st: Petersen (2009) vs Thompson (2011) - Estimating standard errors in panel data sets

["Julia Ke" <julia.ke@gmx.net>]

Dear Statalist, 

I am running a panel regression. It is a rather small sample with multiple firms and a few years. 

I was reading into which method to use, which came down to the following: If one dimension has far more units than the other, clustering on one dimension and using dummies on the other seems to be used (especially in smaller panels). 

My question concerns which unit to cluster and which one to use dummies on. The below two authors seem to state the opposite which is confusing me a bit... 

Petersen (2009): "Since most panel data sets have more firms than years, the most common approach is to include dummy variables each year (to absorb the time effect) and then cluster by firm."
"When there are only a few clusters in one dimension, clustering by the more frequent cluster yields results that are almost identical to clustering by both firm and time."

Thompson (2011): If there are far more firms than time periods, clustering by time eliminates most of the bias unless within-firm correlations are much larger than within-time period correlations.

Many thanks in advance, 
Julia


Papers mentioned:
- Petersen, M. A. (2009)  "Estimating standard errors in finance panel data sets: Comparing approaches" Review of Financial Studies, vol.  22, pp. 435-480.
- Thompson, S. B. (2011) "Simple formulas for standard errors that cluster by both firm and time" Journal of Financial Economics, vol. 99, pp. 1-10.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/