Panel regression

[Mary SMV]

Hello,

I am working with panel data where T=23 and N= 22 (country).
I performed the following steps:

xtreg depvar indepvar (1) indepvar (2) indepvar (dummy), fe
estimate store fe
xtreg depvar indepvar (1) indepvar (2) indepvar (dummy), re
estimate store fe

To decide between the both models I run the Hausman test, whose result indicates that fixed effects is the appropriate model. (chi2 = 86.60 Prob>chi2 = 0.0000)

After this, I applied tests to check problems of cross sectional independence (a), heteroskedasticity (b) and autocorrelation(c):

(a): xtcsd, pesaran abs | Result: Pr = 0.7796 (There is no serial correlation)
(b): xttest3 | Result: Prob>chi2 = 0.0000 (There is Heteroskedasticity)
(c): xtserial | Result: Prob > F = 0.0000 (There is Autocorrelation) | I also estimated this command for each of the independent variables. xtserial depvar indepvar 1
xtserial depvar indepvar 2 ... (always autocorrelation)

I have problems of autocorrelation and heteroskedasticity in my model.

So to solve these problems, according to what I read in https://www.stata.com/manuals/xtxtgls.pdf, once I have T>N, I used xtgls:
xtgls dep indepVar's, panels (heteroskedastic) corr (ar1)

My questions is: What I did is correct? Is the best way to fix these problems?

Thank you.

Best regards

[Carlo Lazzaro]

Mary:
your code looks fine to me.

[Mary SMV]

Carlo,
I really appreciate your feedback. Thank you very much.