Re: st: Method for first stage, in xtivreg

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

<>
On Dec 11, 2011, at 2:33 AM,Daniel wrote:

> You asked in the mentioned post "why would you need to 'roll your own' [1st
> stage regression]".
> The thing is, that I need the predicted values of the instrumented variable
> to integrate those into a xtregar,re. I have no idea how to get the
> predicted values of the instrument from the first stage which Stata uses
> afterwards into the second stage estimation of xtivreg,re.
> The only way that I came up with is to do a 1st stage estimation manually
> and use the predicted values in the xtregar,re.

If you could do this, it would be very challenging to get the correct VCE for the second-step estimator. Furthermore, 
if the standard assumption of a random-effects model, that X \perp u, is often very difficult to support, it is even
less likely to be the case that X \perp u (as it must) in a time-series model with autocorrelated errors: but that 
assumption must be tested for the xtregar, re to have any validity. If you are finding issues of autocorrelation in
the errors, I imagine it is not due to an inherently autocorrelated error process, but rather a reflection of some
sort of misspecification---often omitted dynamics. I would try to deal with the autocorrelation by refining the specification
and then using a cluster-robust VCE, which in the panel context allows for arbitrary within-panel correlations in the
disturbance process.

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/