st: Re: xtreg, fe cluster(id) vs xtabond2 - just to control for serial correlation

["Rodrigo A. Alfaro" <raalfaroa@gmail.com>]

I just posted references for the same problem:
http://www.stata.com/statalist/archive/2007-06/msg00573.html

Shortcut: AB/GMM (xtabond) and Within-Groups or WG (xtreg, fe) are biased
for "large n and large T". In your case WG will be more biased that AB/GMM
if the panel is balanced, but your panel is unbalanced!!

Bruno (2004) shows that bias of WG will depend on the ratio between the
harmonic average of T (H) and the arithmetic avg of T (A): Ahrens- Pincus 
index
(see [5] at http://ideas.repec.org/e/pbr136.html). Maybe a good idea is to 
simulate
your case (n=4000) and (T=10) using -xtarsim-.

I don't know about AB/GMM in unbalanced panels. My guess is that also will
depend on H/A, but it seems hard to show that.

Without more information, I suggest to run both. I preference AB/GMM in
your case, just because T is not too large and because unbalancedness it 
could
be (in average) very-very low.

Rodrigo.
PS: The standard error corrections -cluster(id)- will not affect the 
estimator.


----- Original Message ----- 
From: "Alexander P�tz" <puetza@uni-koeln.de>
To: <statalist@hsphsun2.harvard.edu>
Sent: Friday, June 15, 2007 12:29 PM
Subject: st: xtreg, fe cluster(id) vs xtabond2 - just to control for serial 
correlation



> Dear Statalist members,
>
> I've got a question concerning an unbalanced panel dataset with N about 
> 4000
> and T = 10.
>
> I would like to regress a model with panel fixed effects and control for
> serial correlation in the dependent variable. Now I'm not sure about
> including the lagged dependent variable as an explanatory variable in the
> model and estimating the equation by xtabond2 or just leaving the lagged
> dependent variable out of the model and estimate the equation by xtreg, fe
> cluster(id).
>
> Because I'm not sure about how to use xtabond2 correctly and I'm only
> interested in the signs and inference of the other explanatory variables, 
> I
> would tend to leave the lagged dependent out of the model. Anyway I don't
> know if this would produce biased estimates.
>
> Thank you for your help.
>
> Alexander
>
>
>
>
>
>
> *
> *   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/