st: re: some questions about ivreg2...

[Kit Baum <baum@bc.edu>]

Francesco said

Dear users, I'm a newbie and not a great econometrician...

I'm trying to estimate a linear equation model (unbalanced panel  
database) where one regressor is endogenous:

* yit = x1it x2it x3it uit

yit is the dependent variable
x1it and x2it are exogenous
x3it is endogenous
uit error

By Stata, I run the following model:

use...
sort id year
tsset id year

* ivreg2 y1 x1 x2 (x3 = x4 x5 x6), first bw(2) gmm2s kernel(tru) robust

(note: x4 and x5 are suggested by economic theory)

Then, I was suggested to compute the same equation by "xtivreg2",  
getting very different result (and the insignificance of one  
instrument)!
I know that as estimation methods differ you have something of a  
check on whether models
are stable, efficient and so on, but I really don't understand why  
and how it would be better estimating the equation above by xtivreg2  
(or xtvreg). Moreover:

1) what are the differences and the reason of choice between: ivreg2  
(WITH gmm option) and xtivreg and xtvreg2?
2) What does concretely mean that xtvreg2 is a "wrapper" for ivreg2  
(from the help of Stata)?
3) Would be incomplete an econometric analysis employing an ivreg2  
estimation only?
4) For xtivreg2, the instruments for the endogenous variable could be  
the lagged values of the endogenous or are usually the same of a  
previous ivreg2?
5) More in general are there papers or articles able to explain the  
reason for one choice of computation or for the other? These are new  
methods and I'm a bit confused.



The difference between ivreg2 and xtivreg2 is the same as the  
difference between regress and xtreg, fe. If you have panel data and  
there is significant unobserved heterogeneity per panel, fixed  
effects can 'soak up' those unit-specific factors. If you run  
xtreg,fe, you get an F-test at the bottom that all the fixed effects  
(measured as deviations from their grand mean) are zero. If you  
reject that hypothesis, OLS is misspecified and you should use fixed  
effects.

Exactly the same logic applies to estimating the basic equation with  
IV: should you include fixed effects? If so, then you should use  
xtivreg2 rather than ivreg2, with other options the same.

xtivreg2 is a 'wrapper' for ivreg2 in that xtivreg2 does not function  
independently of ivreg2. xtivreg2 affords access to all of the  
options of ivreg2. Please see the Baum-Schaffer-Stillman working  
paper no. 667 (available below) for details.

Kit Baum


Kit Baum, Boston College Economics and DIW Berlin
http://ideas.repec.org/e/pba1.html
An Introduction to Modern Econometrics Using Stata:
http://www.stata-press.com/books/imeus.html

*
*   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/