MWG Chap 5: Production

Review: undergraduate production theory

• explicit distinction between input and output
• For example, consider two input factors, L and K, and one output, Q.
• First, for output Q, derive C(Q) = minL,K wL + rK s.t. Q = f (L, K).
• The optimization condition for the cost minimization is
w
M RT SL,K = ,
r
i.e., the marginal rate of technological substitution between labor and capital
is equivalent to the ratio of wage and rent.
• Another way to look at the profit maximization is:

max pf (L, K) wL rK.

L,K

Then the first order conditions are:

p ⇥ M PL = w and p ⇥ M PK = r

Production Sets - a more general approach

There are L goods, some of which may be used as inputs to produce other
good(s). A production vector is a vector y 2 RL : Here negative coordinates repre-
sent inputs and the positive coordinates represent outputs. The set of all feasible
production vectors for a firm is known as the firm’s production set Y ⇢ RL : any
y 2 Y is possible and any y 62 Y is not.
The production set Y is taken as a primitive datum of the theory, which is deter-
mined primarily by technological constraints, but possibly also by legal restrictions
and contractual commitments.
Commonly imposed properties of production sets Y :
– Y is closed: for a converging sequence (y n ) 2 Y , limn!1 y n 2 Y .
– Possibility of inaction (shut-down): 0 2 Y .
– No free lunch: Y \ RL+ = {0}.
– Free disposal: y 2 Y and y 0  y imply y 0 2 Y .
Given these properties, we may conveniently represent Y as consisting of vectors y
assuming non-positive values of F : RL ! R, called the transformation function,
with the property that the boundary points of Y assume 0.

16
 • Production set Y := {y 2 RL : F (y)  0} ⇢ RL
• Transformation frontier: {y 2 RL : F (y) = 0}
• Marginal rate of transformation (along the transformation frontier):
@F
@yl
M RTl,k = @F
@yk

(M RT Sl,k is simply a renaming of M RTl,k of input l for input k in the special

case of a single-output, many-input technology.)
Other properties that may be assumed on Y :
– Irreversibility: y 2 Y & y 6= 0 ) y 2 / Y.
– Non-increasing returns to scale: y 2 Y ) ↵y 2 Y for ↵ 2 (0, 1).
– Non-decreasing returns to scale: y 2 Y ) ↵y 2 Y for ↵ 1.
– Constant returns to scale: y 2 Y ) ↵y 2 Y for ↵ 0.
– Additivity: y 2 Y and y 0 2 Y ) y + y 0 2 Y .
– Convexity: y 2 Y and y 0 2 Y ) ↵y + (1 ↵)y 0 2 Y for all ↵ 2 (0, 1).

17
Profit Maximization Problem

L
X
max p · y = p` y` s.t. y 2 Y (P M P )
y
`=1

• firms are price-takers where p = (p1 , · · · , pL ) 0.

• Y is nonempty and closed and satisfies the free disposal property.
• the solution y(p) is the supply function (or correspondence).
• the optimised value ⇡(p) := p · y(p) is the profit function.

Prop 5.C.1: Assume Y is closed and satisfies the free disposal property. Then
(i) ⇡(p) is homogeneous of degree 1 in p.
(ii) ⇡(p) is convex in p.
(iv) y(p) is homogeneous of degree 0 in p.
(v) If Y is convex, then y(p) is a convex set for all p; if Y is strictly convex, then
y(p) is single-valued (if nonempty).
(vi) Hotelling’s Lemma: If y(p̄) is single-valued, then ⇡(·) is di↵erentiable at p̄ and
r⇡(p̄) = y(p̄).
(vii) If y(·) is di↵erentiable at p̄, then Dy(p̄) = D2 ⇡(p̄) is a symmetric and positive
semi-definite matrix.

Proof. (i) and (iv) are straightforward.

(ii) Note that if z 2 y(↵p + (1 ↵)p0 ) for some ↵ 2 (0, 1), then
⇡(↵p + (1 ↵)p0 ) = ↵p · z + (1 ↵)p0 · z  ↵⇡(p) + (1 ↵)⇡(p0 )
where the equality follows from z 2 y(↵p + (1 ↵)p0 ) and the inequality follows
from the definition of ⇡. This verifies convexity of ⇡, i.e., ⇡(↵p + (1 ↵)p0 ) 
↵⇡(p) + (1 ↵)⇡(p0 ) for all p, p0 and ↵ 2 (0, 1).

18
 (v) The first part follows because the objective function of (PMP) is linear.
To prove second part by contradiction, suppose that z, z 0 2 y(p). Then, as z 00 =
(z + z 0 )/2 is in the interior of Y , so is z 00 + ✏ej for sufficiently small ✏ where ej 2 RL
is a vector with 1 in some j-th coordinate such that pj > 0 and 0 in all other
coordinates. This would imply that p · (z 00 + ✏ej ) > y(p), contradicting z, z 0 2 y(p).
(vi) We prove this assuming all relevant di↵erentiability. Di↵erentiating the
identity F (y(p)) = 0 wrt pj , we get
X @yi ⇣ @y @yL ⌘T ⇣ @y @yL ⌘T
1 1
Fi (y(p)) =0 , rF (y(p))· ,··· , =0 , p· ,··· , =0
i
@pj @pj @pj @pj @pj

where the second equivalence (,) follows because rF (y(p)) = p as the graph
above illustrates and is explained below. Then, di↵erentiating ⇡(p) = p · y(p) wrt
pj , we get

@⇡(p) ⇣ X @y ⌘ ⇣ @y @yL ⌘T
i 1
= pi + yj (p) = p · ,··· , + yj (p) = yj (p).
@pj i
@pj @pj @pj

This proves (vi).

Then, (ii) and (vi) imply (vii).

Why rF (y(p)) = p (assuming di↵erentiability)?

Proof. Take any point y on the transformation frontier. The tangent line (more
precisely, the tangent hyperplane) at y consists of points y 0 such that, as y is changed
to the direction of y 0 y by an infinitesimal amount, the rate with which F (·) changes
is 0, that is,

@F (y) 0 @F (y) 0 @F (y) 0

(y1 y1 ) + (y2 y2 ) + · · · + (yL yL ) = 0
@y1 @y2 @yL
() (F1 (y), · · · , FL (y)) · (y10 y1 , · · · , yL0 yL ) T = 0
() rF (y) · (y 0 y) = 0.

Therefore, rF (y) is perpendicular to the tangent hyperplane of F at y.

At y = y(p), py is maximised among all y 2 Y , which implies that p is per-
pendicular to the tangent plane of F at y(p). Therefore, rF (y(p)) and p are both
perpendicular to the same hyperplane, so that rF ((p)) = p. That > 0 fol-
lows because both F (y) and py increases as y moves “outward” direction from y(p)
perpendicular to the hyperplane.

19
Prop: [The law of supply] p· y 0.
Proof. Observe that py(p) py(p0 ) and p0 y(p0 ) p0 y(p). Thus,

py(p) p0 y(p) py(p0 ) p0 y(p0 ) () p · y(p) p · y(p0 ) 0.

Cost Minimization Problem - Single output case:

Let q = f (z) be the output level and w the input price vector. Then, the cost
minimization problem is:

min w · z s.t. f (z) q (CmP )

z 0

Note that this is identical to the consumer’s EmP (except that q has cardinal, instead
of ordinal, meaning).
Kuhn-Tucker condition is:
,
@f @f @f wi
 wi , with equality if zi > 0 ) M RT S = = .
@zi @zi @zj wj

Example: Consider two inputs and one output case. Along the iso-quant curve,
@f @f
f (z1 , z2 ) = q. Total-di↵erentiating, we derive @z 1
dz1 + @z 1
dz2 = dq. On the iso-quant
curve, dq = 0. Thus, we derive
,
dz2 @f @f
= .
dz1 dq=0 @z1 @z2

As a solution to CmP, we derive the conditional factor demand z(w, q), and we
then derive the cost function c(w, q) := w · z(w, q).
Prop 5.C.2:
1. c(w, q) is homogeneous of degree 1 in w and nondecreasing in q.
2. c(w, q) is concave in w.
3. Dw c(w, q) = z(w, q) : Shephard’s lemma
4. Dw z(w, q) = Dw2 c(w, q): symmetric, negative semidefinite.
5. If f (·) is homogeneous of degree 1 (i.e., constant return to scale), then c(w, q)
and z(w, q) are homogeneous of degree 1 in q.
6. If f (·) is concave, then c(w, q) is a convex function in q (i.e., M C increases in
q).

20
 7. If the set {z 0 : f (z) q} is convex for every q, then the production set is
Y = {( z, q) : w · z c(w, q) for all w 0}.
8. If the set {z 0 : f (z) q} is convex, then z(w, q) is a convex set.
9. If the set {z 0 : f (z) q} is strictly convex, then z(w, q) is single-valued.

Profit Maximization Problem using Cost function

max p · q c(w, q) (P M P )
q 0

Prop P: If the producer is a price taker, the necessary first-order condition for the
solution q ⇤ is
@c(w, q ⇤ )
p , with equality if q ⇤ > 0. (5.C.6)
@q
If c(w, q) is convex in q, then the first-order condition (5.C.6) is also sufficient for
q ⇤ to be the firm’s optimal output level.
Example 5.C.1: Consider f (z1 , z2 ) = z1↵ z2 where ↵, > 0.

@f 1 ↵ z2 w1
F OC : = wi , i = 1, 2 ) ↵z1↵ 1 z2 = w1 , z1↵ z2 = w2 ) =
@zi z1 w2
✓ ◆ ✓ ◆ ↵+ ✓ ◆ ↵+↵
w1 1 ↵w2 1 w1
z(w, q) : q = z1↵ z1 ) z1 = q ↵+ , z2 = q ↵+
w2 ↵ w1 ↵w2
✓ ◆
1 ↵ ↵ ↵ ↵
) c(w, q) = w1 z1 + w2 z2 = q ↵+ ( ) ↵+ +( ) ↵+ w1 ↵+ w2 ↵+ .
⇣ ↵
⌘ ↵
For notational simplicity, define ✓ := ( ↵ ) ↵+ + ( ↵ ) ↵+ and (w1 , w2 ) := w1 ↵+ w2 ↵+ .
1
Then c(w, q) = q ↵+ ✓ (w1 , w2 ) and M C(q) is

@c(w, q) 1 1
1
= q ↵+ ✓ (w1 , w2 )
@q ↵+

which is increasing (constant, decreasing, resp.) in q if ↵ + < 1 (= 1, > 1, resp.).

(1) If ↵ + < 1 (DRS), then MC converges to 0 as q ! 0 and converges to 1
as q ! 1. Hence, there exists a unique optimal output level q ⇤ that solves (PMP),
characterised by the FOC:
✓ ◆ 1 ↵+↵
1 1
1 ⇤ p(↵ + )
p= q ↵+ ✓ (w1 , w2 ) ) q (p, w) =
↵+ ✓ (w1 , w2 )

21
This solution q ⇤ (p, w) is the firm’s supply function. Plugging q ⇤ (p, w) into the ob-
jective function of (PMP), we derive the firm’s profit function ⇡(p, w) = pq ⇤ (p, w)
c(w, q ⇤ (p, w)). Plugging q ⇤ (p, w) into the conditional factor demand, we derive (un-
conditional) factor demand zi (p, w).
(2) If ↵ + = 1 (CRS), then MC is a constant. If this is equal to p, any output
level is optimal with profit of 0. If this is larger than p then q = 0 is optimal.
Otherwise, there is no optimal output level, because producing more increases the
profit.
(3) If ↵ + > 1 (IRS), then MC converges to 0 as q ! 1. This means that the
profit increases without bound as q ! 1, so again there is no optimal output level.

Short-run and Long-run

– Short-run means a short span of time for which some inputs may not be
changed.
– Long-run means a long enough span of time for which any input may be
changed.
– Long-run cost function is a low envelope of short-run cost functions.
“Le Chatelier Principle”: The long-run response of a firm’s supply to a price
change is larger than the short-run response.
Proof. Consider a benchmark price p⇤ and the optimal long-run supply y(p⇤ ).
Let the first input, z1 , be fixed at z1⇤ = z1 (p⇤ ) in the short run. When the price
changes to p from p⇤ , let y(p) denote the long-run optimised supply for the changed
price, while y(p|z1⇤ ) denote the short-run optimised supply for the changed price.
We need to show that
dy(p) dy(p|z1⇤ )
() y 0 (p⇤ ) y 0 (p⇤ |z1⇤ ) 0.
dp p=p⇤ dp p=p⇤

As y(p) solves maxy py c(y), the first order condition is that p c0 (y) = 0 when
y = y(p). Hence, by di↵erentiating ⇡(p) = py(p) c(y(p)) with respect to p, we get

d⇡(p)
= y(p) + py 0 (p) c0 (y(p))y 0 (p) = y(p).
dp

(Note that this is Hotelling’s Lemma.) Similarly, as the short-run solution y(p|z1⇤ )
solves maxy py c(y|z1⇤ ), the first order condition is that p c0 (y|z1⇤ ) = 0 when

22
y = y(p|z1⇤ ). Hence, by di↵erentiating ⇡ S (p|z1⇤ ) = py(p|z1⇤ ) c(y(p|z1⇤ )|z1⇤ ) with
respect to p, we get

d⇡ S (p|z1⇤ )
= y(p|z1⇤ ) + py 0 (p|z1⇤ ) c0 (y(p|z1⇤ )|z1⇤ )y 0 (p|z1⇤ ) = y(p|z1⇤ ).
dp
Thus, we need to show
⇣ d2 ⇡(p) d2 ⇡ S (p|z1⇤ ) ⌘
y 0 (p⇤ ) y 0 (p⇤ |z1⇤ ) = 0. (1)
dp2 dp2 p=p⇤

Given any p, ⇡(p) denotes the long-run optimised profit level and ⇡ S (p|z1 ) de-
notes the short-run optimised profit level when the first input is fixed at z1 . Thus,
by definition,
⇡(p) = ⇡ S (p|z1 (p)) ⇡ S (p|z1⇤ )
where z1 (p) is the optimal amount of first input in the long-run. Then,

⇡(p|p⇤ ) := ⇡(p) ⇡ S (p|z1⇤ ) = ⇡ S (p|z1 (p)) ⇡ S (p|z1⇤ ) 0, equality when p = p⇤ .

I.e. ⇡(p|p⇤ ) achieves minimum at p⇤ . Thus,

d ⇡(p|p⇤ ) d2 ⇡(p|p⇤ )
= 0 and 0.
dp p=p⇤ dp2 p=p⇤

This proves (1) because

d2 ⇡(p|p⇤ ) ⇣ d2 ⇡(p) d2 ⇡ S (p|z1⇤ ) ⌘

= .
dp2 p=p⇤ dp2 dp2 p=p⇤

23
Monopoly Pricing: (Section 12.B) Let x(p) be the market demand function which
we assume is downward sloping. The inverse demand function is p(·) = x 1 (·) that
associates for each output level a price at which the entire output is demanded.
Then, the monopolist’s problem is
max p(q)q c(q) (12.B.2)
q 0

Under minor technical conditions, including di↵erentiability of p(·) and c(·), a ver-
sion of Prop P for a monopolist characterizes the interior solution q m as
p0 (q m )q m + p(q m ) = c0 (q m ), (12.B.4)
which states that the firm’s (private) marginal revenue equals the marginal cost.
As a benchmark, calculate the socially optimum production q o that maximizes
the total consumers surplus net of the production cost:
Z q̃
o
q solves max p(q)dq c(q̃) () p(q o ) = c0 (q o ) (SO)
q̃ 0

according to the FOC. Note that p(q) and c0 (q) are the “social marginal benefit”
and the “social marginal cost” of production. Since p0 (q) < 0 (i.e., social marginal
benefit decreases), provided that c0 (q) is non-decreasing in q, the FOC is also a
sufficient condition for social optimum. Hence, from comparing (12.B.4) and (SO),
we deduce that q m < q o [elaborate] and consequently, conclude (see a diagrammatic
illustration of (12.B.4) and (SO) in Figure 12.B.1):
• the monopoly price exceeds marginal cost;
• the monopoly output falls short of the socially optimum level;
• the social welfare is lower under monopoly, which is represented by the dead-
weight loss triangle.

Figure 12.B.1

24
The inefficiency in monopoly stems from the following fact: Increasing production
above q m increases social welfare because the willingness to pay for the additional
output exceeds the production cost, however this would take place at the expense
of the monopolist due to a lower price corresponding to a larger output.
Exercise 1.6: Suppose the demand function is x(p) = a bp and the monopolist’s
cost function is c(q) = cq where a, b, c > 0. Find the monopolist’s optimal quan-
tity (q m ) and price (pm ). What are the socially optimal output level (q o ) and the
corresponding price?

(Chapters in M-W-G summarized in this lecture on production: Sections 5.A and

5.C of Chap 5, and Section 12.B of Chap 12 — Read carefully and absorb fully.)

Problem Set #1
(due date to be determined)

Q1: Q1 of Exercise questions on consume theory (Lecture notes, p14)

Q2: part (3) of Q2 of Exercise questions on consume theory.
Q3: 3.D.6 from MWG.
Q4: part (b) of 5.C.9 from MWG, where p is output price and w = (w1 , w2 ) is
input price vector.
Q5: 5.C.13 from MWG.

Additional Exercises in M-W-G (not to be handed in): 5.C.7. 5.C.9, 5.C.10, 12.B.2,
12.B.6, 12.B.7, 12.B.8, 12.B.9.

25