Maarten--
I don't see the source of endogeneity--but I suppose y3 and e1 are
correlated. I will reiterate the concern I had with Kathleen's
approach. Is the -tobit- the right approach? i.e. is it reasonable
to assume that e2 is normally distributed and y2 is set to zero when
negative to give y3? If y3 is nonnegative, that does not mean that a
-tobit- is the right approach; often a -poisson- gives better results.
Try
lpoly y3 x2
for a sense of the "right" functional form. But in general, running
-ivreg- or -ivreg2- or -ivregress- using a generated regressor
produces consistent estimates, so you can do
qui poisson y3 x1 x2 if y1<.
predict y3hat if e(sample)
ivreg2 y1 x1 (y3=y3hat)
for example, and compare to
ivreg2 y1 x1 (y3=x2)
and
qui tobit y3 x1 x2 if y1<.
predict xbhat if e(sample)
g vhat=y3-xbhat
ivreg2 y1 x1 (y3=x2 vhat)
for example.
*
* 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/