Probit with instrumental variables

[Ellesar1]

Hello,

I have estimated a probit model, and I suspect that one of the binary regressors is endogenous.

So I would like to perform the probit model with an instrument- and the instrument I currently have in mind is also binary.

It seems Stata has as function for that- can it be true that Eviews does not have that, and in that case, do you have any other suggestions for estimation of the model?

I have Eviews 7.

Best regards,
Carsten Andersen

[EViews Glenn]

I believe that the Stata ivprobit (if that's the routine to which you are referring), is not applicable if the endogenous regressors are binary. In EViews, I'd probably recommend doing IV on the linear probability model, and using White robust standard errors.

[Ellesar1]

Thank you for your reply.

If you also wanted to estimate an ordered probit with instruments, what could you then do?

[EViews Glenn]

Not sure. Offhand, I don't think it's an easy problem. You could run separate binary IV LPMs for each threshold. This is obviously not efficient, but the results of each should be consistent. You'd have multiple estimates of the beta which you'd have to handle in some way, perhaps with minimum distance estimation, though I'm not certain of the properties of that approach.

There may be some results in the GLM literature, but I'm not sure as I haven't really worked with the ordered versions of those models.

[Ellesar1]

Thank you. Maybe I will try that.