Arellano-Bond is a linear GMM estimator... not IV. For example, for AR(1)
model y(i,t) = a(i) + b*y(i,t-1) +e(i,t), AB is a weighted average of IV
estimators for t=1,...T-1. All of them using "t" instruments (the lags). You
don't have lags of your dependent in your specification you don't need AB.
Rodrigo.
----- Original Message -----
From: "Christian Arndt" <christian.arndt@iaw.edu>
To: <statalist@hsphsun2.harvard.edu>
Sent: Friday, November 10, 2006 11:58 AM
Subject: st: WG: Strict exogeneity and xtivreg2
Dear statalisters,
I am using lags of an endogenous variable in a macro panel
setting, because I think of it as predetermined.
So I use xtivreg2 and have something like:
xtivreg2 employment = (capital wages = l1.capital l1.wages),
fe robust (with some more instrumented variables), altogether
exactly identified.
When I think of my right hand side variables as
predetermined, past values of them are not influenced by
employment_t. But, current and past values are not strictly
exogenous, and hence endogenous in case of the fe-Estimator
(with time-demeaning).
1. Is xtivreg2 with the fe option robust to this problem, as e.g. the
predetermined option in xtabond (wich uses first differences of the
dependent var.)? Oder do I also need strict
exogeneity of the instruments? (I only found a very general answer to a less
specific question in the faq-list up to now)
2. Why is it, if point 1. should be no problem, that in the xtivreg2.help in
the case of the
fe-option, they use only l2.n and l3., but not l1.n and l2.n as
in the case of the fd option?
Kind regards and thanx for help!
Christian
*
* 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/
*
* 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/