st: RE: System GMM and fixed effects - xtabond2

[Rodolphe Desbordes <rodolphe.desbordes@strath.ac.uk>]

Dear Zohal,

I think that you will find the answers to your question in David Roodman's paper (http://www.stata-journal.com/article.html?article=st0159). The WP is available on http://www.cgdev.org/content/publications/detail/11619

In the absence of external instruments, you can include time-invariant variables in your model as long as you are willing to assume that they are uncorrelated with the error term. However, it would be a mistake to include fixed effects in your level equation, since this implicit within transformation would invalidate the use of lags as internal instruments.

The instruments used for the level equation are the lagged first-differences of the time-varying variables. These instruments are valid as long as a) the idiosyncratic shocks  are not serially correlated b) the correlation between these variables and the individual effects is constant, i.e. a stationarity assumption is met. If the additional moment conditions are valid, the "SYS-GMM" estimator is consistent. It is also likely to be more efficient than the "DIFF-GMM", depending on the degree of persistence in your data.

I hope this helps,

Rodolphe






________________________________________
From: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] On Behalf Of Zohal Hessami [Zohal.Hessami@uni-konstanz.de]
Sent: 28 February 2010 20:19
To: statalist@hsphsun2.harvard.edu
Subject: st: System GMM and fixed effects - xtabond2

Dear Statalisters,

I have a question regarding System GMM and fixed effects in the context of
xtabond2.

Apparently, time constant variables can be included in System GMM in the
level equation. Does that mean that the level equation has no fixed effects
included? And if so, does this imply that the level equation is not robust
to unobserved heterogeneity?

Since my control variables are probably correlated with the fixed effects,
should I use Difference GMM instead of System GMM even though it is less
efficient? (The results between the two estimation methods differ
significantly for my model.)

Thank you very much for your help!

Zohal Hessami
Department of Economics, University of Konstanz

*
*   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/

*
*   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/