[Gretl-devel] Re: time series cmds like corrgm in panel data
Am 04.11.2025 um 11:00 schrieb Sven Schreiber:
Hi all,
I remember a longer while ago that we had to block some filter
functions to spuriously work on panel data, because they were mixing
the units together. Now I wonder if something similar is still
happening with commands like corrgm?
And now I'm also seeing a complementing problem, where lrvar doesn't
work in a panel environment although it should (IMHO):
<hansl>
open grunfeld
# smpl 1 1 --unit
# setobs 1 1 --time-series
eval lrvar(mnormal(200)) # error; notice it doesn't work on the panel
dataset
</hansl>
I think lrvar should be allowed to work on a vector/matrix no matter
whether there's currently a panel dataset in place. And what's more, I
saw this behavior inside a function which didn't get any series context
as arguments - only the caller lived in a panel.
If I "fake" a time-series dataset with the commented lines, then it
works. Funny enough, I can use the "setobs 1 1 --time-series" line even
inside my purely matrix-oriented function, so here's my interim workaround:
...
# workaround for lrvar blocking problem;
# "fake" a time series context
if $datatype # dataset may not always be in place
setobs 1 1 --time-series
endif
...
This is with an Oct 9th snapshot.
cheers
sven