st: two stage least squares vs. sem

[Jonathan Richards <richards.868@buckeyemail.osu.edu>]

Hey everyone - 

I am trying to run a model with three endogenous independent variables. But the instruments for each endogenous variable are not significantly correlated with the other endogenous variables. So when I used xtivreg2, which puts all the instruments into each first stage equation, there was underidentification.

I thought of running an sem model in with a separate first stage equation for each endogenous variable. Then I wouldn't need to include all the instruments in each first stage equation.

Is this a good approach? Specifically, would such a model be asymptotically equivalent to 2SLS (if I included the exogenous variables from the second stage in each first stage equation)? And actually, would I need to include the exogenous variables from the second stage in each first stage?

Any insight would be much appreciated.

Thanks.


Brent

*
*   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/