xtscc POLS or xtscc fixed effects?

[Hakki Polat]

Hi everyone,
I've a data set which N=27, T=12 with 1 depend variable and 9 independent variable. When I check diagnostics there are heteroskedastic, autocorrelated, and cross-sectionally dependent a data set in my hand. According to Hausman test Fixed effects model is appropriate. Hoechle (2007) recommand that using Driscoll and Kraay estimator which is command by xtscc in STATA. I run xtscc and xtscc, (fixed effect) fe for my data set and see xtscc estimation results is better than xtscc, fe. My question is: are there any problem to use xtscc instead of xtscc fe? As I understand there are no evidence to selecting between this two approach, I couldnt find any literature about it. I'll be happy if someone can help me about this selecting criteria.
Kind regards

[Andrew Musau]

As I understand there are no evidence to selecting between this two approach, I couldnt find any literature about it.

You are still dealing with panel data. Nowhere does it say that if \(T > N\), it is no longer a panel. The test for the choice between POLS and FE is the test of the joint significance of the firm dummies. In any case, you have \(N>T\) from your description, so look at xtreg. Additionally, 9 right-hand side variables with that sample size is a bit of a stretch.

[Hakki Polat]

Dear Andrew,
Thanks for your respond. Actually I'm working on tax rates with some macroeconomic indicator and financial theories avoid using dummy variables. I'm agree with you about my data sets is not strong for panel models but I must use 27 euro zone country with 2008 to 2020 date. When I check F test fixed effects is appropriate for my data set. But when I check coefficients and r squares the POLS is giving better results (as I understand its confirm your opinion) and I just wonder is there will any problem if I prefer POLS instead of Fixed effects.

[Andrew Musau]

You get this test automatically using xtreg. See #6 https://www.statalist.org/forums/for...nel-data-tests.

I just wonder is there will any problem if I prefer POLS instead of Fixed effects.

Pooled OLS is in general not a recommendable technique except if you have time-invariant RHS variables that capture the country heterogeneity. The test that I highlight will tell you whether this is the case.

[Hakki Polat]

Thank you for your respond. I'm searching on it.