10 a.m.
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?
Example:
<hansl>
open grunfeld
corrgm invest
</hansl>
I'm getting a default output up to lag 23, while this dataset only has a
time dimension of 20. So this looks a bit suspicious.
What's happening there?
thanks
sven
1:44 p.m.
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
8:08 p.m.
On Tue, 4 Nov 2025, Sven Schreiber wrote:
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
eval lrvar(mnormal(200)) # error
</hansl>
I think lrvar should be allowed to work on a vector/matrix no matter
whether there's currently a panel dataset in place.
Yes, agreed. Given a vector argument we should assume the caller knows
what he's doing (in particular, that he's not passing a panel series
in the guise of a vector, since lrvar doesn't handle panel data), and
that's now the case in git.
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.
By design, a gretl function always gets a pointer to the caller's
dataset. Otherwise it would be impossible to create a series inside a
function unless it calls for a series or list argument. So that
observation is not really valid "more". We should ignore the character
of the dataset when given vector input, is all.
Allin
3:24 p.m.
On Tue, 4 Nov 2025, Sven Schreiber wrote:
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?
Example:
<hansl>
open grunfeld
corrgm invest
</hansl>
The corrgm command is not set up to handle panel data. In the GUI
the "Correlogram" menu item is inactive when a panel dataset is
loaded, but there was no block on invoking corrgm from the command
line. That's now fixed in git: corrgm flags an error if the dataset
is not time series. The same goes for xcorrgm, pergm and fractint.
Allin
4:07 p.m.
Am 04.11.2025 um 16:24 schrieb Allin Cottrell:
On Tue, 4 Nov 2025, Sven Schreiber wrote:
> 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? ...
The corrgm command is not set up to handle panel data. In the GUI the
"Correlogram" menu item is inactive when a panel dataset is loaded,
but there was no block on invoking corrgm from the command line.
That's now fixed in git: corrgm flags an error if the dataset is not
time series. The same goes for xcorrgm, pergm and fractint.
OK, good.
thanks
sven
146
days inactive
146
days old
4 comments
2 participants
participants (2)
-
Allin Cottrell -
Sven Schreiber