Scribd
Upload
105 views

Solution To Exercise 4.4: Econometric Theory and Methods

Uploaded by

Leonard Gonzalo Saavedra Astopilco
Copyright
© Attribution Non-Commercial (BY-NC)
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
105 views

Solution To Exercise 4.4: Econometric Theory and Methods

Uploaded by

Leonard Gonzalo Saavedra Astopilco
Copyright
© Attribution Non-Commercial (BY-NC)
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/ 2

Econometric Theory and Methods Answers to Starred Exercises 7

Solution to Exercise 4.4

4.4 Consider the random variables x


1
and x
2
, which are bivariate normal with
x
1
N(0,
2
1
), x
2
N(0,
2
2
), and correlation . Show that the expectation
of x
1
conditional on x
2
is (
1
/
2
)x
2
and that the variance of x
1
conditional
on x
2
is
2
1
(1
2
). How are these results modied if the means of x
1
and x
2
are
1
and
2
, respectively?
By the denition of the correlation coecient,
=
Cov(x
1
, x
2
)
_
Var(x
1
)Var(x
2
)
_
1/2
,
from which we deduce that Cov(x
1
, x
2
) =
1

2
. Thus x
1
and x
2
are bivariate
normal with mean zero and covariance matrix

_

2
1

1

2

2
2
_
.
At this point, the fastest way to proceed is to note that = AA

, with
A =
_

1
(1
2
)
1/2

1
0
2
_
.
It follows that x
1
and x
2
can be expressed in terms of two independent stan-
dard normal variables, z
1
and z
2
, as follows:
x
1
=
1
(1
2
)
1/2
z
1
+
1
z
2
, and x
2
=
2
z
2
,
from which we nd that x
1
=
1
(1
2
)
1/2
z
1
+(
1
/
2
)x
2
, where z
1
and x
2
are independent, since x
2
depends only on z
2
, which is independent of z
1
by
construction. Thus E(z
1
| x
2
) = 0, and it is then immediate that E(x
1
| x
2
) =
(
1
/
2
)x
2
. For the conditional variance, note that
x
1
E(x
1
| x
2
) =
1
(1
2
)
1/2
z
1
,
of which the variance is
2
1
(1
2
), as required.
If the means of x
1
and x
2
are
1
and
2
, respectively, then we have
x
1
=
1
+
1
(1
2
)
1/2
z
1
+
1
z
2
, and x
2
=
2
+
2
z
2
, (S4.01)
where z
1
and z
2
are still independent standard normal variables. Since z
2
=
(x
2

2
)/
2
, we nd that
E(x
1
| x
2
) =
1
+(
1
/
2
)(x
2

2
). (S4.02)
Copyright c 2003, Russell Davidson and James G. MacKinnon
Econometric Theory and Methods Answers to Starred Exercises 8
It remains true that x
1
E(x
1
| x
2
) =
1
(1
2
)
1/2
z
1
, and so the conditional
variance is unchanged.
The factorization of as AA

is somewhat tricky unless one applies Crouts


algorithm directly. A clumsier, but more direct, way to proceed is to note
that x

1
x is distributed as
2
(2), and can thus be expressed as the sum
of the squares of two independent standard normal variables. By x we mean
the vector with x
1
and x
2
as the only two components. By brute force, one
can compute that

1
=
1

2
1

2
2
(1
2
)
_

2
2

1

2

2
1
_
,
so that
x

1
x =
1

2
1

2
2
(1
2
)
_

2
2
x
2
1
2
1

2
x
1
x
2
+
2
1
x
2
2
_
.
By the operation of completing the square, the right-hand side of this equa-
tion becomes
1

2
1
(1
2
)
_
_
x
1
(
1
/
2
)x
2
_
2
+x
2
2
_

2
1
/
2
2

2
1
/
2
2
_
_
=
_
x
1
(
1
/
2
)x
2
_

2
1
(1
2
)
_
1/2
_
2
+
_
x
2

2
_
2
.
The two variables of which the squares are summed here can now be identied
with z
1
and z
2
of (S4.01), and the rest of the argument is as above.
Since the above calculations are all either tricky or heavy, it is convenient
to have a simpler way to rederive them when needed. This can be done by
making use of the fact that, from (S4.02), the conditional expectation of x
1
is
linear (more strictly, ane) with respect to x
2
. Therefore, consider the linear
regression
x
1
= +x
2
+u. (S4.03)
If both x
1
and x
2
have mean 0, it is obvious that = 0. The population
analog of the standard formula for the OLS estimator of tells us that
=
Cov(x
1
, x
2
)
Var(x
2
)
=

1

2
2
=

1

2
.
Thus we see that
x
1
=

1

2
x
2
+u. (S4.04)
It is then immediate that E(x
1
| x
2
) = (
1
/
2
)x
2
. The conditional variance
follows by the same argument as above, and it can also be seen to be simply
the variance of u in (S4.04).
Copyright c 2003, Russell Davidson and James G. MacKinnon

You might also like