[Gretl-users] Re: Restricted vars
Am 18.11.2020 um 08:41 schrieb Riccardo (Jack) Lucchetti:
On Wed, 18 Nov 2020, Adam Elderfield wrote:
>
> Just wondering if it's possible to estimate restricted vars via EGLS
> as proposed by Lutkephol (New introduction to time series analysis)?
> That is, can you feed a restriction matrix to a var estimation?
>
> Or would you need to estimate the system with SUR and the system
> command?
I think, the latter. You can skip entire blocks of lags for all
equations in the 'var' command (as documented). But for other types of
restrictions (where OLS becomes inefficient) I think we currently
require to switch over to the system/sur.
I do think and agree that we should have more friendly tools to make
this switch easier. This doesn't have to be in core gretl in C, AFAICS,
packages written in hansl would do nicely.
Basically, SUR estimation is entirely possible, provided your
restrictions are not too weird (zero restrictions are absolutely ok).
See this thread:
https://gretlml.univpm.it/hyperkitty/list/gretl-users@gretlml.univpm.it/t...
I'm not sure whether SUR and EGLS are equivalent in this context. I
suspect they are, but I'm not 100% sure. Sven?
I had to look up what Helmut Lütkepohl meant with the "E" here, but it's
just "estimated" (and not "efficient", for example). So yes EGLS
should
be the same as FGLS ("feasible") and SUR. The New Intro book avoids the
old "seemingly unrelated" terminology, that term only appears inside the
literature references.
cheers
sven