Page 1 of 1
option "l=number" in GMM
Posted: Thu Oct 16, 2008 7:11 pm
by maxchen
For the command gmm, the Command Reference says
l=number Set maximum number of iterations on the first-stage iteration to get the one-step weighting matrix.
What is it for? can not find the difference in equation object or system object gmm estimation.
Re: option "l=number" in GMM
Posted: Thu Oct 16, 2008 9:26 pm
by EViews Gareth
That option sets the number of pre-GMM least squares iterations that are performed to get starting values for the GMM procedure in non-linear equations. It is only in rare circumstances that this option will effect the final results.
Re: option "l=number" in GMM
Posted: Fri Oct 17, 2008 7:26 pm
by maxchen
thx
What is the option "e" then?
Code: Select all
pageload cs 'cs.wf1 is an example workfile of EViews
smpl @first 1995q1
equation eq01.tsls(n) log(cs) c log(gdp) @ c log(cs(-1)) log(gdp(-1))
equation eq02.gmm(e) log(cs) c log(gdp) @ c log(cs(-1)) log(gdp(-1))
In my opinion, the coefs and std errs should be equal between eq01 and eq02. However, the std errs are not equal, why?
Re: option "l=number" in GMM
Posted: Sat Oct 18, 2008 1:18 am
by EViews Gareth
GMM estimation doesn't use a degree of freedom correction, whereas TSLS (in a single equation) does not.