N greater than T in panel data, is xtreg cluster (id) appropriate?

[Jovana Ju]

Hi, everyone, and thank you for your opinion and help in advance.

I have data for 79 companies and 7 periods (years), a strongly balanced panel. The appropriate tests indicated the use of a specification with fixed individual effects. There are issues with autocorrelation and heteroscedasticity. My first idea was to use the xtreg, fe cluster(id) or fe robust option to adequately address these problems. However, I read in one paper that this option(s) is fine to deal with heteroskedasticity but not with the arr(1)? I haven’t used the XTGLS or PCSE because most of the texts I’ve read suggest that it is appropriate to apply them when the T dimension is significantly larger than N, although I found that the PCSE option was used even in cases where N was greater than T.

Can you please help me, what is your opinion on this matter? Which approach should I follow for the estimation?

Thank you for your understanding!
Sincerely,
Ju

[Carlo Lazzaro]

Jovana:
in N>T panel datasets, autocorrelation and/or heteroscedasticity-related issues are dealt with cluster robust standard errors (-robust- or -vce(cluster panelid)- options do the cary same job under -xtreg-).
-xtpcse- was created for T>N dataset with residual correlated across panels.

[Maxence Morlet]

In addition to Carlo's great response, You may want to use boottest from ssc if your clusters do not have a similar number of observations (Mackinnon et al., 2023).

[Jovana Ju]

Thank you both very much, Carlo and Maxence!
One more question related to this: in which case would it be justified to include a lagged dependent variable among the regressors in the presence of autocorrelation, in order to examine whether it is necessary to introduce a dynamic component?

[Maxence Morlet]

Very good question, which has led to a lot of debate. As far as I know, clustering standard errors by the id variable largely addresses this. Because of the Nickel bias, acute if you only have T=7, you may want to try GMM. Is your panel balanced?

[Jovana Ju]

Thank you for the exchange of thoughts. Yes it is, it is balanced.

[Carlo Lazzaro]

Jovana:
if you plan to switch from static to dynamic, in line with Maxence's helpful guidance, take a look at the several options for standard errors.