Is it reasonable to use ar1 option in xtgls although there is no autocorrelation?

[Nguyen Tr]

I've run heteroscedasticity and autocorrelation test on the fixed effects model and it turns out that my residuals have heteroscedasticity. Therefore, I used FGLS to fix the problem. The result when using panels(h) option contains 4 significant variables out of 8, whereas the result when panels(h) corr(ar1) is used has all 8 variables significant. Should I use that corr(ar1) option in xtgls although there is no autocorrelation? The signs of the models are the same. I'd really appreciate your help.

[Carlo Lazzaro]

Nguyen:
welcome to this forum.
Are you dealing with a N>T or a T>N panel dataset (N=cross-sectional dimension; T=time-series dimension)?