> It seems that the error is caused by a change in the sample size. Using
> the full range works fine (tested on other data sets) but as soon as I
> change the sample start arbitrarily the error occurs again. Only
> changing the sample end (leaving the start at its first observation)
> does not lead to any error.
So that would seem to confirm my speculation... Actually while trying to
reproduce this behavior I found another bug (this time clearly not a
gretl bug) so I will prepare a new version that tries to squash them both.
Attached is a new version (1.1, without the help text so far, so not for
uploading to the server), it would be nice if you could test it.
(I worked around the subsampling issue by getting the starting
observation via the $t1 accessor. The other bug should also be fixed,
not assuming the presence of conditioning variables any more.)