On Tue, 28 Jan 2014, Stefano Fachin wrote:
I'd be grateful if anybody could answer this very reasonable
question a
referee put to me: why only the ADF-GLS test with constant have
p-values, while those with trend do not? the manual does not explain it,
and I am getting curious...
We use James MacKinnon's (1996) "urcval" apparatus to obtain p-values for
our unit root tests, and this apparatus does not cover the case of ADF-GLS
with a trend (not even asymptotically). So we fall back to the critical
values given by Elliott, Rothenberg and Stock (1996).
When the gretl code in question was written we were not aware of a
(manageable) way of generating p-values for ADF-GLS with trend. I haven't
researched the matter since, but if you know of a method we should now be
using, please speak up.
Allin Cottrell