Panel data random effects with small n and large t autocorrelation problem

[Joao Caseiro]

I have a panel data random effects model with n=14 and t=38.

I have tested for autocorrelation with -xtserial-, resuilting in the presence of autocorrelation. Nonetheless I have read that this test is not trustworthy in my case of small n and large t, is this true?
Also, if it is the case that autocorrelation is present, is the option -vce(cluster)- enough to correct this?

Thank you!

[Arthur Morris]

A couple things:
(1) Being specific about what you've read will help you get a better answer.
(2) Though you are referring to Stata commands, your question is a statistical one. You may find a more active discussion of these topics on https://stats.stackexchange.com.

That said, I don't think vce(cluster) (or any other clustering approach) will solve your problem because these approaches rely on large N and fixed T.

See pages 297 and 298 of Wooldridge (2010) for a discussion of the random effects estimator and the robust variance matrix estimator. In particular, this combination does seal with serial correlation, but requires large N and fixed T to do so.

See also pages 706 and 707 of Cameron and Trivedi (2005) for a nice explanation of why the "robust" or "clustered" estimators require large N and fixed T.

Cameron and Trivedi (2009) also have a very nice discussion of how to approach these models in Stata (see 8.7.1 "RE estimator" starting on page 262).

Wooldridge, Jeffrey M. "Econometric analysis of cross section and panel data 2nd ed." MA: MIT Press Cambridge (2010).
Cameron, A. Colin, and Pravin K. Trivedi. Microeconometrics: methods and applications. Cambridge university press, 2005.
Cameron, Adrian Colin, and Pravin K. Trivedi. Microeconometrics using stata. Vol. 5. College Station, TX: Stata press, 2009.

[Carlo Lazzaro]

Joao:
as an aside to Arthur's helpful reply, please note that Stata commands for large panels are -xtgls- and -xtregar-.

[Joao Caseiro]

Thank you for your replies, Arthur and Carlo!

Best,
João