Re: [Gretl-users] vecm comparisons
On Fri, 22 Jan 2010, John C Frain wrote:
If you include seasonal dummies in the "NO CONSTANT" or
"restricted
constant" it is unlikely that the sum of the seasonals will add to
zero and you will then be imposing a trend on the undifferenced series
in the vecm. In such a case the relevant statistics will not follow
the "Johansen" distributions. Thus it may be more correct to not
allow seasonal dummies in those cases.
Hmm. I haven't given a great deal of thought to this, so I may be mistaken
here, but this point doesn't seem relevant to me. First, the only step in
the Johansen procedure when you need to take care of the deterministic
terms for conducting inference is when you test for the cointegration
rank. And even then, centered seasonal dummies do no harm, since they have
no effect on the _asymptotic_ distribution of the test statistics
($\lambda$-max or trace). The fact that they may not sum to zero in a
finite sample is totally irrelevant; what matters is that their average
goes to 0 asymptotically. But once the cointegration rank is taken as
given, seasonal dummies are completely irrelevant for the distribution of
the statistics.
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti