Re: st: Re: 3SLS and HAC SEs

[Christopher Baum <kit.baum@bc.edu>]

<>
On Mar 29, 2011, at 2:33 AM, Khurana wrote:

> Ok, as pointed out by Kit Baum, this is indeed a 2SLS regression.
> 
> That said, I am still a bit confused about the correct syntax.
> 
> If I run "ivreg2 w (y z = a1 a2 a3 b1 b2 b3), kernel(tru) bw(12) robust",
> Stata assumes that y and z are instrumented with all the 6 "a" & "b"
> variables. What I would like to estimate is a model in which y is only
> intsrumented with "a" variables, z - only with "b" ones.

In 2SLS (limited information) estimation, you can't do that. Al exogenous variables are used to instrument all variables needing instruments.

Conceptually, you can impose the restrictions you consider by estimating three separate equations in 3SLS, a systems (full information) estimator.
But the antediluvian code of -reg3- does not support anything beyond classical standard errors -- not robust, nor cluster-robust, nor HAC, nor bootstrap -- so that you cannot use -reg3- with Newey-West standard errors.

But the bigger question: why do you want to impose these constraints? If you add the -first- option to ivreg2, and do a test for the joint significance of the b variables in the y equation (or the a variables in the z equation), and those terms are jointly significant, that says your a priori restrictions are rejected by the data, and the regressions you want to run for y and z are misspecified. That is testable, and should be tested. If you reject the null hypotheses of exclusion restrictions, then you should just let all a and b variables enter both FSRs.

Kit

Kit Baum   |   Boston College Economics & DIW Berlin   |   http://ideas.repec.org/e/pba1.html
                              An Introduction to Stata Programming  |   http://www.stata-press.com/books/isp.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/