[Gretl-users] Doing forecasts with system equation?
Hi!
One last try before I go back to Eviews :-)
Does anyone - maybe the creators of this gretl feature - know if one can do forecasts with
the system equation tool? Defining macro-models with equations and identities for example
and estimate the variables in mutual dependency? And by chance, has anyone done this
before and can show me the way a little bit?
regards, Jan
________________________________
Von: gretl-users-bounces(a)lists.wfu.edu im Auftrag von Sven Schreiber
Gesendet: So 09.11.2008 15:21
An: Gretl list
Betreff: Re: [Gretl-users] Newbie questions about using systems in Gretl
Hi Jan,
I don't have any experience using gretl's "system" (the
simultaneous-equation, structural thing, as opposed to VARs and VECMs),
so I can only offer some general comments.
First, it seems to me that this is an area that could be better
documented; in the gretl guide there is no chapter about it AFAICS. And
the command reference is as condensed as it probably should be, but then
of course is no substitute for the missing chapter in the guide. Maybe
that should be added as a minor bug to the bug tracker.
On to your questions:
1. In the command reference, page 51 (A4-Version) about using systems
it is mentioned that identities are needed if FIML estimation is
used. But also with ols or other estimations the system has to
know how to create the forecast values of derived variables which
are not estimated inside the system. Did I miss anything in the
reference? And I guess that also with ols-estimation I have to
declare the endogenous variables?
Identitites as such need not be estimated, so for other estimators I'd
say it's really not necessary. Forecasting is the step after estimation
and is a different issue.
With respect to endogenous variables and OLS: with OLS it's really only
the left-hand variables that can be / should be endogenous. If you have
right-hand variables that are known to be endogenous, OLS will be
inconsistent, that's why you have tsls, 3sls, etc.
2. Which command is necessary if I want the system to do
forecasts
for the system as a whole? In the graphic interface there is the
possibility to select only one estimated variable, but as you know
the clue of using systems is all variables are estimated
simultaneous in mutual dependency...
That I don't know at all. Maybe it's really not implemented yet in
gretl? In principle, being only able to pick a single variable for the
forecast graph does not necessarily mean that no system forecast is done
in the background. But I somehow doubt that that's what's really happening.
But as I said, I don't have experience with that area in gretl, I'm only
speculating. Anybody else know the answer?
good luck,
sven
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
Attachments:
- winmail.dat (application/ms-tnef — 6.2 KB)