re:st: Testing the validity of instruments when estimating a GMM model with

[Christopher Baum <kit.baum@bc.edu>]

<>
I estimate a GMM model with Windmeijer corrected robust standard errors 
as follows

xtdpd y l3.y_rain x1 x2 x3 x4 x5, dgmmiv(l3.y_rain, lag(2 3))  dgmmiv(x1 
x2 x3, lag(2) ) lgmmiv(x1 x2 x3,lag(2)) div(x4 x5 ) twostep hascons 
vce(robust)

When using the vce(vcetype) option, the sargan statistic cannot be 
obtained because the distribution of the Sargan test is unknown when the 
disturbances are heteroscedastic. However, I need to test the validity 
of over-identifying restrictions and was thinking whether one can use 
the same model without a Windmeijer correction (without vce(robust)) to 
test for the validity of over-identifying restrictions and then infer on 
the robust model?


The Sargan test assumes iid errors. The xtdpd command refuses to compute same with a robust VCE.
But there is always a Hansen J statistic available, and it is robust to non-iid errors. I suggest you use
David Roodman's xtabond2 (available from SSC) for these models; it will produce both a Sargan
and a Hansen J test, and warn you about their various shortcomings. It will also allow you to estimate all 
DPD models with a single command, rather than Stata's three official commands, and provides several
features missing from Stata's commands.

Kit


Kit Baum   |   Boston College Economics & DIW Berlin   |   http://ideas.repec.org/e/pba1.html
                             An Introduction to Stata Programming  |   http://www.stata-press.com/books/isp.html
  An Introduction to Modern Econometrics Using Stata  |   http://www.stata-press.com/books/imeus.html


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/