Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: AW: RE: ivreg2 or xtabond2 endogeneity of regressors

From   "Dithmer, Jan" <>
To   "" <>
Subject   st: AW: RE: ivreg2 or xtabond2 endogeneity of regressors
Date   Fri, 21 Oct 2011 16:16:00 +0200

Yes, sorry, I was indeed talking about dynamic panel data estimation...
In the meantime I looked it up in the textbook you recommended and also in the xtabond2 documentation.
Thus, I can just test whether some variables might be endogenous if I assume the others are not, right?
I am not sure if you are familiar with the xtabond2 syntax, but maybe someone of the statalisters is.
In order to test if certain regressors can be treated as exogenous with the Difference-in-Hansen test, I think I just have to put each regressor in its own ivstyle option,
which will return the p-value of the test statistic distributed as chi2.
Assuming that the other regressors are specified correctly, if the test returns a p-value bigger than, say 0.2, exogeneity cannot be rejected. Is this the way to proceed?

Thanks for your suggestions!


-----Ursprüngliche Nachricht-----
Von: [] Im Auftrag von Schaffer, Mark E
Gesendet: Wednesday, October 19, 2011 3:33 PM
Betreff: st: RE: ivreg2 or xtabond2 endogeneity of regressors


By "dynamic GMM" I think you probably mean dynamic panel data estimation
a la xtabond, xtabond2, xtdpd, etc.

ivreg2 can do this, but it would be a major hassle because you'd have to
construct all the instrument sets by hand.  Can be done but not a lot of

The easier option is to use xtabond2 (or one of the other estimators)
and do an endogeneity test by hand as a GMM distance test, i.e., a
difference in J stats.  This is discussed briefly in the Baum et al. SJ
papers you cite; if you are looking for a textbook discussion, Hayashi's
(2000) "Econometrics" text covers it.  In a nutshell, the difference
between the J stat with the regressors treated as endogenous, and the J
stat with the regressors treated as exogenous, should be distributed as
chi2 (with dof=number of tested regressors) under the null that the
regressors being tested are exogenous.  The advantage of the ivreg2
implementation is that it uses a version of the test that is guaranteed
to be nonnegative in finite samples, but that's not a big deal.

BTW, you are mistaken about the documentation of the endog option of
ivreg2.  It is discussed in some detail in the Baum et al. (2007) paper.


> -----Original Message-----
> From: 
> [] On Behalf Of 
> Dithmer, Jan
> Sent: 19 October 2011 10:39
> To:
> Subject: st: ivreg2 or xtabond2 endogeneity of regressors
> Dear all,
> can the ivreg2 program of Baum et al. (2003, 2007) also be 
> used for dynamic GMM estimation in STATA or just for static 
> regressions?
> I am especially interested to use the program to test subsets 
> of regressors for endogeneity. According to the paper 
> "Instrumental variables  and GMM: Estimation and testing"
> I would use the orthog option in ivreg2. However, I am a bit 
> confused, because in the Stata program (ivreg2) the endog 
> option is available as well, which is not mentioned in the paper.
> Thus, would the endog option be the right choice for test for 
> endogeneity of regressors and orthog is used to test for 
> exogeneity of instruments?
> Or is there a possibility to do the same thing in xtabond2 of 
> David Roodman? I am working with panel data.
> Thank you very much for your help!
> Best,
> Jan
> *
> *   For searches and help try:
> *
> *
> *

Heriot-Watt University is a Scottish charity
registered under charity number SC000278.

Heriot-Watt University is the Sunday Times
Scottish University of the Year 2011-2012

*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index