please keep the discussion on the mailing list (I added it again here).
From what you describe it sounds as if, yes, gretl has all these tools.
But right now I don't have the time to implement this any further or to
do the bug-hunting. I already gave you something that was working as a
starting point. I'm afraid you're on your own for now. Or perhaps
somebody else on the list is interested.
Am 15.07.2015 um 08:25 schrieb JOSE FRANCISCO PERLES RIBES:
Sorry for disturb you with this issue, but I send this mail only to
update the information and to do a small question.
After my request, professor Angrist very kindly has send me some files
with a SAS programme to calculate the Angrist-Newet (1991) test.
In the pdf_III pages 2 and 3 there is a detailed procedure to get the X2
I have seen the relevant steps as:
1).-Compute the variables in means deviation. (page 2)
2).-regress a fixed effect model using as "endogenous" variables the
means deviation transformation and save the residuals. (page 3)
3).-regress the residuals for each time on lags and leads of all orginal
variables (non-transformed) **computation 3sls statistics** (page 3)
To estimate the fixed-effect model and the test statistics he uses a SAS
procedure based on 3SLS estimation named "PROC SYSLIN SUR VARDEF=n"
So, my question is: Gretl has this kind of system estimation? or its fe
routine leads would lead to simmilar results (as I think)?
Also, looking for the Angrist code, I think the script your provide me
to calculate the X2 statistic was correct. But would you mind to see the
Angrist code to see if I need to adjust some of the script? Because I
found a problem with the number of parameters estimation.
Many thanks for your attention and sorry for any inconvenience.
University of Alicante.