Re: [Gretl-users] New Windows snapshot - Again....
On Tue, January 31, 2006 20:19, john w wrote:
Hmm...I saw omitting endogenous variables in PcGive. It is possibile.
These
are so called Exclusion restrictions and all variables in VAR can be tested.
You can test all the variables in VAR using F test but Exclusion
restrictions are done via Subset Chi^2 (square) test and you can test
endogenous, deterministics and exogenous variables.
Of course it is possible. VARs are rather trivial parametric models for which
standard inference methods apply (at least in the stationary case --- see
below). My point was, general exclusion tests are often pointless in VARs,
since these models don't aim at a structural representation of the data, but
only at encapsulating conveniently their time-series properties. If you feel
like carving out variables and/or lags, you're much better off with a
simultaneous equations system.
Are Granger tests doable for VECMs ? As tere are no F tests in
VECM
output. (my question excludes testing beta)
This brings back a point you raised some time ago (see
http://ricardo.ecn.wfu.edu/pipermail/gretl-users/2006-January/000326.html and
subsequent messages). I had a look at JMulTi. They do Granger causality tests
in cointegrated systems by using a very ingenious method, proposed by
Lütkepohl (usurprisingly) and Dolado, that bypasses most of the inferential
difficulties by estimating a levels VAR of higher order than necessary. It's
not difficult to do the same in a script, I think (but I have to dig a little
deeper into this).
--
Riccardo "Jack" Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"
Ancona