dynamic.

Technology and Wealth Inequality

Even if the harms we identify are real and persistent, they might be over-
shadowed by the procompetitive gains we identify in Chapter 1. Granted,
consumers might lose in some markets, but overall they may benefit under

the new competitive dynamics, with more consumer surplus (and money)
to spend or save. Overall their wealth may increase.
The fact that some online markets are competitive does not excuse anti-
competitive tactics in other markets. But assessing the gains and losses puts
the problems we identify into perspective. Overall, all else being equal, will
wealth inequality increase, decrease, or remain unchanged as we enter the
age of virtual competition?
The current record wealth inequality is a pressing issue. In January 2016,
ahead of the Davos World Economic Forum, Oxfam released a report on
global wealth inequality. According to the report, sixty-two billionaires
own the same wealth as half the world’s population—3.6 billion people.27
As an Oxfam executive director said: “half of the world’s population—
that’s three and a half billion people—own no more than a tiny elite whose
numbers could all fit comfortably on a double-decker bus.”28
One factor contributing to income and wealth inequality is market
power.29 Consider, for instance, the state of competition in the United States.
In 2016, the White House issued an executive order30 and report by its
Council of Economic Advisers (“CEA Report”).31 The CEA Report high-
lighted several signs indicating a decline in competition in the United
States since the 1970s. First, entry appears to be decreasing in many eco-
nomic sectors, including a decades-long decline in new business forma-
tion. The United States is seeing lower levels of firm entry and labor market
mobility. Second, many industries are becoming more concentrated. Third,
increasing industry profits are falling into the hands of fewer firms. Basi-
cally, the CEA Report identifies how more industries are dominated by
fewer firms (increasing concentration). These few powerful firms are ex-
tracting greater profits (and wealth) from workers, sellers, and consumers.
And it is getting harder for new firms to enter markets and for workers to
change employers. The solution is more competition. This includes, the
Copyright © 2016. Harvard University Press. All rights reserved.

CEA Report noted, robust antitrust enforcement.32 The Council of Eco-

nomic Advisors also noted that as more sectors of the economy are digi-
tized, the government needs to consider how digitization is impacting
competition, and whether additional regulation is needed.33
Thus, in assessing whether the emerging digitized hand is reducing
market power and increasing consumer surplus, one metric is wealth
inequality. It should decrease. But this isn’t guaranteed. The increasing
importance of Big Data, a UN panel of experts noted in a 2014 report, is
creating the prospect that “a whole new inequality frontier will open up,

splitting the world between those who know, and those who do not.”34 The
concern is that the boom in Big Data and information asymmetry perpet-
uates wealth inequality.35 As one writer has noted, “the data mining of in-
dividual privacy is fundamentally reshaping markets by transferring so
much knowledge about user interests, behavior and desires into a few cor-
porate hands.”36
We can see how firms extract wealth in our three scenarios. In the col-
lusion scenarios, the money is extracted directly from purchasers’ pockets
in the form of higher prices. Even where the collusion involves interme-
diary goods, the cost increase can translate to higher prices for the fin-
ished goods that we buy.
Technology, in our behavioral discrimination scenario, may increase
wealth inequality. As we saw, one defense for price discrimination is that it
increases market access to poorer customers (such as university grants and
scholarships). But the victims of behavioral discrimination will not always be
the wealthy who are willing to pay more. The victims will often be the poor
whose circumstances limit viable outside options. Take, for example, Staples.
It was discriminating against—rather than in favor of—the poor. The Wall
Street Journal, in examining Staples’ online price discrimination, found that
customers living in wealthier communities generally paid lower prices.37 The
wealthy (who presumably had more outside options, such as more competing
stores nearby or greater ability to drive to alternatives) received discounts.
In our Frenemy scenario, wealth extraction can occur on several levels.
First, the super-platforms extract wealth from users by getting their valu-
able data, including their likes, dislikes, intentions, and so on, without
having to pay its fair market value. Second, the super-platforms extract
wealth by getting content from users for free. (In a competitive market one
might expect rivals to offer valuable consideration to induce users to
provide them content.) Third, the super-platforms help extract wealth by
Copyright © 2016. Harvard University Press. All rights reserved.

advancing behavioral advertising and discrimination. Fourth, the super-

platforms extract wealth when they scrape valuable content from other
websites and post it on their own.38
Thus, as super-platforms expand into personal assistants, the Internet of
Things, and smart technologies, the concern is that their data advantage
increases their competitive advantage and market power. As their capacity
to extract our wealth increases, more money will flow to the few super-
platforms. Interestingly, in early 2016, the growing profitability of one
super-platform, Google, sparked “an after-hours [trading] rally that

propelled it past Apple Inc. as the world’s most-valuable company.”39 Thus

the two mobile super-platforms in early 2016 were the world’s two most
valuable companies.

Privacy and Trust

Wealth transfers, for some economists, are a nonissue. Widows and or-
phans, they observe, could be shareholders in the monopolies. When the
monopoly profits don’t return to the poor and middle class, the govern-
ment, through taxes and other programs, can redistribute the wealth. The
economists’ concern is deadweight welfare loss, namely transactions that
would have occurred in a competitive market but are forgone in the cartel-
ized or monopolized industry.
Competition officials, at times, see the privacy issues we raise as unrelated
to their competition concerns. In the Facebook/WhatsApp merger, the
European Commission recognized that Facebook may weaken WhatsApp’s
privacy commitments and start collecting and using personal data from
WhatsApp users. The Commission, however, focused primarily on adver-
tisers, not individuals:
For the purposes of this decision, the Commission has analysed poten-
tial data concentration only to the extent that it is likely to strengthen
Facebook’s position in the online advertising market or in any sub-
segments thereof. Any privacy-related concerns flowing from the in-
creased concentration of data within the control of Facebook as a result
of the Transaction do not fall within the scope of the EU competition law
rules but within the scope of the EU data protection rules.40

These privacy-related effects, for some, are unrelated to their notion of

any way in which consumer welfare is currently being harmed by Google.”42

Others take the view that merely conducting a relatively narrow cost-
benefit analysis of the issue “diminishes privacy’s status as a right.”43
These views are mistaken. Just because an app is free, and continues to
be free after the merger, doesn’t mean that consumers will necessarily ben-
efit. Privacy protection can be a parameter of non-price competition, namely
quality.44
Moreover, privacy at times fits within neoclassical economic theory’s
concern over deadweight welfare loss. The virtual economy, as Chapter 1

discusses, can promote allocative efficiency. But it can also increase the
deadweight welfare loss by increasing distrust. Market economies rely on
trust.45 Fairness and trust, business and economic research shows, are
highly interrelated. On a macro level, the empirical evidence does not iden-
tify greed as a prerequisite for a market economy.46 Societies with greedier
residents do not necessarily have stronger economies. Instead, norms of
fairness can play a far greater role than greed in supporting a market
economy. As Professor Lynn Stout discussed, societal norms of fairness
and prosocial behav ior are both common in and necessary for a market
economy.47 Violations of social norms of fairness decrease trust and in-
crease retaliation.48 How trusting can you be in a world where people will
seek whenever possible to profit at your expense? The transaction costs in
a world where greed runs amok would be astronomical. Many people in
behavioral economic experiments are trusting. However, their willingness
to trust and cooperate is conditional, depending on the actual or expected
cooperation of others.49
For online markets to deliver their benefits, people must trust firms and
their use of their data. But as technology evolves and more personal data is
collected, we are increasingly aware that companies are using our personal
information for their own benefit, not ours.50 There is growing mistrust of
how firms use Big Data and Big Analytics.51 Take, for instance, the nu-
merous challenges involving Google’s alleged evasion of privacy protec-
tions,52 and the revelation that owners of Nissan Leaf cars were being
tracked (and that other users could access the relevant data).53 Many people,
the U.K. competition agency recently found, are unhappy with how well
firms explain why they collect data.54 As the agency concluded, “Consumer
trust could be fragile and at risk if negative perceptions about new tech-
nologies or the way firms manage data take hold. We are concerned that
future changes in the way that data is collected and used (such as more pas-
Copyright © 2016. Harvard University Press. All rights reserved.

sive collection via the [Internet of Things]) could test how far consumers
would be willing to continue to provide data.”55
If industries advance closer toward perfect behavioral discrimination,
consumers will increasingly distrust ser vice providers, platforms, mobile
devices, and wearables. Consumers may no longer use the technology—
such as search engines, mobile technology, smart electric meters, smart
watches, and other wearables—to the same degree, out of privacy concerns.
They may forgo purchases that they otherwise would have made if they had
greater trust in the seller and the fairness of the market norms. Thus, as

the volume, variety, and complexity of personal data being collected ex-
pands, privacy issues will likely become more pressing.56 As distrust in-
creases, so too will the deadweight welfare loss.

Money, Power, and Politics

Besides the deadweight welfare loss, the rise of market giants in control of
data, analytics, and the ecosystems raises challenging questions about the
concentration of power.57 To what extent does economic concentration
affect enforcement?
Economic power can translate into political power. Corporations and
trade groups spend billions of dollars lobbying the U.S. government.58 And
with the rise of private economic power, the dividing line between corpo-
rate action and the political realm may easily fade.59 The extent of crony
capitalism may surprise some. Many of us knew that the major financial
institutions had clout, but the economic crisis and subsequent U.S. tax-
payer bailouts exposed the extent to which they manipulated the regula-
tory environment in their favor, and how the economically powerful have
every incentive to use the government to protect their economic interests.
Not surprisingly, two-thirds of Americans, including a majority of Repub-
licans, believe that the economy “unfairly favors powerful interests.”60
Now the clout of the tech giants and their influence on government pol-
icies are coming to light. The super-platforms and other interested parties
are investing heavily in shaping the debate. Take, for example, Google’s
lobbying, which captured the media’s attention in recent years. The week
after it became public that the U.S. Federal Trade Commission was inves-
tigating Google for monopolistic abuses, the company hired twelve addi-
tional lobbying firms61 and increased its lobbying expenses in 2012 by
88 percent, becoming among “the top 10 of spenders seeking to influence
Copyright © 2016. Harvard University Press. All rights reserved.

the federal government.”62 During this time, Google, the Wall Street Journal
reported, “had a flurry of meetings with top officials at the White House
and Federal Trade Commission.” 63 Google reportedly established a close
relationship with the Obama administration: “Google representatives at-
tended White House meetings more than once a week, on average, from
the beginning of Obama’s presidency through October 2015. Nearly 250
people have shuttled from government ser vice to Google employment or
vice versa over the course of his administration.”64 We don’t know what
was discussed; these meetings might have been benign or involved an
outsider’s perspective on impor tant policy issues. Still, they reflect one

firm’s unparalleled access to the highest levels of the Executive Branch,

and the opportunities to align state policies with corporate interests.
With such influence in mind, eyebrows were raised when it emerged that
the FTC’s decision to close its Google investigation may have been incon-
sistent with the recommendations of its legal staff.65 The story came to light
in 2015 as the Wall Street Journal inadvertently received portions of the FTC
legal staff report. The legal staff recognized the likely defenses and challenges
in bringing a monopolization case, and opted to forgo challenging some of
Google’s practices. But the FTC legal staff, after a lengthy investigation, con-
cluded that Google’s conduct “has resulted—and will result—in real harm
to consumers and to innovation in the online search and advertising
markets.”66 Thus the legal staff recommended suing Google for abusing its
monopoly power in several ways that harmed Internet users and rivals.67
After the FTC staff report became public, there was an outcry, especially
among European policymakers, as Google was under investigation by the
European Commission. Google’s lobbyist e-mailed the FTC saying the
company was “deeply troubled”68—its rivals were using the FTC staff re-
port to “sow confusion and undermine the FTC’s conclusions, especially
in Europe.”69 It pressed the FTC to defend its decision in a press statement.
“Two days after the email was sent, and after the Wall Street Journal pub-
lished another article about Google’s relationship with Washington,” the
reporter who broke the story noted, “the FTC released a statement that pro-
vided the context [the Google lobbyist] had sought.”70
Moving beyond this story, it is important to understand why lobbying
by leading companies and key economic players will likely intensify with
virtual competition. The stakes are greater, as firms can extract even
more wealth under our collusion, behavioral discrimination, and Frenemy
scenarios. Their increasing economic power will likely translate to political
power, influencing governmental policies to preserve the status quo.71
Copyright © 2016. Harvard University Press. All rights reserved.

Moreover, the legal standards applicable to many of our scenarios invite

lobbying. Let us explain why:
For hard-core antitrust violations, such as the Messenger and perhaps
Hub and Spoke collusion scenarios, the per se illegal antitrust standard
limits defenses (and discretion). But for our other scenarios, greater discre-
tion exists. It is basic economics that the more discretion the government
has in bringing and determining violations, the more prone its policies are
to distortion by lobbyists. The vaguer the legal standard, the more subjec-
tive input it allows from lobbyists. The less transparent the review and its
objectives, the less predictable the enforcement becomes.

Consequently, the problem is not lobbyists. The problem is the combi-

nation of concentrated economic power, weakened limits on corporate po-
litical spending,72 and an amorphous legal standard, such as the Supreme
Court’s “rule of reason” legal standard for most antitrust violations.73 The
amorphous legal standard is attractive to economists, lobbyists, and anti-
trust counsel who “know” and “can work” with the agency to dissuade it
from intervening in our three scenarios.

Intellectual Capture
Closely linked to economic power is the ability to foster intellectual and reg-
ulatory capture. As anthropology professor David Graeber observed, “if 1% of
the population controls most of the disposable wealth, what we call ‘the
market’ reflects what they think is useful or important, not anybody else.”74
Lobbying, discussed above, provides a central tool to shape opinions of
governments and the public—to affect the public debate and our perception
of right and wrong.75 Other means to capture the debate include the funding
of articles, academic initiatives, and think tanks.76 Here one may harness the
credibility of individuals and institutions to propagate certain ideas and
create a pool of supportive media and writing that can cross-reference itself.
Google, for example, from the beginning of the FTC investigation through
the end of 2013, reportedly gave George Mason University’s Law and Eco-
nomics Center $762,000 in donations.77 Why? The center “issued numerous
studies supporting Google’s position that they committed no legal violations,
and hosted conferences on the same issues where Google representatives sug-
gested speakers and invitees.”78 Between 2009 and 2015, at least “66 published
studies by over 45 academics” were reportedly either “ ‘commissioned by
Google,’ ‘funded by Google,’ or ‘supported by a gift from Google, Inc.’ ”79
Google is not alone. We should expect the beneficiaries of our three sce-
Copyright © 2016. Harvard University Press. All rights reserved.

narios to fund articles, academic initiatives, and think tanks to frame the
“virtual competition” debate in ways that support their corporate agenda.
When confronted with market realities—such as declining upward mo-
bility, diminishing rates of small-company creation, increasing market
concentration and power, and widening wealth inequality—they will likely
pivot from the economic realities to the neoclassical economic theories of
self-correcting markets. They will likely argue that any meaningful inter-
vention in our three scenarios would cause greater harm than good. The
dynamic market’s self-policing powers, we’ll be told, will prevent compa-
nies from inflicting significant harm and will safeguard our welfare.

Similarly, arguments in favor of any form of intervention are likely to be

framed as opposing technology or innovation. For example, in his opening
statement in the historic 1998 antitrust trial, Microsoft’s lead attorney com-
pared the Justice Department to nineteenth-century Luddites who smashed
labor-saving machines “to arrest the march of progress driven by science
and technology.”80 The government’s case was “a repudiation of the basic
principle in our society that creative commercial activity should be encour-
aged and rewarded.”81 Of course, as European and U.S. courts have recog-
nized, monopolistic abuses don’t foster innovation; rather, their abuses, left
unchecked, can cause greater harm to technological innovation. But the
doublethink is that hindering monopolistic abuses hinders innovation.
With the dominant firms’ many levers to affect public opinion, we cannot
help but wonder whether competition enforcers and regulators can resist
the intellectual capture skillfully propagated by these giants—through
media, lobbying, political power, and funding. Is the controlled ecosystem
limited to our online environment, or is it already capturing the main
junctions in our political environment—leaving us with only the illusion
of autonomy and independence?

The Road Ahead

We are often told how the technological revolution will increase our wel-
fare through greater transparency, communication, choice, and value.
Computer algorithms are constantly improving. The use of computers and
their interface with humans are rapidly evolving. The Internet of Things
will expand, with more sensors in the home, car, work, and on our body.
The technologies’ potential benefits are often significant and self-evident,
from increasing our health to finding an available parking space. We cer-
tainly agree, and welcome these transformative technologies. But these
Copyright © 2016. Harvard University Press. All rights reserved.

exciting developments and their welfare benefits should not mask the
potential risks.
Our aim was to show that data-driven online markets will not neces-
sarily correct themselves. Even when markets are competitive, this pres-
sure may not necessarily improve our welfare. With the changing market
reality, new dynamics may change our ability to safeguard our welfare. It
may become even harder to effectively resist the flow of our diminishing
wealth to fewer, more powerful firms. As power shifts to the hands of the
few, the risks this will likely have for competition, our democratic ideals,
and our economic and overall well-being will increase accordingly.

Thus, careful and measured intervention will likely be warranted. Main-

stream competition, privacy, and consumer protection enforcement and
regulation all have an important role in safeguarding our welfare. As we
illustrated, however, some remedies that traditionally come to mind are
imperfect. As market dynamics evolve, new and carefully measured instru-
ments will be needed—to deter the abuses by super-platforms and harms
from the Frenemy dynamics, to prevent computers from colluding, and to
block the road to perfect behavioral discrimination. Education should also
play a key role. The notion that “free may be expensive” should not be the
preserve of the privileged few. Competition agencies and our elected gov-
ernments have a responsibility to educate customers as to the promises and
perils of virtual competition.
Importantly, enforcement agencies must devote resources to under-
standing the changing market dynamics and incentives, the role and use
of data and algorithms, and the implications for our welfare. In addition,
faced with the limited utility of current legal doctrines on agreement and
intent in the age of pricing algorithms, agencies will face the challenge of
updating the enforcement toolbox to match the emerging challenges.
This task has never been easy. It is even harder in dynamic markets. Ill-
advised or misguided intervention, without a clear and credible theory of
harm, can carry significant welfare costs. But so will ignoring the scenarios
we raise.
Even more worrying is the intellectual capture that may inhibit a critical
look at market dynamics. Ultimately, it comes down to political will. The
most daunting challenge is getting government agencies to respond. Some
agency officials, in our discussions about this book, were engaged and eager.
Others were somnolent. One was comatose. The siren song of power ful
players both inside and outside the government may undermine change and
discredit attempts to evaluate new market dynamics.
Copyright © 2016. Harvard University Press. All rights reserved.

So political pressure is needed. Antitrust under the Obama administra-

tion is perceived at best as mixed and at worst feeble. Despite having one of
the older antitrust laws, the United States, as one former official noted, “is
no longer viewed as the intellectual leader of antitrust.”82
So to you, our dear readers who took the time to join this endeavor, we
thank you. Few get excited about antitrust anymore. But apathy has a price.
We cannot assume that the digitized hand will always protect our welfare.
It is ultimately up to us to start asking our elected officials and agencies
what they are doing to prevent these scenarios from happening.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Notes
Acknowledgments
Index
Copyright © 2016. Harvard University Press. All rights reserved.

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Copyright © 2016. Harvard University Press. All rights reserved.

1 • The Promise of a Better Competitive Environment

1. Financial Conduct Authority, Price Comparison Website: Consumer Market

Research (June 2014), 7, https://www.fca.org.uk /static/documents/research
/price-comparison-website-consumer-research.pdf.
2. For instance, the U.K.’s Financial Conduct Authority estimates that one-
third of the 26.6 million auto insurance policies written in 2013 were sold
through PCWs; Financial Conduct Authority, Price Comparison Websites
in the General Insurance Sector, Thematic Review 14/11 (July 2014), 3,
https://www.fca.org.uk /static/documents/thematic-reviews/tr14-11.pdf.
3. G. Stigler, “The Economics of Information,” Journal of Political Economy 69,
no. 3 (1961): 213–225; Organisation for Economic Co-operation and Devel-
opment, Price Transparency, DAFFE/CLP(2001)22 (Paris: Organisation for
Economic Co-operation and Development, 2001); R. Nitsche and N. von
Hinten-Reed, “Competitive Impacts of Information Exchange,” Charles
River Associates (June 2004), http://ec.europa.eu/competition/consultations
/2004 _ 6 _reg _4056 _ 86/note _on _information _exchange _en.pdf.
4. G. J. Stigler, “Perfect Competition, Historically Contemplated,” Journal of
Copyright © 2016. Harvard University Press. All rights reserved.

Political Economy 65, no. 1 (February 1957): 1–17.

5. Organisation for Economic Co-operation and Development, Unilateral
Disclosure of Information with Anticompetitive Effects, DAF/COMP(2012)17
(Paris: Organisation for Economic Co-operation and Development, Oc-
tober 11, 2012), 11; see also Organisation for Economic Co-operation and
Development, Roundtable on Information Exchanges between Competitors
under Competition Law, Note by the Delegation of the United States, DAF/
COMP/WD(2010)117 (Paris: Organisation for Economic Co-operation and
Development, October 11, 2010), 11, http://www.justice.gov/sites/default/fi les
/atr/legacy/2014/09/17/269282.pdf.

251

6. Organisation for Economic Co-operation and Development, Roundtable on

Information Exchanges.
7. Ibid.
8. We add the caveat “in general” because competition authorities recognize
that, at times, increases in transparency can foster collusion. See European
Commission, Guidelines on the Assessment of Horizontal Mergers under the
Council Regulation on the Control of Concentrations between Undertakings,
2004/C 03 (February 5, 2004), para. 51 (noting how “practices, such as
meeting-competition or most-favoured-customer clauses, voluntary publica-
tion of information, announcements, or exchange of information through
trade associations, may increase transparency or help competitors interpret
the choices made”).
9. Organisation for Economic Co-operation and Development, Roundtable on
Information Exchanges.
10. See Consumer Futures, Price Comparison Websites: Consumer Perceptions
and Experiences (July 2013), 32 (“Factors Considered”), for some empirical
research as to which features of PCWs consumers value. http://webarchive
.nationalarchives.gov.uk /20140522123945/; http://www.consumerfutures.org
.uk /fi les/2013/07/Price-Comparison-Websites-Consumer-perceptions-and
-experiences.pdf.
11. New York City Labor Market Information Ser vice, Employment in New York
City Grocery Stores (New York: New York City Labor Market Information
Ser vice, May 2009), http://www.nyc.gov/html/sbs/wib/downloads/pdf
/grocery_ stores.pdf (5,111 stores in 2008).
12. Peter A. Diamond, “A Model of Price Adjustment,” Journal of Economic
Theory 3, no. 2 (June 1971): 156–168; Peter A. Diamond, Search Theory,
Working Paper No. 389 (Cambridge, MA: Massachusetts Institute of Tech-
nology, August 1985); J. Stiglitz, “Imperfect Information in the Product
Market,” in Handbook of Industrial Organization, vol. 1, Richard Schmalensee
and R. Willig, eds. (Amsterdam: North-Holland, 1989), 769–847.
13. Ibid.
Copyright © 2016. Harvard University Press. All rights reserved.

14. J. Yannis Bakos, “Reducing Buyer Search Costs: Implications for Electronic
Marketplaces,” Management Science 43, no. 12 (December 1997); John G.
Lynch Jr. and Dan Ariely, “Wine Online: Search Costs Affect Competition
on Price, Quality, and Distribution,” Marketing Science 19, no. 1 (2000):
83–103.
15. Gerald Häubl and Valerie Trifts, “Consumer Decision Making in Online
Shopping Environments: The Effects of Interactive Decision Aids,” Marketing
Science 19, no. 1 (2000): 4–21.
16. Competition Commission, Extended Warranties on Domestic Electrical
Goods: A Report on the Supply of Extended Warranties on Domestic Electrical

Goods within the UK, vols. 1–3, Cm 6089 (December 2003). The OFT
accepted undertakings in lieu of the reference to the Competition Com-
mission. Accordingly, a dedicated information site was launched: www
.compareextendedwarranties.co.uk.
17. U.S. Department of Justice and Federal Trade Commission. Horizontal
Merger Guidelines (August 19, 2010), section 9, https://www.ftc.gov/sites
/default/fi les/attachments/merger-review/100819hmg.pdf.
18. Maurice E. Stucke, “Behavioral Economists at the Gate: Antitrust in the
Twenty-First Century,” Loyola University of Chicago Law Journal 38 (2007):
513, 563–572.
19. European Commission, Guidelines on the Assessment of Horizontal Mergers,
para. 68; see also U.S. Auto Parts Network, Inc v. Parts Geek, LLC, 494 F
App’x 743, 745 (U.S. Court of Appeals [9th Circuit], 2012) (affirming
summary judgment on antitrust counterclaims when Parts Geek failed inter
alia to offer evidence such that a jury could reasonably find significant entry
barriers in that market).
20. Airbnb, Inc., The $1,000,000 Host Guarantee, https://www.airbnb.com
/guarantee.
21. Rob Price, “The Incredibly Simple Way to Find Your ‘Secret’ Uber Passenger
Rating,” Business Insider UK, February 11, 2015, http://uk.businessinsider
.com/uber-passenger-rating-how-to-customer-stars-how-do-i-2015-2.
22. For instance, a PCW dedicated to hotel accommodations would represent
numerous hotels, and is more likely to achieve a higher conversion rate than
a single supplier—that is, converting more clicks on advertisements into
business. So a small Parisian hotel, with a limited marketing budget, will pay
the PCW only if a booking took place—when a referral resulted in a sale. It
spends money on advertising only when the advertising proved successful.
This lowers the hotel’s risks and costs. As a result, more hotels are attracted
to the PCW, and customers benefit from the healthier mix of sellers, greater
competition, greater choice, and lower prices. See A. Ezrachi, “The Competi-
tive Effects of Parity Clauses on Online Commerce,” SSRN Working Paper
Copyright © 2016. Harvard University Press. All rights reserved.

(October 11, 2015), http://ssrn.com/abstract =2672541.

23. Consumer Futures, Price Comparison Websites, 35. Note, for example, the
change that web-aggregators have made to the U.K. energy sector. In
the U.K., since its deregulation in 1998, the energy supply market has been
characterized by the leading presence of the “Big Six” major suppliers. The
operation of web-aggregators in recent years has increased competitive
pressure and facilitated new entry, as with Ovo Energy and Co-operative
Energy. The U.K.’s competition authority has recognized the contribution of
web-aggregators to subtle but significant changes in the structure of the U.K.
domestic energy market. See Competition and Markets Authority, State of

the Market Assessment (March 27, 2014), 26, 78, 79–80, fig. 38, https://
www.ofgem.gov.uk /sites/default/fi les/docs/2014/03/assessment _document
_published _1.pdf.
24. Ibid.
25. Bakos, “Reducing Buyer Search Costs.”
26. The OFT ceased to exist on April 1, 2014, and was succeeded by the Compe-
tition and Markets Authority.
27. OFT Statement of Objections, Annex 1, para. 1.15, as cited in the Competition
Appeal Tribunal judgment in Skyscanner Limited v. Competition and Markets
Authority, Case No. 1226/2/12/14, September 26, 2014, [2014] CAT 16, 31–32.
28. Info Entrepreneurs, Stock Control and Inventory, http://www.infoentre
preneurs.org/en/guides/stock-control-and-inventory/#5.
29. Ismat Sarah Mangla, “3 Tricks to Help You Snag the Best Deals Online,”
Time, September 8, 2014, http://time.com/money/3136612/dynamic-pricing
-amazon-best-buy-walmart/.
30. Ariel Ezrachi, “The Competitive Effects of Parity Clauses on Online Com-
merce,” European Competition Journal 11, 488 (2015).
31. Amazon submission to the House of Lords EU Internal Market Sub-
Committee inquiry into online platforms in the EU Digital Single Market.
http://data.parliament.uk /writtenevidence/committeeevidence.svc/evidence
document/eu-internal-market-subcommittee/online-platforms-and-the-eu
-digital-single-market/written/24005.html; House of Lords, Select Com-
mittee on European Union ‘Online Platforms and the Digital Single Market’
(20 April 2016) 10th Report of Session 2015–16, http://www.publications
.parliament.uk /pa/ld201516/ldselect/ldeucom/129/129.pdf.
32. See, for example, the effects of insurance aggregators: George Maher, Andy
Staudt, and Ryan Warren, “Why Aren’t We Making Money . . . ,” Willis
Towers Watson (February 2011), http://www.towerswatson.com/en-GB
/Insights/IC-Types/Survey-Research-Results/2011/02/Why-arent-we-making
-money.
33. Jeff rey R. Brown and Austan Goolsbee, “Does the Internet Make Markets
Copyright © 2016. Harvard University Press. All rights reserved.

More Competitive? Evidence from the Life Insurance Industry,” Journal of

Political Economy 110, no. 3 (2002): 2 (draft of October 2010), http://faculty
.chicagobooth.edu/austan.goolsbee/research/insure.pdf.
34. C. R. Leslie, “Trust, Distrust, and Antitrust,” Texas Law Review 82, no. 3
(2004): 628.
35. U.S. Department of Justice, Price Fixing, Bid Rigging, and Market Allocation
Schemes: What They Are and What to Look For (June 15, 2015), http://www
.justice.gov/atr/price-fi xing-bid-rigging-and-market-allocation-schemes.
36. Ibid.

37. A. Mitra and J. G. Lynch Jr., “Toward a Reconciliation of Market Power and
Information Theories of Advertising Effects on Price Elasticity,” Journal of
Consumer Research 21 (March 1995): 644–659; A. Kaul and D. Wittink,
“Empirical Generalisation about the Impact of Advertising on Price Sensi-
tivity and Price,” Marketing Science 14, no. 3 (1995): 151–161.
38. P. Nelson, “Advertising as Information,” Journal of Political Economy 78
(March/April 1974): 311–329.

2 • New Economic Reality: The Rise of Big Data and Big Analytics

1. Charles Fishman, The Wal-Mart Effect: How the World’s Most Power ful
Company Really Works—and How It’s Transforming the American Economy
(New York: Penguin, 2006).
2. Citri-Lite Co. v. Cott Beverages, Inc., No. 1:07-CV-01075 OWW, 2011 WL
4751110, 5 (E.D. Cal. Sept. 30, 2011) aff ’d, 546 F. App’x 651 (9th Cir. 2013).
3. Steven Barrison, “Study Proves It: Walmart Super-stores Kill Off Local Small
Businesses,” Daily News, May 4, 2011.
4. Wal-Mart Stores, Inc., 2006 WL 695801, 17 (S.E.C. No-Action Letter,
March 16, 2006) (quoting Anthony Bianco and Wendy Zellner, “Is Wal-Mart
Too Powerful?” BusinessWeek, October 6, 2003).
5. Ibid.
6. Justin Lahart, “How Wal-Mart’s Store Closings Paint Wider Retail Picture:
Shift to Online Sales Shows Difference between Retailing’s Haves and
Have-Nots,” Wall Street Journal, January 15, 2016.
7. Wal-Mart Stores, Inc., Form 10-K (2015), 23, https://www.sec.gov/Archives
/edgar/data/104169/000010416915000011/wmtform10-kx13115.htm.
8. Ibid., 20. “A critical piece of identifying consumer preferences involves price
transparency, assortment of products, customer experience and convenience.
These factors are of primary importance to customers and they continue to
increase in importance, particularly as a result of digital tools and social
media available to consumers and the choices available to consumers for
Copyright © 2016. Harvard University Press. All rights reserved.

purchasing products online, at physical locations or through a combination

of both retail offerings. Failure to timely identify or effectively respond to
changing consumer tastes, preferences (including the key factors described
above) and spending patterns, whether for our physical retail offerings,
digital retail offerings or a combination of these retail offerings, could
negatively affect our relationship with our customers, the demand for our
products and ser vices and our market share.”
9. Wal-Mart Stores, Inc., Annual Report (2014), 7, http://stock.walmart.com
/fi les/doc _ financials/2014/Annual/2014-annual-report.pdf.

10. Kim S. Nash, “Wal-Mart to Pour $2 Billion into E-Commerce over Next Two
Years,” Wall Street Journal, October 14, 2015, http://blogs.wsj.com/cio/2015
/10/14/wal-mart-to-pour-2-billion-into-e-commerce-over-next-two-years/.
11. James B. Stewart, “Walmart Plays Catch-Up with Amazon,” New York Times,
October 23, 2015.
12. Ibid.
13. Amazon.com, Inc., Form 10-K, For the Fiscal Year Ended December 31, 2015
(2016), 17; Amazon.com, Inc., Form 10-K, For the Fiscal Year Ended
December 31, 2014 (2015), 17.
14. Letter from Jeff rey P. Bezos, Founder and Chief Executive Officer, Amazon
.com, to its Shareholders, in Amazon.com 2015 Annual Report.
15. Amazon.com 2014 10-K, supra, at 3.
16. Charlie Osborne, “Amazon Sellers Sold Record Number of Products in 2014,”
ZDNet (January 5, 2015), http://www.zdnet.com/article/amazon-sellers-sold
-record-number-of-products-in-2014/#!.
17. United States v. Apple, Inc., No. 13-3741-cv (2d Cir. 2015), 15.
18. Khadeeja Safdar, “Gap’s Chief Open to Selling on Amazon,” Wall Street
Journal, May 19, 2016, at B6.
19. Michael Eisen, “Amazon’s $23,698,655.93 Book about Flies,” It Is NOT Junk
(April 22, 2011), http://www.michaeleisen.org/blog/?p =358.
20. Andrew Couts, “Why Did Amazon Charge $23,698,655.93 for a Textbook?”
Digital Trends (April 23, 2011), http://www.digitaltrends.com/computing
/why-did-amazon-charge-23698655-93-for-a-textbook /.
21. Greg Besinger, “Boomerang Commerce, a Real-Time Pricing Startup, Raises
$8.5 Million,” Wall Street Journal, July 16, 2014, http://blogs.wsj.comdigits
/2014/07/16/boomerang-commerce-a-real-time-pricing-startup-raises-8-5
-million/.
22. CamelCamelCamel.com, Amazon Price History for Conair Cuisinart ICE-21
1.5 Quart Frozen Yogurt-Ice Cream Maker (White) (2015), http://camel
camelcamel.com/Cuisinart-ICE-21-Frozen-Yogurt-Ice-Sorbet/product
/B003KYSLMW?context=browse.
Copyright © 2016. Harvard University Press. All rights reserved.

23. CamelCamelCamel.com, Amazon Price History for Fossil Women’s ES3733

Stella Crystal-Accented Stainless Steel Watch with Link Bracelet (2015),
http://camelcamelcamel.com/Fossil-ES3733-Crystal-Accented-Stainless
-Bracelet/product/B00NVAYBUQ?active=price_amazon&context=top_drops.
24. Jubin Mehta, “Boomerang Commerce, a Dynamic Price Optimization
Company Raises $8.5 Million,” Your Story (July 16, 2014), http://yourstory
.com/2014/07/boomerang-commerce-raises-8-5-million/.
25. Lizzie O’Leary, “Marketplace Weekend for Friday, December 19, 2014,”
Marketplace Weekend (December 19, 2014), http://www.marketplace.org
/shows/marketplace-weekend/marketplace-weekend-friday-december-19-2014.

26. Ryan Mac, “Ex-Amazon Manager Gets Funding to Help Retailers Battle
His Former Employer,” Forbes, July 16, 2014, http://www.forbes.com/sites
/ryanmac/2014/07/16/boomerang-amazon-battle/.
27. Besinger, “Boomerang Commerce.”
28. See Salil K. Mehra, “Antitrust and the Robo-Seller: Competition in the Time
of Algorithms,” Minnesota Law Review 100 (March 2015), http://ssrn.com
/abstract=2576341 (discussing growth of pricing algorithms). See, for
instance, providers of “dynamic pricing optimizers” such as Boomerang
Commerce, Prisync, Price Maker, RepricerExpress, and others.
29. George Packer, “Cheap Words: Amazon Is Good for Customers. But Is It
Good for Books?” The New Yorker, February 17 and 24, 2014, http://www
.newyorker.com/magazine/2014/02/17/cheap-words.
30. Julian D’Onfro, “13 Interesting Startups Founded by Former Amazon
Employees,” Business Insider UK, June 20, 2015, http://uk.businessinsider
.com/startups-founded-by-former-amazon-employees-2015- 6?r =US.
31. Joe Lindsey, “This Soft ware Company Wants to Help Retailers Compete
against Amazon,” Entrepreneur, October 29, 2014, http://www.entrepreneur
.com/article/238454.
32. Amazon.co.uk. Prime Same-Day Delivery (2015), http://www.amazon.co.uk
/b/?ie =UTF8& node =5782509031& tag =googhydr-21&hvadid= 66544544104
&hvpos=1t1&hvexid=&hvnetw=g&hvrand=12880355850154458096&hvpone
=&hvptwo =&hvqmt =b&hvdev= c& ref = pd _ sl _ 241ngzeoi1_b.
33. Amazon.co.uk, Amazon PrimeNow (2015), http://www.amazon.co.uk /b/ref
= pn _uk _ surl _ lp?node = 6584642031.
34. Neal Ungerleider, “It Has 40 Million Subscribers. Now Amazon Prime
Is Eyeing the Competition,” Fast Company (July 9, 2015), http://www
.fastcompany.com/3048366/it-has-40-million-subscribers-now-amazon
-prime-is-eyeing-the-competition. The influence of Amazon Prime on
brick-and-mortar retailers is arguably demonstrated by Walmart’s piloting
of its “Shipping Pass” ser vice: Walmart, Shipping Pass Pilot (2015),
http://www.walmart.com/cp/ShippingPass/1229595.
Copyright © 2016. Harvard University Press. All rights reserved.

35. For an overview of the four Vs of Big Data and their competitive implica-
tions, see Maurice E. Stucke and Allen P. Grunes, Big Data and Competition
Policy (Oxford: Oxford University Press, 2016). The President’s Council of
Advisors on Science and Technology, in its report to the president, noted the
different definitions of Big Data, including those by business consultants
(“high-volume, high-velocity and high-variety information assets that
demand cost-effective, innovative forms of information processing for
enhanced insight and decision making”); computer scientists (“a term
describing the storage and analysis of large and/or complex data sets using
a series of techniques including, but not limited to, NoSQL, MapReduce,

and machine learning”); and privacy experts (“data about one or a group of
individuals, or that might be analyzed to make inferences about individ-
uals”). President’s Council of Advisors on Science and Technology, Report
to the President, Big Data and Privacy: A Technological Perspective (Wash-
ington, DC: Executive Office of the President, May 2014), 2. https://www
.whitehouse.gov/sites/default/fi les/microsites/ostp/PCAST/pcast _big _data
_ and _privacy_-_may_ 2014.pdf.
36. Organisation for Economic Co-operation and Development, Exploring the
Economics of Personal Data: A Survey of Methodologies for Measuring
Monetary Value, OECD Digital Economy Papers, No. 220 (Paris: Organisa-
tion for Economic Co-operation and Development, April 2, 2013), 7,
http://dx.doi.org/10.1787/5k486qtxldmq-en.
37. Organisation for Economic Co-operation and Development, Data-Driven
Innovation for Growth and Well-Being: Interim Synthesis Report (Paris:
Organisation for Economic Co-operation and Development, October 2014),
http://www.oecd.org/sti/inno/data-driven-innovation-interim-synthesis.pdf;
Executive Office of the President, Big Data: Seizing Opportunities, Preserving
Values (Washington, DC: Executive Office of the President, May 2014), 2,
https://www.whitehouse.gov/sites/default/fi les/docs/big _data _privacy_report
_may_1_ 2014.pdf (noting that most definitions “reflect the growing techno-
logical ability to capture, aggregate, and process an ever-greater volume,
velocity, and variety of data”).
38. For example, on a year-over-year basis, funding for AI start-ups jumped
more than 300 percent. The most sizable deals included Sentient Technolo-
gies’ $103.5 million Series C financing from investors including Tata,
Horizons Ventures, and Access Industries, and Vicarious Systems’ $40
million Series B financing led by Formation 8. ABB Technology Ventures
later extended the round by another $12M. “Artificial Intelligence Startups
See 302% Funding Jump in 2014,” CBInsights (February 10, 2015), https://www
.cbinsights.com/blog/artificial-intelligence-venture-capital-2014/). See also
Robert McMillan, “IBM Turns Up Heat Under Competition in Artificial
Copyright © 2016. Harvard University Press. All rights reserved.

Intelligence,” Wall Street Journal, November 24, 2015, http://www.wsj.com

/articles/ibm-turns-up-heat-under-competition-in-artificial-intelligence
-1448362800.
39. Jo Best, Jo. “IBM Watson: The Inside Story of How the Jeopardy-Winning
Supercomputer Was Born, and What It Wants to Do Next,” TechRepublic
(2013), http://www.techrepublic.com/article/ibm-watson-the-inside-story-of
-how-the-jeopardy-winning-supercomputer-was-born-and-what-it-wants-to
-do-next/.
40. Steve Lohr, “IBM’s AI System Watson to Get Second Home, on West Coast,”
New York Times, September 24, 2015, http://www.nytimes.com/2015/09/25

/technology/ibms-ai-system-watson-to-get-new-west-coast-home.html
?smprod=nytcore-iphone&smid=nytcore-iphone-share& _r = 0.
41. Antonio Regalado, “Is Google Cornering the Market on Deep Learning?”
MIT Technology Review, January 29, 2014, http://www.technologyreview.com
/news/524026/is-google-cornering-the-market-on-deep-learning/; Nicola
Jones, “Computer Science: The Learning Machines,” Nature, January 8, 2014,
http://www.nature.com/news/computer-science-the-learning-machines
-1.14481.
42. European Data Protection Supervisor, Towards a New Digital Ethics: Data,
Dignity and Technology, Opinion 4/2015 (September 11, 2015), 9.
43. Robert D. Hof, “Deep Learning,” MIT Technology Review, April 23, 2013,
http://www.technologyreview.com/featuredstory/513696/deep-learning/.
44. Tereza Pultarova, “Jaguar Land Rover to Lead Driverless Car Research,”
E&T (October 9, 2015), http://eandt.theiet.org /news/2015/oct /jaguar-land
-rover-driverless-cars.cfm; David Talbot, “CES 2015: Nvidia Demos a Car
Computer Trained with ‘Deep Learning,’ ” MIT Technology Review,
January 6, 2015), http://www.technologyreview.com /news/533936/ces-2015
-nvidia-demos-a-car-computer-trained-with-deep-learning /; David
Levitin, 2015. “The Sum of Human Knowledge,” Wall Street Journal,
September 18, 2015, http://www.wsj.com /articles/the-sum-of-human
-knowledge-1442610803.
45. Lohr, “IBM’s AI System Watson to Get Second Home.”
46. European Data Protection Supervisor, Towards a New Digital Ethics.
47. Take, for example, the Rubicon Project, “a leading technology company
automating the buying and selling of advertising.” As its website notes,
“Relentless in its efforts for innovation, Rubicon Project has engineered one
of the largest real-time cloud and Big Data computing systems, processing
trillions of transactions within milliseconds each month”; Rubicon Project.
(2016), http://rubiconproject.com/whoweare/. The company discussed the
interplay among Big Data, machine learning, and data-driven network
effects as a competitive strength: “As we process more volume on our
Copyright © 2016. Harvard University Press. All rights reserved.

automated platform, we accumulate more data, such as pricing, geographic

and preference information, data on how best to optimize yield for sellers
and more. This additional data helps make our machine-learning algorithms
more intelligent and this leads to more effective matching between buyers
and sellers. As a result, more buyers and sellers are attracted to our platform,
from which we get more data, which further reinforces the network effect
and thereby increases market liquidity, which benefits both buyers and
sellers”; Rubicon Project, Amendment No. 3 to Form S-1 Registration Statement
(April 30, 2014). See also Organisation for Economic Co-operation and
Development, Data-Driven Innovation for Growth and Well-Being,

4 (discussing how algorithms can “learn from data of previous situations

and . . . autonomously make decisions based on the analysis of these data”).
48. Viktor Mayer-Schönberger and Kenneth Cukier, Big Data: A Revolution that
Will Transform How We Live, Work, and Think (London: John Murray, 2013),
36–39 (discussing how more data trumps better algorithms in the area of
better language processing); Alon Halevy Peter Norvig, and Fernando
Pereira, “The Unreasonable Effectiveness of Data,” IEEE Intelligent Systems,
March/April 2009, 9, http://static.googleusercontent.com/media/research
.google.com/en//pubs/archive/35179.pdf (observing that “invariably, simple
models and a lot of data trump more elaborate models based on less data. . . .
[O]bservations have been made in every other application of machine learning
to Web data: simple n-gram models or linear classifiers based on millions of
specific features perform better than elaborate models that try to discover
general rules. In many cases there appears to be a threshold of sufficient data”).
49. Suzanne Vranica and Robert McMillan, “IBM Nearing Acquisition of
Weather Co.’s Digital and Data Assets,” Wall Street Journal, October 27,
2015, http://www.wsj.com/articles/ibm-nearing-acquisition-of-weather-co-s
-digital-and-data-assets-1445984616.
50. Ibid.
51. IBM, Mobile App from Octo Telematics Uses IBM and the Weather Com-
pany Data to Help Drivers Score Savings with Pay-As-You-Drive Insurance
(October 26, 2015), http://www- 03.ibm.com /press/us/en /pressrelease
/47949.wss.
52. Ibid.
53. Ibid.
54. Ibid.
55. Josh Constine, “Facebook launches Messenger platform with chatbots,”
April 12, 2016, http://techcrunch.com/2016/04/12/agents-on-messenger/;
Ben Popper, “Mark Zuckerberg Thinks AI Will Start Outperforming Humans
in the Next Decade,” April 28, 2016, http://www.theverge.com/2016/4/28
/11526436/mark-zuckerberg-facebook-earnings-artificial-intelligence
Copyright © 2016. Harvard University Press. All rights reserved.

-future.
56. Madhumita Murgia, “Facebook Messenger’s New Bots Are a Powerful Way
to Target Adverts,” The Telegraph, April 13, 2016, http://www.telegraph.co.uk
/technology/2016/04/12/facebook-messenger-launches-chat-bot-economy-to
-take-on-apps/.
57. Jack Nicas, “Google Touts New AI-Powered Tools,” Wall Street Journal,
May 19, 2016, at B1; Jay Greene and Matthias Verbergt, “Microsoft Cuts
Low-End Phones,” Wall Street Journal, May 19, 2016, at B1.
58. Michael Bowling et al., “Heads-Up Limit Hold’em Poker Is Solved,” Science,
January 9, 2015, 145.

59. Tuomas Sandholm, “Solving Imperfect-Information Games,” Science,

January 9, 2015, 122.
60. Bowling et al., “Heads-Up Limit Hold’em Poker Is Solved,” 148.
61. Amazon Machine Learning (last visited May 3, 2016), https://aws.amazon
.com/machine-learning/.
62. Danny Palmer, “Amazon follows Microsoft and Google with AI tools in
Amazon Machine Learning Ser vice,” Computing, April 10, 2015, http://www
.computing.co.uk /ctg/news/2403533/amazon-follows-microsoft-and-google
-into-offering-customers-ai-tools-with-amazon-machine-learning-service.
63. Ingrid Lunden, “Amazon Launches AWS IoT—A Platform for Building,
Managing and Analyzing the Internet Of Things,” Tech Crunch, October 8,
2015, http://techcrunch.com/2015/10/08/amazon-announces-aws-iot-a
-platform-for-building-managing-and-analyzing-the-internet-of-things/#
.gfgxjj:0nTE.
64. Ibid. Amazon Web Ser vices is a collection of cloud computing ser vices
offered by Amazon. “Amazon Web Ser vices offers a broad set of global
compute, storage, database, analytics, application, and deployment ser vices
that help organizations move faster, lower IT costs, and scale applications”;
Amazon Web Ser vices, Cloud Products (2015), https://aws.amazon.com
/products/?nc2=h _ql _ny_ livestream _blu.
65. Reuters, “Amazon Launches Platform to Build Apps for IoT,” FirstPost,
October 9, 2015, http://www.firstpost.com/business/amazon-launches
-platform-to-build-apps-for-iot-2461852.html.
66. Ibid.
67. With our increasing interaction with digital devices exacerbating the
opportunities for data about how we lead our daily lives to be collected, even
at the dinner table, we may feel the need to reach for a pepper grinder that
can block Wi-Fi and mobile signals; Susmita Baral, “New Pepper Grinder
Blocks All Electronics during Meals while Grinding Spice for Your Food,”
iDigitalTimes, April 1, 2015, http://www.idigitaltimes.com/new-pepper
-grinder-blocks-all-electronics-during-meals-while-grinding-spice-your
Copyright © 2016. Harvard University Press. All rights reserved.

-food-428558.
68. City of New York. Mayor de Blasio Announces Minerva Tantoco as City’s First
Ever Chief Technology Officer (New York: City of New York, September 4,
2014), http://www1.nyc.gov/office-of-the-mayor/news/437-14/mayor-de
-blasio-minerva-tantoco-city-s-first-ever-chief-technology-officer.
69. “Pricing Algorithms: Is the Price You Pay Right?,” Bloomberg (May 12, 2015),
http://www.bloomberg.com/news/videos/b/02d3f0f0-e653-4ca1-8bdd- 0f95a
5a81212.
70. Executive Office of the President, Big Data and Differential Pricing
(Washington, DC: Executive Office of the President, February 2015), 13,

https://www.whitehouse.gov/sites/default/fi les/whitehouse _ fi les/docs/Big

_ Data _ Report _ Nonembargo_v2.pdf.
71. According to Visa, e-commerce spending on Thanksgiving 2015 by its U.S.
account holders increased to $1.5 billion, up 22 percent from 2014. Suzanne
Kapner, “Black Friday Shopping—with Thinner Crowds,” Wall Street Journal,
November 28, 2015, http://www.wsj.com/articles/black-friday-shoppingwith
-thinner-crowds-1448639398.
72. Ibid.
73. Ibid. As the newspaper also noted, 36 percent of consumers planned to shop
online during the 2015 holiday season, up from 19 percent who said so in
2014; only 18 percent planned to shop in physical stores, down from 35 percent
a year ago. Ibid.
74. Andrew McAfee and Erik Brynjolfsson, “Big Data: The Management
Revolution,” Harvard Business Review, October 2012, https://hbr.org/2012/10
/big-data-the-management-revolution/ar/.
75. Behavioral advertising is the “collection of data from a par ticu lar computer
or device regarding Web viewing behaviors over time and across non-affi liated
Web sites for the purpose of using such data to predict user preferences or
interests to deliver advertising to that computer or device based on the
preferences or interests inferred from such Web viewing behav iors”; Digital
Advertising Alliance, Self-Regulatory Principles for Online Behavioral
Advertising Implementation Guide: Frequently Asked Questions (October
2010), http://www.aboutads.info/resource/download/OBA%20Self-Reg%20
Implementation%20Guide%20-%20Frequently%20Asked%20Questions.pdf.
76. European Data Protection Supervisor, Towards a New Digital Ethics, 35.
(“The situation is likely to be compounded by the growth of the Internet of
Things, which will include many technical or embedded devices collecting
personal data, with the result that their users will be unable to consult the
privacy policy on the device itself, but would have to find paper documenta-
tion or more likely browse from another device to the relevant web sites”);
President’s Council of Advisors on Science and Technology, Report to the
Copyright © 2016. Harvard University Press. All rights reserved.

President, Big Data and Privacy, 15: “To some, it seems farfetched that the
typical home will foreseeably acquire cameras and microphones in every
room, but that appears to be a likely trend. What can your cell phone
(already equipped with front and back cameras) hear or see when it is on the
nightstand next to your bed? Tablets, laptops, and many desktop computers
have cameras and microphones.”
77. President’s Council of Advisors on Science and Technology, Report to the
President, Big Data and Privacy.
78. McAfee and Brynjolfsson, “Big Data.”

79. McKinsey Global Institute, Big Data: The Next Frontier for Innovation,
Competition, and Productivity (McKinsey & Company, June 2011), 98,
http://www.mckinsey.com/insights/business _technology/big _data _the _next
_frontier_for_innovation; “More and more sensors are being embedded in
physical devices—from assembly-line equipment to automobiles to mobile
phones—that measure processes, the use of end products, and human
behav ior. Individual consumers, too, are creating and sharing a tremendous
amount of data through blogging, status updates, and posting photos and
videos. Much of these data can now be collected in real or near real time.”
80. Executive Office of the President, Big Data and Differential Pricing, 5.
81. Samuel B. Hwang and Sungho Kim, “Dynamic Pricing Algorithm for
E-Commerce,” in Advances in Systems, Computing Sciences and Software
Engineering, Proceedings of SCSS05, Tarek Sobh and Khaled Elleithy, eds.
(Dordrecht: Springer, 2006), 149–155; N. Abe and T. Kamba. “A Web
Marketing System with Automatic Pricing,” Computer Networks 33 (2000):
775–78; L. M. Minga, Y. Q. Fend, and Y. J. Li, “Dynamic Pricing: E-
Commerce—Oriented Price Setting Algorithm,” Proceedings of the 2nd
International Conference on Machine Learning and Cybernetics, Xi’an,
China, 2003, vol. 2.
82. See Mehra, “Antitrust and the Robo-Seller.”

3 • Light Touch Antitrust

1. Alan Greenspan, “The Effects of Mergers (Testimony before the Committee

on the Judiciary, U.S. Senate),” Federal Reserve Bulletin, 84 (June 16, 1998):
643, 646, at 5.
2. Note, however, differences between the EU and US with respect to abuse of
dominance. The European Commission has taken a rather critical look at
some activities by dominant technology companies. Noteworthy are the EU
Commission focus on activities by Google, Microsoft, Apple, and Intel. Note,
for example, the Statement of Objections issued by the Commission with
Copyright © 2016. Harvard University Press. All rights reserved.

respect to Google’s comparison shopping ser vices (April 2015) and Android
operating system and applications (April 2016); by contrast, the U.S.
Department of Justice (DOJ) and Federal Trade Commission have largely
abstained from prosecuting monopolies in online markets (or any other
market) over the past sixteen years. More generally, on EU enforcement,
see A. Ezrachi, EU Competition Law: An Analytical Guide to the Leading
Cases (Oxford: Hart Publishing, 2016). For statistics on US enforcement, see
Maurice E. Stucke and Allen P. Grunes, Big Data and Competition Policy
(Oxford: Oxford University Press, 2016).

3. United States v. Syufy Enterprises, 903 F.2d 659, 662–63 (9th Cir. 1990).
4. Order of the President of the People’s Republic of China No. 68, Anti-Mono-
poly Law of the People’s Republic of China, Chapter 1, Article 1, http://www
.china.org.cn/china/2011-02/11/content _ 21900178.htm.
5. U.S. Department of Justice Antitrust Division, Antitrust Division Manual,
4th ed., chap. 5 (Washington, DC: U.S. Department of Justice, March 2008).
6. Adam Smith, The Wealth of Nations (1776).
7. Unilateral Conduct Working Group, Report on the Objectives of Unilateral
Conduct Laws, Assessment of Dominance/Substantial Market Power, and
State-Created Monopolies (Moscow: International Competition Network,
May 2007), http://www.internationalcompetitionnetwork.org/uploads
/library/doc353.pdf.
8. See, for example, Robert H. Bork, The Antitrust Paradox: A Policy at War
with Itself (New York: Basic Books, 1978); Richard A. Posner, “The Chicago
School of Antitrust Analysis,” University of Pennsylvania Law Review 127
(1978): 925, 933.
9. Posner, “The Chicago School of Antitrust Analysis.”
10. Justin Fox, The Myth of the Rational Market (New York: Harper Business/
HarperCollins, 2009), 89–107.
11. As President Reagan told the nation, “government is not the solution to our
problem; government is the problem”; Ronald Reagan, First Inaugural
Address (January 20, 1981), http://www.reaganlibrary.com/reagan/speeches
/first.asp.
12. Case No. T-79/12, Cisco Systems Inc. v. Commission [December 11, 2013] 612
TJ 0079, para. 69.
13. Ibid.
14. United States v. Microsoft Corp., 253 F.3d 34, 49 (D.C. Cir. 2001) (noting
“significant debate amongst academics and practitioners over the extent to
which ‘old economy’ § 2 monopolization doctrines should apply to firms
competing in dynamic technological markets characterized by network
effects”); United States v. Bazaarvoice, Inc., No. 13-CV-00133-WHO, 2014
Copyright © 2016. Harvard University Press. All rights reserved.

WL 203966, 76 (N.D. Cal. Jan. 8, 2014) (noting “the debate over the proper
role of antitrust law in rapidly changing high-tech markets” with some
maintaining “that antitrust law is ill-suited to these dynamic markets,
arguing, for example, that it undermines high-tech innovation); see, for
example, Robert J. Barro, “Why the Antitrust Cops Should Lay Off High–
Tech,” BusinessWeek, August 16, 1998, http://www.bloomberg.com/bw
/stories/1998-08-16/why-the-antitrust-cops-should-lay-off-high-tech (“the
best policy for the government in the computer industry is to stay out of it”
or that market power is transitory in high-tech industries where competitive
ideas can overcome entrenchment).

15. Thomas O. Barnett, “Maximizing Welfare through Technological Innova-

tion,” George Mason Law Review 15 (2008): 1191, 1200.
16. See, for example, United States v. Microsoft Corp., 147 F.3d 935, 948 (D.C. Cir.
1998) (“Antitrust scholars have long recognized the undesirability of having
courts oversee product design, and any dampening of technological innova-
tion would be at cross-purposes with antitrust law”); compare John M.
Newman, “Anticompetitive Product Design in the New Economy,” Florida
State University Law Review 39 (2012): 681, proposing a structured, efficient,
and rational method for analyzing design-related conduct in tech markets.
17. Maureen K. Ohlhausen, A Smarter Section 5 (Washington DC: U.S. Federal
Trade Commission, September 25, 2015), 12, https://www.ftc.gov/system/fi les
/documents/public _ statements/804511/150925smartersection5.pdf.
18. J. Thomas Rosch, I Say Monopoly, You Say Dominance: The Continuing
Divide on the Treatment of Dominant Firms, Is It the Economics? (Florence:
International Bar Association, Antitrust Section Conference, September 8,
2007), 5, https://www.ftc.gov/sites/default/fi les/documents/public _ statements
/i-say-monopoly-you-say-dominance-continuing-divide-treatment-dominant
-firms-it-economics/070908isaymonopolyiba _0.pdf.

4 • Looking beyond the Façade of Competition

1. Viktor Mayer-Schönberger and Kenneth Cukier, Big Data: A Revolution

That Will Transform How We Live, Work, and Think (London: John Murray,
2013), 35.
2. C. M. Bishop, Pattern Recognition and Machine Learning (New York:
Springer-Verlag, 2006).
3. “Pricing Algorithms: Is the Price You Pay Right?” Bloomberg (May 12, 2015),
http://www.bloomberg.com/news/videos/b/02d3f0f0-e653-4ca1-8bdd
-0f95a5a81212.
4. For instance, see Sarah Griffiths, “Facebook Ads to Become More Intrusive:
Site Will Soon Show Promotions for Products You’ve Looked at across the
Copyright © 2016. Harvard University Press. All rights reserved.

Web,” Daily Mail, June 13, 2014, http://www.dailymail.co.uk /sciencetech

/article-2657043/Facebook-ads-intrusive-Site-soon-promotions-products
-youve-searched-web.html.
5. Executive Office of the President, Big Data: Seizing Opportunities, Preserving
Values, (May 2014), 5, https://www.whitehouse.gov/sites/default/fi les/docs
/big _data _privacy_report _may_1_ 2014.pdf; see also Facebook, Announcing
New Product Ads on Facebook (February 17, 2015), https://www.facebook.com
/business/news/product-ads.
6. European Data Protection Supervisor, Towards a New Digital Ethics: Data,
Dignity and Technology, Opinion 4/2015 (September 11, 2015), 6.

7. For example, “Crystal creates a unique profi le for anyone with a LinkedIn
account, explaining how to speak, email, work with or sell to them most
effectively. With disconcerting specificity, it tells you the ‘words, phrases,
style and tone you should use to reach the recipient in the way that they like
to communicate, rather than your own’—even their tolerance for sarcasm
and emoticons.” Elle Hunt, “Crystal Knows Best . . . or Too Much? The
Disconcerting New Email Advice Ser vice,” The Guardian, May 19, 2015,
http://www.theguardian.com/media/2015/may/19/crystal-knows-best-or-too
-much-the-disconcerting-new-email-advice-service (referring to Crystal
Project Inc., https://www.crystalknows.com/).
8. Allen Grunes, “Tracking Not Allowed (Unless You’re Google),” Politico
(October 1, 2015), http://www.politico.com/agenda/story/2015/10/tracking
-not-allowed-unless-youre-google-000261.
9. See, e.g., French Autorité de la concurrence and the German Bundeskartellamt,
Competition Law and Data, May 10, 2016, http://www.bundeskartellamt
.de/SharedDocs/Publikation /DE/Berichte/Big%20Data%20Papier.pdf ?_
_ blob = publicationFile& v =2; U.K. House of Lords, Select Committee on
European Union “Online Platforms and the Digital Single Market,” April 20,
2016, 10th Report of Session 2015–16, http://www.publications.parliament
.uk /pa / ld201516/ ldselect / ldeucom /129/129.pdf; Keynote Remarks of FTC
Commissioner Terrell McSweeny, “Competition Law: Keeping Pace in a
Digital Age,” 16th Annual Loyola Antitrust Colloquium, Chicago, IL,
April 15, 2016.
As noted by the House of Lords in its Report on “Online Platforms and
the Digital Single Market,” “[R]apid developments in data collection and data
analytics have created the potential for new welfare reducing and anti-
competitive behaviours by online platforms, including subtle degradations of
quality, acquiring datasets to exclude potential competitors, and new forms
of collusion. While some of these abuses are hy pothetical, they raise
questions as to the adequacy of current approaches to competition enforce-
ment” (paragraph 178).
Copyright © 2016. Harvard University Press. All rights reserved.

10. Following restructuring, a new holding company named Alphabet has been
formed to include the wide range of Google operations. In our discussion we
make reference to Google when discussing all of Alphabet’s operations. On
the activities of the new company, see Google, “G Is for Google,” Google
Official Blog (August 10, 2015), https://googleblog.blogspot.co.uk /2015/08
/google-alphabet.html.

Part II • The Collusion Scenarios

1. Studies of detected cartels find that they are often bimodal: some cartels
last less than a year, but many others last between four and six years. M. C.

Levenstein and V. Y. Suslow, “What Determines Cartel Success?,” Journal of

Economic Literature 44 (2006): 43, 51–52 (noting that duration is bimodal,
with cartels lasting only one year, and twice as many lasting between four
and six years). Many conspiracies, including those with eleven or more
conspirators, can last for years, if not decades. J. M. Connor, “Cartels and
Antitrust Portrayed: Internal Structure—Private International Cartels
1990–2008,” American Antitrust Institute Working Paper No. 09-06 (2009),
4, 8, ssrn.com/abstract =1372849 (finding cartels’ median and mean duration
was fift y-seven and eighty-two months, respectively, and that global cartels
lasted 57 percent longer than the average cartel). The average duration of
international cartels successfully prosecuted between 1983 and 1994 was
approximately ninety months; the average duration declined below eighty
months for the period 1995 to 1999, and trended upward to nearly ninety
months for the period 2005 to 2008. Ibid., 11. Not surprisingly, one DOJ
official took aim at the Wall Street Journal, which surmised that, “If colluders
push prices too high, defectors and new entrants will set things right.” The
DOJ official responded, “Our experience has shown that this is not the case.
Several of the cartels we prosecuted had been in existence for over ten years,
including one (sorbates) that lasted 17 years, from 1979 to 1996.” W. J.
Kolasky (Deputy Assistant Attorney General, Antitrust Division, U.S. DOJ),
Antitrust Compliance Programs: The Government Perspective, Speech at
Corporate Compliance 2002 Conference of the Practising Law Institute
(San Francisco: U.S. Department of Justice, July 12, 2002).
2. United States v. National Turtle Farmers & Shippers Assoc., Inc., Trade Reg.
Rep. Summaries (CCH) ¶ 45,095, at 44,744 (D. La. 1995).
3. United States v. Home City Ice Co., No. 07-CR-140 (S.D. Ohio Nov. 5, 2007),
http://www.justice.gov/atr/cases/f234200/234205.htm.
4. United States v. William Barrett Numismatics Ltd., Trade Reg. Rep. Summa-
ries (CCH) ¶ 45,095, at 44,741 (S.D.N.Y. 1995).
5. Scott D. Hammond (Deputy Assistant Attorney General for criminal enforce-
ment, Antitrust Division, U.S. DOJ), Recent Developments, Trends, and
Copyright © 2016. Harvard University Press. All rights reserved.

Milestones in the Antitrust Division’s Criminal Enforcement Program

(Washington, DC: U.S. Department of Justice, March 26, 2008), http://www
.justice.gov/atr/speech/recent-developments-trends-and-milestones-antitrust
-divisions-criminal-enforcement. Believing that existing criminal penalties
were suboptimal in deterring antitrust offences, the United States over the
past four decades has increased the Sherman Act’s criminal penalties—
conspirators may be jailed to a maximum of ten years, and both the indi-
viduals and the corporations may be subjected to heft y fines.
6. Price fi xers, observed one DOJ official, “tend to be recidivists.” Kolasky,
Antitrust Compliance Programs; J. M. Connor and C. G. Helmers, “Statistics
on Modern International Cartels 1990–2005,” AAI Working Paper No. 07-01

(January 10, 2007), 23 (170 companies in 283 international cartels studied

were price-fi xing recidivists, of which eleven companies were caught ten or
more times fi xing prices). Major cartels (some detected under the leniency
program) continued to operate after the DOJ’s publicized price-fi xing case
against Archer Daniels Midland Company (ADM). As BusinessWeek
observed upon the release of the movie The Informant!, “for all the splashy
headlines, stiff sanctions, and caught-on-tape teaching moments generated
by the ADM case, price-fi xing appears to be as pervasive as ever.” M. Orey,
“Price Fixing, the Perpetual Sequel,” BusinessWeek, September 28, 2009,
http://www.businessweek.com/bwdaily/dnflash/content/sep2009
/db20090928 _ 842438.htm.

5 • The Messenger Scenario

Epigraph: Bill Baer (Assistant Attorney General, Antitrust Division, U.S. DOJ),
Former E-Commerce Executive Charged with Price Fixing in the Antitrust
Division’s First Online Marketplace Prosecution (Washington, DC: U.S. Depart-
ment of Justice, April 6, 2005), 15–421, http://www.justice.gov/atr/public/press
_releases/2015/313011.docx.
1. Scott D. Hammond (Director of Criminal Enforcement, Antitrust
Division, U.S. DOJ), The Fly on the Wall Has Been Bugged— Catching an
International Cartel in the Act (Washington, DC: U.S. Department of
Justice, May 15, 2001), http://www.justice.gov/atr/public/speeches/8280.htm
(ADM case).
2. Baer, Former E-Commerce Executive Charged with Price Fixing.
3. United States v. Topkins, CR 15–00201 WHO (N.D. Cal. Apr. 30, 2015),
Plea Agreement, para. 4, http://www.justice.gov/atr/cases/topkins.html.
4. U.S. Department of Justice, E-Commerce Exec and Online Retailer Charged
with Price Fixing Wall Posters (December 4, 2015), http://www.justice.gov
/opa/pr/e-commerce-exec-and-online-retailer-charged-price-fi xing-wall
-posters.
Copyright © 2016. Harvard University Press. All rights reserved.

5. The banks agreed to pay criminal fines totaling over $2.5 billion. Moreover,
two banks—UBS and Barclays—had to pay an additional $203 million and
$60 million, respectively, for breaching their 2012 non-prosecution agree-
ments resolving the DOJ’s investigation involving the LIBOR benchmark
interest rate. Ibid. See also Elai Katz, “U.S. Brings Computerized Price-
Fixing Charges,” New York Law Journal 254, no. 120 (December 23, 2015).
6. U.S. Department of Justice, Five Major Banks Agree to Parent-Level Guilty
Pleas: Citicorp, JPMorgan Chase & Co., Barclays PLC, the Royal Bank of
Scotland PLC Agree to Plead Guilty in Connection with the Foreign Exchange
Market and Agree to Pay More than $2.5 Billion in Criminal Fines (May 20,

2015), http://www.justice.gov/opa/pr/five-major-banks-agree-parent-level
-guilty-pleas.
7. The DOJ prescribed fines for Citicorp ($925 million), Barclays ($650 million),
JPMorgan ($550 million), and RBS ($395 million). Ibid.
8. Ibid.
9. Since 2009, financial institutions have paid over $204 billion in 175 settle-
ments where fines exceeded $100 million. Jeff Cox, “Misbehaving Banks
Have Now Paid $204B in Fines,” CNBC (October 30, 2015), http://www.cnbc
.com/2015/10/30/misbehaving-banks-have-now-paid-204b-in-fines.html
(Bank of America $77.09 billion, JPMorgan Chase $40.12 billion, Citigroup
$18.39 billion, Wells Fargo $10.24 billion, BNP Paribas $8.90 billion, UBS
$6.54 billion, Deutsche Bank $5.53 billion, Morgan Stanley $4.78 billion,
Barclays $4.23 billion, and Credit Suisse $3.74 billion).
10. RPM is where the manufacturer/distributor agrees with the retailer on the
price charged to consumers. That practice is regarded as anticompetitive by
object in the European Union. In the United State, RPM was also per se
illegal for over ninety-five years until the Supreme Court overruled its earlier
decision and held that vertical price restraints are to be judged by the more
lenient rule of reason standard. Leegin Creative Leather Products, Inc. v.
PSKS, Inc., 551 U.S. 877, 882 (2007).
11. Under the agreements that governed that network, Carrefour provided
recommended resale prices to its franchisees, which, through an annex
expressly stipulated, were normally adhered to as resale prices. The Greek
Hellenic Competition Commission ruled that these provisions constituted
“a general obligatory rule for the franchisees to follow Carrefour’s recom-
mended retail prices, to which only exceptions could exist.” Lia Vitzilaiou,
“The Hellenic Competition Commission Fines a Retailer for Resale Price
Maintenance and Other Infringements within Its Franchise Network
(Carrefour Marinopoulos),” e-Competitions (February 2011), http://www
.lambadarioslaw.gr/publications/2011/en/article _ 33885.pdf.
12. Hellenic Competition Commission. Decision Concerning Infringements of
Copyright © 2016. Harvard University Press. All rights reserved.

Articles 1 of Law 703 / 77 and Article 101 TFEU by the Retailer Carrefour
Marinopoulos S.A. in Connection with the Franchise Network for the Opera-
tion of “5 Marinopoulos” Retail Stores (Athens: Hellenic Competition
Commission, July 15, 2010), http://www.epant.gr/img/x2/news/news270_1
_1279200461.pdf.
13. Vitzilaiou, “The Hellenic Competition Commission Fines a Retailer.”
14. Hellenic Competition Commission, Decision Concerning Infringements.
15. United States v. Airline Tariff Publ’g Co., 836 F. Supp. 9, 12 (D.D.C. 1993).
16. This information exchange greatly facilitated tacit collusion, and as noted by
the DOJ, it was of little benefit to consumers. Some defendants disputed this

claim, submitting numerous affidavits from travel agents praising the

airlines’ policy of advanced notice, and arguing that such signaling was
employed in geographic markets where only one airline had market power.
The travel agents did not, however, have access to some of this information
(such as the footnote designators), and thus could not readily determine all
of the airlines’ contemplated changes to fares. Nor could the agents (unlike
the airlines) readily determine the relationships between proposed fare
increases for certain routes and the elimination of discounted fares on other
routes. Moreover, the pricing information, asserted the DOJ, was unreliable
and misleading, in par ticu lar because the airlines changed the ticket dates
often. The DOJ’s consent decrees attempted to shift the lever toward pro-
moting information of use to the consumers. The decrees did not prohibit
the posting of airfare pricing; rather, the defendants were prohibited from
posting fare information of little significance to the consumer, namely Last
Ticket Dates, with the exception of those used in advertised promotions, and
First Ticket Dates. Thus, the airlines’ posted fares would have some signifi-
cance for the consumer, as the travel agents could immediately purchase the
ticket that day for that fare. Likewise, by restricting the airlines from using
Last Ticket Dates except under advertised commitments, the decrees ended
the “costless communication” among the defendants about which discounts
should be removed. The decrees did not eliminate the possibility of tacit
coordination. Rather, they made such negotiations costlier for the airlines by
imposing some risk on the price leader. Moreover, when one airline violated
this decree by signaling a price increase through a prohibited mechanism, it
resulted in a $3 million civil penalty.
17. Take, for example, the European concept of “concerted practice,” which
refers to forms of coordination between companies, which, without reaching
the level of an agreement, have nevertheless established practical coopera-
tion between them. See Case 40/73, Suiker Unie and Others v. Commission,
[1975] ECR 1663, para. 26, and Case C-89/85 I, [1993] ECR I-1307, para. 63;
commission decision in Case IV/37.614/F3 PO, Interbrew and Alken-Maes,
Copyright © 2016. Harvard University Press. All rights reserved.

[2003] OJ L200/1, para. 221.

18. Article 101(1) TFEU provides that agreements, decisions, or concerted
practices which have as their object or effect the direct or indirect fi xing of
selling price would be prohibited. The European courts and commission
have generally treated price-fi xing, market-sharing, and bid-rigging arrange-
ments as having the object of restricting competition.
19. Agreements among competitors that “tamper” with price structure are per
se illegal. United States v. Socony-Vacuum Oil Co., 310 U.S. 150, 221 (1940).
“Even though members of the price-fi xing group were in no position to

control the market, to the extent that they raised, lowered, or stabilized
prices they would be directly interfering with the free play of market forces.”
20. Power Conversion, Inc. v. Saft Am., Inc., 672 F. Supp. 224, 227 (D. Md. 1987).
“Price-fi xing is per se illegal regardless of whether the objective is to raise or
lower market prices, whether the agreement is successful or not, and
whether the prices were reasonable or not.” Thus, the Sherman Act reaches
combinations formed for the purpose, and with the effect, of raising,
depressing, fi xing, pegging, or stabilizing prices. Antitrust plaintiffs need
not prove that defendants fi xed prices directly or controlled a substantial
part of the commodity, that no competition remained, or that prices as a
result were uniform, inflexible, or unreasonable. Socony-Vacuum, 310 U.S.
at 222, 224.
21. Maurice Stucke, “Morality and Antitrust,” Columbia Business Law Review
(2006): 443.
22. Songfacts, “Milgram’s 37 (We Do What We’re Told),” by Peter Gabriel,
http://www.songfacts.com/detail.php?id=772.
23. S. Milgram, “Behavioral Study of Obedience,” Journal of Abnormal & Social
Psychology 67, no. 4 (1963): 371.
24. DP DenkProducties, “Milgram Experiment—Jeroen Busscher,” YouTube
(June 2012), https://www.youtube.com/watch?v=yr5cjyokVUs.
25. S. Milgram, Obedience to Authority: An Experimental View (New York:
Harper & Row, 1974), 30–31.
26. Milgram, “Behavioral Study of Obedience.”
27. F. Gino et al., “See No Evil: When We Overlook Other People’s Unethical
Behav ior,” HBS Working Paper No 08-045 (January 11, 2008), 11.
28. M. C. Levenstein and V. Y. Suslow, “Breaking Up Is Hard to Do: Determi-
nants of Cartel Duration,” Ross School of Business Paper No. 1150 (Sep-
tember 2009), 11, http://ssrn.com/abstract =1676968; Gino et al., “See No
Evil” (discussing identifiable victim effect where people have greater concern
for identifiable, than statistical, victims).
Copyright © 2016. Harvard University Press. All rights reserved.

6 • Hub and Spoke

1. United States v. Newton, 326 F.3d 253, 255 (1st Cir. 2003).
2. Ibid.
3. Interstate Circuit v. United States, 306 U.S. 208, 227 (1939).
4. United States v. Lapier, No. 13-30279, 2015 WL 4664689, at para. 8 (9th Cir.
August 7, 2015) (internal quotations omitted).
5. United States v. Apple, Inc., No. 13-3741-CV, 2015 WL 3953243, para. 28
(2d Cir. June 30, 2015).

6. Ibid., para. 17; Howard Hess Dental Labs. Inc. v. Dentsply Int’l, Inc., 602 F.3d
237, 255 (3d Cir. 2010); see also Toys “R” Us, Inc. v. FTC, 221 F.3d 928, 932–934
(7th Cir. 2000).
7. Interstate Circuit v. United States, 306 U.S. 208 (1939).
8. On the principal liability of such a facilitator, see, for instance, the European
Commission’s decision in AC-Treuhand AG. There, a consulting group was
found to violate the competition law by helping organize cartel meetings.
The company collected and supplied to the cartel members data on sales on
the relevant markets; offered to act as a moderator in case of tensions
between the cartel members; and encouraged the cartel members to fi nd
compromises, for which it received remuneration. The consulting firm
claimed it wasn’t liable, as the law applied to the competitors who conspired,
and not to those who merely helped to organize the cartel meetings or
provide ser vices in the context of the anticompetitive agreements. The
General Court and the European Court of Justice upheld the Commission’s
decision (T-27/10 AC-Treuhand v. Commission, C-194/14 P AC Treuhand v.
Commission).
9. European Commission, Antitrust: Commission Fines Broker ICAP €14.9
Million for Participation in Several Cartels in Yen Interest Rate Derivatives
Sector, IP/15/4104 (Brussels: European Commission, February 4, 2015),
http://europa.eu/rapid/press-release _IP-15-4104 _en.htm.
10. United States v. Apple, Inc., 952 F. Supp. 2d 638 (S.D.N.Y. 2013), aff ’d,
791 F.3d 290 (2d Cir. 2015). Apple and its coconspirators changed the
wholesale business model used at the time by Amazon, leading to higher
prices for e-books. United States v. Apple, Inc., 791 F.3d 290, 310 (2d Cir.
2015), cert. denied, 136 S. Ct. 1376 (2016) (“Based on data from February
2010—just before the Publisher Defendants switched Amazon to agency
pricing—to February 2011, an expert retained by the Justice Department
observed that the weighted average price of the Publisher Defendants’ new
releases increased by 24.2%, while bestsellers increased by 40.4%, and other
ebooks increased by 27.5%, for a total weighted average ebook price increase
Copyright © 2016. Harvard University Press. All rights reserved.

of 23.9%”).
11. The case concerned Apple’s use of price parity conditions when launching
the iPad and its iBooks Store in 2010. Price parity clauses provide assurance
to the downstream online platform that it has received goods or ser vices
from the supplier at terms that are at least as favorable as those offered to any
other buyers. They are sometimes combined with an agency distribution
model, in which the seller determines the price offered on the platform.
Under such a combination, the platform and the seller agree that the price
charged on the platform will not be lower than the price the seller sets and
charges when selling through other platforms. See generally Okeoghene

Odudu, “Indirect Information Exchange: The Constituent Elements of Hub

and Spoke Collusion,” European Competition Journal 7, no. 2 (2011):
205–242.
12. Indeed, the court condemned the agreements, “not because those [most
favored nation and agency] Contracts themselves were independently
unlawful, but because, in context, they provide strong evidence that Apple
consciously orchestrated a conspiracy among the Publisher Defendants.”
Apple, 791 F.3d at 316. See also a similar statement by the U.S. District
Court, S.D. New York: “If Apple is suggesting that an adverse ruling
necessarily implies that agency agreements, pricing tiers with caps, MFN
clauses, or simultaneous negotiations with suppliers are improper, it is
wrong. As explained above, the Plaintiffs have not argued and this Court has
not found that any of these or other such components of Apple’s entry into
the market were wrongful, either alone or in combination. What was
wrongful was the use of those components to facilitate a conspiracy with the
Publisher Defendants”; United States v. Apple Inc., 952 F. Supp. 2d 638, 708
(S.D.N.Y. 2013).
13. Boomerang Commerce, Our Story, http://www.boomerangcommerce.com
/about/.
14. Jason Del Rey, “Amazon Vet Raises $8.5 Million to Help Retailers Think
More Like Amazon,” Re/Code (July 16, 2014), http://recode.net/2014/07/16
/amazon-vet-raises-8-5-million-to-help-retailers-think-more-like-amazon/.
15. Ibid.
16. Boomerang Commerce, What’s Worse than An 800-Pound Gorilla Under-
cutting Your Prices?, http://www.boomerangcommerce.com/resources/whats
-worse-than-an-800-pound-gorilla-undercutting-your-prices/.
17. Rohit Joshi, “How Does Uber’s Dispatch Algorithm Work?” Quora (De-
cember 13, 2014), http://www.quora.com/How-does-Ubers-dispatch
-algorithm-work; James Surowiecki, “In Praise of Efficient Price Gouging,”
MIT Technology Review, August 19, 2014, http://www.technologyreview.com
/review/529961/in-praise-of-efficient-price-gouging/; Eric Posner, “Why
Copyright © 2016. Harvard University Press. All rights reserved.

Uber Will—and Should—Be Regulated,” Slate (January 5, 2015), http://www

.slate.com/articles/news _ and _politics/view_from _chicago/2015/01/uber
_ surge _pricing _federal _regulation _over_taxis _ and _car_ride _ services
.html.
18. Douglas Macmillan and Telis Demos, “Uber Valued at More than $50
Billion,” Wall Street Journal (London), July 31, 2015, http://www.wsj.com
/articles/uber-valued-at-more-than-50-billion-1438367457.
19. See, for example, Mark Harris, “Uber: Why the World’s Biggest Ride-
Sharing Company Has No Drivers,” The Guardian, November 16, 2015,
http://www.theguardian.com/technology/2015/nov/16/uber-worlds-biggest

-ride-sharing-company-no-drivers; Izabella Kaminska, “If and When Uber

Drivers Unionise . . . ,” Financial Times, January 12, 2016, http://ftalphaville
.ft.com/2016/01/12/2149878/if-and-when-uber-drivers-unionise/ (on the
possibility of Uber drivers receiving low wages and few employment rights);
Tim Bradshaw and Leslie Hook, 2015. “Uber Drivers Win Union ‘Break-
through,’ ” Financial Times (San Francisco), December 15, 2015, http://www
.ft.com/cms/s/0/37930e72-a2c6-11e5-bc70-7ff6d4fd203a.html#axzz3yTDzV6n7
(discussing a new law introduced by Seattle City Council allowing Uber
drivers to unionize to demand better conditions from the company); Leslie
Hook, “Setback for Uber on Drivers’ Class Action Case,” Financial Times
(San Francisco), December 10, 2015, http://www.ft.com/cms/s/0/ddc7b032
-9ec9-11e5-b45d-4812f209f861.html#axzz3yTDzV6n7 (on a California case
on whether Uber drivers should be treated as full employees rather than
contractors).
20. Ian Beetlestone, “Why London’s Black Cab Drivers Are Protesting over
Uber,” The Guardian, June 11, 2014, http://www.theguardian.com
/commentisfree/2014/jun/11/why-london-taxi-drivers-protesting-uber-tfl
(an article by a London black cab driver on how Uber is able to become a taxi
ser vice “without going through any of the regulatory hoops”).
21. Macmillan and Demos, “Uber Valued at More than $50 Billion.”
22. Uber, Always the Ride You Want: The Best Way to Get Wherever You’re
Going, https://www.uber.com/ride.
23. Sarah Ashley O’Brien, “NYC Uber Drivers Protest Rate Cuts,” CNN Money
(February 1, 2016), http://money.cnn.com/2016/02/01/technology/uber-nyc
-protest/index.html?sr =twCNN020116uber-nyc-protest0317PMVODtopPhoto
&linkId=20849630; Lyft, Nashville Drivers Make Up to $6000/Month Driving
Your Car, https://www.lyft.com/drive-for-lyft?im=& inc= 6000& t=month
&kw=Nashville%20Drivers& utm _ source =bing& utm _medium= search& utm
_campaign=Driver_BNA _v2 _ Search _Brand _ All_Lyft& utm _term=lyft%20
com%20driver&adgroup =lyft _driver&device = c& matchtype =b.
24. Uber, “Dynamic Pricing 101 | Uber,” YouTube (December 2014), https://www
Copyright © 2016. Harvard University Press. All rights reserved.

.youtube.com/watch?v=76q7PDnxWuE.
25. Annie Lowrey, “Is Uber’s Surge-Pricing an Example of High-Tech Gouging?,”
New York Times Magazine, January 10, 2014, http://www.nytimes.com/2014
/01/12/magazine/is-ubers-surge-pricing-an-example-of-high-tech-gouging
.html?_r = 0.
26. Jay Hathaway, “Uber Turned on Surge Pricing for People Fleeing Sydney
Hostage Scene,” December 15, 2014, http://gawker.com/uber-turned-on
-surge-pricing-for-people-fleeing-sydney-1671193132; Brian Ries & Jenni
Ryall, “Uber Intros Surge Pricing during Sydney Hostage Siege, Then

Backtracks after User Outcry,” December 15, 2014, http://mashable.com/2014

/12/14/uber-sydney-surge-pricing/#lnLL3YYzXSqM.
27. Min Kyung Lee, Daniel Kusbit, Evan Metsky, and Laura Dabbish, “Working
with Machines: The Impact of Algorithmic and Data-Driven Management
on Human Workers” (Pittsburgh: Human-Computer Interaction Institute,
Heinz College, Carnegie Mellon University, 2015), http://www.cs.cmu.edu
/~mklee/materials/Publication/2015-CHI _ algorithmic _management.pdf.
28. For instance, in Tesco v. Office of Fair Trading, the U.K. Competition Appeal
Tribunal elaborated that an indirect information exchange through a third
party will amount to an objectionable hub-and-spoke conspiracy when two
phases are present: 1. Retailer A discloses to supplier B its future pricing,
with the intention that B will pass that information to other retailers in order
to influence market conditions. 2. Retailer C receives the information from
supplier B, knowing the circumstances in which its competitor retailer A
disclosed it to B, and C makes use of that information in determining its
own future pricing intentions; Case 1188/1/1/11, Tesco v. Office of Fair
Trading, [2012] CAT 31, para. 57, 58.
29. Case C-74/14, Eturas and Others (2016).
30. The Advocate General, who advises the Court, noted in his opinion that
“where the sender of the information is not a competitor but rather a
third party, such interaction may give rise to a horizontal collusion
between competitors only if the addressee may be deemed to appreciate
that the information transmitted by a third party comes from a compet-
itor or at least is also communicated to a competitor.” Ibid., AG Opinion,
para. 50.
31. Case C-74/14, Eturas and Others, para. 45 (holding that “if it cannot be
established that a travel agency was aware of that message, its participation
in a concertation cannot be inferred from the mere existence of a technical
restriction implemented in the system at issue . . . , unless it is established on
the basis of other objective and consistent indicia that it tacitly assented to an
anticompetitive action”).
Copyright © 2016. Harvard University Press. All rights reserved.

32. Maurice E. Stucke, “Is Intent Relevant?” Journal of Law, Economics & Policy
8 (2012): 801; U.S. Department of Justice Antitrust Division, Antitrust Division
Manual, 5th ed. (Washington, DC: U.S. Department of Justice, March 2014),
chap. 3–12 (noting how the Department of Justice would not prosecute an
offense criminally if “there is clear evidence that the subjects of the
investigation were not aware of, or did not appreciate, the consequences of
their action”). In evaluating collaboration among competitors, competition
agencies consider evidence of intent, which “may aid in evaluating market
power, the likelihood of anticompetitive harm, and claimed procompetitive

justifications where an agreement’s effects are other wise ambiguous”; Federal

Trade Commission and U.S. Department of Justice, Antitrust Guidelines for
Collaborations among Competitors (April 2000), p. 12, note 35, https://www
.ftc.gov/sites/default/fi les/documents/public _events/joint-venture-hearings
-antitrust-guidelines-collaboration-among-competitors/ftcdojguidelines-2
.pdf. Likewise, the European Commission assesses “whether or not an
agreement has as its object the restriction of competition,” based on “a
number of factors,” including evidence of the parties’ subjective intent;
European Commission. 2004. Communication from the Commission, Notice
Guidelines on the Application of Article 81(3) of the Treaty, 2004/C 101/08)
(April 27, 2004), para. 22, http://eur-lex.europa.eu/LexUriServ/LexUriServ.do
?uri= OJ:C:2004:101:0097:0118:EN:PDF. See also Federal Trade Commission
and U.S. Department of Justice, Horizontal Merger Guidelines (August 10,
2010), para. 2.2.1, https://www.ftc.gov/sites/default /fi les/attachments
/merger-review/100819hmg.pdf: “Explicit or implicit evidence that the
merging parties intend to raise prices, reduce output or capacity, reduce
product quality or variety, withdraw products or delay their introduction,
or curtail research and development efforts after the merger, or explicit or
implicit evidence that the ability to engage in such conduct motivated the
merger, can be highly informative in evaluating the likely effects of a
merger.”
33. United States v. U.S. Gypsum Co., 438 U.S. 422, 444–446 (1978) (concluding
that “action undertaken with knowledge of its probable consequences and
having the requisite anticompetitive effects can be a sufficient predicate for a
finding of criminal liability under the antitrust laws”).
34. See for example, Case 1188/1/1/11, Tesco v. Office of Fair Trading.
35. Returning to our Uber example, it is impor tant to distinguish our discussion
of an algorithm being used as a hub from an argument that Uber and the
drivers compete horizontally with each other. Such a claim served as the
backbone to a lawsuit launched in Canada in September 2015. Uber was
accused of price fi xing by Edmonton taxi companies. The CAN$150 million
Copyright © 2016. Harvard University Press. All rights reserved.

lawsuit put forward an argument that Uber is a competitor to its drivers

since it is the one setting the prices for fares. Subsequently, it argued that
Uber conspired with its drivers to fi x prices of vehicle-for-hire ser vices.
See “Uber Accused of Price-Fixing in $150M Lawsuit by Edmonton Taxi
Companies,” CBC News (September 14, 2015), http://www.cbc.ca/news
/canada/edmonton/uber-accused-of-price-fi xing-in-150m-lawsuit-by
-edmonton-taxi-companies-1.3228115.
36. Meyer v. Kalanick, Case 1:15-cv-09796-JSR, slip op. (S.D.N.Y. March 31, 2016).
37. Ibid., 15.

38. Maurice E. Stucke, “Does the Rule of Reason Violate the Rule of Law?”
U.C. Davis Law Review 42 (2009): 1375; Maurice E. Stucke, “Antitrust
Marathon: Antitrust and the Rule of Law,” Loyola Consumer Law Review,
22 (2009): 15; Peter C. Carstensen, “The Content of the Hollow Core of
Antitrust: The Chicago Board of Trade Case and the Meaning of the ‘Rule
of Reason’ in Restraint of Trade Analysis,” Research in Law and Economics
15 (1992): 1, 4.
39. Adam Candeub, “Behavioral Economics, Internet Search, and Antitrust,”
MSU Legal Studies Research Paper No. 12-03 (2014), http://ssrn.com/abstract
=2414179.
40. Judy Wajcman, Pressed for Time: The Acceleration of Life in Digital Capi-
talism (Chicago: University of Chicago Press, 2015); see also Hartmut Rosa,
Social Acceleration—A New Theory of Modernity (New York: Columbia
University Press, 2013).

7 • Tacit Collusion on Steroids: The Predictable Agent

1. Brooke Group Ltd. v. Brown & Williamson Tobacco Corp., 509 U.S. 209
(1993); R. S. Khemani and D. M. Shapiro, Glossary of Industrial Organisation
Economics and Competition Law (Paris: Organisation for Economic Co-
operation and Development, 1993), http://www.oecd.org/dataoecd/8/61
/2376087.pdf.
2. Marc Ivaldi, Bruno Jullien, Patrick Rey, Paul Seabright, and Jean Tirole,
“The Economics of Tacit Collusion,” Final Report for DG Competition
(Toulouse: Eu ropean Commission, March 2003), 4, http://ec.europa .eu
/competition/mergers/studies _ reports/the _ economics _of _tacit _ collusion
_ en.pdf.
3. For a review of the economics of tacit collusion and the EU approach, see
Nicolas Petit, “The ‘Oligopoly Problem’ in EU Competition Law” in Research
Handbook in European Competition Law, Ioannis Liannos and Damien
Geradin, eds. (Edward Elgar Publishing, 2013), 259.
Copyright © 2016. Harvard University Press. All rights reserved.

4. White v. R.M. Packer Co., 635 F.3d 571, 579 (1st Cir. 2011).
5. Ibid. Twenty-one cents of that difference is attributable to the higher costs of
transporting gas to the island than to the mainland cape.
6. White v. R.M. Packer Co.
7. Ibid. Would-be competitors attracted to the market by high profit margins
“face a regulatory barrier to entry: they need permission from the Martha’s
Vineyard Commission. The Commission has denied all petitions to open
new gas stations since 1997. This, along with their location on a relatively
small island, insulates the current stations from competition.”

8. Ibid. “Gasoline in general is a nondurable good, so that customers have to

buy it frequently and are not likely to simply stay out of the market until
prices drop. This is particularly true for customers who are summer residents
and are in the market for only limited periods of time.”
9. Ibid.
10. Ibid.
11. Ibid., 579.
12. Ibid.
13. Ibid.
14. Consider, for instance, the German Competition Authority (Bundeskartellamt)
Fuel Sector Inquiry, in which the agency identified a dominant oligopoly in
regional fuel retail markets. The agency concluded that the “retail prices of
the majority of off-motorway petrol stations were higher in the oligopolistic
setting than they would have been if effective competition had been in
place.” Bundeskartellamt, Fuel Sector Inquiry Final Report in Accordance
with § 32e GWB (Bonn: Bundeskartellamt, May 2011), section 5, http://
www.bundeskartellamt.de/SharedDocs/Publikation/EN/Sector%20
Inquiries/Fuel%20Sector%20Inquiry%20-%20Final%20Report.pdf ?_ _blob
= publicationFile& v=14.
15. Federal Trade Commission and U.S. Department of Justice, Horizontal
Merger Guidelines (August 10, 2010), para. 7.2, https://www.ftc.gov/sites
/default/fi les/attachments/merger-review/100819hmg.pdf.
16. Ibid.
17. Case T-342/99, Airtours, [2002] ECR 2585, [2002] 5 CMLR 317, para. 61.
18. European Commission, Guidelines on the Assessment of Horizontal Mergers
under the Council Regulation on the Control of Concentrations between
Undertakings, 2004/C 31/03 (February 5, 2004), para. 49–50.
19. Case T-342/99, Airtours.
20. Ivaldi et al., “The Economics of Tacit Collusion,” 5.
21. Federal Trade Commission and U.S. Department of Justice, Horizontal
Merger Guidelines, para. 7.
Copyright © 2016. Harvard University Press. All rights reserved.

22. Roland Moore-Colyer, “Predictive Analytics Are the Future of Big Data,” V3
(October 9, 2015), http://www.v3.co.uk /v3-uk /analysis/2429494/predictive
-analytics-are-the-future-of-big-data.
23. Ibid., citing Larry Augustine, chief executive at SugarCRM.
24. Samuel B. Hwang and Sungho Kim, “Dynamic Pricing Algorithm for
E-Commerce,” in Advances in Systems, Computing Sciences and Software
Engineering, Proceedings of SCSS05, Tarek Sobh and Khaled Elleithy, eds.
(Dordrecht: Springer, 2006), 149–155; N. Abe and T. Kamba, “A Web
Marketing System with Automatic Pricing,” Computer Networks 33 (2000):
775–788; L. M. Minga, Y. Q. Fend, and Y. J. Li, “Dynamic Pricing:

E-Commerce-Oriented Price Setting Algorithm,” International Conference

on Machine Learning and Cybernetics 2 (2003).
25. Alternatively, the computers can engage in parallel accommodating conduct,
whereby “each rival’s response to competitive moves made by others is
individually rational, and not motivated by retaliation or deterrence nor
intended to sustain an agreed-upon market outcome, but nevertheless
emboldens price increases and weakens competitive incentives to reduce
prices or offer customers better terms.” Federal Trade Commission and U.S.
Department of Justice, Horizontal Merger Guidelines, para. 7.
26. Fuel apps have become a common feature and can be downloaded for free.
27. See Salil K. Mehra, “Antitrust and the Robo-Seller: Competition in the Time
of Algorithms,” Minnesota Law Review 100 (March 10, 2015), http://ssrn.com
/abstract=2576341, on how pricing algorithms can promote tacit collusion
under a Cournot model.
28. The outcome, in any game with set rules and a finite number of sequential
moves, could be determined by the initial move; Avinash Dixit and Barry
Nalebuee, Thinking Strategically: The Competitive Edge in Business, Politics, and
Everyday Life (New York: W. W. Norton, 1991), 41–44. Indeed, E. Zermelo
formulated in the early twentieth century an algorithm that leads to an
equilibrium outcome in chess, but nonetheless, grandmasters still compete
today. Martin J. Osborne and Ariel Rubinstein, A Course in Game Theory
(Cambridge, MA: Massachusetts Institute of Technology Press, 1994), 6.
29. Christopher Chabris, “High-Tech Chess Cheaters Charge Ahead,” Wall Street
Journal, October 9, 2015, http://www.wsj.com/articles/high-tech-chess
-cheaters-charge-ahead-1444404660.
30. Michael Lewis, Flash Boys: A Wall Street Revolt (New York: W. W. Norton,
2014).
31. The plaintiff can allege that the defendant firms collectively agreed to use
these algorithms; specifically, it was their collective agreement to use a
facilitating device that fosters tacit collusion. See Todd v. Exxon Corp.,
275 F.3d 191 (2d Cir. 2001). The benefit of this approach is that it may be
Copyright © 2016. Harvard University Press. All rights reserved.

easier to prove that the industry agreed to use algorithms (especially in order
to ensure their interoperability) and knew that its rival firms’ algorithms had
similar reward structures than it is to prove an agreement to fi x prices. The
downsides of this approach are the cost, duration, and unpredictability of a
rule of reason case, and the difficulty for the court in weighing the procom-
petitive benefits of product developments with the anticompetitive effects.
32. See, for example, Case C-199/92, P Hüls AG v. Commission, [1999] ECR
I-4287, [1999] 5 CMLR 1016; Joined Cases C-89, 104, 114, 116, 117, 125,
129/85, Ahlström Osakeyhtiö and others v. Commission (Wood Pulp II), [1993]
ECR I-1307, [1993] 4 CMLR 407; Cases T-442/08, CISAC v Commission,

[2013] 5 CMLR 15 (General Court). Note that our focus is on non-dominant

firms.
33. See Federal Trade Commission and U.S. Department of Justice, Horizontal
Merger Guidelines, para. 7.
34. In re Text Messaging Antitrust Litig., 782 F.3d 867, 874 (7th Cir.) cert. denied
sub nom.; Aircraft Check Servs. Co. v. Verizon Wireless, 136 S. Ct. 524 (2015).
35. Boise Cascade Corp. v. F.T.C., 637 F.2d 573 (9th Cir. 1980).
36. E. I. du Pont de Nemours & Co. v. F.T.C., 729 F.2d 128 (2d Cir. 1984).
37. Ibid., 128, 139.
38. U.S. Securities and Exchange Commission, Administrative Proceeding File
No. 3-16199 (October 16, 2014), http://www.sec.gov/litigation/admin/2014/34
-73369.pdf.
39. The computer trading program was “placing a large number of aggressive,
rapid-fire trades in the final two seconds of almost every trading day during
a six-month period to manipulate the closing prices of thousands of
NASDAQ-listed stocks.” U.S. Securities and Exchange Commission, SEC
Charges New York–Based High Frequency Trading Firm with Fraudulent
Trading to Manipulate Closing Prices, October 16, 2014, http://www.sec.gov
/News/PressRelease/Detail/PressRelease/1370543184457#.VEOZlfldV8E.
Ibid.
40. Ibid.
41. Ibid. As the SEC alleged Athena’s manipulative scheme focused on trading
in order to create imbalances in securities at the close of the trading day:
“Imbalances occur when there are more orders to buy shares than to sell
shares (or vice versa) at the close for any given stock. Every day at the close of
trading, NASDAQ runs a closing auction to fi ll all on-close orders at the best
price, one that is not too distant from the price of the stock just before the
close. Athena placed orders to fi ll imbalances in securities at the close of
trading, and then traded or ‘accumulated’ shares on the continuous market
on the opposite side of its order.” According to the SEC’s order, Athena’s
algorithmic strategies became increasingly focused on ensuring that the firm
Copyright © 2016. Harvard University Press. All rights reserved.

was the dominant firm—and sometimes the only one—trading desirable

stock imbalances at the end of each trading day. The firm implemented
additional algorithms known as “Collars” to ensure that Athena’s orders
received priority over other orders when trading imbalances. These eventu-
ally resulted in Athena’s imbalance-on-close orders being at least partially
fi lled more than 98 percent of the time. Athena’s ability to predict that its
orders would get fi lled on almost every imbalance order allowed the firm to
unleash its manipulative Gravy algorithm to trade tens of thousands of
shares right before the close of trading. As a result, these shares traded at
artificial prices that NASDAQ then used to set the closing prices for on-close

orders as part of its closing auction. Athena’s high-frequency trading scheme

enabled its orders to be executed at more favorable prices.
42. U.S. Securities and Exchange Commission, Administrative Proceeding File
No. 3-16199, para. 34.
43. Ibid., para. 36.
44. Peter J. Henning, “Why High-Frequency Trading Is so Hard to Regulate,”
New York Times, October 20, 2014, http://dealbook.nytimes.com/2014/10/20
/why-high-frequency-trading-is-so-hard-to-regulate/.
45. Steve Goldstein, “High-Frequency Trading Firm Fined for Wave of Last-
Minute Trades,” Market Watch (October 16, 2014), http://www.marketwatch
.com/story/high-frequency-trading-firm-fined-for-wave-of-last-minute
-trades-2014-10-16.
46. Note that the algorithms may be designed to deter entry, provide complex
signals as to profitability, and engage in limit pricing or other strategies.
47. One would expect tacit collusion to be feasible with a larger number of
participants than commonly assumed. On the common market assump-
tions, see generally R. Selten, “A Simple Model of Imperfect Competition,
Where Four Are Few and Six Are Many,” International Journal of Game
Theory 2 (1973): 141; Steffen Hucka, Hans-Theo Normannb, and Jörg
Oechssler, “Two Are Few and Four Are Many: Number Effects in Experi-
mental Oligopolies,” Journal of Economic Behavior and Organization 53,
no. 4 (2004): 435–446.
48. Sugar Institute, Inc. v. United States, 297 U.S. 553, 598 (1936).
49. United States v. United States Gypsum Co., 438 U.S. 422, 441 n.16 (1978);
See also Richard A. Posner, Antitrust Law, 2nd ed. (Chicago: University of
Chicago Press, 2001), 160. Generally, the more information sellers have about
their competitors’ prices and output, the more efficiently the market will
operate.
50. See, for example, Federal Trade Commission, Funeral Directors Board Settles
with FTC (August 16, 2004), http://www.ftc.gov/opa/2004/08/vafuneral.htm
(a board’s prohibition on licensed funeral directors advertising discounts
Copyright © 2016. Harvard University Press. All rights reserved.

deprived consumers of truthful information); Federal Trade Commission,

Arizona Automobile Dealers Association, FTC C-3497 (February 25, 1994)
(a trade association illegally agreed with members to restrict nondeceptive
comparative and discount advertising and advertisements concerning the
terms and availability of consumer credit); Organisation for Economic
Co-operation and Development, Price Transparency, DAFFE/CLP(2001)22
(September 11, 2001), 183, 185–186 (citing examples of U.S. enforcement
agencies seeking to increase price transparency); compare InterVest, Inc. v.
Bloomberg, L.P., 340 F.3d 144 (3d Cir. 2003) (lack of price transparency in
bond market not illegal if consistent with unilateral conduct).

8 • Artiﬁcial Intelligence, God View, and the Digital Eye

1. Johana Bhuiyan and Charlie Warzel, “ ‘God View’: Uber Investigates Its Top
New York Executive for Privacy Violations,” BuzzFeed News (November 18,
2014), http://www.buzzfeed.com/johanabhuiyan/uber-is-investigating-its-top
-new-york-executive-for-privacy#.fcOoXKDMw.
2. Federal Trade Commission and U.S. Department of Justice, Horizontal
Merger Guidelines (August 10, 2010), para. 7.2, https://www.ftc.gov/sites
/default/fi les/attachments/merger-review/100819hmg.pdf.
3. Eric Brown, “Amazon’s AWS IoT Platform Taps Three Linux SBCs,” Linux-
Gizmos.com (October 9, 2015), http://linuxgizmos.com/amazons-aws-iot
-platform-taps-three-linux-sbcs/.
4. Natalie Mortimer, “Amazon Launches Platform to Build IoT Apps for Cars,
Lightbulbs and More,” The Drum (October 9, 2015), http://www.thedrum
.com/news/2015/10/09/amazon-launches-platform-build-iot-apps-cars
-lightbulbs-and-more.
5. Ibid.
6. Absent such limiting principles, the scenario would be similar to the first
category of Messenger.
7. Bell Atlantic Corp. v. Twombly, 550 U.S. 554 (2007).
8. Avinash Dixit and Barry Nalebuff, Thinking Strategically: The Competitive Edge
in Business, Politics, and Everyday Life (New York: W. W. Norton, 1991), 108.
9. Ibid., 111.
10. Don Ross, “Game Theory,” in The Stanford Encyclopedia of Philosophy
(Winter 2014 ed.), Edward N. Zalta, ed., http://plato.stanford.edu/archives
/win2014/entries/game-theory/.
11. For a comprehensive examination of how cartels facilitate trust, see
Christopher R. Leslie, “Trust, Distrust, and Antitrust,” Texas Law Review
82 (2004): 515.
12. One empirical analysis of successfully prosecuted cartels between 1910
and 1972 showed that cartels on average had many participants: where a
Copyright © 2016. Harvard University Press. All rights reserved.

trade association facilitated collusion, 33.6 was the mean number of fi rms
involved, and fourteen firms was the median; in price-fi xing cartels (without
a trade association involved), 8.3 firms was the mean and six was the
median. Arthur G. Frass and Douglas F. Greer, “Market Structure and Price
Collusion: An Empirical Analysis,” Journal of Industrial Economics 26
(1977): 21, 25, 36–41.
13. Federal Trade Commission and U.S. Department of Justice, Horizontal
Merger Guidelines, para. 7.
14. Daniel Kahneman, Thinking, Fast and Slow (New York: Farrar, Straus and
Giroux, 2011).

15. Tony Curzon Price, “Using Co-Evolutionary Programming to Simulate

Strategic Behaviour in Markets,” Journal of Evolutionary Economics 7, no. 3
(September 1997): 219–254, http://link.springer.com/article/10.1007
/s001910050042.
16. On this point see Federal Trade Commission and U.S. Department of Justice,
Horizontal Merger Guidelines, para. 7; Casey C. Sullivan, “As Machines
Learn, Will They Learn the Law? Will They Follow It?” FindLaw (September
8, 2015), http://blogs.findlaw.com/technologist/2015/09/as-machines-learn
-will-they-learn-the-law-will-they-follow-it.html.
17. A. Ezrachi, “Sponge,” University of Oxford Centre for Competition Law and
Policy Working Paper CCLP (L) 42 (March 1, 2015), http://ssrn.com/abstract
=2572028.

Part III • Behavioral Discrimination

1. Jennifer Valentino-Devries, Jeremy Singer-Vine, and Ashkan Soltani,

“Websites Vary Prices, Deals Based on Users’ Information,” Wall Street
Journal, December 24, 2012, http://www.wsj.com/articles/SB10001424127887
323777204578189391813881534.

9 • Price Discrimination (Brieﬂy) Explained

1. Herbert Hovenkamp, Mark D. Janis, Mark A. Lemley, Christopher R. Leslie,

and Michael A. Carrier, IP and Antitrust: An Analysis of Antitrust Principles
Applied to Intellectual Property Law, 2nd ed. (Frederick, MD: Aspen
Publishers, 2010), Appendix F.
2. Deven R. Desai, Ioannis Lianos, and Spencer Weber Waller, eds., Brands
Competition Law and IP (Cambridge: Cambridge University Press, 2015).
3. Federal Trade Commission and U.S. Department of Justice, Horizontal
Merger Guidelines (August 10, 2010), para. 3, https://www.ftc.gov/sites
Copyright © 2016. Harvard University Press. All rights reserved.

/default/fi les/attachments/merger-review/100819hmg.pdf.
4. R. S. Khemani and D. M. Shapiro, Glossary of Industrial Organisation
Economics and Competition Law (Paris: Organisation for Economic Co-
operation and Development, 1993), http://www.oecd.org/dataoecd/8/61
/2376087.pdf.
5. Peter Schmidt, “At Elite Colleges—Dim White Kids,” Boston Globe, Sep-
tember 28, 2007, http://www.boston.com/news/globe/editorial _opinion/oped
/articles/2007/09/28/at _the _elite _colleges _ _ _dim _white _ kids/?page =full.
6. Federal Trade Commission and U.S. Department of Justice, Horizontal
Merger Guidelines, para. 3.

7. Ibid.
8. Barry C. Smith, John F. Leimkuhler, and Ross M. Darrow, “Yield Manage-
ment at American Airlines,” Interfaces 22, no. 1 (1992): 8–31; Paul Davies,
“Airline Ties Profitability Yield to Management,” SIAM News 27, no. 5 (1994),
cited in R. Preston McAfee and Vera te Velde “Dynamic Pricing in the
Airline Industry”: “This number may be inflated for several reasons. First,
it includes sales of yield management strategy to others, as opposed to
American’s own use of the techniques, although the value of American’s
internal use is put at just slightly less. Second, it incorporates “damaged
good” considerations in the form of Saturday-night stayover restrictions, as
well dynamic pricing. Such restrictions facilitate static price discrimination,
and are reasonably well-understood in other contexts (Deneckere and
McAfee 1996). Nevertheless, there is little doubt that dynamic price discrim-
ination is economically impor tant. The pricing systems used by most major
airlines are remarkably opaque to the consumer, which is not surprising
given one estimate that American Airlines changes half a million prices per
day.” http://www.mcafee.cc/Papers/PDF/DynamicPriceDiscrimination.pdf.
9. See for example Jack Nicas, “Now Prices Can Change from Minute to
Minute,” Wall Street Journal, December 14, 2015, http://www.wsj.com
/articles/now-prices-can-change-from-minute-to-minute-1450057990.
10. An exception is when the incremental costs to price discriminate exceed the
incremental profits: it costs more to implement than the resulting gain.

10 • The Age of Perfect Price Discrimination?

1. Advertising Week XII, “Journey of Data-Driven Marketing,” Advertising

Week, September 28–October 2, 2015, http://www.advertisingweek.com
/replay/#date =2015- 09-30~video-id=227~venue =2; Suzanne Vranica, “Ad
Blocking Is the Latest Hot Topic, Media Executives,” Wall Street Journal,
September 27, 2015), http://www.wsj.com/articles/ad-blocking-is-a-hot-topic
-for-marketing-media-executives-1443259981.
Copyright © 2016. Harvard University Press. All rights reserved.

2. James Surowiecki, “In Praise of Efficient Price Gouging,” MIT Technology

Review (August 19, 2014), http://www.technologyreview.com/review/529961
/in-praise-of-efficient-price-gouging/.
3. J. Turow, L. Feldman, and K. Meltzer, K. “Open to Exploitation: America’s
Shoppers Online and Offline,” Report from the Annenberg Public Policy
Center of the University of Pennsylvania (June 6, 2005), http://repository
.upenn.edu/asc _papers/35; Jennifer Valentino-Devries, Jeremy Singer-Vine,
and Ashkan Soltani, “Websites Vary Prices, Deals Based on Users’ Informa-
tion,” Wall Street Journal, December 24, 2012, http://www.wsj.com /articles
/SB10001424127887323777204578189391813881534.

4. Valentino-Devries et al., “Websites Vary Prices, Deals Based on Users’

Information.”
5. Ibid.
6. Ibid.
7. Ibid.
8. Ibid.
9. Paragraph 290, House of Lords, Select Committee on European Union,
“Online Platforms and the Digital Single Market,” April 20, 2016, 10th
Report of Session 2015–16, http://www.publications.parliament.uk /pa
/ld201516/ldselect/ldeucom/129/129.pdf.
10. Emily Steel and Julia Angwin, “On the Web’s Cutting Edge, Anonymity
in Name Only,” Wall Street Journal, August 4, 2010, http://www.wsj.com
/articles/SB10001424052748703294904575385532109190198.
11. Valentino-Devries et al., “Websites Vary Prices, Deals Based on Users’
Information.”
12. Ibid.
13. Ibid.
14. Consumer Federation of America, “Report Finds Auto Insurers Charge
Higher Premiums in African American Zip Codes,” CFAnews Update
(November 24, 2015), http://consumerfed.org/cfanews-update-11242015/.
15. Sal Thomas, “Does Dynamic Pricing Risk Turning Personalisation into
Discrimination?” Marketing Magazine, October 22, 2014, http://www
.marketingmagazine.co.uk /article/1317995/does-dynamic-pricing-risk
-turning-personalisation-discrimination.
16. Coupons.com Inc., Form 10-K for the Fiscal Year Ended December 31,
2014 (2015), 3, http://www.sec.gov/Archives/edgar/data/1115128
/000156459015001837/coup-10k _ 20141231.htm. In 2015, the company
changed its name to Quotient Technology Inc.; Coupons.com Inc., “Coupons
.com Incorporated Changes Corporate Name to Quotient” (October 6, 2015),
http://investors.coupons.com/investors/press-releases/press-releases-details
/2015/Couponscom-Incorporated-Changes-Corporate-Name-to-Quotient
Copyright © 2016. Harvard University Press. All rights reserved.

/default.aspx.
17. Coupons.com Inc., “Coupons.com Incorporated Changes Corporate Name
to Quotient.”
18. Coupons.com Inc., Form 10-K for the Fiscal Year Ended December 31,
2014, 5.
19. Ibid.
20. Ibid., 4.
21. Executive Office of the President, Big Data and Differential Pricing (Wash-
ington, DC: Executive Office of the President), February 2015, p. 12,
https://www.whitehouse.gov/sites/default/fi les/whitehouse _ fi les/docs/Big

_Data _ Report _ Nonembargo_v2.pdf: “Loyalty programs provided some of

the first applications of big data to personalized pricing. When a buyer joins
a loyalty program, they typically provide some personal information and
consent to a seller tracking their purchases. In return, the buyer typically
receive [sic] some type of benefit such as seat upgrades or free fl ights from
an airline frequent flier program, or price discounts on specific items from a
grocery store. Sellers use loyalty programs to customize their marketing.
Some retailers also partner with companies that aggregate data from loyalty
programs and use it to create customized coupons, which are printed on the
back of receipts that a customer receives at the cash register or point of sale.
Firms that specialize in this type of personalization claim that data-driven
analysis can increase the redemption rates on such coupons from 1 percent
for non-personalized coupon offers to as much as 25 percent for highly-
targeted coupons.”
22. Charles Duhigg, “How Companies Learn Your Secrets,” New York Times
Sunday Magazine, February 16, 2012, http://www.nytimes.com/2012/02/19
/magazine/shopping-habits.html?pagewanted= all& _r = 0.
23. Ibid.
24. Ibid.
25. Ibid.
26. Target, Privacy Policy, October 1, 2015, http://www.target.com/spot/privacy
-policy#section1 (on collecting data, including the “device model, operating
system version, device date and time, unique device identifiers, and mobile
network information”).
27. Ibid.
28. Ibid.
29. Ibid.
30. Ibid.
31. Ibid.
32. Duhigg, “How Companies Learn Your Secrets.”
33. Ibid.
Copyright © 2016. Harvard University Press. All rights reserved.

34. Ibid.
35. Ibid.
36. Brad Howarth, “How Tesco’s Loyalty Card Transformed Customer Data
Tracking,” CMO (May 21, 2015), http://www.cmo.com.au/article/575497/how
-tesco-loyalty-card-transformed-customer-data-tracking/; Krish Swarmy,
“Analyzing Tesco—The Analytics behind a Top-Notch Loyalty Program,” Big
Data Analytics (August 21, 2011), http://stat-exchange.blogspot.com/2011/08
/analyzing-tesco-analytics-behind-top.html.
37. McKinsey & Company, Making Loyalty Pay: Lessons from the innovators
(McKinsey & Company, July 2013).

38. “The Card Up Their Sleeve,” The Guardian, July 19, 2003, http://www
.theguardian.com/lifeandstyle/2003/jul/19/shopping.features.
39. Alex Chisholm (CMA chief executive), Why “Sleepers” Can’t Always Be Left
to “Sleep,” CCRP 2016 Competition Policy Roundtable (London: Competi-
tion Markets Authority, January 25, 2016), https://www.gov.uk /government
/speeches/alex-chisholm-on-consumer-engagement-in-a-digital-world.
40. Coupons.com Inc., “Coupons.com Incorporated Changes Corporate Name
to Quotient.”
41. For tracking of shoppers’ movements (nonpersonalized), see Chris Cooper,
“Nordstrom Stores Tracking Customers’ Cell Phones,” Stop Cell Phone
Tracking (February 11, 2013), http://www.stopcellphonetracking.com
/nordstrom-tracking-customers-cell-phones/; Chris Cooper, “This Is How
Retail Stores Track Your Smartphone,” Stop Cell Phone Tracking (June 3,
2013), http://www.stopcellphonetracking.com/this-is-how-retail-stores-track
-your-smartphone-video/. For use of technology for individualized in-shop
tracking, see Verne Kopytoff, “Stores Sniff Out Smartphones to Follow
Shoppers,” MIT Technology Review (November 12, 2013), http://www.technology
review.com/news/520811/stores-sniff-out-smartphones-to-follow-shoppers/.
42. Clare McDonald, “Almost 30% of Retailers Use Facial Recognition Tech-
nology to Track Consumers in Store,” ComputerWeekly.com (September 15,
2015), http://www.computerweekly.com/news/4500253499/Almost-30-of
-retailers-use-facial-recognition-technology-to-track-consumers-in-store;
Jimmy Rose, “How Facial Recognition Will Change Shopping in Stores,”
Extreme Tech (June 23, 2015), http://www.extremetech.com/mobile/208815
-how-facial-recognition-will-change-shopping-in-stores; James Hercher,
“Shopper Behav ior Begins In-Store—But Brick-and-Mortars Need Tech
to Harness It,” Ad Exchanger (July 8, 2015), http://adexchanger.com/data
-exchanges/shopper-behavior-begins-in-store-but-brick-and-mortars-need
-tech-to-harness-it/ (“It’s notable that many retail startups rely so heavily on
beacons. AdMobilize, which announced a $1.6 million funding round last
week, installs face-recognizing beacons and cameras in brick-and-mortars in
Copyright © 2016. Harvard University Press. All rights reserved.

order to glean shelf-level analytics”); Heather Fletcher, “Facial Recognition:

Ads Target Consumers for You,” Target Marketing (October 5, 2015),
http://www.targetmarketingmag.com/article/facial-recognition-ads-target
-consumers/ (“Transparency Market Research forecasts the global market
for facial recognition as reaching nearly $2.7 billion by 2022.” However, note:
“Seventy-five percent of consumers would not shop at a store using face
recognition technology for marketing purposes, a First Insight survey
found,” reads the September 9 post by Krystal Overmyer. “However,
55 percent said they would be open to the technology if they knew a benefit
was associated with it, such as discounts”).

43. Federal Trade Commission, Spring Privacy Series: Mobile Device Tracking,
(February 19, 2014), https://www.ftc.gov/news-events/events-calendar/2014
/02/spring-privacy-series-mobile-device-tracking.
44. In re Nomi Technologies, Inc., FTC No. 132 3251 (April 23, 2015), https://www
.ftc.gov/system/fi les/documents/cases/150423nomicmpt.pdf.
45. Ibid.
46. Federal Trade Commission, Statement of Chairwoman Ramirez, Commis-
sioner Brill, and Commissioner McSweeny in the Matter of Nomi Technolo-
gies, Inc. (April 23, 2015), https://www.ftc.gov/system/fi les/documents/public
_ statements/638351/150423nomicommissionstatement.pdf.
47. Federal Trade Commission, Data Brokers: A Call for Transparency and
Accountability (May 2014), ii–iii, https://www.ftc.gov/system/fi les/documents
/reports/data-brokers-call-transparency-accountability-report-federal-trade
-commission-may-2014/140527databrokerreport.pdf.
48. Ibid.
49. Kahneman, Thinking, Fast and Slow.
50. Richard Thaler, Misbehaving: The Making of Behavioral Economics (New
York: W. W. Norton, 2015), chap. 7.
51. G. B. Northcraft and M. A. Neale, “Experts, Amateurs, and Real Estate: An
Anchoring-and-Adjustment Perspective on Property Pricing Decisions,”
Organizational Behavior and Human Decision Processes 39 (1987): 84–97.
52. The packet included “a copy of the MLS summary of residential real estate
sales for both the entire city and the immediate neighborhood of the
property for the last 6 months; and information (including listing price,
square footage, characteristics of the property, etc.) about other property
located in the same neighborhood as the property being evaluated (this
information was divided into four categories: property currently for sale,
property recently sold, property sold but the sale not yet completed, and
property previously listed which did not sell); [and] standard MLS listing
information for other property in the immediate neighborhood currently for
sale.” Ibid.
Copyright © 2016. Harvard University Press. All rights reserved.

53. Ibid.
54. Ibid.
55. David Streitfeld, “Some Online Bargains May Only Look Like One,” New
York Times, April 13, 2016, http://www.nytimes.com/2016/04/14/technology
/some-online-bargains-may-only-look-like-one.html?smprod=nytcore
-iphone&smid=nytcore-iphone-share& _r = 0.
56. Ibid.
57. In one experiment, MBA students put down the last two digits of their social
security number (e.g., 14). Dan Ariely, Predictably Irrational: The Hidden

Forces That Shape Our Decisions (New York: HarperCollins Publishers,

2008), 25–28. The students, then participants, monetized it (e.g., $14), and
then answered “Yes or No” for each bidded item if they would pay that
amount for the item. The students then stated the maximum amount they
were willing to pay for each auctioned product. Students with the highest
ending SSN (80–99) bid 216 to 346 percent higher than students with
low-end SSNs (1–20), who bid the lowest. See also Daniel Kahneman,
Thinking, Fast and Slow (New York: Farrar, Straus and Giroux, 2011),
119–128, which discusses anchoring effects generally.
58. European Data Protection Supervisor, Mobile Health: Reconciling Techno-
logical Innovation with Data Protection, Opinion 1/2015 (May 21, 2015), para.
29, https://secure.edps.europa.eu/EDPSWEB/webdav/site/mySite/shared
/Documents/Consultation/Opinions/2015/15-05-21_ Mhealth_EN.pdf.

11 • The Rise of “Almost Perfect” Behavioral Discrimination

1. One business article, for example, outlined how all buying decisions “stem
from the interplay of the following six emotions”:
1. Greed. “If I make a decision now, I will be rewarded.”
2. Fear. “If I don’t make a decision now, I’m toast.”
3. Altruism. “If I make a decision now, I will help others.”
4. Envy. “If I don’t make a decision now, my competition will win.”
5. Pride. “If I make a decision now, I will look smart.”
6. Shame. “If I don’t make a decision now, I will look stupid.”
Geoff rey James, “6 Emotions that Make Customers Buy: Customers Make
Decisions at the Gut Level. Here’s How to Use the Customer’s Emotions to
Your Advantage,” Inc.com (February 8, 2012), http://www.inc.com/geoffrey
-james/6 -emotions-that-make-customers-buy.html.
2. Executive Office of the President, Big Data and Differential Pricing (Wash-
ington, DC: Executive Office of the President, February 2015), 8, https://www
.whitehouse.gov/sites/default/fi les/whitehouse _ fi les/docs/Big _Data _ Report
Copyright © 2016. Harvard University Press. All rights reserved.

_ Nonembargo_v2.pdf.
3. Competition and Markets Authority, The Commercial Use of Consumer
Data: Report on the CMA’s Call for Information, CMA38 (June 2015), 2.86,
https://www.gov.uk /government/uploads/system/uploads/attachment _data
/fi le/435817/The _commercial _use _of_consumer_data.pdf. 2.86.
4. Ibid.
5. Ibid.
6. Ibid., 2.75.
7. Ibid., 2.85.

8. Apple, About Advertising & Privacy (September 14, 2015), https://support

.apple.com/en-us/HT205223.
9. Ibid.
10. Ibid.
11. Ibid. Apple states that its iAd “does not know or make available to adver-
tisers information regarding your sexual orientation, religious beliefs or
political affi liations.” No Apple Pay transactions or Health app data is
accessible to iAd, or is used for advertising purposes. iAd does not sell or
other wise transmit any personally identifiable information to third parties.
12. Ibid. A user, in 2015, to enable Limit Ad Tracking on iOS, would have to
open Settings, then tap on “Privacy,” then “Advertising,” and then slide the
Limit Ad Tracking switch to “on.” To enable Limit Ad Tracking in iTunes,
the user would have to open Preferences in iTunes, then click the “Store”
pane, and then check “Limit Ad Tracking.” To enable Limit Ad Tracking on
Apple TV, open Settings, then select “iTunes Store,” select “Ad Tracking,”
and click “Limit Ad Tracking.”
13. Federal Trade Commission, Data Brokers: A Call for Transparency and
Accountability (Washington, DC: Federal Trade Commission, May 2014),
19–20, https://www.ftc.gov/system/fi les/documents/reports/data-brokers-call
-transparency-accountability-report-federal-trade-commission-may-2014
/140527databrokerreport.pdf.
14. Lifehack Quotes, http://quotes.lifehack.org/edward-norton/we-buy-things
-we-dont-need-with/.
15. Karen Freeman, “Amos Tversky, Expert on Decision Making, Is Dead at 59,”
New York Times, June 6, 1996, http://www.nytimes.com/1996/06/06/us/amos
-tversky-expert-on-decision-making-is-dead-at-59.html.
16. Ned Welch, “A Marketer’s Guide to Behavioral Economics,” McKinsey
Quarterly, February 2010, http://www.mckinsey.com/insights/marketing
_ sales/a _marketers _ guide _to_behavioral _economics.
17. Robert B. Cialdini, Influence: The Psychology of Persuasion (New York:
HarperBusiness, 2007).
Copyright © 2016. Harvard University Press. All rights reserved.

18. Dan Ariely, Predictably Irrational: The Hidden Forces that Shape Our
Decisions (New York: HarperCollins, 2009), 2.
19. Welch, “A Marketer’s Guide to Behavioral Economics”; Sheryl E. Kimes,
Robert Phillips, and Lisabet Summa, “Pricing in Restaurants,” in The Oxford
Handbook of Pricing Management, A. Özer and Robert Phillips, eds.
(Oxford: Oxford University Press, 2012), 106 (noting how the phenomenon
may be related to the so-called “compromise effect” or “context effect”).
20. Ibid.
21. Ibid.
22. Ibid.

23. Kyle James, “Beware of These Pricing Tricks Retailers Use to Fool Your
Brain,” Two Cents (May 22, 2015), http://twocents.lifehacker.com/beware-of
-these-pricing-tricks-retailers-use-to-fool-yo-1706225322.
24. Anshu Jalora, “Applying Consumer Psychology to Soft ware Pricing,” in
Innovation in Pricing: Contemporary Theories and Best Practices, An-
dreas Hinterhuber and Stephan Liozu, eds. (London: Routledge, 2013),
396–399.
25. Ibid., 397.
26. Executive Office of the President, Big Data and Differential Pricing, 11.
27. Aniko Hannak, Gary Soeller, David Lazer, Alan Mislove, and Christo
Wilson, “Measuring Price Discrimination and Steering on E-Commerce
Web Sites,” Proceedings of the 2014 Conference on Internet Measurement
Conference, New York, 305–318, http://www.ccs.neu.edu/home/cbw/pdf
/imc151-hannak.pdf.
28. Competition and Markets Authority, The Commercial Use of Consumer
Data, 93–94.
29. Hannak et al., “Measuring Price Discrimination and Steering on E-Commerce
Web Sites.”
30. Ibid., sections 4.2–4.5.
31. Ibid., section 5.2.
32. Ibid.
33. Ibid.
34. Ibid.
35. See, for example, Xavier Gabaix and David Laibson, “Shrouded Attributes,
Consumer Myopia, and Information Suppression in Competitive Markets,”
Quarterly Journal of Economics 121 (2006): 505–508; Oren Bar-Gill and
Elizabeth Warren, “Making Credit Safer,” University of Pennsylvania Law
Review 157, no. 1 (2008): 27–28; Simon Johnson and James Kwak, 13 Bankers:
The Wall Street Takeover and the Next Financial Meltdown (New York:
Pantheon, 2010), 81, 108.
36. Adi Ayal, “Harmful Freedom of Choice: Lessons from the Cellphone
Copyright © 2016. Harvard University Press. All rights reserved.

Market,” Law and Contemporary Problems 74 (2011): 91, 118; “Contractual

complexity thus acts to raise switching costs, which allows for raising prices
to existing customers while hiding the existence of discrimination among
customers paying different prices for similar consumption.”
37. Eugenio J. Miravete, “The Doubtful Profitability of Foggy Pricing 2–3,” NET
Institute Working Paper No. 04-07 (2004), http:/ssrn.com/abstract= 618465.
38. Ayal, “Harmful Freedom of Choice,” 124.
39. Ellen Peters et al., “More Is Not Always Better: Intuitions about Public Policy
Can Lead to Unintended Health Consequences,” Social Issues & Policy
Review 7, no. 1 (1996): 114, 122.

40. Chris M. Wilson and Catherine Waddams Price, “Do Consumers Switch to
the Best Supplier?” Oxford Economic Papers 62 (2011): 98–131; see also Adi
Ayal, “Harmful Freedom of Choice,” 121. For discussion of the anticompeti-
tive implications of product differentiation in U.S. health care markets, see
William M. Sage and Peter J. Hammer, “Competing on Quality of Care: The
Need to Develop a Competition Policy for Health Care Markets,” University
of Michigan Journal of Law Reform 32 (1999): 1073, 1082.
41. Competition and Markets Authority, Energy Market Investigation: Summary
of Provisional Findings Report (July 7, 2015), para. 123, https://www.gov.uk
/government/uploads/system/uploads/attachment _data/fi le/442500/EMI
_PFs _ Summary.pdf.
42. Ibid. More generally, the CMA found barriers to engagement resulted from:
“lack of access to the internet (or a lack of confidence in using the internet).”
Ibid., para. 124. Also “customers on low income and with low levels of
education” were less likely to use price comparison websites. Ibid., para. 125.
43. Ibid., para. 126.
44. Stefania Sitzia, Jiwei Zheng, and Daniel John Zizzo, “Complexity and
Smart Nudges with Inattentive Consumers,” CCP Working Paper 12–13,
available online: http://competitionpolicy.ac.uk /documents/8158338
/8251737/CCP+Working+Paper+12-13.pdf/f9d4eff 9-daf7- 4244 -acd7
-240b8972bfd7.
45. Anthony Giorgianni, “Avoid These Mattress Store Tricks: Don’t Be Misled
into Paying More or Buying Stuff You Don’t Want,” Consumer Reports, May
8, 2014, http://www.consumerreports.org/cro/news/2014/05/avoid-mattress
-store-tricks/index.htm.
46. U.K. Office of Fair Trading, The Impact of Price Frames on Consumer
Decision Making, OFT1226 (May 2010), 1.5, http://webarchive.national
archives.gov.uk/20140402142426/, http://www.oft.gov.uk /shared_oft/economic
_research/OFT1226.pdf.
47. Executive Office of the President, Big Data and Differential Pricing, 6.
48. Interstate Circuit v. United States, 306 U.S. 208 (1939).
Copyright © 2016. Harvard University Press. All rights reserved.

49. Ismat Sarah Mangla, “3 Tricks to Help You Snag the Best Deals Online,”
Time, September 8, 2014, http://time.com/money/3136612/dynamic-pricing
-amazon-best-buy-walmart/.
50. Daniel Kahneman and Amos Tversky, “Prospect Theory: An Analysis of
Decision under Risk,” Econometrica 47 (1979): 263; U.K. Office of Fair
Trading, Consumer Behavioural Biases in Competition: A Survey, Final
Report, OFT1324 (May 2011), 3.10–3.201.11.
51. E. Vis and J. Toth, “ The Abolition of the No-Discrimination Rule,” (Am-
sterdam: ITM Research, March 2000), 7–10, http://www.creditslips.org/fi les
/netherlands-no-discrimination-rule-study.pdf.

52. Daniel Kahneman, Jack L. Knetsch, and Richard H. Thaler, “Fairness as a

Constraint on Profit Seeking: Entitlements in the Market,” in Advances in
Behavioral Economics, Colin F. Camerer, George Loewenstein, and Matthew
Rabin, eds. (Princeton, NJ: Princeton University Press, December 28, 2003),
252, 257.
53. Organisation for Economic Co-operation and Development, Competition
and Regulation in Agriculture: Monopsony Buying and Joint Selling, DAF/
COMP(2005)44 (December 21, 2005), 8, http://www.oecd.org/competition
/abuse/35910977.pdf.
54. Colin F. Camerer, “Prospect Theory in the Wild: Evidence from the Field,”
in Advances in Behavioral Economics, Colin F. Camerer, George Loewen-
stein, and Matthew Rabin, eds. (Princeton, NJ: Princeton University Press,
December 28, 2003): 148, 152 (many consumers “dislike price increases more
than they like the windfall gain from price cuts and will cut back purchases
more when prices rise compared with the extra amount they buy when prices
fall”); Daniel Kahneman, “Maps of Bounded Rationality: Psychology for
Behavioral Economics,” American Economic Review 93 (December 2003):
1449, 1458.
55. James Surowiecki, “In Praise of Efficient Gouging,” MIT Technology Review
(August 19, 2004), http://www.technologyreview.com/review/529961/in
-praise-of-efficient-price-gouging/.
56. Frank Pasquale, The Black Box Society: The Secret Algorithms that Control
Money and Information (Cambridge, MA: Harvard University Press, 2015).
57. Alex Chisholm (CMA chief executive), Why “Sleepers” Can’t Always Be Left
to “Sleep,” CCRP 2016 Competition Policy Roundtable (London: Competi-
tion Markets Authority, January 25, 2016), https://www.gov.uk /government
/speeches/alex-chisholm-on-consumer-engagement-in-a-digital-world;
quoting In re Text Messaging Antitrust Litig., 782 F.3d 867, 874 (7th Cir.) cert.
denied sub nom and Aircraft Check Servs. Co. v. Verizon Wireless, 136 S. Ct.
524 (2015).
58. Paragraph 79, House of Lords, Select Committee on European Union,
Copyright © 2016. Harvard University Press. All rights reserved.

“Online Platforms and the Digital Single Market,” April 20, 2016, 10th Report
of Session 2015–16, http://www.publications.parliament.uk /pa/ld201516
/ldselect/ldeucom/129/129.pdf.
59. In the next part we consider how price comparison websites may assist
limiting the possibility for discrimination.
60. Executive Office of the President, Big Data and Differential Pricing.
61. In the EU, the Privacy and Electronic Communications Regulations “deal with
the collection of location and traffic data by public electronic communications
ser vices providers (‘CSPs’) and use of cookies (and similar technologies).” For
example, “Traffic Data held by a CSP must be erased or anonymised when it is

no longer necessary for the purpose of the transmission of a communication,”

unless (1) “it is being used to provide a value added ser vice,” and (2) “consent
has been given for the retention of the Traffic Data.” Moreover, “the use and
storage of cookies and similar technologies requires” (1) “clear and compre-
hensive information,” and (2) “consent of the website user.” DLA Piper, Data
Protection Laws of the World, http://dlapiperdataprotection.com/#handbook
/online-privacy-section/c2 _GB.
62. “At first blush, detecting personalization on e-commerce sites seems
conceptually simple: have two users run the same search, and any differences
in the results indicate personalization. Unfortunately, this approach is likely
to have many false positives, as differences between users’ results may exist
for a number of reasons not related to personalization. For example, results
may differ due to changes in product inventory, regional tax differences, or
inconsistencies across data centers. As a result, accurately detecting person-
alization on e-commerce sites remains an open challenge. Second, even
when customers are aware, it is often the case that markets exhibit weak
consumer engagement. Only a minority will likely invest time in countermea-
sures to curtail tracking. To benefit from discounts through loyalty pro-
grams, they must reveal their identity. Moreover, the loyalty discounts are
more salient than the savings in remaining anonymous. An asymmetry in
power is observed between us and those who hold and sell our personal
data.” Hannak et al., “Measuring Price Discrimination and Steering on
E-Commerce Web Sites.”
63. Chisholm, Why “Sleepers” Can’t Always Be Left to “Sleep.”
64. “David Currie speaks about the CMA experience of behavioural economics,”
April 20, 2015. Available on the CMA website: https://www.gov.uk
/government/speeches/david-currie-speaks-about-the-cma-experience-of
-behavioural-economics.

12 • Behavioral Discrimination: Economic and Social Perspectives

1. Michael Eisen, “Amazon’s $23,698,655.93 Book about Flies,” It Is NOT Junk

(April 22, 2011), http://www.michaeleisen.org/blog/?p =358.
2. P. T. Leeson and R. Sobel, “Costly Price Discrimination,” Economics Letters
99, no. 1 (2008): 206–208, http://www.peterleeson.com/Costly_Price
_Discrimination.pdf; “We show this occurs [perfect price discrimination is
often socially inefficient] because firms face costs of enacting price discrimi-
nation. These costs, which include segmenting consumers, identifying
elasticities, and preventing resale, are significant in all industries. This, of
course, is the reason not all firms enact this pricing strategy. The omission of

these transactions costs from existing theories of price discrimination is

impor tant because, as Varian has pointed out: ‘A full welfare analysis of
attempts to engage in [perfect] price discrimination cannot neglect the
transactions costs involved in the negotiation itself.’ ”
3. P. Papandropoulos, “How Should Price Discrimination Be Dealt with by
Competition Authorities?” Revue des droits de la concurrence 3 (2007):
34–38, http://ec.europa.eu/dgs/competition/economist/concurrences _03
_ 2007.pdf.
4. Note, for instance, a study of air travel fares: “The results are consistent with
the hypothesis that, as more carriers operate on a given route, the carriers’
competition for consumers with higher price elasticity of demand increases,
while fares charged to consumers with inelastic demand stay high”; J. Stavins,
“Price Discrimination in the Airline Markets: The Effect of Market Concen-
tration,” Review of Economics and Statistics 83 (2001): 200. In line with this
rationale, the competition agencies in a merger review may seek to maintain
price discrimination when it protects buyers who cannot pay a higher,
uniform price. Alternatively, the U.S. competition authorities give an
example where a powerful buyer benefits from price discrimination: the
buyer can “negotiate lower pre-merger prices than other customers by
threatening to shift its large volume of purchases from one merging firm to
the other. No other suppliers are as well placed to meet Customer C’s needs
for volume and reliability. The merger is likely to harm Customer C. In this
situation, the Agencies could identify a price discrimination market con-
sisting of Customer C and similarly placed customers. The merger threatens
to end previous price discrimination in their favor”; Federal Trade Commis-
sion and U.S. Department of Justice, Horizontal Merger Guidelines (August
19, 2010), para. 8.
5. See, for example, pharmaceutical markets: “In low-income countries the vast
majority are unwilling to pay for effective drugs simply because they are
unable to pay. Low-income nations need more price discrimination—and
vastly lower prices—if they are ever to afford the world’s most effective
Copyright © 2016. Harvard University Press. All rights reserved.

medicines”; Judith L. Wagner and Elizabeth McCarthy, “International

Differences in Drug Prices,” Annual Review of Public Health 25 (2004): 475.
6. See, for example, the pharmaceutical markets. A study by Jerry Hausman
and Jeff rey K. MacKie-Mason indicated price discrimination increased
dynamic welfare because of the positive effects on research and development;
Jerry A. Hausman and Jeff rey McKie-Mason, “Price Discrimination and
Patent Policy,” RAND Journal of Economics 19 (1988): 253. Firms will also
have greater incentives to innovate when they can recover the significant
upfront costs by price discriminating: “if dynamic incentives are taken into

account as well, price discrimination may ensure that long-run incentives to

invest (e.g. in R&D) are preserved by providing fi rms with sufficient returns”;
Papandropoulos, “How Should Price Discrimination Be Dealt with by
Competition Authorities?” When the ability to innovate is present, the
incentive to innovate is driven by the desire to appropriate financial gains
associated with innovation. This economic framework is “often associated
with the eminent economist Professor Joseph Schumpeter,” who suggested,
“firms with market power will have the greatest incentive to innovate
because of their large relative size and dominant position in a market” D. L.
Weisman and R. B. Kulick, “Price Discrimination, Two-Sided Markets, and
Net Neutrality Regulation,” Tulane Journal of Technology and Intellectual
Property 13 (2010): 81, https://www.researchgate.net/profile/Dennis_Weisman
/publication/228307995_Price _Discrimination _Two-Sided _ Markets _ and
_Net _Neutrality_Regulation/links/0deec5187eadf2a5c8000000.pdf.
7. Josh Wright, “Price Discrimination Is Good, Part I,” Truth on the Market
(November 30, 2008), http://truthonthemarket.com/2008/11/30/price
-discrimination-is-good-part-i/. This approach, generally associated with
Nobel laureate economist Kenneth Arrow, suggests “the increased business
generated by an innovation will come mostly from sales that formerly would
have gone to competitors, while monopolists may largely cannibalize their
own business”; Weisman and Kulick, “Price Discrimination, Two-Sided
Markets, and Net Neutrality Regulation.”
8. “The reason why price discrimination may intensify competition is that
with uniform pricing, firms would only compete for ‘marginal consumers’
whereas through price discrimination, firms can compete for all customers,
including those with strong loyalty to a competitor’s brand”; Papandro-
poulos, “How Should Price Discrimination Be Dealt with by Competition
Authorities?”
9. For example, the European car-rental industry. In 2014 the European
Commission wrote to six international car-rental companies, including Avis,
Europcar, and Hertz, after cases in which the price of car rentals varied
Copyright © 2016. Harvard University Press. All rights reserved.

considerably based on the country of residence of the customer; “Car Hire

Prices ‘Unfairly Vary’ across the EU,” BBC News (August 12, 2014),
http://www.bbc.co.uk /news/business-28756674. The U.S. Department of
Justice had this concern in a 2003 merger between Quest Diagnostics, Inc.,
and Unilab Corp., the two leading providers of clinical laboratory testing
ser vices to physician groups in Northern California: “[P]urchasers of these
ser vices cannot economically resell them to other customers, and that
suppliers of the ser vices can potentially identify the competitive alternatives
available to physician group customers according to the group’s base of
physicians and geographic coverage. This information indicated that a

hy pothet ical monopolist could discriminate on price among customer

types. Suppliers’ ability to price discriminate, combined with the fact that
some types of customers had few competitive alternatives to contracting
with suppliers that had a network of locations, led staff to defi ne markets
based on customer categories”; Federal Trade Commission and U.S.
Department of Justice, Commentary on the Horizontal Merger Guidelines
(March 2006), 8, http://www.justice.gov/atr/commentary-horizontal-merger
-guidelines.
10. Commission Decision 85/609 ECS/AKZO, OJ L 374 (December 31, 1985),
para. 83.
11. Competition and Markets Authority, The Commercial Use of Consumer
Data, CMA38 (June 2015), https://www.gov.uk /government/uploads/system
/uploads/attachment _data/fi le/435817/The _commercial _use _of_consumer
_data.pdf, 3.48.
12. Papandropoulos, “How Should Price Discrimination Be Dealt with by
Competition Authorities?” “Price discrimination can also be an instrument
to implement predatory pricing. Indeed, it can reduce the costs of the
strategy and therefore, make it profitable to predate when it would not be
profitable if the dominant firm could not price discriminate. Assume a
dominant firm serves two market segments and there is entry on only one of
the segments. By selectively offering predatory prices to customers in the
segment facing entry, the costs of the strategy would be lower than if the
predatory price had to be charged across the board, to both segments. In
some instances, the ability to target price cuts could reduce the costs of
predation enough to be outweighed by the long-run benefits of impeding
entry. Hence, there may be situations in which, predation would not be
profitable if price discrimination was impossible. Yet, even in this case, it is
not price discrimination as such that may lead to anti-competitive effects but
the predatory strategy. In the context of an effects-based analysis of targeted
price cuts, it is the predatory nature of the price cuts that would cause
foreclosure effects.”
Copyright © 2016. Harvard University Press. All rights reserved.

13. Donald S. Clark (secretary of the Federal Trade Commission), The Robinson-
Patman Act: General Principles, Commission Proceedings, and Selected Issues
(San Jose, CA: Federal Trade Commission, June 7, 1995), https://www.ftc.gov
/public-statements/1995/06/robinson-patman-act-general-principles
-commission-proceedings-and-selected.
14. Office of Fair Trading, The Economics of Personalised Pricing, OFT1488
(May 2013), http://webarchive.nationalarchives.gov.uk /20140402142426/,
http://www.oft.gov.uk /shared _oft/research/oft1488.pdf; see also Office of
Fair Trading, Personalised Pricing: Increasing Transparency to Improve Trust,
OFT1489 (May 2013), http://webarchive.nationalarchives.gov.uk

/20140402142426/, http://www.oft.gov.uk /shared _oft/markets-work

/personalised-pricing/oft1489.pdf.
15. Competition and Markets Authority, The Commercial Use of Consumer
Data, 90 (citing a call for information by the Office of Fair Trading in 2012
on personalized pricing).
16. Laura Bach, “Tobacco Company Marketing to Kids,” Campaign for Tobacco-
Free Kids (June 4, 2015), https://www.tobaccofreekids.org/research/factsheets
/pdf/0008.pdf (discussing research of cigarette manufacturers’ promotional
activities that influenced previously nonsusceptible nonsmokers to become
susceptible to or experiment with smoking).
17. Ibid.
18. United States v. Brown University, 5 F.3d 658 (3d Cir. 1993).
19. Stacy Dale and Alan B. Krueger, “Estimating the Return to College Selec-
tivity over the Career Using Administrative Earning Data,” Princeton
University Working Paper 563 (February 16, 2011), http://i.bnet.com/blogs
/ivy-league-study.pdf (finding that, for black and Hispanic students and for
students who come from less-educated families in terms of their parents’
education, the estimates of the earnings return to college selectivity remain
large, even in models that adjust for unobserved student characteristics).
20. Derek Thompson, “Why Smart Poor Students Don’t Apply to Selective
Colleges (and How to Fix It),” The Atlantic, January 24, 2013, http://www
.theatlantic.com/business/archive/2013/01/why-smart-poor-students-dont
-apply-to-selective-colleges-and-how-to-fi x-it/272490/; Caroline M. Hoxby
and Christopher Avery, “The Missing ‘One-Offs’: The Hidden Supply of
High-Achieving, Low Income Students,” NBER Working Paper 18586
(December 2012), http://www.nber.org/papers/w18586 (showing that “the
vast majority of very high-achieving students who are low-income do not
apply to any selective college or university . . . despite the fact that selective
institutions would often cost them less, owing to generous financial aid, than
the resource-poor two-year and non-selective four-year institutions to which
they actually apply”).
Copyright © 2016. Harvard University Press. All rights reserved.

21. Maurice E. Stucke, “Is Intent Relevant?” Journal of Law, Economics & Policy
8 (2012): 801, 822–828 (collecting some of the literature); see also Lynn A.
Stout, Cultivating Conscience: How Good Laws Make Good People (Princeton,
NJ: Princeton University Press, 2011), 238–240 (discussing how societal
norms of fairness and prosocial behav ior are both common in, and neces-
sary for, a market economy); Thomas J. Horton, “Unraveling the Chicago/
Harvard Antitrust Double Helix: Applying Evolutionary Theory to Guard
Competitors and Revive Antitrust Jury Trials,” University of Baltimore Law
Review 41 (2012): 615, 653–654 (citing research on how “ ‘fairness evolved as
a stable strategy for maintaining social harmony’ in our economic relation-

ships” and how “[n]eurobiological studies have found that ‘the sense of
fairness fundamental to distributive justice’ is rooted in humans’ emotional
processing”), quoting Joan Roughgarden, The Genial Gene: Deconstructing
Darwinian Selfishness (Berkeley: University of California Press, 2009), 160;
Michael Shermer, The Mind of the Market: Compassionate Apes, Competitive
Humans, and Other Tales from Evolutionary Economics (New York: Times
Books, 2008), 11.
22. Daniel Kahneman, Jack L. Knetsch, and Richard H. Thaler, “Fairness as a
Constraint on Profit Seeking: Entitlements in the Market,” American
Economic Review 76, no. 4 (1986): 728, 735.
23. Ellen Garbarino and Sarah Maxwell, “Consumer Response to Norm-
Breaking Pricing Events in E-Commerce,” Journal of Business Research
63 (2010): 1066, 1069.
24. Lan Xia and Kent B. Monroe, “Is a Good Deal Always Fair? Examining the
Concepts of Transaction Value and Price Fairness,” Journal of Economic
Psychology 31 (2010): 884, 891.
25. Joseph Turow, Lauren Feldman, and Kimberly Meltzer, “Open to Exploita-
tion: American Shoppers Online and Offline,” Report from the Annenberg
Public Policy Center of the University of Pennsylvania (June 1, 2005), 4,
http://repository.upenn.edu/cgi/viewcontent.cgi?article =1035&context= asc
_papers.
26. Ibid.
27. Executive Office of the President, Big Data and Differential Pricing (Wash-
ington, DC: Executive Office of the President, February 2015), 16, https://www
.whitehouse.gov/sites/default/fi les/whitehouse _ fi les/docs/Big _Data _ Report
_ Nonembargo_v2.pdf.
28. Federal Trade Commission, Data Brokers: A Call for Transparency and
Accountability, (May 2014), 20, https://www.ftc.gov/system/fi les/documents
/reports/data-brokers-call-transparency-accountability-report-federal-trade
-commission-may-2014/140527databrokerreport.pdf.
29. Ibid.
Copyright © 2016. Harvard University Press. All rights reserved.

30. Ibid.
31. Ibid., v.
32. Ibid.
33. Ibid., 56; Article 21, Charter of Fundamental Rights of the European Union;
Directive 2006/54/EC of the European Parliament and of the Council of
July 5, 2006, on the implementation of the principle of equal opportunities
and equal treatment of men and women in matters of employment and
occupation.
34. Julia Angwin, Surya Mattu, and Jeff Larson, “The Tiger Mom Tax: Asians
Are Nearly Twice as Likely to Get a Higher Price from Princeton Review,”

ProPublica (September 1, 2015), https://www.propublica.org/article/asians

-nearly-twice-as-likely-to-get-higher-price-from-princeton-review.
35. Ibid.
36. Amit Datta, Michael Carl Tschantz, and Anupam Datta, “Automated Experi-
ments on Ad Privacy Settings: A Tale of Opacity, Choice, and Discrimina-
tion,” Proceedings on Privacy Enhancing Technologies 2015, no. 1 (2015):
92–112.
37. Ibid., 92.
38. Ibid.
39. Ibid., 93.
40. Ibid.
41. Ibid.
42. Ibid., 105.
43. University of Washington, “Who’s a CEO? Google Image Results Can Shift
Gender Biases,” ScienceDaily (April 9, 2015), www.sciencedaily.com/releases
/2015/04/150409143143.htm.
44. Latanya Sweeney, Online Ads Roll the Dice, Federal Trade Commission
(September 25, 2014), https://www.ftc.gov/news-events/blogs/techftc/2014/09
/online-ads-roll-dice. The FTC official began by searching “best credit cards”
and “worst credit cards” on the Google search engine and collected the first
twenty-five credit cards for each category, as mentioned in these third-party
sources. “Credit card ads that appeared on omegapsiphi2011.com were offers
for criticized cards or were generic. No ads naming any of the praised cards
appeared.”
45. Frank Pasquale, The Black Box Society: The Secret Algorithms that Control
Money and Information (Cambridge, MA: Harvard University Press, 2015),
38–42.
46. Aniko Hannak, Gary Soeller, David Lazer, Alan Mislove, and Christo
Wilson, “Measuring Price Discrimination and Steering on E-Commerce
Web Sites,” Proceedings of the 2014 Conference on Internet Measurement
Conference, New York, 5, http://www.ccs.neu.edu/home/cbw/pdf/imc151
Copyright © 2016. Harvard University Press. All rights reserved.

-hannak.pdf [noting that they could not “determine why results are being
personalized based on the data from real-world users, since there are too
many confounding variables attached to each . . . user (e.g., their location,
choice of browser, purchase history, etc.)”].
47. Li Xi v. Apple Inc., 603 F. Supp. 2d 464 (E.D.N.Y. 2009) (dismissing end-users’
Robinson-Patman Act claims when the plaintiffs never alleged that they were
“competitors engaged in the business of reselling iPhones, that they are in
actual competition with a favored purchaser, or that they even resold or
attempted to resell their iPhones”); Matthew A. Edwards, “Price and Prejudice:
The Case against Consumer Equality in the Information Age,” Lewis & Clark

Law Review 10 (2006): 559, 580; “Even under the most liberal current
interpretation of the RPA, consumer price discrimination claims fail because
end-use buyers are not in competition with other buyers who are receiving
preferential pricing treatment. Thus, the RPA does not require retailers to
treat these consumers equally” (footnotes omitted).
48. The Act prohibits a seller from discriminating in price between two or more
competing buyers in the sale of commodities of like grade and quality, where
the effect of the discrimination “may be substantially” to “lessen competi-
tion . . . in any line of commerce;” or “tend to create a monopoly in any line
of commerce;” or “injure, destroy, or prevent competition with any person
who grants or knowingly receives the benefit of the discrimination, or with
the customers of either of them”; Federal Trade Commission, The Robinson-
Patman Act: General Principles, Commission Proceedings, and Selected Issues
(June 7, 1995), https://www.ftc.gov/public-statements/1995/06/robinson
-patman-act-general-principles-commission-proceedings-and-selected.
49. Ross E. Elfand, “The Robinson-Patman Act,” American Bar Association
(n.d.), http://www.americanbar.org/groups/young _ lawyers/publications/the
_101_ 201_practice _ series/robinson _patman_act.html.
50. Ibid., citing Feesers, Inc. v. Michael Foods, Inc., 591 F. 3d 191, 198 (3d Cir.
2010) (discussing recent Supreme Court jurisprudence); Ryan Luchs, Tansev
Geylani, Anthony Dukes, and Kannan Srinivasan, “The End of the Robinson-
Patman Act? Evidence from Legal Case Data,” Management Science 56, no.
12 (2010): 2123–2133, http://www-bcf.usc.edu/~dukes/Papers/Dukes13
_ EndofRP_ MgtSci _10.pdf. The decision in Brooke Group Ltd. v. Brown &
Williamson Tobacco Corp., 509 U.S. 209 (1993), raised the standard for
“primary line” cases where “one manufacturer reduces its prices in a specific
geographic market and causes injury to its competitors in the same market,”
requiring evidence that the larger company intended to raise prices for
consumers at a later date to recoup money lost on discounts, shift ing the
focus to the consumer rather the competitor. Similarly, the decision in Volvo
Trucks North America, Inc. v. Reeder-Simco GMC, Inc., 546 U.S. 164 (2006),
Copyright © 2016. Harvard University Press. All rights reserved.

raised the standard for “secondary line” cases, “which occur when favored
customers of a supplier are given a price advantage over competing cus-
tomers,” requiring that the supplier show that the different pricing policies
made it harder to compete for the same customers at the same time; Federal
Trade Commission, Price Discrimination: Robinson-Patman Violations
(n.d.), https://www.ftc.gov/tips-advice/competition-guidance/guide-antitrust
-laws/price-discrimination-robinson-patman; see also Robert J. Toth, “A
Powerful Law Has Been Losing a Lot of Its Punch,” Wall Street Journal, May 1,
2002, http://www.wsj.com/articles/SB100014240527023047466045773801727
54953842.

51. Edwards, “Price and Prejudice,” 596, quoting Howard J. Alperin and
Roland F. Chase, Consumer Law: Sales Practices and Credit Regulation, 2004
Supplement (St. Paul, MN: West Pub. Co., 2004) 39–40.
52. C-209/10, Post Danmark A/S v Konkurrencerådet.
53. Ibid., para. 30.
54. Competition and Markets Authority, The Commercial Use of Consumer
Data, 2.120.
55. Traders are not prevented from charging a different price in a later invitation
to purchase, so long as the total price is clearly displayed before completion.
However, as a result of “dubious commercial behaviour,” the European
Commission is “organising workshops to further assess under which circum-
stances the sudden increase of prices by a trader may become illegitimate
from a consumer protection angle.” Such circumstances include where
“consumers make repeated searches from a device with the same IP address,
leading to the website to offer a higher price on each successive search.” Mac
Macmillan, “European MEP Calls for Investigation of Online Price Discrim-
ination,” Hogan Lovells Chronicle of Data Protection, September 13, 2013,
http://www.hldataprotection.com/2013/09/articles/consumer-privacy
/european-mep-calls-for-investigation-of-online-price-discrimination/.
Rafaele Rivais, “Why the Prices of Trains and Planes Vary from One Minute
to the Next (Continued),” SOS Conso Blog, Le Monde (January 24, 2013),
translation.
56. Directive 95/46/EC of the European Parliament and of the Council of
October 24, 1995, on the protection of individuals with regard to the
processing of personal data and on the free movement of such data. A
consumer’s IP address constitutes personal data, and as such a member of
the European Parliament has stated that “clients . . . should be informed
about the processing [of the IP address information].” European Parliament,
Parliamentary Questions, P-001257/13, E-001574/13, E-000956/13, April 18,
2013, http://www.europarl.europa.eu/sides/getAllAnswers.do?reference =E
-2013-000956&language =EN.
Copyright © 2016. Harvard University Press. All rights reserved.

57. See Dana Mattioli, “On Orbitz, Mac Users Steered to Pricier Hotels,” Wall
Street Journal, August 23, 2012, http://www.wsj.com/articles/SB10001424052
702304458604577488822667325882.
58. Pasquale, The Black Box Society, 32, discusses how runaway data can lead to
cascading disadvantages; Katherine Noyes, “The EU Will Examine Banks’
Use of Customer Data for Profi ling and Marketing Campaigns,” PCWorld,
October 5, 2015, pcworld.com/article/2989048/privacy/banks-use-of-big
-data-to-be-scrutinzed-by-eu-regulators.html.
59. See, for instance, Sydney Ember and Rachel Abrams, “On Instagram and
Other Social Media, Redefining ‘User Engagement,’ ” New York Times,

September 20, 2015, http://www.nytimes.com/2015/09/21/business/media

/retailers-use-of-their-fans-photos-draws-scrutiny.html?smprod=nytcore
-iphone&smid=nytcore-iphone-share; Vindu Goel, “Flipping the Switches on
Facebook’s Privacy Controls,” New York Times, January 29, 2014, http://www
.nytimes.com/2014/01/30/technology/personaltech/on-facebook-deciding
-who-knows-youre-a-dog.html?_r = 0.
60. Mattioli, “On Orbitz, Mac Users Steered to Pricier Hotels”(“Orbitz World-
wide Inc. has found that people who use Apple Inc.’s Mac computers spend
as much as 30% more a night on hotels, so the online travel agency is starting
to show them different, and sometimes costlier, travel options than Windows
visitors see.”). Article referred to in Frederik Zuiderveen Borgesius, “Online
Price Discrimination and Data Protection Law,” Amsterdam Law School
Research Paper No. 2015-32 (August 28, 2015), http://ssrn.com/abstract
=2652665.

13 • The Comparison Intermediaries

1. J. Yannis Bakos, “Reducing Buyer Search Costs: Implications for Electronic

Market Places,” Management Science 43, no. 12 (1997): 1, 5.
2. Ibid., 13; Joseph E. Stiglitz, “Imperfect Information in the Product Market,”
in Handbook of Industrial Organisation, vol. 1, R. Schmalensee and R. D.
Willig, eds. (Dordrecht: Elsevier, 1989), 769; J. Yannis Bakos, “A Strategic
Analysis of Economic marketplaces,” MIS Quarterly 295 (1991).
3. OFT Statement of Objections, para. 1.14–1.15, as cited in the Competition
Appeal Tribunal judgment, Skyscanner Limited v. Competition and Markets
Authority, Case No. 1226/2/12/14 (September 26, 2014), [2014] CAT 16,
31–32.
4. Ariel Ezrachi, “The Competitive Effects of Parity Clauses on Online Com-
merce,” European Competition Journal 11, no. 2–3 (2015).
5. The following section is based on our submission to the House of
Lords; A. Ezrachi and Maurice E. Stucke, “Online Platforms and the EU
Copyright © 2016. Harvard University Press. All rights reserved.

Digital Single Market,” University of Tennessee Legal Studies Research

Paper No. 283 (October 16, 2015), http://ssrn.com /abstract =2677267;
see also Maurice E. Stucke and Ariel Ezrachi, “When Competition Fails
to Optimize Quality: A Look at Search Engines,” Yale Journal of Law and
Technology 18 (2016), http://papers . ssrn.com /sol3/papers .cfm?abstract _ id
=2598128.
6. Marina Lao, “Networks, Access, and ‘Essential Facilities’: From Terminal
Railroad to Microsoft ,” Southern Methodist University Law Review 62
(2009): 557, 560–561; “The defining characteristic of network industries
is the increasing value of their products to users as the number of users

increases, a phenomenon called ‘network effects’ or demand-side economies

of scale.” The increased value can come directly (having more interconnec-
tions as a result of more users (e.g., telephones) or indirectly (having more
supporting complements developed for that product as the number of users
increases (e.g., Windows operating system). See also United States v. Micro-
soft Corp., 84 F. Supp. 2d 9, 20 (D.D.C. 1999) (discussing the “positive
network effect” of Windows); Case T-201/04, Microsoft Corp. v. Commission,
2007 E.C.R. 11-3601 (discussing the indirect network effects of streaming
media players).
7. In January 2016, an Oracle lawyer reportedly told the court in Oakland,
California, that one Google executive testified in a sworn deposition that the
annual sum Google pays Apple to keep its search box as the default on the
iPhone has topped $1 billion; see Mike Swift, “Oracle Pushes Judge to Force
Google to Disclose Terms of Deals with Apple, Other Mobile Competitors,”
mLex Market Insight (January 14, 2016), http://mlexmarketinsight.com
/category_editors/digital-risk /.
8. Organisation for Economic Co-operation and Development, Data-Driven
Innovation for Growth and Well-Being, Interim Synthesis Report (October
2015), 29.
9. See, for example, Ioannis Lianos and Evgenia Motchenkova, “Market
Dominance and Search Quality in the Search Engine Market,” Journal of
Competition Law & Economics 9 (2013): 419, 422 (advance access publication,
April 17, 2013), discussing how search engines “act as ‘information gate-
keepers’: they not only provide information on what can be found on the web
(equivalent to yellow pages), but they also are ‘an essential first-point-of-call
for anyone venturing onto the Internet’ ” and how they differ from other
two-sided platforms, as “search engines detain an impor tant amount of
information about their customers and advertisers (the ‘map of commerce’).”
10. Scott McCartney, “How Booking Sites Influence Which Hotels You Pick,”
Wall Street Journal, January 27, 2016, http://www.wsj.com/articles/how
-booking-sites-influence-which-hotels-you-pick-1453921300.
Copyright © 2016. Harvard University Press. All rights reserved.

11. Ibid.
12. Competitive Enter. Inst. v. U.S. Dep’t of Transp., 856 F.2d 1563, 1564–65 (D.C.
Cir. 1988).
13. Ibid.
14. Doreen Carvajal, “Amazon.com Plans to Revise Its Ad Program,” New York
Times, February 10, 1999, http://www.nytimes.com/1999/02/10/business
/amazoncom-plans-to-revise-its-ad-program.html.
15. Similar allegations were also raised with respect to the way in which
leading PCWs display the results on their default pages. Payment for
placement by leading platforms was criticized as a business model that
misleads consumers, as users are not always aware of whether the display

reflects organic or sponsored results; see James Maguire, “Case Study:

BizRate,” ECommerce- Guide (January 2, 2003), http://www.ecommerce
-guide.com /news/news/article.php/1563011/Case-Study-BizRate.htm;
“How Shopping Bots Really Work,” Loginworks (December 18, 2014),
http://www.loginworks.com /informative/shopping-bots-really-work /;
Leslie Walker, “What Shopping Guides Don’t Advertise,” Washington Post,
November 6, 2003, http://www.washingtonpost.com /archive/ business/2003
/11/06/what-shopping-guides-dont-advertise/9c387768-f228- 4356 -850e
-78a2afc60b47/.
16. Study conducted by the German Verbraucherzentrale. For the full study and
related materials, see: http://www.marktwaechter.de/pressemeldung
/buchungs-und-vergleichsportale-bieten-zu-wenig-nutzen-fuer-verbraucher;
http://www.verbraucherzentrale.de/vergleichsportale#header.
17. Rachel Rickard Straus, “Price Comparison Website Bosses under Attack
from MPs for Not Showing Customers the Best Deals,” This Is Money
(February 4, 2014), http://www.thisismoney.co.uk /money/bills/article
-2939364/Price-comparison-website-bosses-attack-MPs.html.
18. Ibid.
19. European Commission Case No. Comp/M. 5727, Microsoft/Yahoo! Search
Business Regulation (EC) No. 139/2004 Merger Procedure (February 18,
2010), para. 100.
20. Ibid. para. 35, 45.
21. See, for example, Andrea Amelio and Dimitrios Magos, “Economic Back-
ground of the Microsoft/Yahoo! Case,” Competition Policy Newsletter 2
(2010): 51. “For instance, instead of displaying links to additional merchants
in the organic search results, search engines could display links to ‘informa-
tional’ sites or placing the links winning the auctions also in prominent
positions in the organic search results, in order to decrease substitution
between organic and paid searches.” Federal Trade Commission, FTC Staff
Report, Google Inc., File No. 111-0161 (August 8, 2012), 92, released by the
Wall Street Journal, http://graphics.wsj.com/google-ftc-report/img/ftc-ocr
Copyright © 2016. Harvard University Press. All rights reserved.

-watermark.pdf, stating that “Google’s threat (and willingness) to degrade its

own web search product—by banishing high-quality vertical websites from
its web search results altogether—suggests that Google’s motive in scraping
high-quality content from its vertical competitors was not pro-competitive.”
A few caveats about this report, which the FTC released (mistakenly) under
the Freedom of Information Act to the Wall Street Journal. First, only the
report’s even-numbered pages were released, so the missing odd pages may
have contained impor tant qualifications. Second, other reports, including
any prepared by the FTC economists and Google, were not released. Third,
although the competition staff recommended that the FTC fi le a complaint,
the commissioners elected not to.

22. Users, the Commission argued, do not “necessarily see the most relevant
comparison shopping results in response to their queries, and . . .
incentives to innovate from rivals are lowered as they know that however
good their product, they will not benefit from the same prominence as
Google’s product.” Eu ropean Commission, Antitrust: Commission Sends
Statement of Objections to Google on Comparison Shopping Ser vice,
MEMO/15/4781 (April 15, 2015), http://europa .eu /rapid /press-release
_ MEMO -15 - 4781_ en. htm.
23. European Commission, Statement by Commissioner Vestager on Antitrust
Decisions Concerning Google, STATEMENT/15/4785 (April 15, 2015), http://
europa.eu/rapid/press-release _ STATEMENT-15-4785_en.htm.
24. Using these three conditions, we explored elsewhere how a dominant search
engine like Google could degrade quality (by providing less relevant responses
to a search inquiry), even with competition by Bing, Yahoo!, and DuckDuckGo.
Maurice E. Stucke and Ariel Ezrachi, “When Competition Fails to Optimize
Quality: A Look at Search Engines,” Yale Journal of Law and Technology 18
(2016), http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2598128.
25. Under the system Apple devised, publishers would have the freedom to set
e-book prices in the iBookstore and would keep 70 percent of each sale. The
remaining 30 percent would go to Apple as a commission. United States v.
Apple, Inc., 791 F.3d 290, 303 (2d Cir. 2015), cert. denied, 136 S. Ct. 1376 (2016).
Other distribution models may include a combination of both wholesale and
agency elements. For instance, one may design a hybrid wholesale model (such
as the merchant model) that includes a fixed markup. Under such a model the
upstream supplier determines the wholesale price, while the contract between
the parties includes an agreed margin for the retailer.
26. United States v. Apple, Inc., 791 F.3d 290, 304 (2d Cir. 2015), cert. denied,
136 S. Ct. 1376 (2016).
27. Ibid., 305.
28. A. Ezrachi, “The Competitive Effects of Parity Clauses on Online Com-
merce,” Oxford Legal Studies Research Paper No. 55/2015 (October 11, 2015),
Copyright © 2016. Harvard University Press. All rights reserved.

http://papers.ssrn.com/sol3/papers.cfm?abstract _id=2672541.
29. Benjamin Edelman and Julian Wright, “Price Coherence and Excessive
Intermediation,” Harvard Business School Working Paper No. 15-030
(March 2015), 3, http://www.benedelman.org/publications/pricecoherence
-2015-03-12.pdf.
30. Competition and Markets Authority, Private Motor Insurance Market
Investigation: Final Report (September 24, 2014), 8.42, 8.14, https://assets
.digital.cabinet-office.gov.uk /media/5421c2ade5274a1314000001/Final
_report.pdf.
31. Ibid., 8.13.

32. Douglas Fraser, “Heat Is On: Why Are Energy Bills so High?” BBC News
(January 15, 2014), http://www.bbc.co.uk /news/uk-scotland-scotland
-business-25743336.
33. Ibid.
34. According to the figures released, annual profits range from £69.7 million
(Compare The Market), to £53.9 million (MoneySuperMarket), to £19.8mil-
lion (GoCompare), to £16.4 and £9.8 million (Confused.com and uSwitch);
see Straus, “Price Comparison Website Bosses under Attack.”
35. Fraser, “Heat Is On.”
36. Ibid.
37. U.K. Parliament Energy and Climate Change Select Committee, Protecting
Consumers: Making Energy Price Comparison Websites Transparent
(March 26, 2015), http://www.publications.parliament.uk /pa/cm201415
/cmselect/cmenergy/1145/114502.htm.
38. In its submission to the committee, the U.K. Office of Gas and Electricity
Markets noted that the level of commission charged by price comparison
websites does not impact the actual price that the consumer pays when
switching. As noted in the government response, “the cost a consumer will
pay on a tariff is the same regardless of whether they switch through a
comparison site or directly through the supplier, it is difficult to see how
consumers would benefit this requirement.” Ibid., Appendix 1: Government
Response, Recommendation 3.
39. David Ronayne, “Price Comparison Websites,” Warwick Economic Research
Papers 1056 (October 2015), http://www2.warwick.ac.uk /fac/soc/economics
/research/workingpapers/2015/twerp_1056b_ronayne.pdf.
40. Ibid.
41. Ibid.
42. HRS-Hotel Reservation Ser vice B 9-66/10 (December 20, 2013),
http://www.bundeskartellamt .de/SharedDocs/Entscheidung /EN
/Entscheidungen /Kartellverbot /B9 - 66 -10.pdf; United States v. Apple
Inc., 952 F. Supp. 2d 638, 15 647 (S.D.N.Y. 2013); Case COMP/AT-39.847,
Copyright © 2016. Harvard University Press. All rights reserved.

E-Books (July 25, 2013).

43. For further discussion on the effects of narrow and wide MFNs, see Ariel
Ezrachi, “The Competitive Effects of Parity Clauses on Online Commerce.”

Part IV • Frenemies

1. On the role of network effects, note the Report by the House of Lords, Select
Committee on European Union “Online Platforms and the Digital Single
Market” (April 20, 2016) 10th Report of Session 2015–16, paras 65–93, http://www
.publications.parliament.uk/pa/ld201516/ldselect/ldeucom/129/129.pdf.

2. For instance, in the personal computer world two super-platforms are

noticeable: Microsoft’s Windows for operating system soft ware and Intel for
microprocessors.

14 • The Dynamic Interplay among Frenemies

1. Some firms have no preexisting relationship, as when a vulture fund acquires

a distressed newspaper.
2. Marius Schwartz and David Eisenstadt, “Vertical Restraints,” U.S. Depart-
ment of Justice Antitrust Division, EPO Discussion Paper 82-2 (1982), 4, 5;
Business Electronics Corp. v. Sharp Electronics Corp., 485 U.S. 717 (1988), 11.
3. “FTC Looks at Google-Apple Board Ties: Report,” Reuters (May 5, 2009),
http://www.reuters.com/article/us-google-apple-idUSTRE54403Z20090505.
The FTC investigated, and both individuals resigned from Apple’s board;
https://www.ftc.gov/news-events/press-releases/2009/08/statement-bureau
-competition-director-richard-feinstein-regarding; “Statement of FTC
Chairman Jon Leibowitz Regarding the Announcement That Arthur D.
Levinson Has Resigned from Google’s Board,” FTC Press Release (Oc-
tober 12, 2009), https://www.ftc.gov/news-events/press-releases/2009/10
/statement-ftc-chairman-jon-leibowitz-regarding-announcement. See
Section 8 of the Clayton Act, 15 U.S.C. § 19(a)(5).
4. Companies with a complementary relationship may not interact directly. In
their relationship, an increase in demand for one firm’s products or ser vices
increases the demand for the other’s—such is the relationship between
producers of peanut butter and jelly, or ketchup and french fries.
5. European Commission, Guidelines on the Assessment of Non-Horizontal
Mergers under the Council Regulation on the Control of Concentrations
between Undertakings (October 8, 2008), para. 5.
6. Ericsson, Ericsson Mobility Report: On the Pulse of the Networked Society
(June 2015), http://www.ericsson.com/res/docs/2015/ericsson-mobility
-report-june-2015.pdf.
Copyright © 2016. Harvard University Press. All rights reserved.

7. comScore, The 2015 U.S. Mobile App Report (September 22, 2015), 5.
8. Ibid.
9. Note the distinction between the powerful super-platform, which we discuss
in this chapter, and the more general definition of platform, which often
encompasses online companies operating in two- or multisided markets. On
the EU approach to platforms, see European Commission, Public Consulta-
tion on the Regulatory Environment for Platforms, Online Intermediaries,
Data and Cloud Computing and the Collaborative Economy (September 24,
2015), https://ec.europa.eu/digital-agenda/en/news/public-consultation
-regulatory-environment-platforms-online-intermediaries-data-and-cloud.

10. Don Clark and Robert McMillan, “Facebook, Amazon and Other Tech
Giants Tighten Grip on Internet Economy,” Wall Street Journal, November 5,
2015, http://www.wsj.com/articles/giants-tighten-grip-on-internet-economy
-1446771732.
11. David McLaughlin, “Google Said to Be under U.S. Antitrust Scrutiny over
Android,” Bloomberg (September 25, 2015), http://www.bloomberg.com
/news/articles/2015-09-25/google-said-to-be-under-u-s-antitrust-scrutiny
-over-android-iezf41sg.
12. Statistica, “Number of Available Applications in the Google Play Store from
December 2009 to February 2015” (2016), http://www.statista.com/statistics
/266210/number-of-available-applications-in-the-google-play-store/;
Statistica, “Number of Available Apps in the Apple App Store from July 2008
to June 2015” (2016), http://www.statista.com/statistics/263795/number-of
-available-apps-in-the-apple-app-store/.
13. One study of the top 200 apps found that thirty-eight paid and seventy-four
free apps were present on both super-platforms. Mikey Campbell, “Apps No
Longer Differentiator in iOS vs. Android War, Ser vices Next Battleground,”
Apple Insider (January 6, 2014), http://appleinsider.com/articles/14/01/06
/apps-no-longer-differentiator-in-ios-vs-android-war-services-next
-battleground.
14. Vangie Beal, “API—Application Program Interface,” Wedopedia (n.d.),
http://www.webopedia.com/TERM/A/API.html.
15. Ron Amadeo, “Google’s Iron Grip on Android: Controlling Open Source by
Any Means Necessary,” Ars Technica (October 20, 2013), http://arstechnica
.com/gadgets/2013/10/googles-iron-grip-on-android-controlling-open
-source-by-any-means-necessary/4/.
16. Ibid.
17. Ron Adner, Jianqing Chen, and Feng Zhu, “Frenemies in Platform Markets:
The Case of Apple’s iPad vs. Amazon’s Kindle,” Harvard Business School
Working Paper 15-087 (May 6, 2015).
18. Ibid., 2.
Copyright © 2016. Harvard University Press. All rights reserved.

19. Ibid. This paper provides an economic model when platforms have the
incentive to become Frenemies. As they discuss in regard to the e-reader
market: “Apple’s iPad provides many features beyond reading e-books, while
Amazon’s Kindle is almost exclusively an e-book reader. As a result, in
equilibrium, compared to Amazon, Apple’s hardware profits are more
impor tant to its total profits. In contrast, for Amazon, royalties from e-book
sales are more impor tant to its total profits relative to Apple. When this
difference in profit foci is large enough, having the Kindle Reader available
on iPad is agreeable to both Apple and Amazon: Amazon’s e-book sales
increase because iPad users can now purchase e-books from Amazon and

read them via Kindle Reader, and Apple’s hardware sales increase because
greater value accrues to the iPad with access to Kindle Reader than in the
case of incompatibility. The additional profits Apple generates from hard-
ware sales more than compensate its loss in royalties from e-book sales
through its iBooks. Similarly, the additional profits Amazon generates from
e-book sales are greater than its loss in Kindle device sales. In par ticu lar,
when Amazon subsidizes Kindle sales, it is always in Amazon’s interest to
have Kindle Reader on Apple’s iPad. We also show that it is never in Apple’s
or Amazon’s interest to have iBooks available on the Kindle device.”
20. Uber, “The Top 10 Facts You May Not Know about Uber Driver Partners”
(August 5, 2015), http://newsroom.uber.com/2015/08/the-top-10-facts-you
-may-not-know-about-uber-driver-partners/.
21. Brad Stone, “Exclusive: Google Is Developing Its Own Uber Competitor,”
Bloomberg (February 2, 2015), http://www.bloomberg.com/news/articles
/2015-02- 02/exclusive-google-and-uber-are-going-to-war-over-taxis.
22. Ibid.
23. Jack Nicas, “Alphabet Cruises into Ride-Sharing Business,” Wall Street
Journal, May 17, 2016, B4.
24. Chunka Mui, “Google Is Millions of Miles Ahead of Apple in Driverless
Cars,” Forbes, August 21, 2015, http://www.forbes.com/sites/chunkamui
/2015/08/21/google-is-millions-of-miles-ahead-of-apple-in-driverless-cars/.
25. Daisuke Wakabayashi, “Apple Targets Electric-Car Shipping Date for 2019,”
Wall Street Journal, September 21, 2015, http://www.wsj.com/articles/apple
-speeds-up-electric-car-work-1442857105.
26. “Apple Invests in Chinese Uber Rival Didi Chuxing,” BBC News (May 13,
2016), http://www.bbc.co.uk /news/business-36283661; “Apple Invests $1bn in
‘Chinese Uber’ Didi Chuxing,” The Telegraph (May 13, 2016), http://www
.telegraph.co.uk /technology/2016/05/13/apple-invests-1bn-in-chinese-uber
-didi-chuxing/.
27. NDTV Correspondent, “Google Reportedly Wants More of Its Apps Prein-
stalled on Android Devices,” Gadgets360 (September 29, 2014), http://gadgets
Copyright © 2016. Harvard University Press. All rights reserved.

.ndtv.com/mobiles/news/google-reportedly-wants-more-of-its-apps-pre
installed-on-android-devices-599478.
28. Jack Nicas, “Alphabet Cruises into Ride-Sharing Business.”
29. For instance, in Europe, such may be the case when a dominant undertaking
operates at both the upstream and downstream levels and refuses to supply a
competitor operating at the downstream level. Such a refusal to supply is
contrary to Article 102 TFEU when it eliminates competition in the down-
stream market. Another form of refusal to supply that is objectionable occurs
when a dominant party controls the provision of an essential infrastructure,
uses that essential facility, but refuses other companies access to that facility,

with no objective justification for doing so. In other cases a refusal to supply
may involve a refusal to license intellectual property rights. For relevant
cases see Ariel Ezrachi, EU Competition Law—An Analytical Guide to the
Leading Cases, 4th ed. (Oxford: Hart Publishing, 2014), 252.
30. John Markoff, “Toyota Invests $1 Billion in Artificial Intelligence in U.S.,”
New York Times, November 6, 2015, http://www.nytimes.com/2015/11/06
/technology/toyota-silicon-valley-artificial-intelligence-research-center.html
?_r = 0; Mui, “Google Is Millions of Miles Ahead of Apple in Driverless Cars.”
31. See, e.g., Markoff, “Toyota Invests $1 Billion in Artificial Intelligence in U.S.”
32. Jean Tirole, “Comments Made at FT-ETNO Summit 2015,” Financial Times
(October 13, 2015), https://live.ft.com/Events/2015/FT-ETNO-Summit-2015.
33. Stone, “Exclusive.”
34. Matt Weinberger, “Microsoft Could See an Opportunity to Poke Google
in the Eye with Uber Investment,” Business Insider UK (July 31, 2015),
http://www.businessinsider.com/microsoft-and-google-are-uber-investors
-2015-7.
35. Nathaniel Mott, “Uber Should Fear the Company Formerly Known as
Google,” Gigaom (August 11, 2015), https://gigaom.com/2015/08/11/uber-vs
-alphabet-google/.
36. Weinberger, “Microsoft Could See an Opportunity to Poke Google in the Eye
with Uber Investment.”
37. Douglas MacMillan, “GM Invests $500 Million in Lyft, Plans System for
Self-Driving Cars,” Wall Street Journal, January 4, 2016, http://www.wsj.com
/article _email/gm-invests-500-million-in-lyft-plans-system-for-self-driving
-cars-1451914204-lMyQjAxMTI2NTA2NDEwODQyWj.
38. Coupons.com, Form 10-K for 2014 (2014), 17; Yelp Inc., Form 10-Q for the
Quarterly Period Ended June 30, 2015 (2015), 33, http://www.sec.gov
/Archives/edgar/data/1345016/000120677415002479/yelp_10q.htm. “The
number of people who access information about local businesses through
mobile devices, including smartphones, tablets and handheld computers, has
increased dramatically over the past few years and is expected to continue to
Copyright © 2016. Harvard University Press. All rights reserved.

increase. Although many consumers access our platform both on their

mobile devices and through personal computers, we have seen substantial
growth in mobile usage. We anticipate that growth in use of our mobile
platform will be the driver of our growth for the foreseeable future and that
usage through personal computers may continue to decline worldwide. As a
result, we must continue to drive adoption of and user engagement on our
mobile platform, and our mobile app in par ticu lar.” LinkedIn Corp., Form
10-Q, for the Quarterly Period Ended June 30, 2015 (2015), 47, http://www
.sec.gov/Archives/edgar/data/1271024/000127102415000020/a20150630
-10qdocument.htm. “Many individuals use mobile devices to access online

ser vices. . . . The number of people who access online ser vices through
mobile devices, such as smart phones, handheld tablets and mobile tele-
phones, as opposed to personal computers, has increased dramatically in the
past few years and is projected to continue to increase.”
39. They noted, for instance, that the Independent Music Companies Associa-
tion “claimed that YouTube had threatened to remove content and block
access to its ser vices “unless non-negotiable licensing conditions were
accepted,” and had tried to impose a “ ‘least-favoured nation’ clause ensuring
the royalty rate of all independents could be aligned with the lowest rate
agreed with any label worldwide.” Paragraph 126, House of Lords, Select
Committee on European Union, “Online Platforms and the Digital Single
Market” (April 20, 2016), 10th Report of Session 2015–16, http://www
.publications.parliament.uk /pa/ld201516/ldselect/ldeucom/129/129.pdf. The
Association of Authors’ Agents said that Amazon asked “suppliers and
customers to agree to terms and conditions that are liable to change without
notice.” The British Booksellers Association agreed. Amazon’s contracts
enabled it “to change the terms whenever it liked,” and added that many
publishers “had been asked by Amazon to ring fence stock . . . without
receiving a guaranteed order.” Ibid.
40. Coupons.com, Form 10-K for 2014, 15, 17.
41. Facebook, Form 10-K for 2014 (2014), 11.
42. Ibid.
43. Ibid.
44. Ibid.
45. In finding that the Facebook/WhatsApp merger was unlikely to be anticom-
petitive, the European Commission inquired, among other things, whether:
(1) users of the consumer communications apps are locked in to any par ticu lar
physical network, hardware solution, or anything else that needs to be
replaced in order to use competing products; (2) consumers had control over,
and there were any significant limits on, the portability of their data; and (3)
the parties had any means to preclude competitors from recreating a user’s
Copyright © 2016. Harvard University Press. All rights reserved.

network on the parties’ applications. Case COMP/M.7217, Facebook/

WhatsApp, Commission Decision, 2014 O.J. (C 7239), 24–25, 134. Presum-
ably, if the answer was “yes,” then the risks of anticompetitive, unilateral
conduct increase. These three factors, which identify several more potential
abuses of a dominant position, involve consumers’ switching costs. The basic
premise is that as the more the time and cost needed to switch products or
ser vices increase, the more the consumer is locked in, and the greater the
dominant firm’s ability to increase price or, for our purposes, reduce other
parameters of competition, such as quality, including the level of privacy

protection. This is especially the case where consumers cannot readily

predict costs or quality levels over time.
46. These exclusionary practices are explored in greater detail in Maurice E.
Stucke and Allen P. Grunes, Big Data and Competition Policy (Oxford:
Oxford University Press, 2016).
47. European Commission, “Antitrust: Commission Sends Statement of Objec-
tions to Google on Android Operating System and Applications,” press release
(April 20, 2016), http://europa.eu/rapid/press-release_IP-16-1492 _en.htm.
48. Ibid.
49. If a manufacturer wished to preinstall Google proprietary apps, including
Google Play Store and Google Search, on any of its devices, Google required
the manufacturer not to sell any devices running on “Android forks”—
basically, any modified Android mobile operating system. Forgoing Google
apps, as some phone makers have found, is difficult. As the Wall Street Journal
reported, Amazon.com in 2014 launched its customized version of an Android
smartphone. Because Amazon’s version was an Android fork, it couldn’t
include the more popu lar apps like Google Search, YouTube, Maps, or the
Play Store. Amazon’s smartphone “sold poorly, which some Google detractors
blamed on the lack of Google apps.” Natalia Drozdiak and Sam Schechner,
“EU Set to Charge Google over Android Phone Apps: European Commission
Focusing on Demand That Phones Load Google Apps,” Wall Street Journal,
April 19, 2016, http://www.wsj.com/article _email/eu-set-to-charge-google
-over-android-1461067383-lMyQjAxMTI2MDE5OTAxMjk4Wj.
50. One inducement is Google Play Store, where smartphone users can access
and download apps. Google’s Play Store accounts for more than 90 percent of
apps downloaded on Android devices in Europe. In its licensing contracts
with the smartphone manufacturers, Google basically told them that if they
wanted to install Google Play Store on their Android devices, then they had
to preinstall Google Search and make it the default search engine.
51. European Commission, “Antitrust: Commission Sends Statement of
Objections to Google on Android Operating System and Applications—
Copyright © 2016. Harvard University Press. All rights reserved.

Factsheet” (April 20, 2016), http://europa.eu/rapid/press-release _ MEMO-16

-1484 _en.htm.

15 • Extraction and Capture

1. See Part III and Chapter 8.

2. Riley v. California, 134 S. Ct. 2473, 2489, 189 L. Ed. 2d 430 (2014).
3. “First, a cell phone collects in one place many distinct types of information—
an address, a note, a prescription, a bank statement, a video—that reveal

much more in combination than any isolated record. Second, a cell phone’s
capacity allows even just one type of information to convey far more than
previously possible. The sum of an individual’s private life can be recon-
structed through a thousand photographs labeled with dates, locations, and
descriptions; the same cannot be said of a photograph or two of loved ones
tucked into a wallet. Third, the data on a phone can date back to the pur-
chase of the phone, or even earlier. A person might carry in his pocket a slip
of paper reminding him to call Mr. Jones; he would not carry a record of all
his communications with Mr. Jones for the past several months, as would
routinely be kept on a phone. Finally, there is an element of pervasiveness
that characterizes cell phones but not physical records. Prior to the digital
age, people did not typically carry a cache of sensitive personal information
with them as they went about their day. Now it is the person who is not
carry ing a cell phone, with all that it contains, who is the exception. Ac-
cording to one poll, nearly three-quarters of smart phone users report being
within five feet of their phones most of the time, with 12% admitting that
they even use their phones in the shower. . . . [I]t is no exaggeration to say
that many of the more than 90% of American adults who own a cell phone
keep on their person a digital record of nearly every aspect of their lives—
from the mundane to the intimate.” Ibid., at 2489–2490 (internal footnote
omitted).
4. Ibid., at 2490.
5. Ibid.; see, for example, Brian X. Chen, “Why and How Apple Is Collecting
Your iPhone Location Data,” Wired, April 21, 2001, http://www.wired.com
/2011/04/apple-iphone-tracking/.
6. Australian Communications and Media Authority. Here, There and
Everywhere— Consumer Behaviour and Location Services (December 2012),
1, http://www.acma.gov.au/webwr/_ assets/main/lib310665/location _ services
-dec2012.pdf.
7. Ibid.
8. Jack Nicas, “Google to Increase Frequency, Size of Ads,” Wall Street Journal,
Copyright © 2016. Harvard University Press. All rights reserved.

May 25, 2016, B4.

9. Brian See Voo, “Smartphones & You: Who Is (Really) in Control?” Hongkiat
(December 3, 2013), http://www.hongkiat.com/blog/smartphone-you-who-in
-control/; Andrew Munchbach, “Your Smartphone Is Tracking You, and You
Said It Was Okay,” BGR (April 20, 2011), http://bgr.com/2011/04/20/your
-smartphone-is-tracking-you-and-you-said-it-was-okay/.
10. Australian Communications and Media Authority, Here, There and Every-
where, 18
11. Alun Williams, “Google Invites Developers to Its Brillo IoT Platform,”
Electronics Weekly, October 28, 2015, http://www.electronicsweekly.com

/blogs/eyes-on-android/embedded-android/google-invites-developers-to-its
-brillo-iot-platform-2015–10/.
12. Marcia Coyle, “Justices Hang Up on Call for Cellphone Location Protection,”
National Law Journal, November 9, 2015, http://www.nationallawjournal
.com/id=1202741937880/Justices-Hang-Up-on-Call-for-Cellphone-Location
-Protection#ixzz3r2OQU3pN.
13. “[M]any users don’t realise that some app providers and developers sell their
users’ location data to marketing companies, allowing profi les to be built for
targeted advertising and other purposes not necessarily apparent from use of
the original app. . . . Nine times out of ten, [a location-based ser vice] app
simply tells you that the application wants to use your location and then asks
you whether to allow, or not allow; hardly sufficient to be considered
‘informed’ consent, as most users wouldn’t understand the full implication
of pressing ‘allow.’ ” Taylor Wessing, “ ‘Toto, We’re Not in Satnav Anymore’:
Does the Law Protect Mobile Users from a Misuse of Their Location Data?”
(April 2011), http://united-kingdom.taylorwessing.com/download/article
_ satnav.html#.Vhw3oRNViko.
14. Craig Timberg, Nancy Scola, and Andrea Peterson, “Uber Executive Stirs
Up Privacy Controversy,” Washington Post, November 18, 2014, http://www
.washingtonpost.com / business/technology/uber-executive-stirs-up-privacy
-controversy/2014/11/18/d0607836 - 6f61-11e4 -ad12-3734c461eab6 _ story
.html.
15. Ibid.
16. Uber, Privacy Statement (effective July 15, 2015), https://www.uber.com/legal
/privacy/users/en.
17. Ibid.
18. Ibid.
19. Ibid.
20. Uber, iOS App Permissions, https://www.uber.com/ios/permissions.
21. Ibid.
22. Uber, Android App Permissions, https://www.uber.com/android
Copyright © 2016. Harvard University Press. All rights reserved.

/permissions.
23. Kenneth Olmstead and Michelle Atkinson, “Apps Permissions in the Google
Play Store,” Pew Research Center (November 10, 2015), http://www.pew
internet.org/2015/11/10/apps-permissions-in-the-google-play-store/.
24. Daniel Eran Dilger, “After Abandoning iAd Revenue, Apple Inc Can
Reintroduce an Ad-Free Internet,” Apple Insider (January 17, 2016), http://
appleinsider.com/articles/16/01/18/after-abandoning-iad-revenue-apple-can
-reintroduce-an-ad-free-internet; see also Jennifer LeClaire, “Apple May Pull
Plug on Its iAd Advertising Business,” CIO Today (January 14, 2016),
http://www.cio-today.com/article/index.php?story_id=11300AJK1JYB.

25. Uber, Privacy Statement.

26. Ibid.
27. Uber, Cookie Statement (Global) (effective July 15, 2015), https://www.uber
.com/legal/privacy/cookies/en.
28. In re Hulu Privacy Litig., 86 F. Supp. 3d 1090, 1094 (N.D. Cal. 2015) (internal
citations omitted): “Examples include how the website should be displayed,
how many times a user has visited the website, what pages he visited, and
authentication information. Each web browser on a computer (e.g., Internet
Explorer or Chrome) stores the cookies that are created during a user’s use of
the browser in a folder on the user’s computer that is unique to that browser.
When a user types a website address into her browser, the browser sends:
(a) a request to load the page to the webserver for that website address; and
(b) any cookies on the user’s computer that are associated with the website
(such as the cookies for hulu.com or facebook.com). The remote website
server returns the requested page and can update the cookies or write
new ones. The only servers that can access a particular cookie are those
associated with the domain that wrote the cookie. In other words, Hulu can
read only hulu.com cookies, while Facebook can read only facebook.com
cookies; the companies cannot read or write to cookies associated with the
other ser vice.”
29. Julia Angwin and Tom McGinty, “Sites Feed Personal Details to New
Tracking Industry,” Wall Street Journal, July 30, 2010, http://www.wsj.com
/articles/SB10001424052748703977004575393173432219064.
30. Ibid. The word “party” refers to the “website that is placing the cookie”; Open
Tracker, “Third-Party Cookies vs First-Party Cookies” (April 15, 2013),
http://www.opentracker.net/article/third-party-cookies-vs-first-party
-cookies. “So, for example, if you visit widgets.com and the domain of the
cookie placed on your computer is widgets.com, then this is a first-party
cookie. If, however, you visit widgets.com and the cookie placed on your
computer says stats-for-free.com, then this is a third-party cookie”; WhatIs
.com, “Third-Party Cookie” (n.d.), http://whatis.techtarget.com/defi nition
Copyright © 2016. Harvard University Press. All rights reserved.

/third-party-cookie. “As it affects their survival, firms have tried to under-

mine these changes by using other techniques such as respawning cookies,
Flash cookies, entity tags (Etags) and canvas fingerprinting.”
31. Angwin and McGinty, “Sites Feed Personal Details to New Tracking
Industry.”
32. Ibid.
33. Ibid. (“For example, we and our ad partners may rely on information gleaned
through these cookies to serve you ads that may be interesting to you on
other websites. Similarly, our partners may use a cookie, attribution ser vice

or another similar technology to determine whether we’ve served an ad and

how it performed or provide us with information about how you interact
with them.”)
34. Uber, Privacy Statement.
35. Google, “How Google Uses Data When You Use Our Partners’ Sites or
Apps,” Google Privacy & Terms, http://www.google.com/intl/en/policies
/privacy/partners/.
36. Ibid.
37. Ibid.
38. Ibid.
39. Ibid.
40. Google, Google Analytics Opt-Out Browser Add-On, https://tools.google
.com/dlpage/gaoptout.
41. In re Nickelodeon Consumer Privacy Litig., No. CIV.A. 12–07829, 2014 WL
3012873, 1–2 (D.N.J. Jul. 2, 2014).
42. Ibid., 2.
43. For a contrary view on the Video Privacy Protection Act, 18 U.S.C. § 2710,
see Yershov v. Gannett Satellite Info. Network, Inc., No. CIV.A. 14-13112-
FDS, 2015 WL 2340752, 8 (D. Mass. May 15, 2015), on appeal to the First
Circuit.
44. In re Nickelodeon Consumer Privacy Litig.
45. Steve Stecklow, “On the Web: Children Face Intensive Tracking,” Wall Street
Journal, September 17, 2010, http://www.wsj.com/articles/SB10001424052748
703904304575497903523187146.
46. Ibid.
47. Ibid.
48. Ibid.
49. Ibid.
50. Ibid.
51. Ibid.
52. Federal Trade Commission, Data Brokers: A Call for Transparency and
Copyright © 2016. Harvard University Press. All rights reserved.

Accountability (May 2014), v, https://www.ftc.gov/system/fi les/documents

/reports/data-brokers-call-transparency-accountability-report-federal-trade
-commission-may-2014/140527databrokerreport.pdf.
53. Angwin and McGinty, “Sites Feed Personal Details to New Tracking
Industry”; “The Web’s New Gold Mine: Your Secrets,” Wall Street Journal,
July 30, 2010, http://www.wsj.com/articles/SB100014240527487039409045753
95073512989404?cb =logged0.6130286159459502.
54. Alex Chris, “Can You Still Make Money with AdSense?” reliablesoft.net,
https://www.reliablesoft.net/can-you-still-make-money-with-adsense/.

55. In its first quarter of 2016, for example, Google collected $18.02 billion in ad
revenue: $14.328 billion from its own websites and $3.692 billion from its
network members’ websites. Alphabet, Inc., Form 10-Q, for the quarterly
period ended March 31, 2016 (2016), 32, https://abc.xyz/investor/pdf
/20160331_ alphabet _10Q.pdf.
56. Jacob Weisberg, “We Are Hopelessly Hooked,” New York Review of Books,
February 25, 2016, 9.
57. Ibid.
58. Ibid.
59. Lara O’Reilly, “There Is One Key Reason Why People Are Spending More
Time on Facebook’s Apps Than Google’s,” Business Insider, June 29, 2015,
http://www.businessinsider.com/facebook-vs-google-on-time-spent-in-apps
-according-to-forrester-2015- 6.
60. Daisuke Wakabayashi and Jack Marshall, “Apple’s Ad Blockers Rile Pub-
lishers: New iOS Lets Users Halt Ads on Mobile Devices, Posing a Challenge
to Publishers and Google,” Wall Street Journal, August 30, 2015, http://www
.wsj.com /articles/apples-ad-blockers-raise-tensions-1440974849?cb =logged
0.8844516936223954.
61. Coupons.com, Form 10-K (2014), 21.
62. Allen Grunes, “Tracking Not Allowed (Unless You’re Google),” Politico
(October 1, 2015), http://www.politico.com/agenda/story/2015/10/tracking
-not-allowed-unless-youre-google-000261.
63. Ibid.
64. Ibid.
65. Ibid.
66. Andrew Griffin, “Facebook ‘Reactions’ Begin Global Roll-Out, Offering Six
Different Emotions Instead of Dislike Button,” The Independent, January 28,
2016; Marie Brewis, “What Is Facebook Dislike? How to Show Support for or
Dislike a Facebook Post. Plus: Facebook to Test ‘Reactions,’ ” Tech Advisor
(January 29. 2016), http://www.pcadvisor.co.uk /how-to/social-networks
/what-is-facebook-dislike-reactions-cox-3625549/.
Copyright © 2016. Harvard University Press. All rights reserved.

67. Deepa Seetharaman and Elizabeth Dwoskin, “Facebook’s Restrictions on

User Data Cast a Long Shadow,” Wall Street Journal, September 21, 2015,
http://www.wsj.com/article _email/facebooks-restrictions-on-user-data-cast
-a-long-shadow-1442881332-lMyQjAxMTE1MzIwMjEyMTIzWj.
68. Ibid.
69. Ibid.
70. Ibid.
71. Madhumita Murgia, “Facebook Messenger’s New Bots Are a Powerful Way
to Target Adverts,” The Telegraph (April 13, 2016), http://www.telegraph.co

.uk /technology/2016/04/12/facebook-messenger-launches-chat-bot-economy
-to-take-on-apps/ .
72. Bogdan Petrovan, “How Does Google Make Money from Android?” Android
Authority (January 22, 2016), http://www.androidauthority.com/how-does
-google-make-money-from-android- 669008/.
73. Microsoft, Form 10-K for the Fiscal Year Ended June 30, 2014 (2014), 19.
74. Amir Mizroch, “In Belgium, an Encryption Powerhouse Rises,” Wall Street
Journal, December 10, 2015, http://www.wsj.com/articles/in-belgium-an
-encryption-powerhouse-rises-1449791014.
75. Alistair Barr, “Google Mobile Apps Grab Almost as Much User Time as
Facebook’s Apps,” Wall Street Journal, June 29, 2015, http://blogs.wsj.com
/digits/2015/06/29/google-mobile-apps-grab-almost-as-much-user-time-as
-facebooks-apps/.
76. Petrovan, “How Does Google Make Money from Android?”
77. Catherine Stupp and Jorge Valero, “Commission Mulls New Measures in
Divisive Inquiry of Online Giants,” EurActiv.com (September 8, 2015),
http://www.euractiv.com/sections/infosociety/commission-mulls-new
-measures-divisive-inquiry-online-giants-317409.
78. “Desktop Operating System Market Share,” Net Market Share (December
2015), http://netmarketshare.com/operating-system-market-share.aspx
?qprid= 8&qpcustomd= 0.
79. Microsoft, Form 10-K, for the Fiscal Year Ended June 30, 2015 (2015), 18,
http://apps.shareholder.com/sec/viewerContent.aspx?companyid=MSFT
&docid=10834537#D918813D10K _HTM _TX918813_ 3.
80. Ibid.
81. Disconnect, Inc., Complaint of Disconnect, Inc., Regarding Google’s Infringe-
ment of Article 102 TFEU through Bundling into the Android Platform and
the Related Exclusion of Competing Privacy and Security Technology, Case
COMP/40099 (June 2015), para. 45.
Copyright © 2016. Harvard University Press. All rights reserved.

16 • “Why Invite an Arsonist to Your Home?”

Understanding the Frenemy Mentality

1. Feng Zhu and Qihong Liu, “Competing with Complementors: An Empirical

Look at Amazon.com,” Harvard Business School Technology and Operations
Management Unit Working Paper No. 15-044 (August 21, 2015), http://ssrn
.com/abstract=2533616 or http://dx.doi.org/10.2139/ssrn.2533616 (findings
from data from Amazon.com the patterns of Amazon’s entry into its
third-party sellers’ product spaces; how the likelihood of Amazon’s entry is
positively correlated with the popularity and customer ratings of third-party

sellers’ products; that Amazon’s entry reduces the shipping costs of affected
products and hence increases their demand; and how small third-party
sellers affected by Amazon’s entry appear to be discouraged from growing
their businesses on the platform subsequently).
2. In re Goldenshores Technologies, FTC File No. C-4446 (2014), Complaint 4,
https://www.ftc.gov/system/fi les/documents/cases/140409goldenshorescmpt
.pdf .
3. Ibid., Complaint 5.
4. Disconnect, Inc. Complaint of Disconnect, Inc., Regarding Google’s Infringe-
ment of Article 102 TFEU through Bundling into the Android Platform and
the Related Exclusion of Competing Privacy and Security Technology, Case
COMP/40099 June 2015).
5. In re Goldenshores Technologies, Complaint 5.
6. Donald S. Clark, Letter to Isaac Buckman, Commonwealth of Pennsylvania,
Re: In the Matter of Goldenshores Technologies, LLC, File No.1323087, Federal
Trade Commission (March 31, 2014), https://www.ftc.gov/system/fi les
/attachments/goldenshores-technologies-llc-et-al.commission-letters
-commenters-april-9-2014/140409goldenshoresltr-buckman.pdf.
7. Donald S. Clark, Decision and Order, In the Matter of Goldenshores Technolo-
gies, LLC and Erik M. Geidl, Docket No: C-4446, Federal Trade Commission
(March 31, 2014), https://www.ftc.gov/system/fi les/documents/cases/140409
goldenshoresdo.pdf.
8. Ibid.
9. Goldenshores Technologies, LLC, Privacy Policy (2014), http://www.golden
shorestechnologies.com/privacypolicy.html.
10. As the FTC found, while the “Android ‘permissions’ ” provided “notice to
consumers regarding what sensitive information (e.g., location information)
or sensitive device functionality (e.g., the ability to take photos with the
device’s camera) an application may access,” the permissions never explained
“whether the application shares any information with third parties.” In re
Goldenshores Technologies, Complaint, para. 7.
Copyright © 2016. Harvard University Press. All rights reserved.

11. Google, Google Play Developer Policy Center, https://play.google.com/about

/developer-content-policy.html.
12. Ibid.
13. Alphabet, Inc., Form 10-K, for the Fiscal Year Ended December 31, 2015
(2016) at 25, https://abc.xyz/investor/pdf/20151231_ alphabet _10K.pdf.
14. Ibid.; Google Inc., Form 10-K, for the Fiscal Year Ended December 31, 2014
(2015) at 24, https://abc.xyz/investor/pdf/20141231_ google _10K.pdf.
15. AdMob by Google, Monetize Your Apps Intelligently, https://www.google
.com/admob/.

16. AdMob by Google, Maximize Ad Revenue, https://www.google.com/admob

/monetize.html.
17. Article 29 Data Protection Working Party, Opinion 2/2012 on Online
Behavioural Advertising, 00909/10/EN WP 171 (June 22, 2010), http://ec
.europa.eu/justice/policies/privacy/docs/wpdocs/2010/wp171_en.pdf.
18. Google Ads, Find New Customers Now, with Google AdWords, http://www
.google.com/ads/.
19. “Moving Targets: What Advertisers Love, and What They Hate, about
Mobile Devices,” The Economist, September 13, 2014, http://www.economist
.com/news/special-report/21615870-what-advertisers-love-and-what-they
-hate-about-mobile-devices-moving-targets.
20. Julia Angwin, “The Web’s New Gold Mine: Your Secrets,” Wall Street
Journal, July 30, 2010, http://www.wsj.com/articles/SB1000142405274870394
0904575395073512989404?cb =logged0.6130286159459502.
21. Julia Angwin and Tom McGinty, “Sites Feed Personal Details to New
Tracking Industry,” Wall Street Journal, July 30, 2010, http://www.wsj.com
/articles/SB10001424052748703977004575393173432219064.
22. Luigi Vigneri, Jaideep Chandrashekar, Ioannis Pefk ianakis, and Olivier
Heen, “Taming the Android AppStore: Lightweight Characterization of
Android Applications,” Eurecom Research Report RR-15-305 (April 27,
2015), http://arxiv.org/pdf/1504.06093v2.pdf.
23. Ibid., 7.
24. Ibid., 9.
25. Ibid., 19.
26. Disconnect, Inc., Complaint of Disconnect, Inc., para.18.
27. Ibid., para. 19.
28. Ibid., para. 19.
29. Ibid., para. 20.
30. Ibid., para. 27.
31. Ibid., para. 38.
32. Ibid., para. 38.
Copyright © 2016. Harvard University Press. All rights reserved.

33. https://disconnect.me/.
34. Julie Bort, “Why Google Banned a Privacy Tool Called ‘Disconnect Mobile’
from the Android App Store,” Business Insider (August 28, 2014), http://www
.businessinsider.com/why-google-banned-connect-mobile-2014-8.
35. Ibid.
36. Reed Albergotti, Alistair Barr, and Elizabeth Dwoskin, “Why Some Privacy
Apps Get Blocked from the Android Play Store,” Wall Street Journal,
August 28, 2014, http://blogs.wsj.com/digits/2014/08/28/why-some-privacy
-apps-get-blocked-from-the-android-play-store/.

37. Ibid.
38. Ibid.
39. Ibid.
40. As the Wall Street Journal reported, “Dozens of apps aimed at protecting
users’ privacy are available in the Play store, promising to help users opt out
of being tracked for ad targeting or spot which companies are tracking
them.” But Ghostery’s Ad Control, which wasn’t kicked off, “asks companies
to opt users out of targeted advertising on mobile devices. However, users
still see ads, and the app doesn’t block tracking without first asking adver-
tisers.” Ibid. Ghostery’s owner, Evidon, also reportedly “helps companies that
want to improve their use of tracking code by selling them data collected from
the eight million Ghostery users who have enabled a data-sharing feature in
the tool”; Tom Simonite, “A Popu lar Ad Blocker Also Helps the Ad Industry,”
MIT Technology Review (June 17, 2013), http://www.technologyreview.com
/news/516156/a-popular-ad-blocker-also-helps-the-ad-industry/; Hanqing
Chen, “Privacy Tools: How to Block Online Tracking,” ProPublica (July 3,
2014): “Ghostery users are encouraged to opt in to Ghostrank, a ser vice that
sends anonymous information to a Ghostery server about where and how
users encounter trackers. Ghostery is a for-profit company that analyzes the
Ghostrank information and sells it to companies that want to manage their
tracking businesses.” As Ghostery explains on its website, “We rely on
Ghostery users who opt-in to send us anonymous information about the data
collection technology they see, and where they see them. We take that
information, add our analysis, and sell it to companies to help them audit
and manage their relationships with these marketing tools. None of the
information we share is about our users, nor is it stored in a way that could
be used to trace back to our users”; https://www.ghostery.com/support/faq
/ghostery-add-on/how-does-ghostery-make-money-from-the-add-on/.
41. Google Chrome., Turn “Do Not Track” On or Off, https://support.google.com
/chrome/answer/2790761?hl= en.
42. You can control your settings for ads served by Google that you see within
Copyright © 2016. Harvard University Press. All rights reserved.

a browser by visiting the Ads Settings page in that browser. Some apps
may allow you to view web pages without launching your mobile device’s
default web browser. Such pages will not recognize your default browser’s
settings, including your ads settings for ads served by Google on web
pages. If the app allows you to navigate to the Ads Settings page, you can
control your settings for ads served by Google on web pages that you see
within that app.
43. Google Ads Help, Opt Out, https://support.google.com/ads/answer/2662922
?hl= en; “Ads are essential to fund many websites. When you opt out, you’ll
still see ads by Google—they just won’t be based on your interests, your visits

to advertiser websites, or demographics. Instead, they’ll be based on factors

such as the content of the page, your general location, or your recent
searches.”
44. As Google states, “Opting out doesn’t stop interest-based ads that aren’t
served by Google. So, when you browse the Internet and see a website, video,
or app that’s outside of Google’s Display Network, you may still see interest-
based ads from other companies.” Ibid.
45. Ibid. “Even if you opt out of interest-based ads by Google on one browser
(like Chrome), you may still see interest-based ads by Google on your other
browsers (like Internet Explorer and Safari). The same applies if you use
multiple computers or devices.”
46. Ibid. As Google states, “Do you delete or clear your browser’s cache and
cookies? If you’ve cleared your cookies after opting out you may have undone
your opt out for that browser.” Each time the user clears the cookies on the
browser, they would have to opt out again of interest-based ads on their
browser, or install an extension like Protect My Choices.
47. Ibid.
48. Disconnect, Inc., Complaint of Disconnect, Inc., para. 14–15.
49. Ingrid Lunden, “Disconnect.Me Files Antitrust Case against Google in
Europe over Banned Anti-Malware Android App,” TechCrunch (June 2,
2015), http://techcrunch.com/2015/06/02/disconnect-me-fi les-antitrust-case
-against-google-in-europe-over-banned-anti-malware-android-app/.
50. Ibid.
51. Disconnect, Inc., Complaint of Disconnect, Inc., para. 13.
52. Google, 2014, Form 10-K, 16.
53. Ibid.
54. Facebook, Form 10-K, for the Fiscal Year Ended December 31, 2014 (2015),
10, http://fi les.shareholder.com/downloads/AMDA-NJ5DZ/650609882x0x
S1326801%2D15%2D6/1326801/fi ling.pdf.
55. Ibid., 10–11.
56. Coupons.com, Form 10-K (2014), 21.
Copyright © 2016. Harvard University Press. All rights reserved.

57. Daisuke Wakabayashi and Jack Marshall, “Apple’s Ad Blockers Rile Pub-
lishers: New iOS Lets Users Halt Ads on Mobile Devices, Posing a Challenge
to Publishers and Google,” Wall Street Journal, August 30, 2015, http://www
.wsj.com/articles/apples-ad-blockers-raise-tensions-1440974849?cb =logged0
.8844516936223954.
58. Ibid.
59. United States v. Alcoa, 148 F.2d 416, 430 (2d Cir. 1945).
60. Aspen Skiing v. Aspen Highlands Skiing, 472 U.S. 585 (1985); Verizon
Communications Inc. v. Law Offices of Curtis V. Trinko, LLP, 540 U.S.
398 (2004).

61. Case C-52/09, Konkurrensverket v. TeliaSonera Sverige AB, [2011] ECR I-527,
[2011] 4 CMLR 18, para. 24.
62. Barack Obama, Statement of Senator Barack Obama for the American
Antitrust Institute, (September 7, 2007), http://www.antitrustinstitute.org
/fi les/aai-%20Presidential%20campaign%20-%20Obama%209- 07
_092720071759.pdf.
63. Ibid.
64. Ibid.
65. Brent Kendall, “Justice Department Doesn’t Deliver on Promise to Attack
Monopolies: Obama Administration Arrived Promising a Tougher Stance,
but Few Antitrust Cases Have Been Pursued in U.S. and Enforcement Has
Shifted to Europe,” Wall Street Journal, November 7, 2015, http://www.wsj
.com/articles/justice-department-doesnt-deliver-on-promise-to-attack
-monopolies-1446892202.

17 • The Future of Frenemy: The Rise of Personal Assistants

Epigraph: Internet Movie Database, “Quotes for HAL 9000 (Character)

from 2001: A Space Odyssey (1968)” (2015), http://www.imdb.com/character
/ch0002900/quotes.
1. Danny Yadron, “Google Assistant Takes on Amazon and Apple to Be the
Ultimate Digital Butler,” The Guardian, May 18, 2016, https://www
.theguardian.com/technology/2016/may/18/google-home-assistant-amazon
-echo-apple-siri.
2. Christopher Mims, “Ask M for Help: Facebook Tests New Digital Assistant:
Single Interface Could Replace Web Searches and Apps on Mobile Devices,”
Wall Street Journal, November 9, 2015, http://www.wsj.com/articles/ask-m
-for-help-facebook-tests-new-digital-assistant-1447045202.
3. Yadron, “Google Assistant Takes on Amazon and Apple.”
4. Ibid.
5. Mims, “Ask M for Help.”
Copyright © 2016. Harvard University Press. All rights reserved.

6. Mims, “Ask M for Help.”

7. Jack Nicas, “Google’s New Products Reflect Push into Machine Learning,”
Wall Street Journal, May 18, 2016, http://www.wsj.com/articles/googles-new
-products-reflect-push-into-machine-learning-1463598395?mod= ST1.
8. Margrethe Vestager, “How Competition Supports Innovation,” speech,
Regulation4Innovation, Brussels, May 24, 2016, http://ec.europa.eu
/commission/2014-2019/vestager/announcements/how-competition
-supports-innovation _en.
9. Ibid.

10. Amy Mitchell, Jeff rey Gottfried, and Katerina Eva Matsa, “Millennials and
Political News: Social Media—The Local TV for the Next Generation?” Pew
Research Center, June 1, 2015, http://www.journalism.org/2015/06/01
/millennials-political-news/.
11. Katerina Eva Matsa, “Facebook, Twitter Play Different Roles in Connecting
Mobile Readers to News,” Pew Research Center, May 9, 2016, http://www
.pewresearch.org/fact-tank /2016/05/09/facebook-twitter-mobile-news/.
12. Deepa Seetharaman, “Uproar over Bias Claims Ignites Fears over Facebook’s
Influence,” Wall Street Journal, May 11, 2016, A1.
13. Jonathan Zittrain, “Facebook Could Decide an Election without Anyone
Ever Finding Out—The Scary Future of Digital Gerrymandering—and How
to Prevent It,” New Republic, June 2, 2014, https://newrepublic.com/article
/117878/information-fiduciary-solution-facebook-digital-gerrymandering.
Also note more generally, Jonathan Zittrain, The Future of the Internet and
How to Stop It (New Haven, CT: Yale University Press, 2008), http://blogs
.harvard.edu/futureoft heinternet/download/.
14. Zittrain, “Facebook Could Decide an Election.”
15. Robert Epstein, “How Google Could End Democracy,” U.S. News & World
Report, June 9, 2014, http://www.usnews.com/opinion/articles/2014/06/09
/how-googles-search-rankings-could-manipulate-elections-and-end
-democracy; Robert Epstein, “How Google Could Rig the 2016 Election,”
Politico (August 19, 2015), http://www.politico.com/magazine/story/2015/08
/how-google-could-rig-the-2016-election-121548.
16. Epstein, “How Google Could End Democracy.”
17. Ibid.
18. Jack Smith, “Uber Is Using Its App to Rally New Yorkers to Protest Their
Own Mayor,” Tech.Mic (July 16, 2015), http://mic.com/articles/122419/uber
-rallies-new-yorkers-to-protest-mayor-bill-de-blasio#.OHyU15P1r.
19. Zittrain, “Facebook Could Decide an Election without Anyone Ever Finding
Out.”
20. “Facebook and Your Brain: The Inside Dope on Facebook,” Psychology Today,
Copyright © 2016. Harvard University Press. All rights reserved.

May 24, 2012, https://www.psychologytoday.com/blog/vitality/201205/face

book-and-your-brain.
21. Warren and Brandeis in their seminal article on privacy argued that
“modern enterprise and invention have through invasions upon his privacy,
subjected him to mental pain and distress, far greater than could be infl icted
by mere bodily injury.” See further, Samuel D. Warren and Louis D. Brandeis,
“The Right to Privacy,” Harvard Law Review 4, no. 5 (December 15, 1890),
http://groups.csail.mit.edu/mac/classes/6.805/articles/privacy/Privacy_brand
_warr2.html.

22. George Orwell, 1984 (1949), chap. 1, https://ebooks.adelaide.edu.au/o/orwell

/george/o79n/chapter1.1.html.
23. President’s Commission on Law Enforcement and Administration of Justice,
The Challenge of Crime in a Free Society (Washington, D.C.: U.S. Govern-
ment Printing Office, 1979), 202, https://www.ncjrs.gov/pdffi les1/nij/42.pdf.
24. Ibid.
25. Jay Greene and Matthias Verbergt, “Microsoft Cuts Low-End Phones,” Wall
Street Journal, May 19, 2016, B1. The proposed acquisition of LinkedIn will
give Microsoft access to the leading professional social network with more
than 430 million members.
26. Jack Nicas, “Google Touts New AI-Powered Tools,” Wall Street Journal,
May 19, 2016, B1, B4.
27. Ibid.
28. Yadron, “Google Assistant Takes on Amazon and Apple.”

18 • To Regulate or Not to Regulate

1. See, for example, Credit Suisse Research Institute, Global Wealth Report 2014
(October 2014), https://publications.credit-suisse.com/tasks/render/file/?fileID
= 60931FDE-A2D2-F568-B041B58C5EA591A4 (finding that the U.K. was the
only country in the G7 to have recorded rising inequality in the twenty-first
century); Organisation for Economic Co-operation and Development, In It To-
gether: Why Less Inequality Benefits All (Paris: Organisation for Economic Co-
operation and Development, 2015), http://dx.doi.org/10.1787/9789264235120-en
(suggesting the gap between the rich and poor “keeps growing”).
2. F. A. Hayek, The Road to Serfdom (Chicago: University of Chicago Press,
2007), 85.
3. Jeff ry A. Frieden, Global Capitalism: Its Fall and Rise in the Twentieth
Century (New York: W. W. Norton, 2007), 204.
4. Ibid., 215.
5. F. A. Hayek, “The Use of Knowledge in Society,” American Economic Review
Copyright © 2016. Harvard University Press. All rights reserved.

35, no. 4 (September 1945): 519–530.

6. F. A. Hayek, “Competition as a Discovery Procedure” (Marcellus S. Snow,
trans.), Quarterly Journal of Austrian Economics 5, no. 3 (2002): 11, https://
mises.org/sites/default/fi les/qjae5_ 3_ 3.pdf.
7. Ibid.
8. Campbell R. Harvey, “Financial Glossary: Efficient Market Hypothesis,”
Nasdaq (2011), http://www.nasdaq.com/investing/glossary/e/efficient-market
-hypothesis.
9. W. Paul Cockshott and Allin F. Cottrell, “Information and Economics:
A Critique of Hayek,” Research in Political Economy 16 (1997): 177–202.

10. Hayak, “The Use of Knowledge in Society,” 519–530.

11. Hayek, “Competition as a Discovery Procedure,” 9.
12. Ibid., 10.
13. The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred
Nobel 1975, Leonid Vitaliyevich Kantorovich, Tjalling C. Koopmans,
“Mathematics in Economics: Achievements, Difficulties, Perspectives,”
http://www.nobelprize.org/nobel _prizes/economic-sciences/laureates/1975
/kantorovich-lecture.html.
14. Travis Kalanick, “NYE Surge Pricing Explained,” Uber (December 31, 2011),
http://newsroom.uber.com/2011/12/nye-surge-pricing-explained/; see also
Annie Lowrey, “Is Uber’s Surge-Pricing an Example of High-Tech Gouging?”
New York Times, January 10, 2014, http://www.nytimes.com/2014/01/12
/magazine/is-ubers-surge-pricing-an-example-of-high-tech-gouging.html;
Uber’s CEO defended its surge pricing: “Higher prices are required in order
to get cars on the road and keep them on the road during the busiest times.
This maximizes the number of trips and minimizes the number or people
stranded. The drivers have other options as well. In short, without Surge
Pricing, there would be no car available at all.”
15. Nicholas Diakopoulos, “How Uber Surge Pricing Really Works,” Washington
Post, April 17, 2015, http://www.washingtonpost.com/news/wonkblog/wp
/2015/04/17/how-uber-surge-pricing-really-works/.
16. Min Kyung Lee, Daniel Kusbit, Evan Metsky, and Laura Dabbish, “Working
with Machines: The Impact of Algorithmic and Data-Driven Management
on Human Workers,” Proceedings of the 33rd Annual ACM Conference on
Human Factors in Computing Systems (New York: ACM, 2015), http://www
.cs.cmu.edu/~mklee/materials/Publication/2015-CHI_ algorithmic
_management.pdf.
17. Ibid.
18. Uber, Interested in Driving with Uber? https://get.uber.com/drive/.
19. John Kenneth Galbraith, The Essential Galbraith (Boston: Mariner Books,
2010), 72.
Copyright © 2016. Harvard University Press. All rights reserved.

20. Ibid.
21. Eden Medina, Cybernetic Revolutionaries: Technology and Politics in
Allende’s Chile (Cambridge, MA: MIT Press, 2011).
22. Evgeny Morozov, “The Planning Machine: Project Cybersyn and the Origins
of the Big Data Nation,” New Yorker, October 13, 2014, http://www.newyorker
.com/magazine/2014/10/13/planning-machine.
23. Eden Medina, “The Cybersyn Revolution,” Jacobin 17 (Spring 2015),
https://www.jacobinmag.com/2015/04/allende-chile-beer-medina-cybersyn/.
24. Eden Medina, “Designing Freedom, Regulating a Nation: Socialist Cyber-
netics in Allende’s Chile,” Journal of Latin American Studies 38 (2006):

571–606, http://www.informatics.indiana.edu/edenm/EdenMedinaJLAS
August2006.pdf.
25. Laura Tam, “Smart Cities, Limited Resources,” SPUR (October 10, 2012),
http://www.spur.org/publications/article/2012-10-10/smart-cities-limited
-resources.
26. Ibid.
27. San Francisco Municipal Transportation Agency, SFpark Sensors (2016),
http://sfpark.org/how-it-works/the-sensors/.
28. San Francisco Municipal Transportation Agency, SFpark Pilot Project
Evaluation Summary (June 2014), http://sfpark.org/wp-content/uploads/2014
/06/SFpark _ Eval _ Summary_ 2014.pdf.
29. San Francisco Municipal Transportation Agency, SFpark Pricing (2016),
http://sfpark.org/how-it-works/pricing/.
30. San Francisco Municipal Transportation Agency, SFpark Sensors.
31. San Francisco Municipal Transportation Agency, SFpark Pilot Project
Evaluation Summary.
32. Ibid. “In SFpark pilot areas, the amount of time most people reported that
it took to find a space decreased by 43 percent, compared to a 13 percent
decrease in control areas.”
33. Ibid. “Drivers generated 7 metric tons of greenhouse gas emissions per day
looking for parking in pilot areas. This dropped by 30 percent by 2013,
compared to a decrease of 6 percent in control areas.”
34. Ibid. “SFpark encouraged people to drive at non-peak times and improved
parking availability when it mattered most. On-street parking availability
improved by 22 percent during peak periods, compared to 12 percent during
off-peak. In SFpark garages, morning peak entries rose 1 percent while
off-peak entries rose 14 percent, and evening peak exits rose 3 percent
while off-peak exits rose 15 percent. This suggests that SFpark helped to
reduce peak-period congestion, which makes the roads flow more smoothly
for drivers and transit.”
35. Ibid. “In both pilot and control areas, where parking availability improved,
Copyright © 2016. Harvard University Press. All rights reserved.

traffic volume decreased by approximately 8 percent, compared to a

4.5 percent increase in areas where parking availability worsened.”
36. Ibid. “As a result of less circling, pilot areas saw a 30 percent decrease in
vehicle miles traveled from 8,134 miles per day in 2011 to 5,721 miles per day
by 2013. Control areas saw a 6 percent decrease.”
37. Ibid. “In pilot areas, double parking decreased by 22 percent versus a 5 percent
decrease in control areas.”
38. Michael Emmett Brady, “Comparing J. M. Keynes’s and F. Von Hayek’s
Differing Definitions of Uncertainty as It Relates to Knowledge: Keynes’s
Unavailable or Missing Knowledge Concept versus Hayek’s Dispersal of

Knowledge Concept,” International Journal of Applied Economics and

Econometrics 19, no. 3 (January 2011), http://ssrn.com/abstract=1751569.
39. Ibid.
40. Steve Lohr, “Can Apple Find More Hits Without Its Tastemaker?” New York
Times, January 18, 2011, http://www.nytimes.com/2011/01/19/technology
/companies/19innovate.html?_r = 0.
41. Peter Noel Murray, “How Steve Jobs Knew What You Wanted,” Psychology
Today, October 13, 2011, https://www.psychologytoday.com/blog/inside-the
-consumer-mind/201110/how-steve-jobs-knew-what-you-wanted.
42. Sara Stefanini, “Think Tank Urges FERC to Reform Merger Policies,” Law360
(March 15, 2007), http://competition.law360.com/Secure/ViewArticle.aspx
?id=20553.
43. François Moreau, “The Role of the State in Evolutionary Economics,”
Cambridge Journal of Economics 28 (2004): 847, 850.
44. Adrienne LaFrance, “People’s Deepest, Darkest Google Searches Are Being
Used against Them: On the Internet, Search Queries Are Used to Target
Vulnerable Consumers,” The Atlantic, November 3, 2015, http://www
.theatlantic.com/technology/archive/2015/11/google-searches-privacy
-danger/413614/.
45. Ibid.
46. U.S. Bureau of Labor Statistics, Consumer Expenditure Survey (September
2015), http://www.bls.gov/cex /2014/standard/multiyr.pdf.

19 • The Enforcement Toolbox

1. House of Lords, Select Committee on European Union, “Online Platforms

and the Digital Single Market” (April 20, 2016), 10th Report of Session
2015–16, para. 373, http://www.publications.parliament.uk /pa/ld201516
/ldselect/ldeucom/129/129.pdf.
2. For more on the book, see Michael Lewis, “About the Author” (2014),
http://michaellewiswrites.com/index.html#top.
Copyright © 2016. Harvard University Press. All rights reserved.

3. In addition, note the lack of predictive accuracy that characterizes tacit

collusion models and may further chill the agencies’ willingness to inter-
vene. See generally Nicolas Petit, “The ‘Oligopoly Problem’ in EU Competi-
tion Law,” in Research Handbook in European Competition Law, Ioannis
Liannos and Damien Geradin, eds. (Cheltenham: Edward Elgar, 2013), 259.
4. See, for instance, Ramsi Woodcock, “Inconsistency in Antitrust,” University
of Miami Law Review 68 (2013), http://ssrn.com/abstract =2514030; Ariel
Ezrachi and David Gilo, “Excessive Pricing, Entry, Assessment and
Investment—Lessons from the Mittal Litigation,” Antitrust Law Journal 76,
no. 3 (2010): 873–898; Ariel Ezrachi and David Gilo, “Are Excessive Prices

Really Self-Correcting?” Journal of Competition Law & Economics 5, no. 2

(2009): 249–268.
5. United States v. Microsoft Corp., 253 F.3d 34 (D.C. Cir. 2001); Case T-201/04,
Microsoft Corp. v. Comm’n, 2007 E.C.R. II-3601.
6. European Commission, Guidance on the Commission’s Enforcement Priori-
ties in Applying Article 82 of the EC Treaty to Abusive Exclusionary Conduct
by Dominant Undertakings (February 24, 2009), http://eur-lex.europa.eu
/legal-content/EN/TXT/PDF/?uri= CELEX:52009XC0224(01)& from=EN;
“The emphasis of the Commission’s enforcement activity in relation to
exclusionary conduct is on safeguarding the competitive process in the
internal market and ensuring that undertakings which hold a dominant
position do not exclude their competitors by other means than competing on
the merits of the products or ser vices they provide. In doing so the Commis-
sion is mindful that what really matters is protecting an effective competitive
process and not simply protecting competitors.” U.S. Department of Justice,
Single-Firm Conduct and Section 2 of the Sherman Act: An Overview (June 25,
2015), http://www.justice.gov/atr/competition-and-monopoly-single-firm
-conduct-under-section-2-sherman-act-chapter-1; “Section 2 thus aims
neither to eradicate monopoly itself, nor to prevent firms from exercising the
monopoly power their legitimate success has generated, but rather to protect
the process of competition that spurs firms to succeed. The law encourages all
firms—monopolists and challengers alike—to continue striving.”
7. See J. McCarthy and P. Hayes, “Some Philosophical Problems from the
Standpoint of Artificial Intelligence,” Machine Intelligence 4 (1969): 463–
505; G. F. Luger, Artificial Intelligence: Structures and Strategies for Complex
Problem Solving, 5th ed. (New York: Addison-Wesley, 2005), chap. 1 and 17.
8. Claire Cain Miller, “When Algorithms Discriminate,” New York Times, July 9,
2015, http://www.nytimes.com/2015/07/10/upshot/when-algorithms
-discriminate.html?_r = 0.
9. United States v. Ulbricht, 31 F. Supp. 3d 540, 559 (S.D.N.Y. 2014).
10. Verizon Commc’ns Inc. v. Law Offices of Curtis V. Trinko, LLP, 540 U.S. 398,
Copyright © 2016. Harvard University Press. All rights reserved.

415 (2004), quoting Phillip Areeda, Essential Facilities: An Epithet in Need of

Limiting Principles, 58 Antitrust L.J. 841, 853 (1989).
11. For a discussion of the susceptibility of competition law, see Ariel Ezrachi,
“Sponge,” Oxford Legal Studies Research Paper No. 16/2015 (March 1, 2015),
http://ssrn.com/abstract=2572028; Harry First and Spencer Weber Waller,
“Antitrust’s Democracy Deficit,” Fordham Law Review 81 (2013): 2543, 2544
n.5 (“We take as a given that antitrust has political goals and reflects political
value judgments”); John B. Kirkwood, “The Essence of Antitrust: Protecting
Consumers and Small Suppliers from Anticompetitive Conduct,” Fordham
Law Review 81 (2013): 2425, 2453 (addressing and critiquing total welfare

standard); Robert H. Lande, “A Traditional and Textualist Analysis of the

Goals of Antitrust: Efficiency, Preventing Theft from Consumers, and Con-
sumer Choice,” Fordham Law Review 81 (2013): 2349, 2360 n.54 (noting Bork’s
“deceptive use of the term ‘consumer welfare,’ instead of the more honest term
‘total welfare,’ was a brilliant way to market the efficiency objective”); Barak
Orbach, “How Antitrust Lost Its Goal,” Fordham Law Review 81 (2013): 2253,
2273 (noting that “[f]or Bork, the phrase ‘consumer welfare’ meant ‘allocative
efficiency’ ” but a “few years after Bork presented his thesis of the legislative
intent of the Sherman Act, the phrase ‘consumer welfare’ acquired a popular
[and different] cultural meaning referring to the buyer’s well being: the benefits
a buyer derives from the consumption of goods and services, or more casually,
the individual’s well being”); Maurice E. Stucke, “Should Competition Policy
Promote Happiness?” Fordham Law Review 81 (2013): 2575.
12. U.S. Department of Justice, Competition and Monopoly: Single-Firm
Conduct under Section 2 of the Sherman Act (2008), https://www.justice.gov
/atr/competition-and-monopoly-single-fi rm-conduct-under-section-2
-sherman-act.
13. In their statement, three FTC commissioners “strongly distance[d them-
selves] from the enforcement positions stated in the Report.” Pamela Jones
Harbour et al., Statement of Commissioners Harbour, Leibowitz and Rosch
on the Issuance of the Section 2 Report by the Department of Justice (2008), 5,
http://www.ftc.gov/os/2008/09/080908section2stmt.pdf. Moreover, the three
commissioners vowed that they were ready “to fi ll any Sherman Act
enforcement void that might be created if the Department actually imple-
ments the policy decisions expressed in its Report.” Ibid., 11.
14. U.S. Department of Justice, Justice Department Withdraws Report on
Antitrust Monopoly Law: Antitrust Division to Apply More Rigorous Standard
with Focus on the Impact of Exclusionary Conduct on Consumers, Press
Release (May 11, 2009), http://www.justice.gov/opa/pr/justice-department
-withdraws-report-antitrust-monopoly-law.
15. U.S. Department of Justice, “Justice Department Reaches Settlement with
Copyright © 2016. Harvard University Press. All rights reserved.

Texas Hospital Prohibiting Anticompetitive Contracts with Health Insurers

Department Says United Regional’s Contracts Unlawfully Maintain
Monopoly Power,” press release (February 25, 2011), https://www.justice.gov
/opa/pr/justice-department-reaches-settlement-texas-hospital-prohibiting
-anticompetitive-contracts.
16. See Maurice E. Stucke and Allen P. Grunes, Big Data and Competition Policy
(Oxford: Oxford University Press, 2016).
17. Barry C. Lynn, “Amazon’s Book Monopoly: A Threat to Freedom of Expres-
sion?” New America (January 27, 2016), https://www.newamerica.org/open
-markets/amazons-book-monopoly/.

18. Margrethe Vestager, Competition in a Big Data World (Munich: European

Commission, January 17. 2016), https://ec.europa.eu/commission/2014-2019
/vestager/announcements/competition-big-data-world _en.
19. Take, for example, the powers of the U.K. Competition and Markets
Authority to initiate market investigations, gather and appraise evidence,
and, where necessary, impose structural or behavioral remedies; Competi-
tion Commission, Guidelines for Market Investigations: Their Role, Proce-
dures, Assessment and Remedies, CC3 (Revised) (April 2013), https://www
.gov.uk /government/uploads/system/uploads/attachmen _data/file/284390
/cc3_revised.pdf (adopted by the CMA Board). Other authorities possess
similar, albeit more limited, powers to engage in an investigation into sectors
of the economy, following which they may publish reports and initiate
action. See, for example, the European Commission’s power to carry out
sector inquiries pursuant to Article 17 of Regulation 1/2003. Noteworthy is
the commission’s sector inquiry into e-commerce in the EU, launched on
May 6, 2015. The commission plans to publish its preliminary inquiry report
in mid-2016.
20. Another instrument that may be used to address the asymmetric bargaining
power between suppliers (such as app developers) and gatekeepers (such as
super platforms) is the possible use of codes of practice. Such codes could be
designed by the state and provide a voluntary or binding framework, which
can be tailored to specific market realities. Such instruments could ensure
that “all parts of the supply chain were treated equally,” and prevent online
platforms from engaging in “unfair trading practices in their dealings with
SME suppliers.” A code of practice could be designed in collaboration with
the industry and might include a business-to-business dispute resolution
mechanism. It could take into account many of the dynamics discussed in
our Frenemy scenario, consider them in specific market context, and address
them. Written evidence to the House of Lords from IMPALA, and Com-
ments by Hon. Ed Vaizey MP. House of Lords, Select Committee on Euro-
pean Union, “Online Platforms and the Digital Single Market,” para. 129,
Copyright © 2016. Harvard University Press. All rights reserved.

130, 133.
21. Douglass C. North, Understanding the Process of Economic Change
(Princeton, NJ: Princeton University Press, 2005), 67.
22. Stucke and Grunes, Big Data and Competition Policy.
23. Yannis Bakos, Florencia Marotta-Wurgler, and David R. Trossen, “Does
Anyone Read the Fine Print? Consumer Attention to Standard-Form
Contracts,” Journal of Legal Studies 43, no. 1 (January 2014); Florencia
Marotta-Wurgler, “Consumer Behav ior and Disclosure in Online Con-
tracts,” presentation at the Conference on Behavioral Industrial Organ-

ization and Consumer Protection, University College London,

October 18, 2014.
24. Alex Chisholm (CMA chief executive), Why “Sleepers” Can’t Always Be
Left to “Sleep,” CCRP 2016 Competition Policy Roundtable (London:
Competition Markets Authority, January 25, 2016), https://www.gov.uk
/government /speeches/alex-chisholm-on-consumer-engagement-in-a
-digital-world.
25. As the FTC summarizes the Act: “The primary goal of COPPA is to place
parents in control over what information is collected from their young
children online. The Rule was designed to protect children under age 13
while accounting for the dynamic nature of the Internet. The Rule applies to
operators of commercial websites and online ser vices (including mobile
apps) directed to children under 13 that collect, use, or disclose personal
information from children, and operators of general audience websites or
online ser vices with actual knowledge that they are collecting, using, or
disclosing personal information from children under 13. The Rule also
applies to websites or online ser vices that have actual knowledge that they
are collecting personal information directly from users of another website or
online ser vice directed to children. Operators covered by the Rule must: Post
a clear and comprehensive online privacy policy describing their informa-
tion practices for personal information collected online from children;
Provide direct notice to parents and obtain verifiable parental consent, with
limited exceptions, before collecting personal information online from
children; Give parents the choice of consenting to the operator’s collection
and internal use of a child’s information, but prohibiting the operator from
disclosing that information to third parties (unless disclosure is integral to
the site or ser vice, in which case, this must be made clear to parents); Provide
parents access to their child’s personal information to review and/or have the
information deleted; Give parents the opportunity to prevent further use or
online collection of a child’s personal information; Maintain the confidenti-
ality, security, and integrity of information they collect from children,
Copyright © 2016. Harvard University Press. All rights reserved.

including by taking reasonable steps to release such information only to

parties capable of maintaining its confidentiality and security; and Retain
personal information collected online from a child for only as long as is
necessary to fulfi ll the purpose for which it was collected and delete the
information using reasonable measures to protect against its unauthorized
access or use.” Federal Trade Commission, Complying with COPPA: Fre-
quently Asked Questions (March 20, 2015), https://www.ftc.gov/tips-advice
/business-center/guidance/complying-coppa-frequently-asked-questions
#General%20Questions.

26. T-201/04, Microsoft Corp v. Commission, Court of First Instance (September

17, 2007).
27. Eu ropean Commission, Agreement on Commission’s EU Data Protection
Reform Will Boost Digital Single Market (Brussels: Eu ropean Commission,
December 15, 2015), http://europa .eu /rapid /press-release _ IP-15 - 6321_ en
.htm. The new regulation followed the recognition that “Rapid technolog-
ical developments have brought new challenges for the protection of personal
data. The scale of data sharing and collecting has increased dramatically.
Technology allows both private companies and public authorities to make
use of personal data on an unprecedented scale in order to pursue their
activities.” Regulation of the European Parliament and of the Council on the
Protection of Natural Persons with Regard to the Processing of Personal Data
and on the Free Movement of Such Data, and Repealing Directive 95/46/EC
(General Data Protection Regulation), Brussels (April 27, 2016), http://data
.consilium.europa .eu /doc/document /PE-17-2016 -INIT/en/pdf.
28. Article 83, EU Data Protection Regulation (“up to 20 000 000 EUR, or in
the case of an undertaking, up to 4 % of the total worldwide annual
turnover”).
29. Samuel Gibbs, “EU Agrees Draft Text of Pan-European Data Privacy Rules,”
The Guardian, December 16, 2015, http://www.theguardian.com/technology
/2015/dec/16/eu-agrees-draft-text-pan-european-data-privacy-rules.
30. European Commission, Agreement on Commission’s EU Data Protection
Reform Will Boost Digital Single Market.
31. Pedro Domingos, “Get Ready for Your Digital Model: Algorithms Will
Build Data-Driven Alter Egos for Us That Can Do Job Interviews, Shop for
Cars and Go on Dates,” Wall Street Journal, November 12, 2015, http://www
.wsj.com /articles/get-ready-for-your-digital-model-1447351480?alg = y.
32. For discussion of the applicability of the EU data protection law to price
discrimination, see Frederik Zuiderveen Borgesius, “Online Price Discrimi-
nation and Data Protection Law,” Amsterdam Law School Research Paper
No. 2015-32 (August 28, 2015), http://ssrn.com/abstract =2652665.
Copyright © 2016. Harvard University Press. All rights reserved.

33. Simon Birch, “Collective Buying: The Emergence of a New Co-Operative

Movement,” The Guardian, June 15, 2002, http://www.theguardian.com
/social-enterprise-network /2012/jun/15/collective-buying-big-switch
-cooperative-movement.
34. Martin Lewis, “Group Buying Is NOT Collective Purchasing,” MoneySaving
Expert.com (February 10, 2011), http://blog.moneysavingexpert.com/2011/02
/10/group-buying-is-not-collective-purchasing/.
35. Patrick Collinson, “Group-Buying—Does It Deliver?” The Guardian,
January 29, 2011, http://www.theguardian.com/money/2011/jan/29/group
-buying-does-it-deliver.

36. See, for example, Autoebid, Reverse Auctions (2016), https://www.autoebid

.com/reverse-auctions.asp, or Legal BenchMarket International.
37. As an example, in Petroleum Products, the defendant oil companies publicly
announced, at times in advance of the effective date, the discounts (or
decisions to withdraw discounts) to their franchisee gasoline stations;
Coordinated Pretrial Proceedings in Petroleum Products Antitrust Litigation,
906 F.2d 432, 445 (9th Cir. 1990). The public dissemination of the discount
information was of little value to the defendants’ franchisees or the end
consumer. The franchisees could not shop around for the best oil prices; they
could only purchase from their franchisor. Nor did the consumers care what
the gas station paid for the gasoline. They cared only about the retail price. The
purpose and effect, then, of publicly announcing changes in discounts to the
franchisees were, as several defendants’ executives admitted, to quickly
inform their competitors of the price change, in the express hope that these
competitors would follow the move and align their prices. Without such
transparency, the other defendants might not have readily detected one
defendant’s withdrawal of its discount and followed accordingly, because the
individual branded gas stations’ retail prices varied considerably.
38. President’s Council of Advisors on Science and Technology, Big Data and
Privacy: A Technological Perspective (Washington, DC: Executive Office of
the President, May 2014), x, https://www.whitehouse.gov/sites/default/fi les
/microsites/ostp/PCAST/pcast _big _data _ and _privacy_-_may_ 2014.pdf;
Organisation for Economic Co-operation and Development, Exploring
Data-Driven Innovation as a New Source of Growth: Mapping the Policy
Issues Raised by “Big Data” (Paris: Organisation for Economic Co-operation
and Development, June 18, 2003), 12, http://www.oecd.org/officialdocuments
/publicdisplaydocumentpdf/?cote=DSTI/ICCP(2012)9/FINAL&docLanguage
=En, observing that “In some cases, big data is defined by the capacity to
analyse a variety of mostly unstructured data sets from sources as diverse as
web logs, social media, mobile communications, sensors and financial
transactions. This requires the capability to link data sets; this can be
Copyright © 2016. Harvard University Press. All rights reserved.

essential as information is highly context-dependent and may not be of value

out of the right context. It also requires the capability to extract information
from unstructured data, i.e. data that lack a predefined (explicit or implicit)
model.”
39. Stanford Graduate School of Business Staff, “Sharing Information to Boost
the Bottom Line,” Insights by Stanford Business (March 1, 1999), http://www
.gsb.stanford.edu/insights/sharing-information-boost-bottom-line.

Final Reﬂections

1. Ludwig von Mises, Bureaucracy, Bettina Bien Greaves, ed. (Indianapolis:

Liberty Fund, 2007 [1944]), 17.
2. Federal Trade Commission and U.S. Department of Justice, Antitrust
Guidelines for Collaborations among Competitors (April 2000),
https://www.ftc .gov/sites/default /fi les/documents/public _ events/joint
-venture-hearings-antitrust-guidelines-collaboration-among-competitors
/ftcdojguidelines-2 .pdf.
3. See paragraph 204 of the House of Lords, Select Committee on European
Union, “Online Platforms and the Digital Single Market” (20 April 2016)
10th Report of Session 2015–16, http://www.publications.parliament.uk /pa
/ld201516/ldselect/ldeucom/129/129.pdf.
4. Facebook, Form 10-K for the Fiscal Year Ended December 31, 2014
(2015), 33.
5. Statistics Portal, Facebook’s Average Revenue per User from 2010 to 2014, by
Region (in U.S. Dollars), http://www.statista.com/statistics/251328/facebooks
-average-revenue-per-user-by-region/.
6. Digital Strategy Consulting, “How Much Are You Worth? Average Revenue
per User at Google, Facebook and Twitter” (June 18, 2014), http://www
.digitalstrategyconsulting.com/intelligence/2014/06/ad _revenue _per_user
_ google _facebook _twitter.php.
7. Google’s net income in 2014 was $14.4 billion. Google, Form 10-K for the
Fiscal Year Ended December 31, 2014, 22. Facebook’s net income was
$2.925 billion. Facebook, Form 10-K for the Fiscal Year Ended De-
cember 31, 2014, 30.
8. Vindu Goel, “Flipping the Switches on Facebook’s Privacy Controls,” New
York Times, January 29, 2014, http://www.nytimes.com/2014/01/30/technology
/personaltech/on-facebook-deciding-who-knows-youre-a-dog.html?_r =1.
9. Facebook, Form 10-K for the Fiscal Year Ended December 31, 2014, 9.
10. Deepa Seetharaman, “Facebook Prods Users to Share a Bit More,” Wall Street
Copyright © 2016. Harvard University Press. All rights reserved.

Journal, November 2, 2015, http://www.wsj.com/articles/facebook-prods

-users-to-share-a-bit-more-1446520723.
11. Ibid.
12. Ibid.
13. Jacob Weisberg, “We Are Hopelessly Hooked,” New York Review of Books,
February 25, 2016, 9.
14. Organisation for Economic Co-operation and Development, Data-Driven
Innovation for Growth and Well-Being: Interim Synthesis Report (Paris:
Organisation for Economic Co-operation and Development, October 2014),
29, http://www.oecd.org/sti/inno/data-driven-innovation-interim-synthesis

.pdf; reference within quotation is to C. Shapiro and H. R. Varian, Informa-

tion Rules: A Strategic Guide to the Network Economy (Cambridge, MA:
Harvard Business Press, 1999).
15. Case T-201/04, Microsoft Corp. v. Commission, 2007 E.C.R. II-3601 (Court of
First Instance), para. 1061.
16. “Business in America: Too Much of a Good Thing: Profits Are Too High.
America Needs a Giant Dose of Competition,” The Economist, March 26,
2016, http://www.economist.com/node/21695385/print.
17. For the potential data-driven network effects, see Maurice E. Stucke and
Allen P. Grunes, Big Data and Competition Policy (Oxford: Oxford Univer-
sity Press, 2016).
18. Brad Brown, Michael Chui, and James Manyika, “Are You Ready for the Era
of ‘Big Data’?” McKinsey Quarterly (October 2011), 2, http://www.t-systems
.com/solutions/download-mckinsey-quarterly-/1148544 _1/blobBinary/Study
-McKinsey-Big-data.pdf.
19. Maurice E. Stucke and Ariel Ezrachi, “When Competition Fails to Optimize
Quality: A Look at Search Engines,” Yale Journal of Law & Technology 18
(2016): 70.
20. Organisation for Economic Co-operation and Development, Data-Driven
Innovation for Growth and Well-Being, 29; see also Federal Trade Commis-
sion, Google Inc., File No. 111-0163 (August 8, 2012) (published by the Wall
Street Journal), 76 (discussing this “virtuous cycle” and how it represents a
“significant barrier for any potential entrant”).
21. Executive Office of the President, Big Data and Differential Pricing
(Washington, DC: Executive Office of the President, February 2015),
https://www.whitehouse.gov/sites/default /fi les/docs/Big _ Data _ Report
_ Nonembargo_v2 .pdf .
22. According to the study, “differential pricing based on demographics (whereby
Netflix would adjust prices based on a customer’s race, age, income, geographic
location, and family size) could increase profit by 0.8 percent, while using 5,000
Web browsing variables (such as the amount of time a user typically spends
Copyright © 2016. Harvard University Press. All rights reserved.

online or whether she has recently visited Wikipedia or IMDB) could increase
profits by as much as 12.2 percent.” Ibid., citing Benjamin Shiller, “First-Degree
Price Discrimination Using Big Data” (2014), http://benjaminshiller.com
/images/First_Degree_PD_Using _Big _Data_Apr_8,_2014.pdf.
23. For a discussion of the “nowcasting radar,” see Stucke and Grunes, Big Data
and Competition Policy.
24. Yoko Kubota, “Toyota Aims to Make Self-Driving Cars by 2020,” Wall Street
Journal, October 6, 2015, http://www.wsj.com/articles/toyota-aims-to-make
-self-driving-cars-by-2020 -1444136396; Yoko Kubota, “Behind Toyota’s
Late Shift into Self-Driving Cars,” Wall Street Journal, January 12, 2016,

http://www.wsj.com/articles/behind-toyotas-late-shift-into-self-driving-cars
-1452649436 (“In the battle for global pre-eminence, traditional car makers
fear soft ware makers will steal the auto’s soul and profitability, putting
incumbents in a similar position to Chinese factories making smartphones
for global brands.”).
25. Joseph Menn, “Data Collection Arms Race Feeds Privacy Fears,” Reuters
(February 19, 2012), http://www.reuters.com/article/us-data-collection
-idUSTRE81I0AP20120219.
26. Evgeny Morozov, “Socialize the Data Centres!” New Left Review, January–
February 2015, http://newleft review.org/II/91/evgeny-morozov-socialize-the
-data-centres.
27. Oxfam, “David Cameron: End the Era of Tax Havens So That We Can End
Poverty” (2016), https://act.oxfam.org/great-britain/tax-havens-2016- 644e5810
-f58e-40f5-8162-d09b2392efa6?sid=2016-01-18 _ogbsite _ homepage.
28. Graeme Warden, “Oxfam: 85 Richest People as Wealthy as Poorest Half of
the World,” The Guardian, January 20, 2016, http://www.theguardian.com
/business/2014/jan/20/oxfam-85-richest-people-half-of-the-world (quoting
Winnie Byanyima).
29. “Business in America,” The Economist; Jonathan B. Baker and Steven C.
Salop, “Antitrust, Competition Policy, and Inequality,” Georgetown Law
Journal 104 (2015): 1–28, http://scholarship.law.georgetown.edu/facpub/1462
/; Greg Ip, “Behind Rising Inequality: More Unequal Companies,” Wall
Street Journal, November 4, 2015, http://www.wsj.com/articles/behind-rising
-inequality-more-unequal-companies-1446665769. (“Mounting evidence
suggests the prime driver of wage inequality is the growing gap between the
most- and least-profitable companies, not the gap between the highest- and
lowest-paid workers within each company. That suggests policies that have
focused on individuals, from minimum wages to education, may not be
enough to close the pay gap; promoting competition between companies
such as through antitrust oversight may also be impor tant.”)
30. The White House, Office of the Press Secretary, “Executive Order—Steps
Copyright © 2016. Harvard University Press. All rights reserved.

to Increase Competition and Better Inform Consumers and Workers to

Support Continued Growth of the American Economy” (April 15, 2016),
available at https://www.whitehouse.gov/the-press-office/2016/04/15
/executive-order-steps-increase-competition-and-better-inform
-consumers.
31. Council of Economic Advisers, “Benefits of Competition and Indicators of
Market Power,” Issue Brief (May 2016), https://www.whitehouse.gov/sites
/default/fi les/page/fi les/20160502 _competition _ issue _brief _updated _cea
.pdf.
32. Ibid., 14; “Competition policies and robust reaction to market power abuses
can be an impor tant way in which the government makes sure the market

provides the best outcomes for society with respect to choice, innovation,
and price as well as fair labor and business markets.”
33. Ibid., 12–13. One specific concern is whether “big data” is a “critical resource,
without which new entrants might have a difficult time marketing to or
other wise attracting customers.” Ibid., 13.
34. UN Secretary General’s Independent Expert Advisory Group on a Data
Revolution for Sustainable Development, A World That Counts: Mobilising
the Data Revolution for Sustainable Development (United Nations, November
2014), 7, http://en.unesco.org/un-sab/system/fi les/Data%20Revolution%20
Report%20A-World-That-Counts _0.pdf.
35. Joseph E. Stiglitz, “Information and the Change in the Paradigm in Eco-
nomics,” American Economic Review 92, no. 3 (June 2002): 460–501.
36. Nathan Newman, “The Costs of Lost Privacy: Consumer Harm and Rising
Economic Inequality in the Age of Google,” William Mitchell Law Review 40,
no. 2 (2014), http://ssrn.com/abstract=2310146.
37. Jennifer Valentino-Devries, Jeremy Singer-Vine, and Ashkan Soltani,
“Websites Vary Prices, Deals Based on Users’ Information,” Wall Street
Journal, December 24, 2012, http://www.wsj.com/articles/SB10001424127887
323777204578189391813881534 (finding that “the weighted average income
among ZIP Codes that mostly received discount prices was roughly $59,900,
based on Internal Revenue Ser vice data. ZIP Codes that saw generally high
prices had a lower weighted average income, $48,700”).
38. Tom Fairless and Alistair Barr, “EU Lays Groundwork for Antitrust Charges
against Google,” Wall Street Journal, April 11, 2015, http://www.wsj.com
/articles/eu-lays-groundwork-for-antitrust-charges-against-google
-1427928793.
39. Alistair Barr, “Alphabet Reports Rising Profits at Core Google Businesses,”
Wall Street Journal, February 1, 2016, http://www.wsj.com/articles/alphabet
-reports-rising-profits-at-core-google-businesses-1454361634.
40. Case No. COMP/M.7217, Facebook/Whatsapp (October 3, 2014), para. 164.
41. For examples and a critique, see Stucke and Grunes, Big Data and Competi-
Copyright © 2016. Harvard University Press. All rights reserved.

tion Policy; Maurice E. Stucke, “Better Competition Advocacy,” St. John’s

Law Review 82, no. 3 (2008): 951.
42. Adam Thierer, “Can There Be a Market for Unpaid Search Results and Could
Google Be Classified as a Public Utility?” Antitrust & Competition Policy Blog
(May 21, 2012), http://lawprofessors.typepad.com/antitrustprof_blog/2012/05
/can-there-be-a-market-for-unpaid-search-results-and-could-google-be
-classified-as-a-public-utility-c-1.html.
43. Frank Pasquale, “Beyond Innovation and Competition: The Need for
Qualified Transparency in Internet Intermediaries,” Northwestern University
Law Review 104 (2010): 105, 143.
44. Stucke and Grunes, Big Data and Competition Policy.

45. Sapienza Paola and Luigi Zingales, “Trust and Finance,” NBER Reporter
2 (2011): 16 (“For the development of anonymous markets, though, what
matters is generalized trust: the trust that people have in a random member
of an identifiable group”); Lynn A. Stout, “Trust Behav ior: The Essential
Foundation of Securities Markets,” in Behavioral Finance: Investors, Corpo-
rations, and Markets, H. Kent Baker and John R. Nofsinger, eds. (Hoboken,
NJ: Wiley, 2010), 513 (“Faith—or more accurately, trust—is the foundation
on which successful public securities markets are built”); see also Thomas J.
Horton, “The Coming Extinction of Homo Economicus and the Eclipse of
the Chicago School of Antitrust: Applying Evolutionary Biology to Struc-
tural and Behavioral Antitrust Analyses,” Loyola University Chicago Law
Journal 42 (2011): 474, 476, 502, 520 (arguing that fundamental human
values of fairness and reciprocity not only enhance trust but create a
healthier, more stable, more efficient economic ecosystem); Stephen Knack
and Philip Keefer, “Does Social Capital Have an Economic Payoff ? A
Cross-Country Investigation,” Quarterly Journal of Economics 112, no. 4
(November 1997):1251, 1252, 1260 (regression analysis of a twenty-nine-
market economy sample suggests that trust and civic cooperation are
associated with stronger economic performance); Stephan M. Wagner, Linda
Silver Coley, and Eckhard Lindemann, “Effects of Suppliers’ Reputation on
the Future of Buyer-Supplier Relationships: The Mediating Roles of Outcome
Fairness and Trust,” Journal of Supply Chain Management 47 (April 2011): 42
(noting that empirical findings support other research that “trust is the most
impor tant mediator in business-to-business relationships”).
46. See Maurice E. Stucke, “Is Intent Relevant?” Journal of Law, Economics &
Policy 8 (2012): 801 (collecting studies).
47. Lynn Stout, Cultivating Conscience: How Good Laws Make Good People
(Princeton, NJ: Princeton University Press, 2010).
48. Ellen Garbarino and Sarah Maxwell, “Consumer Response to Norm-
Breaking Pricing Events in E-Commerce,” Journal of Business Research 63
(2010): 1067 (“[T]rust will be destroyed when a trusted seller does not behave
Copyright © 2016. Harvard University Press. All rights reserved.

according to the social norms of fairness”); Wagner et al., “Effects of

Suppliers’ Reputation on the Future of Buyer-Supplier Relationships,” 35
(describing literature on importance of fairness and trust in business-to-
business relationships).
49. Devesh Rustagi, Stefani Engel, and Michael Kosfeld, “Conditional Cooperation
and Costly Monitoring Explain Success in Forest Commons Management,”
Science 330 (2010): 964.
50. U.K. Competition and Markets Authority, “The Commercial Use of Con-
sumer Data: Report on the CMA’s Call for Information” (June 2015), 103
(“CMA Report”) (citing studies). In a recent global survey by the Boston

Consulting Group (“BCG Global Consumer Sentiment Survey 2013”), a

global average of 76 percent of people surveyed agreed with the proposition
that they had to be cautious about sharing their personal information online.
51. CMA Report, 103–106.
52. For example, in the United Kingdom, the Court of Appeal has recently
upheld the right of users to sue Google in tort for its alleged exploitation of a
workaround that allowed it to bypass Safari users’ privacy settings: Google
Inc. v. Vidal-Hall & Ors, [2015] EWCA Civ 311.
53. Darlene Storm, “Nissan Leaf Secretly Leaks Driver Location, Speed to
Websites,” Computer World, June 14, 2011, http://www.computerworld.com
/article/2470123/endpoint-security/nissan-leaf-secretly-leaks-drier-location
—speed-to-websites.html.
54. CMA Report, 11.
55. Ibid., 12.
56. Nir Kshetri, “Big Data’s Impact on Privacy, Security and Consumer Welfare,”
Telecommunications Policy 38, no. 11 (2011): 1134–1145.
57. On economic power and antitrust, see Adi Ayal, “The Market for Bigness:
Economic Power and Competition Agencies’ Duty to Curtail It,” Journal of
Antitrust Enforcement 1, no. 2 (2013): 221–246.
58. Center for Responsive Politics, “Lobbying Database” (2016), http://www
.opensecrets.org/lobby.
59. “UBER vs. Mayors: When Astroturfing Becomes Political Browbeating,”
Who’s Driving You? (January 15, 2016), http://www.whosdrivingyou.org/blog
/uber-vs-mayors-astroturfing-becomes-political-browbeating.
60. “Business in America,” The Economist.
61. Brody Mullins, “Google Makes Most of Close Ties to White House: Search
Giant Averages a White House Meeting a Week during Obama Administra-
tion,” Wall Street Journal, March 24, 2015.
62. Jonathan D. Salant, “Google’s Increased Lobbying Belies Cut in Total
Spending,” Bloomberg Technology, January 30, 2013, http://www.bloomberg
.com/news/articles/2013-01-30/google-s-increased-lobbying-belies-cut-in
Copyright © 2016. Harvard University Press. All rights reserved.

-total-spending.
63. Mullins, “Google Makes Most of Close Ties to White House.”
64. David Dayen, “The Android Administration,” The Intercept, (April 22, 2016),
https://theintercept.com/2016/04/22/googles-remarkably-close-relationship
-with-the-obama-white-house-in-two-charts/.
65. In early 2013, the FTC closed its antitrust investigation after Google agreed to
voluntarily change its business practices; Federal Trade Commission, “Google
Agrees to Change Its Business Practices to Resolve FTC Competition
Concerns in the Markets for Devices Like Smart Phones, Games and Tablets,
and in Online Search: Landmark Agreements Will Give Competitors Access

to Standard-Essential Patents; Advertisers Will Get More Flexibility to Use

Rival Search Engines,” Press Release (January 3, 2013), https://www.ftc.gov
/news-events/press-releases/2013/01/google-agrees-change-its-business
-practices-resolve-ftc.
66. Federal Trade Commission, FTC Staff Report, Google Inc., File No. 111-0161
(August 8, 2012), 92, released by the Wall Street Journal, http://graphics.wsj.
com/google-ftc-report/img/ftc-ocr-watermark.pdf. “Google has strengthened
its monopolies over search and search advertising through anticompetitive
means, and has forestalled competitors’ and would-be competitors’ ability to
challenge those monopolies, and this will have lasting negative effects on
consumer welfare.” Ibid.
67. Mullins, “Google Makes Most of Close Ties to White House.”
68. William Alden, “Email Shows How Google Gets Things Done in Wash-
ington,” BuzzFeed, (May 14, 2015), https://www.buzzfeed.com/williamalden
/how-googles-lobbyists-get-things-done-in-washington?utm _term=.qsKDM
jLnX#.ypLAW9PNn .
69. Ibid.
70. Ibid. Also note reports on how Google employed “several former EU officials
as in-house lobbyists, and has funded European think tanks and university
research favorable to its position as part of its broader campaign.” See
“Revealed: How Google Enlisted Members of US Congress It Bankrolled to
Fight $6bn EU Antitrust Case,” The Guardian, 17 December 2015. theguardian
.com/world/2015/dec/17/google-lobbyists-congress-antitrust-brussels-eu.
71. Jeff ry A. Frieden, Global Capitalism: Its Fall and Rise in the Twentieth
Century (New York: W. W. Norton, 2006), 102; Darren Bush, “Too Big to
Bail: The Role of Antitrust in Distressed Industries,” Antitrust Law Journal
77 (2010): 277, 286.
72. In Citizens United, the limitations on corporate political spending were
substantially weakened, thereby vastly increasing the importance of pleasing
large donors in order to win elections. The majority of justices placed great
faith in modern technology. Shareholder objections raised through the
Copyright © 2016. Harvard University Press. All rights reserved.

procedures of corporate democracy, the Court believed, “can be more effective

today because modern technology makes disclosures rapid and informative. . . .
With the advent of the Internet, prompt disclosure of expenditures can provide
shareholders and citizens with the information needed to hold corporations
and elected officials accountable for their positions and supporters.” Citizens
United v. Fed. Election Comm’n, 558 U.S. 310, 370, 130 S. Ct. 876, 916, 175 L. Ed.
2d 753 (2010). The Court found that the appearance of influence or access,
furthermore, will not cause the electorate to lose faith in democracy. The
dissenting justices, on the other hand, found, “[g]oing forward, corporations
and unions will be free to spend as much general treasury money as they wish

on ads that support or attack specific candidates, whereas national parties will
not be able to spend a dime of soft money on ads of any kind. The Court’s
ruling thus dramatically enhances the role of corporations and unions—and
the narrow interests they represent—vis-à-vis the role of political parties—and
the broad coalitions they represent—in determining who will hold public
office.” Citizens United, 558 U.S. at 412.
73. Despite its label, the rule of reason is not a directive that businesses and
consumers can readily understand and internalize (such as clear prohibitions
on agreeing with one’s competitors to fi x prices). Instead, the term embraces
antitrust’s most open-ended principles, namely a “flexible” factual inquiry
into a restraint’s overall competitive effect and “the facts peculiar to the
business, the history of the restraint, and the reasons why it was imposed.”
Am. Needle, Inc. v. Nat’l Football League, 130 S. Ct. 2201, 2217 (2010)
[quoting Board of Trade of Chicago v. United States, 246 U.S. 231, 238 (1918)].
The rule of reason also “varies in focus and detail depending on the nature of
the agreement and market circumstances.” Federal Trade Commission
and U.S. Department of Justice, Antitrust Guidelines for Collaborations
among Competitors (2000) §1.2, at 4, http://www.ftc.gov/os/2000/04
/ftcdojguidelines.pdf.
74. David Graeber, “On the Phenomenon of Bullshit Jobs,” Strike! Magazine,
August 17, 2013, http://strikemag.org/bullshit-jobs.
75. See generally, interview with Barry C. Lynn, senior fellow at New America
Foundation, “What We Have Is Capture of the Regulators’ Minds, a Much
More Sophisticated Form of Capture Than Putting Money in Their Pockets,”
published in Pro-Market Blog, Stigler Center, University of Chicago Booth
School of Business, https://promarket.org/what-we-have-is-capture-of-the
-regulators-minds-a-much-more-sophisticated-form-of-capture-than
-putting-money-in-their-pockets/.
76. Dayen, “The Android Administration.”
77. David Dayen, “Google’s Insidious Shadow Lobbying: How the Internet Giant
Is Bankrolling Friendly Academics—and Skirting Federal Investigations,”
Copyright © 2016. Harvard University Press. All rights reserved.

Salon (November 24, 2015), http://www.salon.com/2015/11/24/googles

_insidious _ shadow_ lobbying _ how_the_internet _ giant _is _bankrolling
_friendly_ academics _ and _ skirting _federal _investigations/.
78. Ibid.
79. Ibid.
80. Rajiv Chandrasekaran, “Microsoft Fights Back,” Washington Post, October 21,
1998, A01.
81. Ibid.
82. Leah Nylen and Lewis Crofts, “To Many, Europe Has Overtaken US as
Intellectual Leader on Monopolization,” mLex (January 29, 2016).

Ezrachi, Ariel, and Maurice E. Stucke. Virtual Competition : The Promise and Perils of the Algorithm-Driven Economy, Harvard
University Press, 2016. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/kcl/detail.action?docID=4742341.
Created from kcl on 2022-08-29 12:27:17.
Copyright © 2016. Harvard University Press. All rights reserved.

One joy in writing this book were the discussions with people within and outside
the antitrust circle. Our ideas benefitted from many exchanges with competition
officials, regulators, practitioners, and academics. Helpful meetings with statisti-
cians, mathematicians, engineers, and IT specialists further widened our horizons
and contributed to our understanding of the interfaces between technology and
competition. We are grateful for their taking the time to listen, their thoughtful
comments, their criticism, and their support.
We would like to note our special gratitude to insights and criticism received
along the way from Allen Grunes, Frank Pasquale, Nicolas Petit, Ingmar Posner,
Tony Curzon Price, Greg Taylor, Frank Wood, and the anonymous referees.
Since this project’s beginning, we were privileged to present parts of it in various
fora. The presentations and the discussion which followed enriched our under-
standing of virtual competition’s many facets. We are particularly grateful to the
organizers and participants of the following conferences and hearings: The Organ-
isation for Economic Co-operation and Development hearing “Across Platform
Parity Agreements”; the House of Lords hearing “Online Platforms and the Digital
Single Market”; the Canadian Competition Bureau’s “Workshop on Emerging
Competition Issues: Keeping Pace in a Changing World”; the European Data Pro-
Copyright © 2016. Harvard University Press. All rights reserved.

tection Supervisor & Academy of Eu ropean Law’s conference “Competition Re-

booted: Enforcement and Personal Data in Digital Markets”; the Federal Trade
Commission’s workshop “The ‘Sharing’ Economy: Issues Facing Platforms, Partici-
pants, and Regulators”; Bar Ilan University’s conference “Fairness in Antitrust”; the
British Institute of International and Comparative Law’s event “The Role of ‘Big
Data’ in Competition and Privacy Law”; Lund University’s conference “21st Century
Challenges and Antitrust: Thinking Ahead”; and Loyola University Chicago and
University of Haifa’s joint event “Antitrust in Transitional Markets.”
Our project began in Oxford while Maurice visited the Centre for Competition
Law and Policy. We thank the University of Oxford for facilitating the visit. Special
345

thanks are due to the Master of Pembroke College, Dame Lynne Brindley, the
professors, and college staff who were all gracious and hospitable. Maurice would
like to thank the University of Tennessee, Doug Blaze, Carol Parker, and Greg
Stein for supporting our research, including his sabbatical at the University of
Oxford to undertake this research.
Martin Dickson provided valuable assistance in the last editing phase. Deb-
orah Grahame-Smith proficiently oversaw the editing of our book. Ian Malcolm
at Harvard University Press was supportive in shepherding this book through the
publication process. We are very much indebted to them for their tireless and self-
less effort.
Finally, we thank our families for all of their support.
Copyright © 2016. Harvard University Press. All rights reserved.

Page numbers followed by f indicate figures.

Ad-blocking technology, 234; Apple and, platform, 18; Machine Learning ser vice,
171, 188; Coupons.com and, 187; Google 18; as super-platform, 149, 209; voice
and, 185–186 activation and, 17; Wal-Mart stores
Adner, Ron, 151 contrasted, 12–15, 257n34
Advertising expenses, reduced by online American Airlines, 87, 112
markets, 7 American Hotel and Lodging Association,
Advertising revenue. See Super-platforms, 137
advertising revenue and privacy issues Android devices, location tracking by, 164,
Advertising Week, 89 165f, 166
Agency model, comparison intermediaries Anticompetitive dynamics. See Behavioral
and, 139–142, 143 discrimination entries; Collusion;
Airbnb, 6–7 Frenemy dynamics
Airlines: behavioral discrimination and, Anticompetitive intent: digital eye scenario
112, 294n4; collusion and, 40–41, and, 77–80; predictable agent collusion
269n16; comparison intermediaries and scenario, 65–69, 279n31, 280nn39,41;
price distortion, 137; price discrimina- tacit collusion scenario and, 71–81
tion and, 87 Antitrust issues: hub and spoke collusion
Airline Tariff Publishing case, 40–41 scenario, 52–55, 276nn33,34; light touch
AKZO, 118–119 antitrust policies, 22–26; seller’s power
Copyright © 2016. Harvard University Press. All rights reserved.

Algorithms, proposal to audit, 230–231. and, 8–9. See also Enforcement issues;
See also Self-learning algorithms Sherman Act
Allstate, 90 Apple: decoy products and, 106; Frenemy
Alphabet. See Google dynamics, 149–151, 150f, 158, 309n19;
Amazon.com: Alexa personal assistant, Frenemy dynamics and Uber, 151–155;
191, 194, 201; Amazon Prime, 15, 257n34; Google and searches, 304n7; iAd
Big Data and Big Analytics and business and personal information, 103–104,
practices of, 12–15, 255n8, 257n34; 290nn11,12; iOS of, 30; location tracking
control over third-party retailers, 178, by, 164, 164f, 166; Siri personal assistant,
319n1; differential pricing and, 90; Digital 191, 193, 194; as super-platform, 149; United
Eye and, 74; e-books and pricing, 137, States v. Apple, Inc., 12, 47, 139, 272nn10,11,
140, 150–151, 306n23, 309n19; IoT 273n12; voice activation and, 17

347

Arbitrage, price discrimination and issues, 127–128, 219, 221, 300n47,

limiting of, 86–87 301nn48,50, 302nn55,56; fairness and
Artificial intelligence (AI), 15–17, 37, equality lacking with, 123–124, 129;
258n38; emerging trends and, 21; neoclassical economists and price
self-learning algorithms and autono- discrimination, 117–119, 295nn4–9,
mous pricing decisions, 74–77, 78 297n12; personal assistants and, 191,
Aston, Daniel William, 40 195–196; positive feedback loop and
Asymmetric bargaining power, 225, private data, 236–238; privacy concerns
332n20 and, 129–130, 227; social acceptance
Asymmetric information, 4, 31, 132 and, 121–122, 129, 130, 303n60; wealth
Asymmetric power, 155–158, 312nn39,45, inequality and, 241
313nn51,52 Behavioral economics, 97–98, 105
Asymmetric price elasticity, behavioral Behavioral experiments, intermediaries
discrimination and, 112–115, 294n62 and, 42–44
Athena Capital Research, 68–69, Benchmark interest rates/exchange rates,
280nn39,41 Messenger collusion scenario and, 40,
Audits, proposed for algorithms, 230–231 269nn7,9
Australian Communications and Media Biases, exploiting to increase consumer
Authority, 162 demand, 105–113; decoy products,
Austria, 229–230 106–107; drip pricing, 109–110; framing
Average revenue per user (ARPU), effects, 111–113; imperfect willpower,
super-platforms and, 236, 236f 110–111; increased complexity, 108–109;
price steering, 107–108
B&Q, 91 Big Data and Big Analytics, competitive
Baer, Bill, 40 environment changes and, 11–21;
Bank of America, 269n9 Amazon’s business practices and, 12–15,
Barclays PLC, 40, 268n5, 269nn7,9 255n8, 257n34; Big Analytics defi ned,
Barriers to entry: price discrimination 15; Big Data’s four Vs, 15, 20; cloud
and, 119, 297n12; reduced by online computing and Internet of Things, 18;
markets, 6–7 emerging trends and industry transfor-
Baymard Institute, 110 mations, 19–21; as mutually reinforcing,
Behavioral advertising, 20, 262n75 16, 259n47; online marketplace growth
Behavioral discrimination (general), and, 15–18
viii, 32, 83–84. See also Comparison BlackBerry, 149
intermediaries; Price discrimination Black Friday shopping, 19–20
Behavioral discrimination, “almost Bloomberg, 155
Copyright © 2016. Harvard University Press. All rights reserved.

perfect,” 30, 101–116, 289n1; biases used BNP Paribas, 269n9

to increase consumer demand, 105–113; Boise Cascade Corp. v. F.T.C., 68
personal data and informational Booking.com, 6
asymmetries, 112–115, 294n62; Boomerang Commerce, 13–14, 48–49, 52, 54
self-learning algorithms and, 101–103; Brick and mortar stores, 6, 8–9; Big Data
targeting and categorization and, and Big Analytics and competition for,
103–105 1, 4, 11–15, 19, 257n34; differential
Behavioral discrimination, economic and pricing and, 90, 91; tracking of
social perspectives on, 117–130, 234; customers and, 94–95
actual discrimination and, 124–127, 129; Brightest Flashlight Free app, 179, 180–184
complexity of, 119–121; economic Brooke Group Ltd. v. Brown & Williamson
incentives and, 228–229; enforcement Tobacco Corp., 57

Bush (G. W.) administration, 189, 224 and, 133–135, 303n6; super-platforms
BusinessWeek, 11 and asymmetry of power, 157–158,
313nn51,52; wide parity clauses and
Capital One Financial Corp., 90 agency model, 139–142
Capture of data, strategies for, 170–173 Competition and Markets Authority
Car dealers, behavioral discrimination and (CMA), in U.K., 103, 109, 119, 140, 243
framing effects, 111 Competitive environment: benefits of
Carnegie Mellon study, 125, 127 online markets to, 3–10; imperfections
Carrefour Marinopoulos, 40, 269n11 of new market dynamics, 27–31; light
Car rentals, behavioral discrimination touch antitrust policies and, 22–26;
and, 296n9 possible interventions, 31–33; techno-
Cartels, 22, 39, 263n2; durability and logical developments and changes in,
persistence of, 35–36, 266n1, 267n6 11–21
Cell phones. See Mobile devices Complementary relationships, 308n4
CheapTickets, 90 Complexity of pricing, increasing to
Chen, Jianqing, 151 exploit consumer biases, 108–109, 110
Chesler, Stanley R., 169 Concerted practice concept, 36, 39, 42, 65,
Chicago School theory of free markets, 22, 78, 270nn17,18
24–25, 26. See also Neoclassical Conglomerate relationship theory, 147
economics Conscious parallelism. See Tacit collusion
Children, Google and ads targeted to, 169 Consumer-provided content, super-
Children’s Online Privacy Protection Act, platforms and revenue from, 233–236,
of U.S., 227, 333n25 236f, 241
Chile, Project Cybersyn in, 213–214 Consumer surplus, 86, 88; producer
China: Anti-Monopoly Law, 23; Apple’s welfare surplus competition and, 234
strategy in, 152 Cookies, data extraction and, 167–170,
Chisholm, Alex, 94 316nn28,30, 316n33
Cialdini, Robert B., 105 Council of Economic Advisers Report,
Citicorp, 40, 269n7 240
Citigroup, 269n9 Coupons.com, 91, 94, 156, 171–172, 187,
Citizens United v. Fed. Election Comm’n, 311n38
342n72 Credit cards, behavioral discrimination
Cloud computing, 18 and framing effects, 111
Codes of practice, asymmetric bargaining Credit Suisse, 269n9
and, 225, 332n20 Currie, David, 116
Collusion, viii, 29–30, 32, 35–36; enforce-
Copyright © 2016. Harvard University Press. All rights reserved.

ment issues, 222; illegality of express, 57, Data advantage, competitive value of,
58; personal assistants and, 191, 196; 20–21, 263n79
pricing algorithms, 9. See also Digital Data brokers, 95, 104–105, 124–125, 170
Eye collusion scenario; Hub and spoke Data extraction. See Extraction and
collusion scenario; Messenger collusion capture strategies; Personal data
scenario; Predictable agent collusion Deadweight welfare loss, privacy and,
scenario; Tacit collusion 242–244
Comcast, 173 Decoy products, behavioral discrimination
Comparison intermediaries, 131–143; and, 106–107
benefits of, 132–133; distortions possible Deep learning. See Self-learning
from, 136–139, 304n15, 305n21; market algorithms
power and, 135–136; network effects Demand-responsive pricing program, 214

Department of Justice, in U.S.: light touch Enforcement issues, 218–232; behavioral

antitrust policies, 24, 263n2; Messenger discrimination and, 127–128, 219, 221,
collusion scenario and, 39–41; report on 300n47, 301nn48,50, 302nn55,56;
competition and monopoly, 224 challenges, 219–224; Digital Eye and,
Deutsche Bank, 269n9 77–80; Frenemy scenario, 221–222; hub
Didi Chuxing, 6, 152 and spoke collusion scenario, 52–55;
Differential pricing. See Price discrim- intervention timing and methods issues,
ination 223–224; predictable agent collusion
Digital Eye collusion scenario, 37, 71–81; scenario, 65–69, 279n31, 280nn39,41;
enforcement challenges and lack of problem not seen, 219–220; self-learning
anticompetitive intent, 77–80; “God algorithms, 223; tacit collusion, 65–69,
View” and real-time data processing, 219, 221, 279n31, 280nn39,41; tool
72–75; self-learning algorithms and framework proposals, 225–231; tools
autonomous pricing decisions, 74–77, 78 available, but with challenges, 222–223;
Digitized hand: competitive prices and tools lacking, 220–222
questions of regulations, 205–217; Epstein, Robert, 198
replacing of invisible hand of competi- Ericsson, 148
tion by, viii, 27, 203 Ethyl Gasoline Corporation v. United
Disconnect app, 179–180, 184–186, 187, States, 68
239 Eturas and Others, 52, 275n31
Discrimination, blurring with behavioral European Commission: Article 102(c)
discrimination, 124–127, 129 TFEU, 119–120, 128, 221, 270n18,
“Do Not Track” browser features: 310n29; collusion and, 47; Disconnect
Facebook and, 173; Google and, 186; app and, 185, 187; Facebook and, 242;
proposed policy for, 172; Target and, 92 Google and, 222; Microsoft and, 25; tacit
Drip pricing, behavioral discrimination collusion and, 60
and, 109–110 European Data Protection Supervisor, 16,
Driverless cars, Frenemy dynamics and 28, 100
Uber, Apple, and Google, 151–155 European Union: Article 101 TFEU, 221;
DuckDuckGo, 157, 200 Court of Justice of, 52; Data Protection
Dynamic and allocative efficiencies, of Directive, 128, 302n56; European Data
online markets, 7–8 Protection Regulation, 227, 333n27;
Dynamic pricing: hub and spoke collusion Unfair Commercial Practices Directive,
scenario, 47–52; online advantages over 128, 302n55
brick-and-mortar stores, 13–14; price Expedia, 107–108
discrimination and, 87–88, 112 Extraction and capture strategies, 159–177;
Copyright © 2016. Harvard University Press. All rights reserved.

data capture, 170–173; data extraction,

Economist, The, 106, 238 159–170. See also Super-platforms,
Efficiencies, of online markets, 7–8, 50 advertising revenue and privacy issues
Efficient markets hypothesis, 207–208.
See also Neoclassical economics Facebook, 50, 171, 173; advertising
Einstein, Albert, 220 revenue, 187; artificial intelligence and,
Electricity markets, in U.K., complexity 17; average revenue per user, 235, 236f;
of pricing and, 109 do not track standard and, 173;
Electric shock experiments, 43–44 Messenger text ing platform, 173;
Energy and Climate Change Select M personal assistant, 191, 192, 193,
Committee, in U.K., 138, 142 194, 200, 201; network effects and, 133;
Energy costs, in U.K., 141–142, 307n36 political news and, 198; as super-

platform, 149, 156, 189, 312n45; tracking General Motors, 155

and, 184; user-provided content and, Geofencing, 183
234–236; WhatsApp merger with, 242, George Mason University, 246
312n45 German Competition Authority, 278n14
Fairness, lacking with price discrimina- Ghostery, Ad Control of, 322n40
tion, 123–124 God View: data processing and tacit
Federal Energy Regulatory Commission, collusion, 72–75; of Uber, 72, 209, 229
in U.S., 216 Goldenshores Technologies. See Brightest
Federal Trade Commission (FTC), in U.S., Flashlight Free app
221, 224; data brokers and, 104–105, Google, 18, 147, 241–242, 243, 266n10;
124–125; Google and independent apps, Adblock Plus, 185; AdMob, 182;
179, 180–181, 320n10, 341n65, 342n66; AdSense, 168–169; Ad Settings, 125–126;
Google’s lobbying and, 244–246; light AdWords, 7; Analytics, 168–169;
touch antitrust policies, 263n2; mobile Android soft ware, 30; antitrust issues
device tracking and, 94–95; price and, 263n2; Apple iPhone searches and,
discrimination and, 119; tacit collusion 304n7; Assistant personal assistant, 191,
and, 68 193, 195, 201; Brain, 16; capture of data
Feedback loops: between machine learning and, 170–172; comparison intermedi-
and Big Data, 18–19; privacy and, aries and price distortion, 138, 305n21;
236–239 Deep Q network, 15–16; direct response
First Circuit Court of Appeals, 57 advertising and, 192; do not track
First-degree (perfect) price discrimination, standard and, 186; Doubleclick, 182,
85–86 184; Frenemy dynamics and, 149–151,
Flash Boys (Lewis), 65, 220 150f, 158; Frenemy dynamics and Uber,
Forbes, 153 151–155; FTC and, 341n65, 342n66;
Framing effects, behavioral discrimination limitations of opt-out feature, 186, 243,
and, 111–113 322nn41,43, 323nn44–46; lobbying by,
“Free” (consumer) side, of comparison 244–246; Ngram Viewer, 205, 206f; Play
intermediaries, 131, 134–135, 304n7 Store, 149, 150f, 151, 174, 179–186,
Frenemy dynamics, viii, 30–31, 32, 313nn51,52; revenues from advertising,
147–158, 234; enforcement issues, 182, 187; as super-platform, 149;
221–222; personal assistants, 191–202; tracking of users’ locations, 162–163;
positive feedback loop and private data, voice activation and, 17. See also
236–238; privacy concerns and, 227; Android devices
super-platform versus independent apps Governmental intervention. See Enforce-
and advertising revenue versus privacy ment issues; Light touch antitrust policies
Copyright © 2016. Harvard University Press. All rights reserved.

issues, 178–190. See also Extraction and Gravy algorithm, 68–69, 280n41
capture strategies Greek Hellenic Competition Commission,
Frieden, Jeff rey, 206–207 40
Greenspan, Alan, 22
Gabriel, Peter, 43 Grunes, Allen, 172
Galbraith, John Kenneth, 212
Gap stores, 12 Hausman, Jerry, 295n6
Gas prices, tacit collusion and algorithms, Hayek, Friedrich A., 32, 206–211, 213, 215
57–59, 62–63, 277nn5,7, 278n8 Home Depot, 108
Gender inequality, price discrimination Horizontal competitive relationships, 147
and, 126–127 Hotel bookings, comparison intermedi-
Genentech, 147 aries and price distortion, 137

Hub and spoke collusion scenario, 36, Jobs, Steve, 215

46–55, 147; enforcement challenges, JPMorgan Chase & Co., 40, 269nn7,9
52–55; Messenger scenario differs from,
48; pricing algorithms and, 47–50; Kantorovich, Leonid, 208
traditional, 46–47; Uber and, 50–55, Kayak, 149
276n35 Knowledge problem, of Hayek, 213, 215;
Hulu, 316n28 free market’s handling of, 206–208;
Hunter, David, 140 technology and illusion of competitive
prices, 208–211
ICAP (U.K. broker), 47
Independent applications: Brightest Lavery, Ian, 142
Flashlight Free app, 179, 180–184; Leeson, P. T., 118, 120, 294n2
Disconnect app, 179–180, 184–186, 187, Levinson, Arthur, 147
239; extraction strategies, 159–170; Lewis, Michael, 65, 220
interdependence between super- LIBOR, manipulation of, 47, 268n5
platforms and, 173–174; intervention Light touch antitrust policies, 22–26
and antitrust issues, 188–190; super- Lobbying, by super-platforms, 244–245
platforms, advertising revenue, and Location data, extraction of, 162–163,
privacy issues, 178–190 315n13
Inequality. See Wealth inequality Loyalty programs, differential pricing and,
Inflated list prices, price discrimination 91–93, 229–230, 285–286n21
and, 97–99 Lyft, 6, 51, 155, 210
Information flow: hub and spoke collusion
scenario, 52, 275nn28,30; improved by Maas, Jenna, 170
comparison intermediaries, 132; online Machine-to-machine communication
markets benefits and, 4, 50 (M2M), 18
Intel Corp., 174, 263n2 Marcus, David, 17
Intellectual capture, 223, 246–248 Market-clearing prices, planned econo-
Interbrand competition, 8–9 mies and, 211–214
Interlocking competitive relationships, 147 Mattress industry, complexity of pricing
International Business Machines Corp., 15, and, 109
16–17 Maverick firms, as threat to collusion, 45,
International Data Corp, 18–19 65, 73, 80–81, 228–229, 230, 237–238
Internet Advertising Bureau, in U.K., 103 McKie-Mason, Jeff rey, 295n6
Internet of Things, 18, 20, 102, 163, McKinsey & Company, 106
262n76 Messenger collusion scenario, 36, 39–45;
Copyright © 2016. Harvard University Press. All rights reserved.

Interstate Circuit v. United States, 47, 292n8 algorithms as intermediaries in, 42–44;
Intrabrand competition, 8–9 hub and spoke collusion scenario differs
Invisible hand of competition, 205, 206f; from, 48
displaced by digitalized hand, viii, 27, Metasearch engines. See Comparison
203; light touch antitrust policies, 24; intermediaries
praised by Ninth Circuit Court, 23; Microsoft, 18, 138, 174, 227; antitrust
technology and illusion of competitive issues and, 222, 247, 263n2; decoy
prices, 209–211. See also Neoclassical products and, 106; personal assistants
economics and, 201; Skype and, 25; as super-
platform, 148, 175–176; Uber and, 155;
Jacobson, Scott, 13 voice activation and, 17; Windows
Jet.com, 14, 48–49 phone platform, 149

Milgram, Stanley, 43–44 allocative efficiencies, 7–8; lower

“Milgram’s 37 (We Do What We’re Told)” barriers to entry, 6–7; market transpar-
(song), 43 ency and flow of information, 4; reduced
Misener, Patrick, 90 search costs, 5–6; reduction in seller
Mobile devices, 163–170; ad blocking power, 8–9
technology and, 171; data extraction Oracle, 174, 304n7
and, 162–170, 313n3; differential pricing Orbitz, 90, 107, 149
and, 90, 108; do not track standards and, Organisation for Economic Co-operation
172; Frenemy dynamic and, 148–151, and Development (OECD), 4, 112
158; tracking of customers and, 94–96 Orwell, George, 199
Morgan Stanley, 269n9 Oxfam, 240
Morozov, Evgeny, 239
Most Favored Nation (MFN) clauses, “Paid” (provider) side, of comparison
comparison intermediaries and, intermediaries, 131, 134–135, 304n7
140–142, 143 Parallel accommodating conduct,
computers and, 279n25
Neoclassical economics, 205; behavioral Parity clauses, 139–142, 143
discrimination and, 120–121; price Pasquale, Frank, 127
discrimination and, 117–119, 295nn4–9, Pay-for-placement fees, comparison
297n12 intermediaries and, 136–139, 304n15,
Nest Labs, 154 305n21
Netfl ix, 238, 337n22 “Perfect competition,” theoretical model
Network effects: of comparison intermedi- of, 4
aries, 133–135, 303n6; market power Perfect information games, 17–18
and, 237–238; super-platforms and, 175 Per se illegality, 35, 42, 54–55, 58, 80, 245,
New market dynamics, imperfection of, 269n10, 270n19, 271n20
27–31 Personal assistants: advantages of, 191;
New York Times, 92–93, 137, 215 concerns about, 191–193; possible
1984 (Orwell), 199 proactivity of, 197–199; purist assistant
Ninth Circuit Court of Appeals, 23 anticipated, 199–201; transformation of
Nissan Leaf, 243 competitive environment and, 194–197
Nomi Technologies, 95 Personal data: behavioral discrimination
North, Douglass, 225 and lack of transparency, 112–115,
Notice-and-consent privacy model, broken 294n62; Big Data and Big Analytics and,
nature of, 226–228 15–18; competitive advantage and, 14,
20–21, 263n79; cost of “ free” ser vices
Copyright © 2016. Harvard University Press. All rights reserved.

Obama, Barack, 189 and, 28–29, 266n1; extraction and

Obama administration, 224, 244, 248 capture of, 159–177; network effect of
Object illegality, 35, 42, 80, 269n10, comparison intermediaries, 134. See also
270n18, 271n20 Extraction and capture strategies;
Octo Telematics, 16–17 Privacy issues
Office Depot, 90 Pharmaceutical markets, behavioral
Office of Fair Trading (OFT), in U.K., 7 discrimination and, 295nn5,6
Oligopolistic price coordination. See Tacit Planned economies: continuing faith in
collusion free market and, 205–206; drawbacks of,
Omega Psi Phi fraternity, 126–127 214–216; Hayek’s argument against,
Online markets: benefits to competitive 206–208; smart regulation by govern-
environment, 3–10; dynamic and ment and, 212–214, 328nn32–37;

Planned economies (continued) Price transparency: data and reduction of,

technology and illusion of competitive 56–60, 229–230; online markets benefits
prices, 208–211 and, 4
Political information, personal assistants Pricing algorithms, 8, 9; hub and spoke
and editing of, 197–198 collusion scenario, 47–50; Messenger
Political power, super-platforms and, collusion scenario and, 39–44; online
244–246 advantages over brick-and-mortar
Post Danmark A/S v. Konkurrencerådet, 128 stores, 13–14; predictive agent scenario
Power Conversion, Inc. v. Saft Am., Inc., and, 36–37
271n20 Privacy issues: algorithms and planned
Predictability, and insufficient data for economies, 215–216; broken nature of
perfect price discrimination, 97–98 notice-and-consent model, 226–228;
Predictable agent collusion scenario, customer empowerment proposal,
36–37, 56–70, 221; enforcement 226–228; positive feedback loop
challenges and anticompetitive intent and, 236–239; proposal for enhance
issues, 65–69, 279n31, 280nn39,41; disclosure, 226–228; trust and, 242–244.
predictive analytics and algorithms, See also Super-platforms, advertising
61–62; speed of information and revenue and privacy issues
algorithm arms race, 62–65, 279n25; Project Cybersyn, in Chile, 213–214
transparency and tacit collusion, 56–60
Pregnancy prediction, Target tracking and, Racial discrimination, 124–125
92–93 Rather-be-shopping.com, 111
President’s Commission on Law Enforce- Reagan administration, 25, 264n11
ment and Administration of Justice, Real estate market, price discrimination
in U.S., 199 and, 96, 97–98, 99, 288n52
Price comparison websites (PCWs), 3, Regulation and intervention: aims of,
5, 7, 253n22. See also Comparison 203–204; smart regulation, 212–214, 218,
intermediaries 322nn32–37; technology and planned
Price discrimination, 85–88; behavioral economy versus free market setting
discrimination’s complexity and, prices, 205–217. See also Enforcement
119–121; differential pricing, 85–86, issues
89–94; dynamic pricing, 87–88; Resale price maintenance (RPM), 40, 269n10
hindrances to perfect, 96–99; limited Reselling. See Arbitrage
arbitrage, 86–87; neoclassical economics Reservation price, of customers, 88, 99;
and concerns about, 118–119, 295n9, behav ior and difficulty of estimating,
297n12; neoclassical economics and 97–99; differential pricing and, 85–86;
Copyright © 2016. Harvard University Press. All rights reserved.

efficiencies of, 118, 295nn4–8; profit insufficient data for perfect price
maximization and, 88; reduced by discrimination, 96–97
online markets, 9; social acceptance Ride-sharing apps, 6–7, 8. See also Uber
and, 121–122, 129, 130, 303n60; tracking Technologies, Inc.
of customers and, 94–96 Road to Serfdom, The (Hayek), 206–208
Price fi xing. See Cartels; Collusion Robinson-Patman Act, 119, 127–128,
Price parity, collusion and, 140–141, 272n11 300n47, 301nn48,50
Prices: allocative function of, 208; Rosetta Stone, 90
signaling function of, 210; technology Royal Bank of Scotland plc, 40, 47, 269n7
and illusion of competitive, 208–211 Rubicon Project, 259n47
Price steering, behavioral discrimination “Rule of Reason” standard, 54–55, 246,
and, 107–108, 130 279n31, 343n73

Salem, Enrique, 149 Stone, Merlin, 93–94

Sample size, and insufficient data for Stop Online Piracy Act (SOPA), Google
perfect price discrimination, 99 protest about, 198
San Francisco parking regulation, 217, Stout, Lynn, 243
328nn32–37; SFpark program, 214, Supermarkets, loyalty programs and
328nn32–37 tracking of customers, 93–94
Scarcity marketing, 110 Super-platforms: capture of data and, 160,
Schmidt, Eric, 147 170–173; enforcement issues, 219–220;
Schumpeter, Joseph, 295–296n6 extraction strategies, 159–170; interde-
Search costs, reduced by online markets, pendence between applications and,
5–6, 50 173–174; multi-dimensional competition
Search engines, as “information gate- between, 147–151, 148f, 150f; power
keepers,” 136, 304n9 asymmetries between apps and,
Securities and Exchange Commission 155–158, 311n38, 312nn39,45,
(SEC), 68–69 313nn51,52; privacy and threats to
Self-learning algorithms, 11, 14, 15–18; revenues, 239; technology and illusion
“almost perfect” behavioral discrimina- of competitive prices, 211–212;
tion and, 101–103; autonomous pricing Uber, Apple, Google and Frenemy
and tacit collusion, 74–77, 78; emerging dynamic, 151–155. See also Personal
trends and, 21; enforcement issues, 223; assistants
hub and spoke collusion scenario, 49 Super-platforms, advertising revenue and
Seller’s power, reduced by online markets, privacy issues, 178–190; ad revenues from
8–9 customer-created content, 233–236, 236f,
Seventh Circuit Court of Appeals, 67 241; Brightest Flashlight Free app and,
Sheffi, Yossi, 89 179, 180–184; Disconnect app and,
Sherman Act, 221, 267n5, 271n20, 179–180, 184–186, 187, 239; intervention
330–331n11, 331n13 and antitrust issues, 188–190
“Sleepers,” 67, 114 Supracompetitive prices, 41, 83, 228,
Small and medium size enterprises 230–231, 335n37
(SMEs), behavioral discrimination and, Supreme Court, of U.S., 342n72; cell
115–116 phones and privacy and, 162–163, 313n3;
Smartphones. See Mobile devices on conspiracy, 46; enforcement issues,
Smart regulation, 212–214, 218, 328nn32–37. 223; price discrimination and, 119; rule
See also Enforcement issues of reason of, 245–246, 343n73; tacit
Snowball effect, as network effect of collusion and, 57, 70
comparison intermediaries, 134–135, Surge pricing, Uber’s market power and,
Copyright © 2016. Harvard University Press. All rights reserved.

304n7 51–52, 209–214, 216, 327n14

Sobel, R., 118, 120, 294n2
Social acceptance, of behavioral discrimi- Tablets. See Mobile devices
nation, 121–122, 129, 130, 303n60 Tacit collusion, 36–37, 56–70, 234;
Social networks, 16, 92, 103, 123, 133, 152, algorithms, 231; Digital Eye, Artificial
161–162, 198, 236 Intelligence, and lack of human intent,
Social welfare. See Behavioral discrimina- 71–81; economic incentives and,
tion, economic and social perspectives 228–229; enforcement challenges and
on; Wealth inequality anticompetitive intent issues, 65–69,
Spillovers, as network effect of comparison 219, 221, 279n31, 280nn39,41; key
intermediaries, 134–135, 304n7 market conditions for, 59–60; positive
Staples, Inc., 14, 90, 241 feedback loop and private data, 236–238;

Tacit collusion (continued) Varney, Christine A., 224

predictive analytics and algorithms, Vertical agreements, hub and spoke
61–62; price transparency and, 229–230; collusion scenario, 36, 47, 52–55
speed of information and algorithm Vertical competitive relationships, 147
arms race, 62–65, 279n25; transparency Vestager, Margrethe, 195, 225
and, 56–60; wealth inequality and, 241 Viacom Inc., 169
Target, differential pricing and, 91–93 Volvo Trucks North America, Inc. v.
Telecommunications market, complexity Reeder-Simco GMC, Inc., 301n50
of pricing and, 108–109 Von Mises, Ludwig, 233–234
Tesco, 93, 237 Vs, of Big Data, 15, 20
Tesco v. Office of Fair Trading, 275n28,
276n34 Wall Street Journal, 20, 90, 137, 149, 168,
Third-degree (imperfect) price discrimina- 169–170, 173, 185, 200, 241, 244, 245,
tion, 86 313n51, 322n40
Tit-for-tat pricing wars, self-learning Wal-Mart Stores, Inc., Amazon’s business
algorithms and avoidance of, 74–76 practices contrasted, 12–15, 257n34
Topkins, David, 39–40 Watson computer, of IBM, 15, 16
Tracking of customers, 99; behavioral Waze, 152, 153, 190
discrimination and, 103–104, 290nn11,12; Wealth inequality, 120, 123–125,
differential pricing and, 94–96. See also 239–242
Privacy issues; Super-platforms, Weather Co., 16–17
advertising revenue and privacy issues Web-aggregators, 9, 253n23; reduced
Travel agents, 52, 107–108, 275n31 search costs and, 5–6. See also Com-
Travelocity, 108 parison intermediaries
Trial and error, as network effect, 133–134 Website owners, extraction strategies,
Truman Show (fi lm), 27, 197, 202 159–170
Trust, privacy and, 242–244 Welfare effects, of behavioral discrimina-
Tuition, at colleges: differential pricing tions. See Behavioral discrimination,
and, 86; price discrimination and, economic and social perspectives on
121–122 Wells Fargo, 269n9
Tversky, Amos, 105 WhatsApp, 28, 242, 312n45
White House reports, 19, 28, 107, 114–115,
Uber Technologies, Inc., 6–7, 10; data 124, 238, 240
extraction and tracking by, 163–170; Wide parity clauses, comparison interme-
Frenemy dynamics and, 151–155; God diaries and, 139–142
View of, 72, 209, 229; hub and spoke Willpower, behavioral discrimination and
Copyright © 2016. Harvard University Press. All rights reserved.

collusion scenario, 50–55, 276n35; surge imperfect, 110–111

pricing and market power of company,
51–52, 209–214, 216, 327n14 Yahoo!, 138, 173
UBS AG, 40, 268n5, 269n9
United Nations, 240–241 Zermelo, E., 279n28
United States v. Apple, Inc., 12, 47, 139, Zhu, Feng, 151
173n12, 272n10, 272nn10,11 Zittrain, Jonathan, 198
United States v. U.S. Gypsum Co., 276n33 Zuckerberg, Mark, 17

Massimo Motta, Martin Peitz, Heike Schweitzer (Eds.) - Market Investigations. A New Competition Tool For Europe - Cambridge University Press (2022) 3
No ratings yet
Massimo Motta, Martin Peitz, Heike Schweitzer (Eds.) - Market Investigations. A New Competition Tool For Europe - Cambridge University Press (2022) 3
410 pages
Continental Carriers Case Analysis
100% (3)
Continental Carriers Case Analysis
2 pages
Copyright Protection For Artificial Intelligence Output Boetticher
No ratings yet
Copyright Protection For Artificial Intelligence Output Boetticher
81 pages
Haapio, 2014, DL
No ratings yet
Haapio, 2014, DL
8 pages
Lina Khan - Brandeisian Antitrust
No ratings yet
Lina Khan - Brandeisian Antitrust
2 pages
OCDE Rethinking Antitrust Tools For Multi Sided Platforms 2018
No ratings yet
OCDE Rethinking Antitrust Tools For Multi Sided Platforms 2018
230 pages
General Theory of Law and Development: Yong-Shik Lee
100% (1)
General Theory of Law and Development: Yong-Shik Lee
58 pages
The Politics of Digital (Human) Rights
No ratings yet
The Politics of Digital (Human) Rights
31 pages
Commercial Law - Module Guide (University of London) 2022nd Edition Rick Canavan & Philip Rawlings - The special ebook edition is available for download now
100% (1)
Commercial Law - Module Guide (University of London) 2022nd Edition Rick Canavan & Philip Rawlings - The special ebook edition is available for download now
72 pages
(Routledge Research in The Law of Emerging Technologies) Edoardo Celeste - Digital Constitutionalism - The Role of Internet Bills of Rights-Routledge (2022)
No ratings yet
(Routledge Research in The Law of Emerging Technologies) Edoardo Celeste - Digital Constitutionalism - The Role of Internet Bills of Rights-Routledge (2022)
255 pages
2021 - Law, Technology and Innovation v. II
No ratings yet
2021 - Law, Technology and Innovation v. II
349 pages
Download full Q A Commercial Law 8th Edition Jo Reddy ebook all chapters
100% (3)
Download full Q A Commercial Law 8th Edition Jo Reddy ebook all chapters
78 pages
EdTechJan2013Report PDF
No ratings yet
EdTechJan2013Report PDF
88 pages
The Open Innovation Model
No ratings yet
The Open Innovation Model
28 pages
Platform Justice: Content Moderation at An Inflection Point
No ratings yet
Platform Justice: Content Moderation at An Inflection Point
24 pages
Handbook of Innovation and Regulation
No ratings yet
Handbook of Innovation and Regulation
429 pages
Multi-Sided Platforms Hagiu 2011
No ratings yet
Multi-Sided Platforms Hagiu 2011
38 pages
(Global Competition Law and Economics Policy) BjÃ RN Lundqvist - Regulating Access and Transfer of Data-Cambridge University Press (2023)
No ratings yet
(Global Competition Law and Economics Policy) BjÃ RN Lundqvist - Regulating Access and Transfer of Data-Cambridge University Press (2023)
269 pages
Expropriation Clauses in International Investment Agreements and The Appropriate Room For Host States To Enact Regulations, 2009
100% (1)
Expropriation Clauses in International Investment Agreements and The Appropriate Room For Host States To Enact Regulations, 2009
66 pages
Evolution of Society Through Artificial Intelligence A Review
No ratings yet
Evolution of Society Through Artificial Intelligence A Review
9 pages
Unilateral Conduct by Dominant Firms at The IP Antitrust Intersection in The EU and The Us
No ratings yet
Unilateral Conduct by Dominant Firms at The IP Antitrust Intersection in The EU and The Us
54 pages
EXPROPRIATION
No ratings yet
EXPROPRIATION
3 pages
Big Data, Political Campaigning and The Law - NORMANN WITZLEB - Routledge - 2020
No ratings yet
Big Data, Political Campaigning and The Law - NORMANN WITZLEB - Routledge - 2020
259 pages
Economides Net Neutrality PDF
No ratings yet
Economides Net Neutrality PDF
25 pages
Artificial Intelligence White Paper Draft 5 Oct 20171
No ratings yet
Artificial Intelligence White Paper Draft 5 Oct 20171
48 pages
European Commission Glossary of Terms Used in Eu Competition Policy - Antitrust and Control of Concentrations
No ratings yet
European Commission Glossary of Terms Used in Eu Competition Policy - Antitrust and Control of Concentrations
59 pages
[Ebooks PDF] download State aid law of the European Union 1st Edition Hofmann full chapters
100% (2)
[Ebooks PDF] download State aid law of the European Union 1st Edition Hofmann full chapters
50 pages
Competition Law And Antitrust David J Gerber download
100% (2)
Competition Law And Antitrust David J Gerber download
86 pages
The Laws of Man Over Vehicles Unmanned: A Special Issue of The JLIS.
100% (3)
The Laws of Man Over Vehicles Unmanned: A Special Issue of The JLIS.
265 pages
Artificial Intelligence and The Humanitie PDF
No ratings yet
Artificial Intelligence and The Humanitie PDF
7 pages
Conjuring Up New Human Rights Alston
No ratings yet
Conjuring Up New Human Rights Alston
16 pages
Privacy Antitrust and Power
No ratings yet
Privacy Antitrust and Power
16 pages
Competition Lecture 6 - Article 102
No ratings yet
Competition Lecture 6 - Article 102
30 pages
Josef Drexl - Research Handbook On Intellectual Property Law and Competition Law (2008, Edward Elgar Pub)
No ratings yet
Josef Drexl - Research Handbook On Intellectual Property Law and Competition Law (2008, Edward Elgar Pub)
510 pages
Gary Born - International Arbitration, Law and Practice, 2012
No ratings yet
Gary Born - International Arbitration, Law and Practice, 2012
535 pages
Global Competition Law and Economics Elhauge and Geradin 2007 44 Common Market Law Review 1839 1841 Dec 2007
No ratings yet
Global Competition Law and Economics Elhauge and Geradin 2007 44 Common Market Law Review 1839 1841 Dec 2007
3 pages
Competition Act, 2002
No ratings yet
Competition Act, 2002
11 pages
Itu-T: Smart Ubiquitous Networks - Overview
No ratings yet
Itu-T: Smart Ubiquitous Networks - Overview
20 pages
23 Colum JTransnatl L561
No ratings yet
23 Colum JTransnatl L561
19 pages
Copyright and Trademark in Cyberspace
No ratings yet
Copyright and Trademark in Cyberspace
6 pages
Anti-Corruption Module 4 Public Sector Corruption
100% (1)
Anti-Corruption Module 4 Public Sector Corruption
49 pages
Competition Law in Digital Markets in TH
No ratings yet
Competition Law in Digital Markets in TH
19 pages
ColetaneaDireitoDigital1 PDF
No ratings yet
ColetaneaDireitoDigital1 PDF
456 pages
Copyright Protection For Ai Generated Works Exploring Originality and Ownership in A Digital Landscape
No ratings yet
Copyright Protection For Ai Generated Works Exploring Originality and Ownership in A Digital Landscape
24 pages
CSR and Ethics
100% (1)
CSR and Ethics
24 pages
Cross-Border Data Transfers Between The EU
No ratings yet
Cross-Border Data Transfers Between The EU
46 pages
International Trade Dumping Impact On Competition
No ratings yet
International Trade Dumping Impact On Competition
35 pages
Judgement-Proof Robots and Artificial Intelligence (A Comparative Law and Economics Approach) __
No ratings yet
Judgement-Proof Robots and Artificial Intelligence (A Comparative Law and Economics Approach) __
159 pages
What Is 'Digital Transformation' of Government Services
No ratings yet
What Is 'Digital Transformation' of Government Services
9 pages
Censorship in China
No ratings yet
Censorship in China
5 pages
Justifying Ipr Module 2
No ratings yet
Justifying Ipr Module 2
62 pages
Artificial Intelligence - Indian Law
No ratings yet
Artificial Intelligence - Indian Law
25 pages
Internet Platforms: Observations On Speech, Danger, and Money
No ratings yet
Internet Platforms: Observations On Speech, Danger, and Money
44 pages
The Artificial Intelligence Act (EU AI Act)
No ratings yet
The Artificial Intelligence Act (EU AI Act)
7 pages
5
No ratings yet
5
30 pages
AI and Judicial Decision-Making
No ratings yet
AI and Judicial Decision-Making
18 pages
SSRN-id2180847 Patents and Innovation: Evidence From Economic History
No ratings yet
SSRN-id2180847 Patents and Innovation: Evidence From Economic History
31 pages
AI and Privacy Concerns in Data Security
No ratings yet
AI and Privacy Concerns in Data Security
21 pages
Virtual Competition The Promise And Perils Of The Algorithmdriven Economy Ariel Ezrachi instant download
No ratings yet
Virtual Competition The Promise And Perils Of The Algorithmdriven Economy Ariel Ezrachi instant download
76 pages
Virtual Competition_ The Promise and Perils of the Algorithm-Driv
No ratings yet
Virtual Competition_ The Promise and Perils of the Algorithm-Driv
8 pages
Antitrust Congress Summary
No ratings yet
Antitrust Congress Summary
4 pages
U10 Etnics
No ratings yet
U10 Etnics
8 pages
Introduction To Public Relations
No ratings yet
Introduction To Public Relations
26 pages
An Overview of Marketing in The Hospitality Industry 1
No ratings yet
An Overview of Marketing in The Hospitality Industry 1
4 pages
CV Safwan Ahmed 2.2.22 PDF
No ratings yet
CV Safwan Ahmed 2.2.22 PDF
5 pages
Autocad Draughtsman: Date: 06.08.2019 Mr. Tushar Taral Subject: Offer of Employment
No ratings yet
Autocad Draughtsman: Date: 06.08.2019 Mr. Tushar Taral Subject: Offer of Employment
2 pages
Euphoria Genx - Summer & Winter Training - Job Oriented Training
No ratings yet
Euphoria Genx - Summer & Winter Training - Job Oriented Training
10 pages
DC - DCFC - Brochure
No ratings yet
DC - DCFC - Brochure
4 pages
pl-900 4
No ratings yet
pl-900 4
49 pages
Justice LEONEN Cases Compilation
100% (2)
Justice LEONEN Cases Compilation
42 pages
System Analysis & Design (PL/SQ Prog) (OER) : Dashboard MIS6231 Week 001: Data and Information Assignment 001
No ratings yet
System Analysis & Design (PL/SQ Prog) (OER) : Dashboard MIS6231 Week 001: Data and Information Assignment 001
13 pages
Food Corner on Wheels Rs. 7.60 million Feb-2021
No ratings yet
Food Corner on Wheels Rs. 7.60 million Feb-2021
24 pages
SFTY 100 - Questions With Answers
No ratings yet
SFTY 100 - Questions With Answers
5 pages
첨부 - 2-1 - GS E - C HSE Management Program
No ratings yet
첨부 - 2-1 - GS E - C HSE Management Program
33 pages
Analysis UPSA
No ratings yet
Analysis UPSA
37 pages
Mango Graham Shake Project 1
No ratings yet
Mango Graham Shake Project 1
4 pages
4 Ms of Production and Business Model
75% (4)
4 Ms of Production and Business Model
21 pages
VADIVEL
No ratings yet
VADIVEL
43 pages
S2 Lmis
No ratings yet
S2 Lmis
29 pages
SBL 1 Assignment
No ratings yet
SBL 1 Assignment
5 pages
16-11-2022 Amaala Prequalification
No ratings yet
16-11-2022 Amaala Prequalification
64 pages
Losses From Wash Sales
No ratings yet
Losses From Wash Sales
7 pages
(Triplicate For Supplier) : Sl. No Description Unit Price Qty Net Amount Tax Rate Tax Type Tax Amount Total Amount
No ratings yet
(Triplicate For Supplier) : Sl. No Description Unit Price Qty Net Amount Tax Rate Tax Type Tax Amount Total Amount
1 page
Chapter 6 The Philippine Financial Market
No ratings yet
Chapter 6 The Philippine Financial Market
19 pages
Project On: Submitted To
No ratings yet
Project On: Submitted To
13 pages
Engineering Insurance 211012 PDF
89% (9)
Engineering Insurance 211012 PDF
160 pages
Roadmarking Quotation
No ratings yet
Roadmarking Quotation
4 pages
REACH Compliance EU
No ratings yet
REACH Compliance EU
6 pages
Resourcing and Talent Management Strategies
No ratings yet
Resourcing and Talent Management Strategies
8 pages
Brs 2 Questions
No ratings yet
Brs 2 Questions
13 pages