[Gretl-users] Re: About the PMG package
Am 24.03.2021 um 16:15 schrieb Riccardo (Jack) Lucchetti:
On Wed, 24 Mar 2021, Sven Schreiber wrote:
>>> Maybe the way the sort run coefficients are not obtained by OLS in
>>> Eviews. I don't know, everything is quite confusing for me.
>>
>>
>> I have the feeling (but I may well be wrong) that Eviews uses the PMG
>> method for estimating the long-run parameters and the loadings and
>> then runs a pooled model for producing the short-run parameters.
>
> No no, the short run is heterogeneous of course. I'm quite sure Eview
> also just runs separate second-stage OLS regressions, because that's the
> best thing one can do.
Uhm, but in that case I don't really understand what eviews reports:
it looks as if theh have only _one_ coefficient per short-run variable
(as opposed to having N). Maybe those are cross-sectional averages?
And if so, how are the standard errors computed?
Again, it's hard to discuss this without having seen the screenshot. But
if we're just talking about the plain direct Eviews output, then these
are the MG estimates. The SEs are the so-called non-parametric estimates
from the cross-sectional dispersion of the heterogeneous coefficients.
(Applying the plain Eviews output to an individual unit would be wrong,
given the short-run heterogeneity. Eviews also provides the
heterogeneous coefficients in some post-estimation view.)
cheers
sven