At the suggestion of Decio Coviello, I have changed xtabond2 so that in
system GMM, it automatically does a difference-in-Sargan test for the
joint validity of the "GMM-style" instruments for the levels equations,
the ones that must be valid for system GMM to be consistent. Also it now
automatically tests each group of instruments defined by a gmm() or iv()
option. So replacing "gmm(x y)" with "gmm(x) gmm(y)" in a command line
won't change the estimates from a regression, but will trigger two,
narrower difference-in-Sargan tests instead of one broader one.
These tests are performed only by the Mata version of xtabond2, which
requires Stata 9. They can be suppressed for the sake of speed with a
new -nodiffsargan- option.
Note that these tests are weak if there are a lot of instruments
relative to the number of observations.
Also, one can do difference-in-Sargan tests manually, running the
estimate with and without the suspect instruments. Under the null of
validity, the change in the regular Sargan/Hansen statistic will be
asymptotically chi-squared with degrees of freedom equal to the number
of suspect instruments. However, doing it manually will give slightly
different test statistics, as is also the case with ivreg2's orthog()
option. Baum, Schaffer, and Stillman explain why at the bottom of p. 18
of their working paper (http://fmwww.bc.edu/ec-p/wp545.pdf).
To install the latest xtabond2, type "ssc install xtabond2, replace" in
Stata. Or install the adoupdate command and use that.
I've also revised the "How to Do xtabond2" paper
(http://www.cgdev.org/content/publications/detail/11619) in accordance,
and fixed some other errors in it pointed out by readers.
David Roodman
Center for Global Development
*
* 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/