st: re: heteroskedasticity and fixed effects

[Kit Baum <baum@bc.edu>]

< >
Mark said

-whitetst- (or, pardon the self-plug, -ivhettest-) after -regress- with
fixed effects inserts as dummies by hand probably won't be appropriate,
for the reason I gave in my previous email.

>From what I know about -xttest3-, it won't be affected by the
Stock-Watson results, but this is just a quick guess.  Kit Baum is the
author of -xttest3- and he or someone else would be better placed than
me to say (as I gather he just did!).


Actually, per my prior posting, -whitetst- is one of those tests that  
only makes use of residuals from a regression, and is thus immune to  
problems with the variance-covariance matrix. All that is needed are  
consistent point estimates. The White general test is a special case  
of the Breusch and Pagan test (q.v. -bpagan-, or more generally - 
ivhettest-).

Note, though, that the alternative hypotheses of the White / Breusch- 
Pagan and -xttest3- tests differ considerably. The latter evaluates  
whether the squared residuals are correlated with a set of dummies  
for the individual panels. The White test considers whether there is  
correlation between squared residuals and all regressors, squared  
regressors and cross-products of regressors. The B-P test allows you  
to choose the variables in that list (which may thus include  
variables not in the original regression).

I wholeheartedly agree that the cluster-robust VCE is the most  
sensible thing to apply in the -xtreg, fe- context. Of course, the vce 
(bootstrap) could be used to evaluate whether the cluster-robust SEs  
look reasonable.

Kit

Kit Baum, Boston College Economics and DIW Berlin
http://ideas.repec.org/e/pba1.html
An Introduction to Modern Econometrics Using Stata:
http://www.stata-press.com/books/imeus.html


*
*   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/