About i.year and i.industry in the XTGLS regression

[Marwa missaoui]

Hi everyone! I need your help with Stata 13. My sample includes around 628 firm-year observations over 4 years in one country.
Depending on the Hausman test, the fixed effect is the appropriate one for all regressions. To correct heteroskedasticity and autocorrelation problems, I use the XTGLS command.
I am inquiring about two things.
1. Is it possible to include only i.year and exclude i.industry in an XTGLS regression since using only i.year yields a significant result?
2. What to do when industry effects take most of the variability in a regression model?

I would appreciate it if you could help me!

[Carlo Lazzaro]

Marwa:
welcome to this forum.
1) with a N>T (that is, short) panel dataset, you should switch to -xtreg,fe-; -xtgls- was conceived for T>N (that is, long) panel datatsets (and is not an -fe- estimator);
2) just use -robust- or -vce(cluster panelid)- (the do teh very same job under -xtreg-) to take both heteroskedasticity and autocorrelation into account;
3) if firms (as it is usually the case) do not change -industry- within panels, being -industry- a time-invariant predictor the -fe- estimator will wipe it out;
4) to reply positively to your query #2, as per FAQ, please post what you typed nad what Stata gave you back. Thanks.