st: ivprobit estimation with weak instruments

["Danny Cohen-Zada" <danoran@bgu.ac.il>]

It is known that when one uses weak instruments, the standard error of the 
endogenous regressor is baised.

Suppose for example that i have the following model in which y2 is the 
endogenous regressor:


1)  y1= ao + a1*y2 + a2*x1 (y1 - is a dichotomous variable)
2)  y2=b0+b1*x1+b2*z1 (z1 is the excluded instrument0

(one instrument, one endogenous regresssor).



In order to obtain unbaised standard errors (due to the problem of weak 
instruments) i thought to do the following.

regress y2 x1 z1
predict, ppu
probit y1 ppu z1, cluster(x3) vce(boots)

Am i correct that by applying this procedure i can obtain the correct 
standard errors of the coef. a1?

In addition, i want to graph the predicted probability as a function of y2 , 
including in the graph the  correct confidence interval (I keep  all the 
other variables are at their mean). I thought to run the following commend 
after the probit estimation:


prgen ppu , f(30) t(38) rest(mean) gen(pp) gap(1) ci bootstrap


Am i correct that thus i get the correct confidence interval for the 
predicted probability (even when my instruments are weak)?

Thanks,

Danny



Dr. Danny Cohen-Zada
Department of Economics
Ben-Gurion University of the Negev
P.O.Box 653
Beer-Sheva 84105
Israel

Tel: +972-8-6472301
Fax: +972-8-6472941

http://www.econ.bgu.ac.il/facultym/danoran/main.htm 
*
*   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/