On Wed, 18 Jan 2012, Sven Schreiber wrote:
I think I was the one to suggest a warning in the Johansen test
when including exogenous variables. Now I have noticed (have been
alerted) the following:
* gretl now also outputs p-values for the case of exogenous I(1)
variables -- I vaguely seem to remember a discussion about this, and
it's cool of course. I think it would be helpful, however, to give a
reference in the output to what setup (article) this refers. (Or maybe
not in the output but in the command reference for coint2, but I
couldn't find anything there, either.)
Yes, it should be documented. But see
The relevant tests are based on Harbo, Johansen, Nielsen and Rahbek,
"Asymptotic Inference on Cointegrating Rank in Partial Systems"
(Journal of Business and Economic Statistics 16/4, October 1998);
critical values re-simulated in gretl with 50,000 replications.
* The displayed warning says "Warning: the p-values shown are
case of a restricted trend", which is kind of strange. Does it actually
mean what it says? (It can't, really, because this would contradict the
output relating to exo I(1) variables...)
Hmm. The intent (and I think the actuality) is that that warning is
printed only when the user has selected the "unrestricted constant"
case. But since that case is statistically weird we drop back to the
"closest" less-weird case, namely the restricted trend (as does
CATS, I believe). The assumption is that the "exo I(1)" variables in
question are something other than constant or trend.