Re: st: GMM: first-differencing vs. orthogonal deviations

["Bernadette Durz" <bernadette.durz@gmx.de>]

Hi Marteen, 

Thank you very much for your answer and sorry for the inconvenience. The two commands I used for the GMM are as follows: 

First-differencing: 
xtabond2  y y(t-1) x1 x2 x3 l.x4 year????c , h(3) noleveleq ///
iv(x1 x2 year????c) ar(2) twostep robust ///
gmm(y(t-1), lag(1 1) c) ///
gmm(x3, lag(2 2) c) gmm(l.x4, lag(1 3) c)


Orthogonal deviations: 
xtabond2  y y(t-1) x1 x2 x3 l.x4 year????c , h(3) orthogonal noleveleq ///
iv(x1 x2 year????c) ar(2) twostep robust ///
gmm(y(t-1), lag(1 1) c) ///
gmm(x3, lag(2 2) c) gmm(l.x4, lag(1 3) c)

???? refer to years 1998 to 2007. Neither the estimated coefficients nor the test statistics are the same for both calculations. In both calculations I have a balanced panel of 799.320 observations and 99.915 groupd. 

The articles I refered to are Roodman (2009): How to do xtabond 2: An introduction to system and difference GMM in Stata and Arellano/Bover (1995): Another look at the instrumental variable estimation of error-components models.

Thank you very much!

Bernadette

-------- Original-Nachricht --------
> Datum: Wed, 27 Jun 2012 11:44:55 +0200
> Von: Maarten Buis <maartenlbuis@gmail.com>
> An: statalist@hsphsun2.harvard.edu
> Betreff: Re: st: GMM: first-differencing vs. orthogonal deviations

> On Wed, Jun 27, 2012 at 11:21 AM, Bernadette Durz
> <bernadette.durz@gmx.de> wrote:
> > I am using the difference GMM estimator with a balanced panel. I read in
> different papers (i.e. Arellano/Bond 1995, Roodmann 2009) that
> transformations with first-differencing and orthogonal deviations should lead to the
> same results when working with a balanced panel.
> >
> > Unfortunately, this is not the case. Does anybody know why the results
> differ? Which transformation should be preferred? I am using xtabond2 in
> Stata.
> 
> In order for this question to be answerable you need to add complete
> references and you need to state exactly (which really really means
> exactly) what you typed in Stata that led you to the conclusion that
> these two methods do not lead to the same results.
> 
> All this is clearly explained in the Statalist FAQ, which you were
> asked to read before posting to Statalist. This FAQ does not exist to
> trick you, it is there to help you ask answerable question. This is
> also in your interest. So please read the FAQ (again). A link can be
> found at the bottom of this post (and any other post on the
> Statalist).
> 
> -- Maarten
> 
> --------------------------
> Maarten L. Buis
> Institut fuer Soziologie
> Universitaet Tuebingen
> Wilhelmstrasse 36
> 72074 Tuebingen
> Germany
> 
> 
> http://www.maartenbuis.nl
> --------------------------
> 
> *
> *   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/

-- 
Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir
belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de
*
*   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/