Hi Monica:
First of all, it is my understanding, in contrast to the posting by Jacob in response to Kate (see below), that the underlying model that the xtabond estimator first-differences takes the individual effects are taken to be random effects, not fixed effects. This can be verified in the Stata reference manual for Su-Z, p.291 bottom. Given that, one cannot get an estimate of the individual effect alpha(i), since it is not taken to be a fixed unknown parameter but a random variable.
But a related question has been preoccupying me too: When the structural model that the Arellano-Bond estimator (which is what the xtabond is) estimates is differenced, not only is the individual effect swept out, but also any constant (ie. intercept) that the structural model may have had. However, xtabond produces estimates not only for the slope coefficients, but also for a constant. What should I make of that? Is it "sneaking" in an intercept in the differenced model before estimating it, i.e. is it estimating
y(it) = alpha + D x(it) * beta + D e(it) (where D is the Delta sign)
instead of estimating (what I think it should be doing):
y(it) = D x(it) * beta + D e(it) ?
or is it estimating the latter, but computing the constant alpha from the estimates of beta? Anyone know where xtabond is getting the intercept estimate from?
Thanks,
Tewodaj
Message: 16
Date: Tue, 20 Apr 2004 09:54:50 -0400
From: "Monica L. Parra Torrado" <[email protected]>
Subject: st: xtabond fixed effect prediction
Dear Stata List,
I have some xtabond estimations and need to predict the fixed effects
coefficient but the command only allows predicting delta Y (the difference
of dependent variable Y) and delta e (the difference of the error term). I
am not sure it is possible to predict such coefficient since the estimation
is done in differences and the fixed effect term disappears. Does someone
have an idea of what could I do?
Thanks,
M�nica L.
From: Kate Ivanova <kivanova@u...>
Date: Tue Mar 30, 2004 9:11 am
Subject: st: xtabond command
I am trying to use the xtabond command but am puzzled whether the estimator that Stata produces is designed for fixed or random effects? Also, I do not understand whether the xtabond command automatically first differences all the variables in the model or whether I have to do that separately before I use the command? I would appreciate any help with that
From: "Jacob, Jeffry Ankur" <jajacob@m...>
Date: Tue Mar 30, 2004 10:34 am
Subject: st: RE: xtabond command
Hi, Xtabond is for fixed effects estimation. Also, the command automatically first differences the data. You can however specify the lags you want to include as explanatory variables in your model.
~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~.~
Tewodaj Mogues
Dept. of Agricultural and Applied Economics
University of Wisconsin - Madison
427 Lorch St. #317, Taylor Hall
Madison, WI 53706
*
* 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/
