December 27, 2003

Hayashi Econometrics

Solution to Chapter 3 Analytical Exercises

1. If A is symmetric and idempotent, then A0 = A and AA = A. So x0 Ax = x0 AAx =
x0 A0 Ax = z0 z 0 where z Ax.
2. (a) By assumption, {xi , i } is jointly stationary and ergodic, so by ergodic theorem the first
term of () converges almost surely to E(x2i 2i ) which exists and is finite by Assumption
3.5.
(b) zi x2i i is the product of xi i and xi zi . By using the Cauchy-Schwarts inequality, we obtain
q
E(|xi i xi zi |) E(x2i 2i ) E(x2i zi2 ).
E(x2i 2i ) exists and is finite by Assumption 3.5 and E(x2i zi2 ) exists and is finite by Assumption 3.6. Therefore, E(|xi zi xi i |) is finite. Hence, E(xi zi xi i ) exists and is finite.
(c) By ergodic stationarity the sample average of zi x2i i converges in probability to some finite
number. Because b is consistent for by Proposition 3.1, b converges to 0 in probability.
Therefore, the second term of () converges to zero in probability.
(d) By ergodic stationarity and Assumption 3.6 the sample average of zi2 x2i converges in probability to some finite number. As mentioned in (c) b converges to 0 in probability.
Therefore, the last term of () vanishes.
3. (a)
Q

0xz S1 xz 0xz Wxz (0xz WSWxz )1 0xz Wxz

0xz C0 Cxz 0xz Wxz (0xz WC1 C0

=
=
=
=

Wxz )1 0xz Wxz

0xz Wxz (G0 G)1 0xz Wxz

0
0
1 0

HH
H0 H H G(G G)
0

H [IK G(G G)
H0 MG H.

GH

G0 ]H

(b) First, we show that MG is symmetric and idempotent.

MG 0

MG MG

=
=
=

IK G(G(G0 G)1 )0
IK G((G0 G)10 G0 )
IK G(G0 G)1 G0

MG .

=
=

IK IK G(G0 G)1 G0 IK IK G(G0 G)1 G0 + G(G0 G)1 G0 G(G0 G)1 G0

IK G(G0 G)1 G0

MG .

Thus, MG is symmetric and idempotent. For any L-dimensional vector x,

x0 Qx =
=

x0 H0 MG Hx
z0 MG z (where z Hx)
0

(since MG is positive semidefinite).

Therefore, Q is positive semidefinite.

1

4. (the answer on p. 254 of the book simplified) If W is as defined in the hint, then
WSW = W

and

0xz Wxz = zz A1 zz .

So (3.5.1) reduces to the asymptotic variance of the OLS estimator. By (3.5.11), it is no smaller
than (0xz S1 xz )1 , which is the asymptotic variance of the efficient GMM estimator.
bS
e (given in
b 1 ) (given in (3.5.12)) and the expression for gn ()
5. (a) From the expression for (
1
b
b
b
b
b
(3.4.2)), it is easy to show that gn ((S )) = Bsxy . But Bsxy = Bg because
b xy = (IK Sxz (S0 S
b 1 Sxz )1 S0 S
b 1 )sxy
Bs
xz
xz
b 1 Sxz )1 S0 S
b 1 )(Sxz + g)
= (IK Sxz (S0 S

(since yi = z0i + i )
b 1 Sxz )1 S0 S
b 1 Sxz ) + (IK Sxz (S0 S
b 1 Sxz )1 S0 S
b 1 )g
= (Sxz Sxz (S0xz S
xz
xz
xz
b
= (Sxz Sxz ) + Bg
b
= Bg.
xz

xz

b 1 = C0 C, we obtain B
b 0S
b 1 B
b =B
b 0 C0 CB
b = (CB)
b 0 (CB).
b But
(b) Since S
b
CB

=
=

b 1 Sxz )1 S0 S
b 1 )
C(IK Sxz (S0xz S
xz
C CSxz (S0xz C0 CSxz )1 S0xz C0 C

= C A(A0 A)1 A0 C
= [IK A(A0 A)1 A0 ]C

(where A CSxz )

MC.

b 1 B
b = (MC)0 (MC) = C0 M0 MC. It should be routine to show that M is symmetb 0S
So B
b 0S
b 1 B
b = C0 MC.
ric and idempotent. Thus B
The rank of M equals its trace, which is
trace(M)

= trace(IK A(A0 A)1 A0 )

= trace(IK ) trace(A(A0 A)1 A0 )
=
=
=

trace(IK ) trace(A0 A(A0 A)1 )

K trace(IL )
K L.

b 1 . Let D be such that D0 D = S1 . The choice of C and D

(c) As defined in (b), C0 C = S
is not unique, but it would be possible to choose C so that plim C = D. Now,

v n(Cg) = C( n g).

By using the Ergodic Stationary Martingale Differences CLT, we obtain n g d N (0, S).
So

v = C( n g) N (0, Avar(v))
d

where
Avar(v)

=
=

DSD0
D(D0 D)1 D0

=
=

DD1 D10 D0
IK .

(d)
bS
b 1 ), S
b 1 )
J((

=
=
=
=
=

bS
bS
b 1 )) S
b 1 gn ((
b 1 ))
n gn ((
b 0S
b 1 (Bg)
b
n (Bg)
(by (a))

b 0S
b 1 Bg
b
n g0 B

n g0 C0 MCg
(by (b))

v0 Mv
(since v nCg).

Since v d N (0, IK ) and M is idempotent, v0 Mv is asymptotically chi-squared with

degrees of freedom equaling the rank of M = K L.
b 0S
b 1 Bg.
b Also from Exercise 5, Bg
b = Bs
b xy .
6. From Exercise 5, J = ng0 B

7. For the most parts, the hints are nearly the answer. Here, we provide answers to (d), (f), (g),
(i), and (j).
(d) As shown in (c), J1 = v10 M1 v1 . It suffices to prove that v1 = C1 F0 C1 v.

v1 nC1 g1

= nC1 F0 g

= nC1 F0 C1 Cg

= C1 F0 C1 nCg

= C1 F0 C1 v
(since v nCg).
(f) Use the hint to show that A0 D = 0 if A01 M1 = 0. It should be easy to show that A01 M1 = 0
from the definition of M1 .
(g) By the definition of M in Exercise 5, MD = D A(A0 A)1 A0 D. So MD = D since
A0 D = 0 as shown in the previous part. Since both M and D are symmetric, DM =
D0 M0 = (MD)0 = D0 = D. As shown in part (e), D is idempotent. Also, M is idempotent
as shown in Exercise 5. So (M D)2 = M2 DM MD + D2 = M D. As shown in
Exercise 5, the trace of M is K L. As shown in (e), the trace of D is K1 L. So the
trace of M D is K K1 . The rank of a symmetric and idempotent matrix is its trace.
b 0S
b 1 B.
b
(i) It has been shown in Exercise 6 that g0 C0 MCg = s0xy C0 MCsxy since C0 MC = B
0 0
0
0
Here, we show that g C DCg = sxy C DCsxy .
g0 C0 DCg = g0 FC01 M1 C1 F0 g
(C0 DC = FC01 M1 C1 F0 by the definition of D in (d))
b 0 (S
b 11 )1 B
b 1 F0 g
b 0 (S
b 11 )1 B
b 1 from (a))
= g0 FB
(since C0 M1 C1 = B
1

b 0 (S
b 11 )1
g01 B
1

b 1g
B
1

(since g1 = F g).

b 1 and the fact that sx y = Sx z + g , it follows that B

b 1g =
From the definition of B
1
1
1
1
b 1 sx y . So
B
1
b 0 (S
b 11 )1 B
b 1 g1 = s0 B
b 0 b 1 B
b 1 sx y
g01 B
1
x1 y 1 (S11 )
1

b 0 (S
b 11 )1 B
b 1 F0 sxy
= s0xy FB
1
= s0xy FC01 M1 C1 F0 sxy
= s0xy C0 DCsxy .
3

(since sx1 y = F0 sxy )

b 0 (S
b 11 )1 B
b 1 = C0 M1 C1 from (a))
(since B
1
1

(j) M D is positive semi-definite because it is symmetric and idempotent.

8. (a) Solve the first-order conditions in the hint for to obtain
b W)
c 1 (S0 WS
c xz )1 R0 .
= (
2n xz
Substitute this into the constraint R = r to obtain the expression for in the question.
Then substitute this expression for into the above equation to obtain the expression for
in the question.
(b) The hint is almost the answer.
b W)
b W)
c )0 (S0 WS
c xz )((
c ) equals the Wald
(c) What needs to be shown is that n((
xz
statistic. But this is immediate from substitution of the expression for in (a).
9. (a) By applying (3.4.11), we obtain
"
# "
#
b1 )
c 1 Sxz )1 S0 W
c1
n(
(S0xz W
xz
=
ng.
b
c 2 Sxz )1 S0 W
c2
n( 1 )
(S0 W
xz

xz

By using Billingsley CLT, we have

ng N (0, S).
d

Also, we have
"

#
 1 0

c 1 Sxz )1 S0 W
c1
(S0xz W
Q1 xz W1
xz

.
0
c 2 Sxz )1 S0 W
c 2 p Q1
2 xz W2
(S0 W
xz

xz

Therefore, by Lemma 2.4(c),

"
#

b1 )
n(
d N 0,
b
n( 1 )

= N 0,
(b)



0
.
Q1
1 xz W1 S (W Q1 .. W Q1 )
1 xz 1
2 xz 2
0
Q1
2 xz W2


A11 A12
.
A21 A22

nq can be rewritten as


b1
b2 ) = n(
b1 ) n(
b2 ) = 1
nq = n(
Therefore, we obtain

"
#
b1 )
n(
1
.
b2 )
n(


nq N (0, Avar(q)).
d

where

Avar(q) = 1


 A11
1
A21

A12
A22




1
= A11 + A22 A12 A21 .
1

(c) Since W2 = S1 , Q2 , A12 , A21 , and A22 can be rewritten as follows:

Q2

=
=

0xz W2 xz
0xz S1 xz ,

A12

0
1
Q1
xz Q1
1 xz W1 S S
2

=
=

0
1
Q1
1 (xz W1 xz )Q2
1
Q1
1 Q1 Q2

Q1
2 ,

1
Q1
SW1 xz Q1
2 xz S
1

Q1
2 ,

A21

= (0xz S1 xz )1 0xz S1 SS1 xz (0xz S1 xz )1

= (0xz S1 xz )1

A22

Q1
2 .

Substitution of these into the expression for Avar(q) in (b), we obtain

Avar(q)

A11 Q1
2

A11 (0xz S1 xz )1
bW
bS
b 1 )).
c 1 )) Avar((
Avar((

=
10. (a)
xz E(xi zi )

b
(b) From the definition of ,
b =

E(xi (xi + vi ))
E(x2i ) + E(xi vi )
(by assumptions (2), (3), and (4)).
x2 6= 0

1X
xi zi
n i=1

!1

1X
1X
xi i = s1
xi i .
xz
n i=1
n i=1

We have xi zi = xi (xi + vi ) = x2i + xi vi , which, being a function of (xi , i ), is ergodic

stationary by assumption (1). So by the Ergodic theorem, sxz p xz . Since xz 6= 0 by
1
(a), we have s1
xz p xz . By assumption (2), E(xi i ) = 0. So by assumption (1), we have
Pn
1
xi i p 0. Thus b p 0.
n

(c)

=
=
=

i=1

sxz

1X
xi z i
n i=1

1X 2
(x + xi vi )
n i=1 i

1 1X 2 1X

x +
xi vi
n i=1
n n i=1 i

p
=

0 E(x2i ) + E(xi vi )
0
5

1
(since = )
n

(d)
n
n

1 X
1X 2
xi vi .
x +
nsxz =
n i=1 i
n i=1

By assumption (1) and the Ergodic Theorem, the first term of RHS converges in probability
to E(x2i ) = x2 > 0. Assumption (2) and the Martingale Differences CLT imply that
n

1 X

xi vi a N (0, s22 ).
d
n i=1
Therefore, by Lemma 2.4(a), we obtain

nsxz x2 + a.
d

(e) b can be rewritten as

b = ( nsxz )1 ng 1 .

From assumption (2) and the Martingale Differences CLT, we obtain

ng 1 b N (0, s11 ).
d

where s11 is the (1, 1) element of S. By using the result of (d) and Lemma 2.3(b),
b (x2 + a)1 b.
d

(a, b) are jointly normal because the joint distribution is the limiting distribution of
"
#

ng 1
n g = 1 Pn
.
n( n i=1 xi vi )
(f) Because b converges in distribution to (x2 + a)1 b which is not zero, the answer is No.