xtscc or xtreg (fe)

[Jihane bou zakhem]

Hello everyone,

I am testing my panel data with fixed effects (N=360, T=4), it included autocorrelation, heteroskedasticity, cross sectional dependence. How I can make sure which command I can use xtscc or xtreg (fe vce(cluster))? And how I can test whether either of these commands has fixed the problems in my data.

Thank you,

Jihane

[Carlo Lazzaro]

Jihane:
go -xtscc- (which is a community-contributed module, as FAQ kindlky ask you to define it).
There's no need to check whether -xtscc- fixed the issue, because there's no problem to solve, but to choose the right estimator for your dataset.
Indeed, -xtreg,fe vce(cluster panelid)- does not take cross-sectional dependence of epsilon into account.

[Joao Santos Silva]

Dear Carlo Lazzaro,

You may have missed that the panel Jihane is working with only has T=4, and therefore the standard errors implemented in -xtscc- are not suitable because they rely on large T asymptotics.

Best wishes,

Joao

[Carlo Lazzaro]

Thanks Joao for pointing this out.
You're absolutely right (I've misread T=40 instead of T=4).
Unfortunately, I've no other advice for the OP.

[Jihane bou zakhem]

Thank you for your replies! So the best thing to change variables?

[Joao Santos Silva]

Dear Jihane bou zakhem,

I am not an expert on this, but I guess your best option is to use clustered standard errors. You may want to define relatively wide clusters to minimize the problem of cross-sectional dependence, or maybe do multi-way clustering. Anyway, others may have better advice for you.

Best wishes,

Joao

[Jihane bou zakhem]

Dear Joao Santos Silva ,

Kindly I have changed my model, it gave me random effects, autocorrelation, heteroskedasticity. with N=360 and T=4 , the best command is xtreg(re), vce and not xtgls or xtpcse right?

[Carlo Lazzaro]

Jihane:
since you're dealing with a N>T panel dataset, -xtgls- and -xregarr, re- are out of debate.
I would go -xtreg,re vce(cluster panelid)-.

[Jihane bou zakhem]

Thank you

[Carlo Lazzaro]

Jihane:
in my previous reply I should have written:
-xtregar,re-.
Sorry for the typos.

[Jihane bou zakhem]

Dear Carlo Lazzaro ,

I know that xtregar fix the autocorrelation first order, but does it fix heteroskedasticity? In addition could we use xtregar if we have independent variables pca components?

Thank you,

Jihane

[Jihane bou zakhem]

Dear Carlo Lazzaro ,

I know that xtregar fix the autocorrelation first order, but does it fix heteroskedasticity? In addition could we use xtregar if we have independent variables pca components?

Thank you,

Jihane

[Carlo Lazzaro]

Jihane:
sympathetic with your willingness of knowing, but please consider that we're all busy people. So be patient.
1) actually heteroskedasticity-robust standard errors do not seem to be mentioned in -xtregar- entry, Stata .pdf manual;
2) Iìve necer challenges myself with -pca- components as regressors, but it may work.

[Jihane bou zakhem]

Carlo Lazzaro , I apologize for any inconvenience. I have an error in my browser (using the chrome) , it posted twice. Thank you for the reply

[Carlo Lazzaro]

Jihane:
no worries.
Enjoy staying with us.