Panel data estimation methods

[Rick Last]

Hi,

I have a question with regards to panel data analysis. I am not really sure which estimator is the best for me.

I have a panel data set of approximately 5000 firms for 15 years (annual data), so my cross sectional dimension is relative large compared to the time-series dimension. One of the explanatory variables is time-varying, but not cross sectionally varying! It is an interest rate. This interest rate is the same for each firm in the same year.

I am doubting between a fixed or random effects model, and if i need time or cross sectional effects.

I red online that fixed effects are in general not possible with a time-invariant variable, but I have a cross-sectional invariant variable, is fixed effects than also not possible? So the question is basicly: if there is a time-invariant variable explanatory variable, is there an prefered estimation method?

I would really appreciate anybody's help. I am having this problems for a while now and it only confuses me more every time I try to find a solution.

Kind regards,

[Carlo Lazzaro]

Rick:
welcome to the list.
Assuming your dependent variables is continuous, you should compare -xtreg, fe- vs -xtreg, re- via -hausman- .
See related help files and entries in Stata .pdf manual for further details.

[Rick Last]

Thank you very much for your reply. I will do.

I have one additional question though: do you have any clue about why the pooled OLS and RE gives me the same results? Can it be that it has something to do with my cross-sectional invariant variable?

Regards

[Carlo Lazzaro]

Rick:
looking at what you typed and what Stata gave you back with POLS and -xtreg, re- would allow helpful replies.

[Marcos Almeida]

Please present command and output. This is the best way to entice helpful replies.

That said, with regards to getting similar results in OLS and RE, you may wish to check the rho. Hazarding a guess, it will be considered statistically "not different from zero"

Crossed with Carlo's advice.