[Gretl-users] Re: About Pooled Mean Group estimation
Am 12.01.2021 um 18:26 schrieb Reynaldo Senra:
PMG_estimate(&b, , {0;0;0},1000)
Most of the time, I achieve the exact results of Eviews by setting {1;0;0}.
Unfortunately, in one case it has been impossible and I really tried
several combinations. Here I just attached a csv file with the data
involved in the problem. I also attached the Eviews output and the Gretl
sample script.
Hi Reynaldo, that's an interesting case study for the PMG package. First
let me note that the automatic setting for the initial values yields a
better likelihood than either {0,0,0} or {1,0,0} in this case.
Anyway, if you try:
PMG_estimate(&b, , {0.5,0.5,-1}, 1000)
then you will get the same results as Eviews. But that's of course
cheating, because I picked those values in the neighborhood of the
results I saw. So that means that in general we're not clever enough
choosing our initial values, and here we apparently got stuck in a local
maximum.
I'm also noticing, however, that the parameters of the mean-group
estimator {0.89, -0.46, -1.8} lead us to the (presumably) correct PMG
result again. Maybe we should routinely also try MG as an
initialization. Jack?
cheers
sven