Dear all,
I'm confused about what overidentifying restrictions xtabond2's
Sargan-Hanen test really tests? Does it test for the validity of
instruments used by system GMM in addition to those used by
differenced GMM? i.e. It tests the validity of differenced variables
as instrument for levels equations? or Does it test for the validity
of both variables lagged 3 or more periods as instruments for
differenced equations and the differenced variables as instruments for
levels equations?
Another confusion what gmm( ,lag(a b)) in xtabond2 really specifies?
For example my model is
Y_{it}=aY_{it-1}+bX_{it-1}+u_{i}+v_{t}+w_{it}
where X_{it-1} is predetermined with respect to w_{it} and T=6
I run xtabond2 Y l.Y l.X yr*, gmm(l.(Y X), lag(1 4)) iv(yr*), robust small
Does this mean that I use Y_{it-2},...,Y_{i1} and X_{it-2},...,X_{i1}
as instruments for differenced equations and use deltaY_{it-1} and
deltaX_{it-1} as instruments for levels equations?
However, with this lag(1 4) specification, Hansen test is rejected (
Prob > chi2 = 0.000). I also tried many other specifications, for
example lag(1 2), lag (2 4) and all possible combination that I can
think of, but the Hansen tests are all rejected. The only exception is
that I don't not sepcify lag(a b), then the Hansen test is not
rejected. Then I'll get the warning that " Two-step estimated
covariance matrix of moment conditions is singular, Number of
instruments may be large relative to number of groups ." What does
this warning really means? The number of my instruments is about 1/2
of the number of groups. Is this a problem?
The weird thing is when I use another Y variable (Y2) that is almost
100% correlated with the one I originally test (Y1), then Hansen test
is never rejected no matter how I specify lag(a b). The only
difference between Y1 and Y2 is that Y1 is more volatile than Y2. I
cann't understand why using Y1 or Y2 could yield such difference in
Hansen test.
Could anyone kindly help me understand what Hansen test in xtabond2
really tests? Why Hansen test is rejected if I specify lag(a b)? Did I
make any mistake when using xtabond2? Sorry, I've asked too many
questions. If you can answer any of them, I'll be appreciated.
Thank you very much in advance,
Caroline
*
* 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/