Hello,

I am running regressions on an unbalanced dataset with large N small T. I have already performed xtreg, fe, and the F tests indicated that Pooled OLS is the better option. I am now struggling a lot on to how I should specify this pooled OLS. I know that by clustering the standard errors at CompanyID will take into account the panel structure. When is, and when is this not, a good thing to do?

With respect to the specification, I also have a lot of doubts:
1. should I include year dummies, as I did in the fixed effects model?
2. should I include i.Companyid, and if so why? I do not see why this would have to be done.
3. How do I take into account problems with heteroscedasticity and autocorrelation? Is clustering at Companyid sufficient? Or should I add robust?
4. To test for heteroscedasticity and specification errors, can I use the same methods as in normal OLS?

Thank you very much!

Kind regards,
Timea De Wispelaere