Endogeneity: Yusep Suparman
Endogeneity: Yusep Suparman
Yusep Suparman
ENDOGENEITY
When the error term correlates to one or
more regressors and the OLS estimates can
be used.
THE CAUSES OF ENDOGENNEITY:
LAGGED DEPENDENT VARAIBEL
The model
yt 0 1 xt 2 yt 1 t
Assuming
t t 1 t
We have
yt 0 1 xt 2 yt 1 t 1 t
and
yt 1 0 1 xt 1 2 yt 2 t 1
THE CAUSES OF ENDOGENNEITY:
LAGGED DEPENDENT VARAIBEL
Since yt-1 is a function of εt-1 , both are correlated.
Hence as ρ≠0, yt-1 and εt in yt’s regression are correlated too.
Accordingly
E yt xt , yt 1 0 1 xt 2 yt 1 E t xt , yt 1
In which
E t xt , yt 1 0
OLS does not correspond to the conditional expectation.
OLS is biased and inconsistent.
THE CAUSES OF ENDOGENNEITY:
LAGGED DEPENDENT VARAIBEL
A lagged dependent variable combined with autocorrelation of
the error terms results in endogeneity.
Note
In the present of a lagged dependent variable as a regressor,
Durbin-Watson test for autocorrelation is not valid. One may
use Breusch-Godfrey Lagrange Multiplier test for
autocorrelation.
THE CAUSES OF ENDOGENNEITY:
MEASURMENT ERROR
The Model
yt 0 1wt t , t ~ iid 0, 2 and E t wt 0
Assume that wt cannot be measured accurately
xt wt ut
Hence we have
yt 0 1 xt t , with t t 1ut
Both xt and εt depend on ut, hence they are correlated, the OLS
estimator is inconsistent.
THE CAUSES OF ENDOGENNEITY:
MEASURMENT ERROR
ˆ 2
plim 1 1 1 2 u
2
w u
Note
Inconsistency of one regression coefficient estimator is usually
carrid over to all other regression coefficient estimators
THE CAUSES OF ENDOGENNEITY:
SIMULTANEITY
The Keynesian model
Ct 0 1Yt t
Yt Ct I t
1 1 2
cov Yt , t cov I t , t var t
1 1 1 1 1 1
plim ˆ1 1
cov Yt , t 1 1 1
2
assuming
cov I t , t 0
INSTRUMENTAL VARIABLE
ESTIMATOR
Consider
yi x1i β1 x2i 2 i
It requires
E i x1i 0 and E i x2i 0
to associate OLS estimator of the model to conditional
expectation or best linear approximation of yi given xs
INSTRUMENTAL VARIABLE
ESTIMATOR
E y x ˆ x ˆ x 0
i 1i 1 2i 2 2i
E y x ˆ x ˆ z 0
i 1i 1 2i 2 2i
INSTRUMENTAL VARIABLE
ESTIMATOR
The instrumental varaibles estimator can be solved from
y x βˆ
N
1
i 1i 1,iv x2i ˆ2,iv x1i 0
N i 1
y x βˆ
N
1
i 1i 1,iv x2i ˆ2,iv z2i 0
N i 1
We obtain
1 N
N
x1i x1i
βˆ iv z i xi z y i i
xi , z i
i 1 i 1 x2i z2i
INSTRUMENTAL VARIABLE
ESTIMATOR
N βiv β ~ N 0, Σ xz Σ zz Σ zx
ˆ 2 1 1
In practice
XZ ZZ
1 1
cov βˆ iv ˆ 2
ZX
ˆ
2 1
N p 1
y - Xβiv y - Xβˆ iv
ˆ
GENERAL INSTRUMENTAL VARIABLE
ESTIMATOR
Consider
yi xi β i
E i zi E yi xi β z i 0
INSTRUMENTAL VARIABLE
ESTIMATOR
and
1 N
N
βˆ iv z i xi z y i i
i 1 i 1
INSTRUMENTAL VARIABLE
ESTIMATOR
When R>K, there are more restrictions than the unknowns. The
estimators are derived by choosing β such that the sample
moment condition
y xβˆ z
N
1
i i iv i 0
N i 1
Here, we minimizing
1 N
1 N
QN β
N
i 1
yi xi β z i WN
N
i 1
i i i
y xβ z
INSTRUMENTAL VARIABLE
ESTIMATOR
In matrix formulation
1 1
QN β Z y - Xβ WN N Z y - Xβ
N
Differentiating with respect to β gives first order conditions
2XZW Zy 2XZW ZXβˆ 0
N N iv
and
βˆ iv XZWN ZX XZWN Zy
1
INSTRUMENTAL VARIABLE
ESTIMATOR
N
z z
i 1
i i
1
βˆ iv XZ ZZ ZX XZ ZZ Zy
1 1
N βiv β ~ N 0, Σ xz Σ zz Σ zx
ˆ 2 1 1
In practice,
1 N
ˆ βˆ iv ˆ 2 XZ ZZ ZX
1 1 2
var ˆ
2
yi xi βˆ iv
N i 1
TESTING FOR INSTRUMENTS
VALIDITY
2
1
N
N N
2 NQN βˆ iv ei z i ˆ z i zi ei z i ~ R2 K
i 1 i 1 i 1
GIVE as
TWO-STAGE LEAST SQAURE EST.
1
βˆ iv X
ˆ X
ˆ ˆ y
X
Package:
ivmodel
card.data