Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: ivreg2

From   "Mark Schaffer" <[email protected]>
To   [email protected]
Subject   Re: st: ivreg2
Date   Thu, 2 Oct 2003 10:50:14 +0100


I'm a co-author of ivreg2, so let me try to answer some of your 

From:           	[email protected]
To:             	[email protected]
Subject:        	st: ivreg2
Date sent:      	Wed, 01 Oct 2003 14:41:39 +0200 (CEST)
Send reply to:  	[email protected]

> Dear all,
> I have a  panel household data set. I am using the command ivreg2
> which allows one to estimate instrumental variables models using GMM.
> I would like to obtain standard errors that are efficient in presence
> of arbitrary conditional and unconditional heteroskedasticity.
> I estimated the following model : 	
>  ivreg2 $Y($X1 $X2 $Z2 year = $D_X1 $D_X2 $B_X1 year), gmm
> cluster (indiv2) ffirst

It looks like year is in your list of endogenous variables, and also 
in your list of excluded instruments.  Should it not be listed as an 
exogenous regressor, e.g.,

ivreg2 $Y year ($X1 $X2 $Z2 = $D_X1 $D_X2 $B_X1), gmm
cluster (indiv2) ffirst

 > because I assumed that the disturbance covariance matrix had a
> clustering form (correlation between observations from the same
> household). With this specification, the test of overidentifying
> restriction is rejected and the estimated coefficients are quite
> differents from that one obtains with the following estimation :
>  ivreg2 $Y($X1 $X2 $Z2 year = $D_X1 $D_X2 $B_X1 year), gmm
> robust ffirst
> In addition, with this last estimation, the test of overidentifying
> restriction is not rejected.
> I do not understand well what is going on. Is it a problem of
> conditional  heteroskedasticity which explains that the disgnostic
> tests, whith the latter estimate, are not valid...? Why the results
> (coefficients estimated,diagnostic tests) are so differents?

The -cluster- option makes your estimates robust to intra-group 
correlation of arbitrary form as well as robust to 
heteroskedasticity.  Since you have a panel, it is very likely that 
you have intra-group correlation.  In other words, in a pooled 
estimation such as yours (no fixed or random effects), the error term 
for Mr. J in year 1995 is very likely to be correlated with the error 
term for Mr. J. in 1998.

This means the independence assumption that you need for your second 
(non-cluster) estimation is almost certainly violated.  The standard 
errors are probably wrong.  Furthermore, you are doing 2-step 
feasible efficient GMM, and the 2nd step uses an estimate of the 
covariance matrix of the moment conditions.  Since this estimated 
matrix is also probably wrong, the coefficients you get from the 2nd 
step are probably wrong too.  The overid stat is wrong as well, and 
so forth.

Your first set of estimates doesn't suffer from this problem, since 
they do not use the independence assumption.  But if you fail the 
overid test, then you have evidence that your orthogonality 
conditions are violated, and you need to investigate further.

Hope this helps.

> Thanks a lot for your help,
> B´┐Żatrice d'Hombres
> *
> *   For searches and help try:
> *
> *
> *

Prof. Mark E. Schaffer
Centre for Economic Reform and Transformation
Department of Economics
School of Management & Languages
Heriot-Watt University, Edinburgh EH14 4AS  UK
44-131-451-3494 direct
44-131-451-3008 fax
44-131-451-3485 CERT administrator

*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index