[Gretl-users] Re: About the PMG package
On Wed, 24 Mar 2021, Sven Schreiber wrote:
Exactly like we do in our MG routine (only that we don't apply it
to the
short-run coeffs in the case with homogeneous long-run coeffs).
Of course we could add all this at some point, when we have the time.
But again, I'd like to know for which purpose. I've always been
skeptical about the meaning and interpretation of MG estimates, given
the assumption of no common structure.
I completely share your point of view. I guess they can be treated as some
sort of descriptive statistic, at best[*]. However, I _could_ see the
point of providing a special function for printing out the full set of
parameters for a given unit in a more nicely formatted way (with standard
errors etc).
Of course, one could always argue that being able to replicate the results
you get from another package is of some intrinsic value, a point which I
wouldn't find entirely baseless.
[*] The only way I could see those numbers as meaningful estimates of
something would be by making some funky assumption about random
coefficients, and even then I have no idea whether they would enjoy
desirable asymptotic properties, and under which circumstances.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------