Re: st: RE: Testing for instrument relevance and overidentification when the endogeneous variable is used in interaction terms

[Jason Wichert <jasonw8907@gmail.com>]

Fellow statalisters,

At the risk of spamming this list, I’ve encountered another issue
related to 2SLS with endogenous interaction terms in an unbalanced
panel. The question concerns clustering of the standard errors.

Say I want to cluster the standard errors at the industry and year
level. Alternatively, I might want to cluster the standard errors at
the firm level. Using ivreg2, I would normally add “, cluster(industry
year)” as option to get heteroskedasticity robust two-way clustered
standard errors.  In the case of firm-level clustering, I’d specify “,
cluster(id)”.

Since I have multiple endogenous interaction terms, I have decided to
follow a procedure outlined earlier ( as “c)” in
http://www.stata.com/statalist/archive/2013-06/msg00462.html ). In a
nutshell, I predict my endogenous variable in a preliminary
regression. I then generate interaction terms of this fitted value
with the exogenous variables used in the second stage interaction
terms. Those generated interaction terms are used as instruments for
the second stage interaction terms.

Now the question is, do I need to cluster the standard errors in the
preliminary regression as well?

Any feedback is highly appreciated.
-Jason

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/