Hello Statalist members,
I’m in need of some help, I have a model off differences in differences where I’m testing for the impact of a policy change in the bank industry, it’s an unbalanced panel.
I run the model as it follows:
P_dummy is a time dummy, T_dummy is a treatment dummy, DifDif is the product of T_dummy and P_dummy, the others are control variables
The xtoverid command suggest that I should use Fixed effects, which I did, as it follows:
After this I tested for heteroskedasticity and serial correlation:
And so I get to the conclusion the model have both. I have searched some things and apparently if I go with robust errors this will deal with those problems, and that is what i did:
And so I test again for both the problems:
. xttest3
And nothing changes, is the serial correlation and the heteroskedasticity to severe to correct? There are no other alternatives? If so, the model has no use?
Hope someone can give me good news...
Thank you all in advance!
I’m in need of some help, I have a model off differences in differences where I’m testing for the impact of a policy change in the bank industry, it’s an unbalanced panel.
I run the model as it follows:
P_dummy is a time dummy, T_dummy is a treatment dummy, DifDif is the product of T_dummy and P_dummy, the others are control variables
Code:
. xtset
panel variable: ID (unbalanced)
time variable: time, 2002m3 to 2012m4
delta: 1 month
. xtreg CRk DifDif P_dummy T_dummy DRLP Un, re
Random-effects GLS regression Number of obs = 3340
Group variable: ID Number of groups = 64
R-sq: within = 0.2215 Obs per group: min = 3
between = 0.6283 avg = 52.2
overall = 0.4601 max = 103
Wald chi2(5) = 930.00
corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0000
------------------------------------------------------------------------------
CRk | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
DifDif | 1.18954 .0496293 23.97 0.000 1.092268 1.286811
P_dummy | -.3073857 .0482846 -6.37 0.000 -.4020218 -.2127497
T_dummy | .7410848 .1687651 4.39 0.000 .4103114 1.071858
DRLP | -.4023052 .0925115 -4.35 0.000 -.5836245 -.2209859
Un | -.0908159 .016692 -5.44 0.000 -.1235315 -.0581002
_cons | 3.8108 .5965272 6.39 0.000 2.641628 4.979972
-------------+----------------------------------------------------------------
sigma_u | 4.6378604
sigma_e | .55184455
rho | .98603977 (fraction of variance due to u_i)
------------------------------------------------------------------------------
. xtoverid
Test of overidentifying restrictions: fixed vs random effects
Cross-section time-series model: xtreg re
Sargan-Hansen statistic 88.754 Chi-sq(5) P-value = 0.0000
Code:
. xtreg CRk DifDif P_dummy T_dummy DRLP Un, fe
Fixed-effects (within) regression Number of obs = 3340
Group variable: ID Number of groups = 64
R-sq: within = 0.2217 Obs per group: min = 3
between = 0.6347 avg = 52.2
overall = 0.4463 max = 103
F(5,3271) = 186.38
corr(u_i, Xb) = 0.6223 Prob > F = 0.0000
------------------------------------------------------------------------------
CRk | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
DifDif | 1.184449 .0490231 24.16 0.000 1.08833 1.280568
P_dummy | -.3048414 .0476903 -6.39 0.000 -.3983473 -.2113355
T_dummy | .5766794 .1679742 3.43 0.001 .2473342 .9060246
DRLP | -.400709 .0913726 -4.39 0.000 -.5798624 -.2215556
Un | -.0895465 .0164898 -5.43 0.000 -.1218778 -.0572152
_cons | 4.075789 .1099719 37.06 0.000 3.860168 4.29141
-------------+----------------------------------------------------------------
sigma_u | 7.1052858
sigma_e | .55184455
rho | .99400404 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(63, 3271) = 5213.34 Prob > F = 0.0000
Code:
. xttest3
Modified Wald test for groupwise heteroskedasticity
in fixed effect regression model
H0: sigma(i)^2 = sigma^2 for all i
chi2 (64) = 4.5e+07
Prob>chi2 = 0.0000
. xtserial CRk DifDif P_dummy T_dummy DRLP Un
Wooldridge test for autocorrelation in panel data
H0: no first-order autocorrelation
F( 1, 63) = 377.268
Prob > F = 0.0000
Code:
. xtreg CRk DifDif P_dummy T_dummy DRLP Un, fe robust
Fixed-effects (within) regression Number of obs = 3340
Group variable: ID Number of groups = 64
R-sq: within = 0.2217 Obs per group: min = 3
between = 0.6347 avg = 52.2
overall = 0.4463 max = 103
F(4,63) = .
corr(u_i, Xb) = 0.6223 Prob > F = .
(Std. Err. adjusted for 64 clusters in ID)
------------------------------------------------------------------------------
| Robust
CRk | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
DifDif | 1.184449 .5629363 2.10 0.039 .0595108 2.309388
P_dummy | -.3048414 .1498798 -2.03 0.046 -.6043522 -.0053306
T_dummy | .5766794 .1579478 3.65 0.001 .261046 .8923129
DRLP | -.400709 .2100127 -1.91 0.061 -.8203858 .0189678
Un | -.0895465 .0442645 -2.02 0.047 -.1780021 -.0010909
_cons | 4.075789 .3456725 11.79 0.000 3.385018 4.766561
-------------+----------------------------------------------------------------
sigma_u | 7.1052858
sigma_e | .55184455
rho | .99400404 (fraction of variance due to u_i)
------------------------------------------------------------------------------
. xttest3
Code:
Modified Wald test for groupwise heteroskedasticity
in fixed effect regression model
H0: sigma(i)^2 = sigma^2 for all i
chi2 (64) = 4.5e+07
Prob>chi2 = 0.0000
. xtserial CRk DifDif P_dummy T_dummy DRLP Un
Wooldridge test for autocorrelation in panel data
H0: no first-order autocorrelation
F( 1, 63) = 377.268
Prob > F = 0.0000
Hope someone can give me good news...
Thank you all in advance!
Comment