Professional Documents
Culture Documents
+
|
|
.
|
\
|
=
V U
Y Q
19
Comparing the above two methods, the following
relationship between s and the structural parameters
must hold good, i.e.,
0 1 0 1
2
0 1
0 0
| o | o
|
| o
o |
+
|
|
.
|
\
|
=
V U
Y P
0 1
1 0 0 1
11
| o
| o | o
t
=
0 1
1 2
12
| o
o |
t
=
0 1
1 1
11
| o
| o
=
V U
e
20
As it may be observed s bear a definite relationship
with the structural parameters. The reduced form
parameters measure the total effect (direct and indirect) of
a change in the exogenous variables on the endogenous
variable. For instance in the above model,
12
measures the
total effect of unit change in the disposable income on the
quantity.
0 1
0 0
21
| o
o |
t
=
0 1
2
22
| o
|
t
=
0 1
22
| o
=
V U
e
21
4) Recursive Models
Because of the interdependence between the
disturbance term and the endogenous variables, the OLS
technique is not appropriate for the estimation of an
equation in the simultaneous equations model. However, in
a special type of simultaneous equations model called
Recursive, Triangular or Casual model, the use of OLS
procedure of estimation is appropriate.
22
Consider following three equation system.
Xs and Ys are the exogenous and endogenous
variables respectively. The disturbance term follow the
following assumption:
3 2 32 1 31 2 32 1 31 30 3
2 2 22 1 21 1 21 20 2
1 2 12 1 11 10 1
U X X Y Y Y
U X X Y Y
U X X Y
+ + + + + =
+ + + + =
+ + + =
| | o o o
| | o o
| | o
0 ) , ( ) , ( ) , (
3 2 3 1 2 1
= = = U U E U U E U U E
23
The above assumption is most crucial assumption that
defines recursive model. If this does not hold good, the above
system is no longer recursive and recursive OLS is also no longer
valid.
The first equation given above contains only the exogenous
variable right hand side. Since by assumption, the exogenous
variables are independent of U
1
, first equation satisfy the
crucial assumption of OLS and hence OLS can be applied to this
equation.
1 2 12 1 11 10 1
U X X Y + + + = | | o
24
Second equation contains endogenous variable Y
1
as
one of explanatory variables along with the non-stochastic
Xs. OLS can be applied to this equation only if it can be
shown that Y
1
and U
2
are independent of each other.
This is true because U
1
which affects Y
1
is by
assumption uncorrelated with U
2
, i.e., E(U
1
, U
2
) = 0. Y
1
in
fact act as a predetermined variable insofar Y
2
is
concerned. Hence OLS can be applied to this equation also.
2 2 22 1 21 1 21 20 2
U X X Y Y + + + + = | | o o
25
Similar argument can be stretched to the third
equation because Y
1
and Y
2
are independent of U
3
.
In this way, in recursive system OLS can be applied to
each equation separately.
3 2 32 1 31 2 32 1 31 30 3
U X X Y Y Y + + + + + = | | o o o
26
Fig. 3. Unidirectional flow in the recursive system
All the predetermined variables and U
1
determines Y
1
.
Y
1
and all the predetermined variables and U
2
determines Y
2
.
Y
1
, Y
2
and all the predetermined variables and U
3
determines
Y
2
.
27
SIMULTANEOUS EQUATION METHODS
Following are the frequently used single equation
methods:
1) Indirect Least Squares (ILS)
2) Two Stage Least Squares (2 SLS)
28
1. Indirect Least Squares (ILS)
This method is designed to estimate one
equation at a time. The method is named indirect
least squares since it estimates the parameters
indirectly by estimating the reduced form
equations, in which endogenous variables are
expressed only as a function of exogenous variables
and of the error term.
29
Steps of Indirect Least Squares
Step 1:
Obtain the reduced form equations. These reduced
form equations are obtained from structural equations in
such a manner that the dependent variable in each
equation is the only endogenous variable and is a function
solely of the predetermined (exogenous) variable and the
stochastic error term(s).
Step 2:
We apply OLS to the reduced form equations
individually. This is permissible since the explanatory
variables in this equations predetermined and hence
uncorrelated with the stochastic disturbances. The
estimates thus obtained are consistent.
30
Step 3:
We obtain estimates of the original structural
coefficients from the estimated reduced form coefficients
obtained in step 2. If an equation is exactly identified,
there is a one - to - one correspondence between the
structural and reduced form coefficients; that is, one can
derive unique estimates of the former from the latter.
Let us take an exactly identified simple demand-
supply model.
31
Let us take an exactly identified simple demand-supply
model.
Where,
D = Quantity demanded
S = Quantity supplied
P = Price of the commodity
Y = Income
W = Weather index
t t
t t t
t t t
S D
U W P S
U Y P D
=
+ + + =
+ + + =
2 2 1 0
1 2 1 0
| | |
o o o
32
The above model is mathematically complete , i.e.,
there are three endogenous variables (D, S, P) and three
equations.
First step of ILS is to obtain reduced form of model.
1 1
2 1
1 1
2
1 1
2
1 1
0 0
o | o |
|
o |
o
o |
| o
+
|
|
.
|
\
|
|
|
.
|
\
|
=
U U
W Y P
t t t
1 1
1 2 1 1
1 1
2 1
1 1
1 2
1 1
0 1 1 0
o |
o |
o |
| o
o |
| o
o |
| o | o
+
|
|
.
|
\
|
|
|
.
|
\
|
=
U U
W Y D
t t t
2 22 21 20
1 12 11 10
V W Y P
V W Y D
t t t
t t t
+ + + =
+ + + =
t t t
t t t
33
Where s and Vs depict the coefficients and disturbances
of the reduced form model. Since the reduced form equations
do not contain endogenous variables, the application of the OLS
method to each reduced form equation lead to unbiased
estimates of s. Hence using sample data on D, P, Y and W, we
may obtain estimates of s.
We are interested in the original structural parameters of
the model. Since the estimates if s are function of these
parameters, we estimate them indirectly from the s through
the following manipulation.
34
(a) Calculation of
0
o
0
0 0
1 0 1 0
1 1
0 0
22
12
20
10
20
1
22
12
0 0
0 1 1 0
20
10
o
| o
o o | o
o |
| o
t
t
t
t
t
o
t
t
| o
| o | o
t
t
=
|
|
.
|
\
|
=
|
|
.
|
\
|
=
Hence
and
in s estimated
relevant the substitute therefore we obtain to
'
,
0
t
o
(
|
|
.
|
\
|
22
12
20
10
20
t
t
t
t
t
35
(b) Calculation of
0
|
(
|
|
.
|
\
|
=
|
|
.
|
\
|
=
|
|
.
|
\
|
=
21
11
20
10
20
0
0
0 0
1 0 1 0
1 1
0 0
21
11
20
10
20
1
21
11
0 0
0 1 1 0
20
10
'
t
t
t
t
t
t |
|
| o
o | | |
o |
| o
t
t
t
t
t
|
t
t
| o
| o | o
t
t
in s substitute we obtain to Theref ore
Hence
and
36
(c) Calculation of
2
o
( )
(
|
|
.
|
\
|
=
|
|
.
|
\
|
= =
22
12
21
11
21
2
2 1 1
1 1
2
22
12
21
11
21
1
22
12
1
21
11
'
t
t
t
t
t
t o
o o |
o |
o
t
t
t
t
t
o
t
t
|
t
t
in s substitute we obtain to Theref ore
37
(d) Calculation of
2
|
( )
.
'
Re
2
21
11
22
12
22
2 1 1
1 2
2
21
11
22
12
22
1
21
11
1
22
12
|
t
t
t
t
t
t
| | o
o |
|
t
t
t
t
t
|
t
t
o
t
t
obtain to
in d substitute be to are s levant
(
|
|
.
|
\
|
=
|
|
.
|
\
|
= =
38
(c) Calculation of
1 1
| o and
. identified exactly is model equation us simultaneo
the only when s ' and s ' of values unique to
rise give will s and ts coefficien structural
e between th relation that noted be should It
21
11
1
22
12
1
| o
t
t
t
|
t
t
o = = and
39
The following are the assumptions of ILS procedure:
(1) The structural equation must be exactly identified.
(2) The disturbance term of the reduced form equation must
satisfy all the assumption of OLS. This is essential because
this method is merely application of OLS to the reduced
form equations.
If these assumptions are not fulfilled, the bias in s
will be transmitted to the estimates of the structural
parameters.
It may, therefore, be said that ILS method is based on
all the assumptions of OLS along with additional
assumption that the model be exactly identified.
40
2. Two Stage Least Squares
This method also being a single equation method seek
to remove the defect of existence of the correlation
between the disturbance term and the independent
variable(s) so that when we apply OLS technique to each
structural equation separately, the simultaneity bias gets
eliminated. Therefore, 2SLS may be considered as an
extension of ILS method.
41
In ILS interdependence between explanatory variable
and disturbance term is bypassed by applying OLS to
reduced form equation; here we purge the explanatory
variable(s) which is correlated with the error term with its
own estimated value. This is done in two stages.
Let us take demand-supply model.
t t
t t t
t t t
S D
U W P S
U Y P D
=
+ + + =
+ + + =
2 2 1 0
1 2 1 0
| | |
o o o
42
Reduced form equations are:
Where s and Vs are estimated by applying OLS to
these reduced form equations. This is the first stage of
estimation.
Having estimated s, now replace for different
values of Y
t
and W
t
. Now replace P
t
in the structural model
by obtained in the first stage as follows:
2 22 21 20
1 12 11 10
V W Y P
V W Y D
t t t
t t t
+ + + =
+ + + =
t t t
t t t
t
P
t
P
43
This is now a transformed model.
Since is based on the estimates from the reduced
form equations, it act as an instrumental variable for the
original data on .
All the structural parameters are estimated by applying
OLS to these transformed equations. This is second stage of
estimation.
2 2 1 0
1 2 1 0
U W P S
U Y P D
t t t
t t t
+ + + =
+ + + =
| | |
o o o
P
t
P
44
Features of 2SLS
Following are the features of 2SLS
(1) Unlike ILS, which provides multiple estimates of the
parameters when applied to over-identified equations,
2SLS provides only one estimate per parameter.
(2) Although 2SLS has been specially designed to handle over-
identified equations, this method can also be applied to
exactly identified equations. ILS and 2SLS give identical
results in such situation.
45
(3) In the application ILS, there is no simple method of
estimation of standard errors of the structural coefficients
from the standard errors of the reduced form coefficients.
But this can be done easily in case of 2SLS estimates
because the structural coefficients are directly estimated
from the second stage (OLS) regressions. However, the
estimates standard errors in the second stage regression
need to be modified.
46
Table 2. 2SLS estimates for simultaneous equation model
Variables
PUBINV PVTINV PROD
Intercept
4144.952***
(11.147)
-272.164***
(3.567)
-1304.541
(1.276)
PUBINV
0.089***
(3.368)
0.837**
(2.182)
PVTINV
0.997***
(2.763)
PROD
0.033***
(3.831)
0.007**
(1.977)
POVR
-1.950***
(3.367)
POPGR
42.561***
(3.393)
2.836
(0.377)
166.236***
(3.096)
LITR
1.979***
(7.409)
0.308*
(1.673)
7.633***
(5.843)
47
***, **, * Significant at 1, 5, and 10 per cent level respectively.
Figures in parentheses are t values
ICAR (New Delhi)
Roy and Pal (2002)
Table 2. Continue.
Variables PUBINV PVTINV PROD
GOVREV 0.007***
(3.688)
GRANTS 0.526***
(3.673)
SUBSG -0.169***
(4.0647)
SUBTOT 0.509**
(2.413)
SUBINP 0.181***
(7.167)
ToT 4.206***
(6.022)
20.770***
(4.059)
CREDIT 0.072
(1.127)
1.975***
(4.986)
48
***, **, * Significant at 1, 5, and 10 per cent level respectively.
Figures in parentheses are t values
ICAR (New Delhi)
Roy and Pal (2002)
Table 2. Continue..
Variables PUBINV PVTINV PROD
Cropping Intensity 8.861*
(1.949)
STORE 14.245***
(5.595)
RAIND -30.745***
(10.403)
Adjusted R
2
0.832 0.898 0.941
F Value 110.10*** 174.85*** 292.09***
D-W statistics 1.857 1.815 2.081
49
***, **, * Significant at 1, 5, and 10 per cent level respectively.
Figures in parentheses are t values
ICAR (New Delhi)
Roy and Pal (2002)
CONCLUSION
The single equation methods are more popular to solve
simultaneous equation models. A unique feature of this
method is that one can estimate a single equation in a multi
equation model without worrying too much about other
equations in the model.
Although OLS is, in general, inappropriate in the context of
simultaneous equation models, it can be applied to the
recursive models where there is a definite but unidirectional
cause-and-effect relationship among the endogenous variables
50
Unique feature of both ILS and 2SLS is that the estimates
obtained thereof are consistent, that is, as the sample
increase indefinitely the estimates tend tot heir true
population values.
51
52