Hi there, I have a question concerning p.146 in the user-guide: Is this correct? "dz = diff(z) dpanel 1 ; y dz dpanel 1 ; y dz ; GMM(*z*,0,0)" Or should it be dpanel 1 ; y dz ; GMM(*dz*,0,0) It the first one is right, then it's because the differences of the instruments are computed after using them as instruments, while in the latter specification they are not altered any more. I tried both variants in a few examples and found that the results of dpanel 1 ; y dz dpanel 1 ; y dz ; GMM(*dz*,0,0) are very similar, but not the same, while dpanel 1 ; y dz dpanel 1 ; y dz ; GMM(*z*,0,0) gives 'totally' different coeffietints and especially s.e. for dz. I think it's quite crucial to understand this methodology and hence I'd be happy for clarification. Cheers Leon

On Sun, 2 Jun 2013, Pindar wrote: > Hi again, > > so, is it 'dz'? I'll pick that one up in context, below. > I've got another question to an example of the user-guide: > <hansl> > open abdata.gdt > dpanel 2 ; n X ; GMM(w,2,3) GMM(k,2,3) Ivars --time --two-step --dpd > # equal to > dpanel 2 ; n X ; GMM(w,2,3) GMM(k,2,3) *GMM(w_1,2,2)* Ivars --time --two-step > --dpd > <hansl> > > Does gretl automatically add these instruments or is by construction of the > estimation procedure already included? It's by construction of the procedure. > 25.05.2013 16:31, Pindar: >> >> I have a question concerning p.146 in the user-guide: >> >> Is this correct? >> >> "dz = diff(z) >> dpanel 1 ; y dz >> dpanel 1 ; y dz ; GMM(*z*,0,0)" [ quoted from the Guide ] >> >> Or should it be >> dpanel 1 ; y dz ; GMM(*dz*,0,0) >> >> It the first one is right, then it's because the differences of the >> instruments are computed after using them as instruments, >> while in the latter specification they are not altered any more. >> >> I tried both variants in a few examples and found that the results of >> >> dpanel 1 ; y dz >> dpanel 1 ; y dz ; GMM(*dz*,0,0) >> >> are very similar, but not the same, while >> >> dpanel 1 ; y dz >> dpanel 1 ; y dz ; GMM(*z*,0,0) >> >> gives 'totally' different coeffietints and especially s.e. for dz. >> >> I think it's quite crucial to understand this methodology and hence I'd be >> happy for clarification. I'd welcome Jack's contribution on this -- it's tricky, and it's a long time since I thought about this. But here's my current take on the example you cite (the purpose of which is to highlight the difference between two specifications which may look at first glance as if they should be equivalent but which are in fact different). First, I'm not sure why we explicitly use the first difference of z (dz = diff(z)) as a regressor in the example, since dpanel does differencing automatically (so we'd be using the second difference of z here, which is not necessarily wrong but seems like an unnecessary complication). Second, the standard procedure with the DIF estimator is to use levels as instruments, and this is what happens when you give the name of the level variable in a GMM() spec. So -- although I may be missing something -- I think the example should read like this: # first case dpanel 1 ; y z # second case dpanel 1 ; y z ; GMM(z,0,0) But if we really wanted the second difference of z in the first spec, then it would read: dz = diff(z) dpanel 1 ; y dz dpanel 1 ; y dz ; GMM(dz,0,0) where we use the first difference as instrument for the second difference. Allin Cottrell _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users