st: RE: Is Hansen test in XTABOND2 really robust?

["Ruiqing Miao" <ruiqingmiao@gmail.com>]

Dear Stata Users,

I am working on a project that involves dynamic panel data analysis. Since
the manual of XTABOND states that the Sargan test is not valid in the
presence of heteroskedasticity, I switch to XTABOND2 that presents Hansen
test that is claimed to be robust. But I realized that for the same model,
the statistic value of XTABOND?s two-step Sargan test is exactly equal to
the value of XTABOND2?s Hansen test; and the value of XTABOND?s one-step
Sargan test is very close to the value of XTABOND2?s Sargan test. 

In the manual of another software (page 168, the last paragraph,
http://gretl.sourceforge.net/gretl-help/gretl-guide.pdf), it reads,
?Specifically, xtabond2 computes both a ?Sargan test? and a ?Hansen test?
for overidentification, but what it calls the Hansen test is, apparently,
what DPD calls the Sargan test. (We have had difficulty determining from the
xtabond2 documentation (Roodman, 2006) exactly how its Sargan test is
computed.)? This may provide a support to my finding above.

So, I guess either the Stata manual about XTABOND or Mr Roodman?s XTABOND2
has something unclear on this issue? If the Stata manual is correct (i.e.,
Sangan test is not robust), then is Hansen test, which is equal to Hansen
test in two-step XTABOND, really robust?

On the manual for ?xtdpdsys postestimation ? Postestimation tools for
xtdpdsys?, page 107 in Stata 11, it reads,  ?Although performing the Sargan
test after the two-step estimator is an alternative, Arellano and Bond
(1991) found a tendency for this test to underreject in the presence of
heteroskedasticity.? What does this sentence mean? Why is it ?an
alternative?? Is the Sargan test after the two-step estimator robust?

Thank you very much for your help!

Ruiqing


*
*   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/