Panel est: Hausman chi**2 positive, but V_b-V_B not positive definite

[Guido Heineck <guido.heineck@oif.ac.at>]

Dear all,

I use the Hausman-Taylor IV estimator to account for possible 
correlation between time-invariant regressors and the unobservables in 
my panel data.

Running RE too and applying the Hausman test, I get a large positive 
Chi**2 which suggests for rejection of the Null, but I also get a note 
that V_b-V_B is not positive definite.

There are threads on negative chi squares due to finite sample sizes. Is 
this a similar problem here? And how to deal with this? Do I just ignore 
it or is there a problem the way I use the test? To my knowledge, HT-IV 
is consistent both under Ho and Ha, so I do the following:

xtreg mymodel, re
est store re
xthtaylor mymodel
est store htiv
hausman htiv re

Any ideas? Thanks in advance.

Guido

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/