Hi all,
I'm using the xtpcse command which accounts for serial correlation in
the errors and contemporaneous error correlation over cross sections.
This is similar to xtgls, but accounts for the fact that my time
dimension (35), though longer than the cross-sectional one (ca. 20,
depending on the specification), is not hugely larger. According to
Beck and Katz (1995), this should yield more conservative standard
errors.
Has anyone ever applied this estimation technique in a two-stage IV
setting? While there is of course no problem in taking the predicted
value from a first-stage regression and plugging it into the
second-stage regression, I have so far been unsuccessful in
determining how correct second-stage standard errors obtain. This is
particularly true because of the aforementioned AR(1) component and
contemporaneous correlation of the errors, which I would very much
like to preserve.
Thanks a lot,
Mark
*
* 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/
