st: re: ivreg2 2sls, gmm2s and autocorrelation test

[Kit Baum <BAUM@BC.EDU>]

< >
Marie-Helene said

so, to finish with this question, and because nobody seems to have any  
idea
about why I have such results with abar after 2SLS robust or GMM2S  
robust, I
have a few very simple questions:

1). to estimate a regression with one endogenous regressor AND  
PROBABLY H , what
is best: 2SLS H robust or GMM2S H robust? what would you do?
2). to test for AC after that, what would you use: ivactest or abar?
3). to estimate a regression with one endogenous regressor AND  
PROBABLY H AND
AC, what is best: 2SLS HAC robust or GMM2S HAC robust? what would you  
do?

In this example, ça ne fait rien:

use http://fmwww.bc.edu/ec-p/data/wooldridge/phillips.dta
tsset year, yearly
ivreg2 cinf (unem = L(1/2).unem), bw(3) kernel(bartlett) robust
abar, lags(6)
ivactest, s(6)
ivreg2 cinf (unem = L(1/2).unem), bw(3) kernel(bartlett) robust gmm2s
abar, lags(6)
ivactest, s(6)

Re (2): Note that abar tests for each level of AC whereas ivactest is  
a portmanteau-style test like the Q test; s(6) says test for AC of  
orders 1,2,3,4,5,6 jointly.

Re (1) and (3): In large samples (which admittedly this example is  
not) the linear IV-GMM estimator is more efficient than the IV-2SLS  
estimator if the errors are not considered i.i.d. So if you consider  
H, AC or HAC, assuming non-i.i.d. errors, you should prefer the IV-GMM  
estimator. In -ivreg2-, as you can see from the example, they both  
produce a J statistic when HAC is invoked.

For more details please see Baum-Schaffer-Stillman, Stata Journal 7:4,  
available in preprint form from my homepage below.


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/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/