Re: st: Do 2sls, ivreg etc. test the rank condition of identification?

["David M. Drukker" <ddrukker@stata.com>]

Thomas Cornelissen <cornelissen@ewifo.uni-hannover.de> gave an example of
-reg3- estimating the parameters of a model that fails the rank condition.

Kit Baum <baum@bc.edu> and "Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>
commented on this topic as well.  Kit privately sent me an example using the
auto dataset in which -reg3- produced estimates of the parameters of model
that fails the rank condition.

    . webuse auto
    . reg3 (price mpg weight displacement turn) (mpg price displacement)
           (weight displacement trunk turn length)

It is unusual for the order conditions to hold and the rank conditions to
fail.  In fact, both Thomas' and Kit's examples rely on introducing an
equation that is not simultaneous with the other equations to achieve this
result.  See, for example, Kit's last equation.

Another situation that gives to this problem is when one equation is a
linear combination of other equations in the system.

Regardless, once we have confirmed that the rank condition can be properly
tested, even when linear constraints are applied to the model using the
option -constraint()-, we will add a rank-condition test to -reg3-.

In the interim, if you are concerned that your model fails the
rank-condition test, you can specify the -ireg3- option.  This option will
trap the condition because when the rank condition is not met, the model is
not identified, and the iterated feasible generalized least squares used by
-ireg3- will be unable to converge.


     --David			 --Vince
       ddrukker@stata.com          vwiggins@stata.com

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/