Panel data tests

[Carla Ferro]

Hi, I'm doing some estimates with panel data. I have some questions. Can I use -hettest- with panel data? I run -xtreg, fe- and -xtreg,re- and to test what was the best choice i did the hausman test, but because from -hettest- resulted that there was heteroskedaticity i used -xtoverid- after -xtreg, re cluster (clustvar)- . Is that correct? I got a p-value= 0,0013 so i rejected the null hypothesis, and prefer the fixed effect model.
After that does it make sense to run -xttest0- to investigate if there are random effects? I did and I got a p-value= 0,1889 so I can't reject the null right?
Best regards,
Carla.

[Andrew Musau]

Once you reject random effects as inconsistent, you cannot go back and compare it with anything else. So do not waste your time running xttest0. If you have \(N>>T\), estimate the fixed effects model and cluster on the panel variable. Don't even bother testing for heteroscedasticity or serial correlation.

[Carla Ferro]

Thank you for your answer. Fon N and T you mean time and countries right? Sorry I'm a beginner. Fon N I have 15 countries and for T 40 quarters so T>N. So in my case is correct to run -hettest- -xtreg, fe cluster- and than -xtoverid-? I deleted -xttest-

[Andrew Musau]

As you have \(T>N\), look at estimation with xtgls.

help xtgls

Also, as your \(T\) dimension is large enough, you may want to allow cross-sectional correlation. Therefore, consider the command xtscc from SSC (ivreghdfe from SSC also allows Driscoll and Kraay standard errors, and is an alternative to xtscc, especially when you have
multiple levels of fixed effects).

ssc install xtscc
help xtscc

[Carla Ferro]

Ok thank you, I try this new calculations. Also I have another question. I know that I should delete it, but isn't weird that with the xttest0 gets a p value> 0,05 so there aren't panel wise effects? I should preferer the OLS, so is correct then compare fixed vs random knowing that there aren't panel wise effect?

[Andrew Musau]

You still have the choice between pooled OLS and fixed effects. This is the Wald test that the country dummies are jointly equal to 0 (in xtreg, you find the statistic at the foot of the table). It may still happen that you conclude that pooled OLS is the way to go.

webuse grunfeld
regress invest mvalue kstock i.company
testparm i.company
xtreg invest mvalue kstock, fe

Res.:

. testparm i.company

 ( 1)  2.company = 0
 ( 2)  3.company = 0
 ( 3)  4.company = 0
 ( 4)  5.company = 0
 ( 5)  6.company = 0
 ( 6)  7.company = 0
 ( 7)  8.company = 0
 ( 8)  9.company = 0
 ( 9)  10.company = 0

       F(  9,   188) =   49.18
            Prob > F =    0.0000

. 
. xtreg invest mvalue kstock, fe

Fixed-effects (within) regression               Number of obs     =        200
Group variable: company                         Number of groups  =         10

R-sq:                                           Obs per group:
     within  = 0.7668                                         min =         20
     between = 0.8194                                         avg =       20.0
     overall = 0.8060                                         max =         20

                                                F(2,188)          =     309.01
corr(u_i, Xb)  = -0.1517                        Prob > F          =     0.0000

------------------------------------------------------------------------------
      invest |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      mvalue |   .1101238   .0118567     9.29   0.000     .0867345    .1335131
      kstock |   .3100653   .0173545    17.87   0.000     .2758308    .3442999
       _cons |  -58.74393   12.45369    -4.72   0.000    -83.31086     -34.177
-------------+----------------------------------------------------------------
     sigma_u |  85.732501
     sigma_e |  52.767964
         rho |  .72525012   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(9, 188) = 49.18                     Prob > F = 0.0000

.

[Carla Ferro]

Thank you so so much!! In your example we can say that the fixed model is the one correct, right?

[Andrew Musau]

Correct.

[Carla Ferro]

Thank you!!