Panel regression

[Mary SMV]

Hello,

I have some doubts about panel regression in fixed effects:

1) When I run fixed effects should I always test for heteroskedasticity, , autocorrelation and serial correlation independence? Or can I draw my conclusions through the interpretation of fixed effects?

2) In the case the three tests are positive (applied to fixed effects model), how should I solve when N is greater than T and T is greater than N?

Thanks,
Mary

[Carlo Lazzaro]

Mary:
1) yes, you have to. Otherwise, your conclusion on the ground of fixed effect estimates might be biased.
2) please note that serial correlation and autocorrelation (of the idiosyncratic error) meand the same thing. That said, if N>T: see -cluster- or -robust- options for standard errors under -xtreg-; see -xtregar- if T>(or equal to) N.

[Mary SMV]

Thanks for your answer.

I have another question, When using xtreg (robust) for N>T and xtregar for T>=N, does the cross sectional independence problem (xtcsd, pesaran abs) also get resolved? If not, how can i solve this?

Thank you very much for your help.

Best regards,
Mary

[Carlo Lazzaro]

Mary:
see -search driscoll-.

[Mary SMV]

For N>T, if I use xtscc depvar indpvars, fe instead xtreg, fe cluster(firm) are my errors (cross sectional dependence, heteroskedasticity and autocorrelation) solved?

Thanks

[Carlo Lazzaro]

Yes.

[Mary SMV]

Thank you very much for your help, Mr Carlo.

[Mary SMV]

What is the difference between xtpcse and xtscc?

My data is the same (as I mentioned above):

- N>T
- Problems: cross sectional dependence, heteroskedasticity and autocorrelation (the tests were performed n fixed effects model)

1) I was using xtcss depvar indpvars, fe, but I read that this "solution" is indicate for T>N and I don´t know if I need to do more autocorrelation tests to check Ma (q) or if xtserial is enough.

2) Can I ignore cross sectional dependece and just correct heteroskedasticity and autocorrelation?

3) I made some research in order to understand better this topic and I also found the command xtpcse depvar indpvars, corr (ar1). Can I use xtpcse for N>T? In this case the model is no longer fixed effect?

Thanks.

Regards,
Mary

[Carlo Lazzaro]

Mary:
see https://www.statalist.org/forums/for...d-effect-panel.
As an aside, please do not post the same thread twice under different titles! Please, give intersted listers the time to reply to your threas, if the want to. We are all busy people like you and do our best to match academic/business/personal duties with contributing to the list positively, when we have knowledge and time to. Thanks.

[Mary SMV]

Thank you and sorry, I'm not going to do it again.