st: xtgls, xtpcse or xtreg when the dataset is very small and N > T

[Florian Stahl <stahl@isu.uzh.ch>]

Dear Statalist,

as I don't know what is exactly going on behind some commands I have 
concerns of using the wrong command.

1. My dataset contains 40 firms and 10 time points (yearly): Dataset is 
very small and N > T
2. The dependent variable is log-centric and the independent variables 
are defined as the difference from the mean: That means my model is 
nonlinear.

My idea is to apply a fixed effects model with dummy variables for each 
firm. As the dataset is very small and N > T, which alternative model do 
you suggest to use:

GLS?
a) xtgls depvar indepvar firmdummies, pannel(hetero) corr(ar1)
b) xtpcse depvar indepvar firmdummies, corr(ar1) hetonly

OLS?
b) regress depvar indepvar firmdummies, vce(robust) cluster(firmid)

Or should I use (due to the nonlinearity in the model)  glm or  mle 
(xtreg depvar indepvar, mle)?

Thanks a lot for your help!

Florian

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/