st: Re: GMM: 2SLS with over-identification

[Kit Baum <baum@bc.edu>]

On Apr 21, 2010, at 2:33 AM, Ed wrote:

> In Stata 11 -gmm-, is it possible to estimate an over-identified 2SLS
> model in which some observations have fewer than the full number of
> instruments?  This is possible when estimating a dynamic panel model
> (with either -gmm- or -xtabond-).  The online Stata manual says:
> 
> "-gmm- automatically excludes observations for which there are no valid
> observations for the panel-style instruments. However, it keeps in the
> estimation sample those observations for which fewer than the maximum
> number of instruments you requested are available. For example, if you
> specify the lags(2/4) suboption, you have requested three instruments,
> but -gmm- will keep observations even if only one or two instruments are
> available."
> 
> I would like to be able to do something similar, but for 2SLS
> (over-identified).  Is that possible?

Presumably you can 'roll your own' version of the trick used in the Arellano-Bond estimator: in the constructed instruments, nonexistent values (e.g., higher lags than are available at that point in time) are set to zero. So instead of using a lags() option to create the instruments, create them explicitly and fill their missing values with zeros. 

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/