I am happy to announce the release of a paper on xtabond2, which is
intended as a pedagogic introduction to linear GMM, the difference and
system GMM estimators, and xtabond2 (as well as abar). I do hope readers
will scour it for errors and confusing bits and send me suggestions. The
working paper is at
http://www.cgdev.org/content/publications/detail/11619. It is the
substance behind the "How to Do xtabond2" PowerPoint presentation on
RePec, from last summer's Stata users group meeting in Boston
(http://ideas.repec.org/p/boc/asug06/8.html).
Also, the latest version of xtabond2 does forward orthogonal deviations.
Thanks to Kit Baum and Oleksandr Talavera for persuading me to do this.
"Forward orthogonal deviations," proposed by Arellano and Bover (1995),
is an alternative transformation to the usual differencing, which also
purges fixed effects. Instead of subtracting the previous observation,
it subtracts the average of future available observations. The main
advantage is that it preserves sample size in panels with gaps. Under
differencing, if some x_it is missing, then both D.x_it and D.x_i,t+1
are missing. But under orthogonal deviations, the transformed x_i,t+1
need not go missing.
At this point, the orthogonal deviations feature is available only in
the Mata version of xtabond2, which automatically runs in Stata 9, but
not in Stata 8 or earlier. Just put "orthog" in the command line after
the comma.
Get the latest version by typing "ssc install xtabond2, replace". Or,
install and use the new adoupdate command (start with "findit
adoupdate") which can automatically download updates for all the
user-written commands you have installed.
--David
*
* 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/
