Hello! I apologize if some of the questions in this post seem simple, but this has to do with my thesis and I appreciate any assistance!
I am dealing with panel data involving some banks over 63 time periods(N=4736 T=63) balanced. I used the Hausman test to determine I needed to use a FE model. But, this is where the trouble begins and my questions begin! So I run a basic xtreg fe model and get this result
so now I go about running some test to check for serial correlation and heteroscedasticity.
Which if I interpret this correctly means my model doesn't suffer from hetero.
Now running xtserial I get
Meaning my model does suffer from autocorrelation. So what now with this information do I run a vce(robust) fe or is there some other tests I should run or other model or option I should use. I am outside my statistical chops currently but I am trying to learn.
Secondly if I were to use a VCE(Robust) model why when I run it using areg as such do I get such a different significance on some of my variables than using xtreg. It was my impression they were so similar that they should not differ by much?
Results from areg note that cert is just a unique identifier for each individual bank:
Now using a fe model:
I do not understand exactly why in particular lnassetsq became so insignificant in the FE model with robust errors but not in the areg? I am sorry if I am missing something elementary here.
I am dealing with panel data involving some banks over 63 time periods(N=4736 T=63) balanced. I used the Hausman test to determine I needed to use a FE model. But, this is where the trouble begins and my questions begin! So I run a basic xtreg fe model and get this result
Code:
xtreg zscore lnasset lnassetsq diverse leverage eeffqr DGS10 CPIAUCSL_PCH GDPC1_PC1, fe
Fixed-effects (within) regression Number of obs = 298,355
Group variable: cert Number of groups = 4,736
R-sq: Obs per group:
within = 0.0179 min = 62
between = 0.0242 avg = 63.0
overall = 0.0227 max = 63
F(8,293611) = 670.17
corr(u_i, Xb) = -0.0196 Prob > F = 0.0000
------------------------------------------------------------------------------
zscore | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lnasset | .4193168 .0420687 9.97 0.000 .3368634 .5017702
lnassetsq | -.0071421 .0016294 -4.38 0.000 -.0103357 -.0039485
diverse | .0000122 .000058 0.21 0.834 -.0001015 .0001259
leverage | -.013206 .0004068 -32.46 0.000 -.0140033 -.0124087
eeffqr | -.0001518 8.33e-06 -18.23 0.000 -.0001681 -.0001355
DGS10 | .0478444 .0021882 21.86 0.000 .0435557 .0521332
CPIAUCSL_PCH | .0676272 .0032808 20.61 0.000 .0611969 .0740575
GDPC1_PC1 | .0310941 .0008967 34.68 0.000 .0293366 .0328516
_cons | -1.795671 .2740217 -6.55 0.000 -2.332746 -1.258596
-------------+----------------------------------------------------------------
sigma_u | 1.7745564
sigma_e | .99459682
rho | .76095759 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(4735, 293611) = 196.07 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 (4736) = 379.43 Prob>chi2 = 1.0000
Now running xtserial I get
Code:
xtserial zscore lnasset lnassetsq diverse leverage eeffqr DGS10 CPIAUCSL_PCH GDPC1_PC1
Wooldridge test for autocorrelation in panel data
H0: no first-order autocorrelation
F( 1, 4735) = 226.237
Prob > F = 0.0000
Secondly if I were to use a VCE(Robust) model why when I run it using areg as such do I get such a different significance on some of my variables than using xtreg. It was my impression they were so similar that they should not differ by much?
Results from areg note that cert is just a unique identifier for each individual bank:
Code:
areg zscore lnasset lnassetsq diverse leverage eeffqr DGS10 CPIAUCSL_PCH GDPC1_PC1, a(cert) vce(robust)
Linear regression, absorbing indicators Number of obs = 298,355
Absorbed variable: cert No. of categories = 4,736
F( 8, 293611) = 343.16
Prob > F = 0.0000
R-squared = 0.7691
Adj R-squared = 0.7654
Root MSE = 0.9946
------------------------------------------------------------------------------
| Robust
zscore | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lnasset | .4193168 .0686047 6.11 0.000 .2848535 .5537801
lnassetsq | -.0071421 .0025061 -2.85 0.004 -.0120541 -.0022301
diverse | .0000122 .0000432 0.28 0.778 -.0000725 .0000969
leverage | -.013206 .0102436 -1.29 0.197 -.0332831 .0068711
eeffqr | -.0001518 .0001045 -1.45 0.146 -.0003566 .000053
DGS10 | .0478444 .0046113 10.38 0.000 .0388064 .0568824
CPIAUCSL_PCH | .0676272 .0041492 16.30 0.000 .0594948 .0757596
GDPC1_PC1 | .0310941 .0011823 26.30 0.000 .0287769 .0334113
_cons | -1.795671 .4056216 -4.43 0.000 -2.590678 -1.000664
------------------------------------------------------------------------------
Code:
xtreg zscore lnasset lnassetsq diverse leverage eeffqr DGS10 CPIAUCSL_PCH GDPC1_PC1, fe vce(robust)
Fixed-effects (within) regression Number of obs = 298,355
Group variable: cert Number of groups = 4,736
R-sq: Obs per group:
within = 0.0179 min = 62
between = 0.0242 avg = 63.0
overall = 0.0227 max = 63
F(8,4735) = 184.31
corr(u_i, Xb) = -0.0196 Prob > F = 0.0000
(Std. Err. adjusted for 4,736 clusters in cert)
------------------------------------------------------------------------------
| Robust
zscore | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lnasset | .4193168 .134761 3.11 0.002 .1551224 .6835111
lnassetsq | -.0071421 .0051472 -1.39 0.165 -.0172331 .0029488
diverse | .0000122 .0000445 0.27 0.784 -.000075 .0000994
leverage | -.013206 .0102906 -1.28 0.199 -.0333804 .0069684
eeffqr | -.0001518 .0001059 -1.43 0.152 -.0003594 .0000558
DGS10 | .0478444 .0067429 7.10 0.000 .0346253 .0610636
CPIAUCSL_PCH | .0676272 .0037723 17.93 0.000 .0602317 .0750227
GDPC1_PC1 | .0310941 .0014933 20.82 0.000 .0281667 .0340216
_cons | -1.795671 .8540467 -2.10 0.036 -3.47 -.1213422
-------------+----------------------------------------------------------------
sigma_u | 1.7745564
sigma_e | .99459682
rho | .76095759 (fraction of variance due to u_i)
------------------------------------------------------------------------------
Comment