Hi, 

first I would suggest to check out the classic Stock/Wright/Yogo reference given in the help for tsls. Basically, all the standard IV tools only work under the assumption of homoscedasticity. Otherwise, one quickly enters the GMM world.

A somewhat more recent relevant reference is Hausman, Newey, Woutersen, Chao, Swanson 2012: https://onlinelibrary.wiley.com/doi/abs/10.3982/QE89

where the abstract starts with "This paper gives a relatively simple, well behaved solution to the problem of many instruments in heteroskedastic data." This is not available in gretl, however, AFAIK.

cheers

sven

Just found out that it is not possible with robust standard errors, but doable with ordinary SE. Why?

Best, Andreas

On Wednesday, July 9, 2025 at 03:54:08 PM GMT+2, Andreas Zervas <anzervas@yahoo.com> wrote:

Hi all, 

I was running IV regressions in a loop using gretl 2024d in windows 10, and saw that while it calculates first stage F, it does not do the same with Cragg - Donald statistic. Is there a reason for it?

How can we calculate it the Cragg - Donald with e.g. matrices? Is there a simple formula to do it?

Best,

Andreas

_______________________________________________
Gretl-users mailing list -- gretl-users@gretlml.univpm.it
To unsubscribe send an email to gretl-users-leave@gretlml.univpm.it
Website: https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/