Scribd
Upload
34 views

3sls Question

The document discusses the Hausman test statistic for testing endogeneity in a simultaneous equations model. It shows that the Hausman test statistic m can be written as the R-squared from regressing the residual of one equation on the instruments from both equations. This establishes that m has a chi-squared distribution with degrees of freedom equal to the number of excluded instruments, which enables its use as a test statistic. It also provides examples of calculating OLS and 2SLS estimates for the structural equations.

Uploaded by

jps.mathematics
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
34 views

3sls Question

The document discusses the Hausman test statistic for testing endogeneity in a simultaneous equations model. It shows that the Hausman test statistic m can be written as the R-squared from regressing the residual of one equation on the instruments from both equations. This establishes that m has a chi-squared distribution with degrees of freedom equal to the number of excluded instruments, which enables its use as a test statistic. It also provides examples of calculating OLS and 2SLS estimates for the structural equations.

Uploaded by

jps.mathematics
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

268 Badi Baltagi

Q 1 ˝
The latter term is obtained by using the partitioned inverse of Z0 .†
PX /Z. Using (11), the Hausman test statistic becomes
 2 2  0    
m D ¢Q 12 =¢Q 22 uQ 2 PX Z1 ¢Q 11  1=¢Q 11 Z01 PX Z1
 2   1  0 1  0 
 ¢Q 12 =¢Q 22 Z01 PX Z2 Z02 PX Z2 Z2 PX Z1 Z1 PX uQ 2 .
 2 
However, .Q¢11  1=¢Q 11/ D ¢Q 12 =¢Q 22 ; this enables us to write
  1 1
m D uQ 02 ZO 1 ZO 01 ZO 1  ZO 01 ZO 2 ZO 02 ZO 2 ZO 02 ZO 1 ZO 01 uQ 2 =¢Q 22 , (13)

where ZO i D PX Zi is the matrix of second stage regressors of 2SLS, for


i D 1, 2. This statistic is asymptotically distributed as ¦23 , and can be given
the following interpretation:
Claim: m D TR2 , where R2 is the R-squared of the regression of uQ 2 on
the set of second stage regressors of both equations, ZO 1 and ZO 2 .
Proof. This result follows immediately using the fact that ¢Q 22 D
uQ 02 uQ 2 =T D total sums of squares/T, and the fact that the regression sum
of squares is given by
" #1 " #
h i ZO 0 ZO O0 O ZO 01
0 O O 1 1 Z1 Z2
RSS D uQ 2 Z1 Z2 uQ 2 , (14)
ZO 02 ZO 1 ZO 02 ZO 2 ZO 02
  1 1
D uQ 2 Z1 ZO 01 ZO 1  ZO 01 ZO 2 ZO 02 ZO 2
0 O
ZO 02 ZO 1 ZO 01 uQ 2 ,

where the last equality follows from partitioned inverse and the fact that
ZO 02 uQ 2 D 0.

11.18 a. Using the order condition of identification, the first equation has two
excluded exogenous variables x2 and x3 and only one right hand sided
included endogenous variable y2 . Hence,

k2 D 2 > g1 D 1

and the first equation is over-identified with degree of over-identification


equal to one. Similarly, the second equation has one excluded exogenous
Chapter 11: Simultaneous Equations Model 269

variable x1 and only one right hand side included endogenous variable y1 .
Hence, k2 D 1 D g1 and the second equation is just-identified. Using the
rank condition of identification
" #
1 “12 ”11 0 0
A D ŒB  D
“21 1 0 ”22 ”23
with y0t D .y1t , y2t / and x0t D .x1t , x2t , x3t /. The first structural equation has
the following zero restrictions:
2 3
0 0
6 7
60 07
6 7
¥1 D 6 7 0
60 07 so that ’1 ¥1 D .”12 , ”13 / D 0
6 7
41 05
0 1
where ’01 is the first row of A. Similarly, the second structural equation has
the following zero restriction:
2 3
0
6 7
607
6 7
¥2 D 6 7 0
617 so that ’2 ¥2 D ”21 D 0
6 7
405
0
" #
0 0
where ’02is the second row of A. Hence, A¥1 D which has
”22 ”23
rank !one provided either ”22 or ”23 is different from zero. Similarly, A¥2 D
”11
which has rank one provided ”11 ¤ 0. Hence, both equations are
0
identified by the rank condition of identification.
b. For the first equation, the OLS normal equations are given by
!
 0  “O 12
Z1 Z 1 D Z01 y1
”O 11 ols
270 Badi Baltagi

where Z1 D Œy2 , x1 . Hence,


" # " #
0 y02 y2 y02 x1 8 10
Z1 Z1 D 0 D
x1 y2 x01 x1 10 20
! !
y0 y1 4
Z01 y1 D 20 D
x1 y1 5
obtained from the matrices of cross-products provided by problem 11.18.
Hence, the OLS normal equations are
" # ! !
8 10 “O 12 4
D
10 20 ”O 11 ols 5
solving for the OLS estimators, we get
! " # ! !
“O 12 1 20 10 4 0.5
D D
”O 11 60 10 8 5 0
ols
OLS on the second structural equation can be deduced in a similar fashion.
c. For the first equation, the 2SLS normal equations are given by
!
 0  “O 12
Z1 PX Z1 D Z01 PX y1
”O 11 2SLS
where Z1 D Œy2 , x1  and X D Œx1 , x2 , x3 . Hence,
" # " #
0 y02 x1 y02 x2 y02 x3 10 20 30
Z1 X D 0 D
x1 x1 x01 x2 x01 x3 20 0 0
2 3 2 3 0 1
1=20 0 0 x01 y1 5
0 1 6 7 0 6 0 7 B C
.X X/ D 4 0 1=20 0 5 and X y1 D 4x2 y1 5 D @40A
0 0 1=10 x03 y1 20
" #
0 0 1 0.5 1 3
Z1 X.X X/ D and
1 0 0
2 3
" # 10 20 " #
0 0.5 1 3 6 7 115 10
Z1 PX Z1 D 420 0 5 D
1 0 0 10 20
30 0
with
0 1
" # 5 !
0.5 1 3 B C 102.5
Z01 PX y1 D @40A D .
1 0 0 5
20
Chapter 11: Simultaneous Equations Model 271

Therefore, the 2SLS normal equations are


" # ! !
115 10 “O 12 102.5
D
10 20 ”O 11 2SLS 5
solving for the 2SLS estimates, we get
! " # ! ! !
“O 12 1 20 10 102.5 1 2000 10=11
D D D
”O 11 2200 10 115 5 2200 250 9=44
2SLS
2SLS on the second structural equation can be deduced in a similar fashion.
d. The first equation is over-identified and the reduced form parameter esti-
mates will give more than one solution to the structural form parameters.
However, the second equation is just-identified. Hence, 2SLS reduces to
indirect least squares which in this case solves uniquely for the structural
parameters from the reduced form parameters. ILS on the second equation
yields
0 1
“O 21
B C 0 1 0
@”O 22 A D .X Z1 / X y2
”O 23 ILS
where Z1 D Œy1 , x2 , x3  and X D Œx1 , x2 , x3 . Therefore,
2 3 2 3
x01 y1 x01 x2 x01 x3 5 0 0
6 7 6 7
X0 Z1 D 4x02 y1 x02 x2 x02 x3 5 D 440 20 0 5 and
x03 y1 x03 x2 x03 x3 20 0 10
0 1 0 1
x01 y2 10
B C B C
X0 y2 D @x02 y2 A D @20A
x03 y2 30
so that
0 1 2 31 0 1
“O 21 5 0 0 10
B C 6 7 B C
@”O 22 A D 440 20 0 5 @20A
”O 23 ILS 20 0 10 30
Note that X0 Z1 for the first equation is of dimension 3  2 and is not even
square.

You might also like