7:13 p.m.
Hi all,
I just observed the following unwanted behavior; this is with gretl
2024d, but probably not brand new:
- I have a short yearly dataset spanning 2013-2023. (Confidential,
cannot share it here.)
- There are 14 time series in it, and somewhere in the middle is a
binary dummy, although its "discrete" flag in the attributes is _not_
set. Don't know whether that matters.
- Via the menu I add another obs to the dataset, for the year 2024.
- All variables have missings for the new obs, as expected, _except_ the
dummy: it has a (spurious) value 1 for the new obs. This is wrong.
thanks
sven
12:35 p.m.
Am 19.12.2024 um 20:13 schrieb Sven Schreiber:
Hi all,
I just observed the following unwanted behavior; this is with gretl
2024d, but probably not brand new:
- I have a short yearly dataset spanning 2013-2023. (Confidential,
cannot share it here.)
- There are 14 time series in it, and somewhere in the middle is a
binary dummy, although its "discrete" flag in the attributes is _not_
set. Don't know whether that matters.
- Via the menu I add another obs to the dataset, for the year 2024.
- All variables have missings for the new obs, as expected, _except_
the dummy: it has a (spurious) value 1 for the new obs. This is wrong.
Hi and happy new year to all devels,
I tried to replicate this problem with a public dataset, but that wasn't
so easy. So I looked again at the dummy in my private dataset, and the
values are:
2013 0
2014 0
2015 1
2016 0
2017 0
2018 1
2019 0
2020 0
2021 1
2022 0
2023 0
So could it be that gretl is applying a heuristic that every third obs
of this dummy should be 1? Effectively treating it as some kind of
seasonality. But again, this is wrong for what this dummy represents. So
I'd suggest that seasonal extrapolation shouldn't be tried if the
dataset's $pd is ==1.
thanks
sven
5:40 p.m.
On 02/01/2025 13:35, Sven Schreiber wrote:
Am 19.12.2024 um 20:13 schrieb Sven Schreiber:
I tried to replicate this problem with a public dataset, but that
wasn't so easy.
Here's a script that replicates the problem:
<hansl>
nulldata 11
x = normal()
d = !(index % 3)
dataset addobs 1
print -o
<hansl>
Note that the dataset is not even declared as time-series. I guess we
should have to revisit the real_periodic_dummy() function in
lib/src/dataset.c.
One of the things I'd do (but wouldn't fix the problem you spotted) is
to change the code block
maybe_extend_dummies(dset, oldn);
if (dataset_is_time_series(dset)) {
maybe_extend_lags(dset, oldn, dset->n - 1);
}
at line 1121 by moving the call to "maybe_extend_dummies(dset, oldn)"
inside the if block: I don't think this should be done for a
cross-sectional dataset (maybe for panel datasets...).
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
9:43 p.m.
On Thu, Jan 2, 2025 at 12:40 PM Riccardo (Jack) Lucchetti
<p002264(a)staff.univpm.it> wrote:
On 02/01/2025 13:35, Sven Schreiber wrote:
> Am 19.12.2024 um 20:13 schrieb Sven Schreiber:
>
> I tried to replicate this problem with a public dataset, but that
> wasn't so easy.
Here's a script that replicates the problem:
<hansl>
nulldata 11
x = normal()
d = !(index % 3)
dataset addobs 1
print -o
<hansl>
Note that the dataset is not even declared as time-series. I guess we
should have to revisit the real_periodic_dummy() function in
lib/src/dataset.c.
One of the things I'd do (but wouldn't fix the problem you spotted) is
to change the code block
maybe_extend_dummies(dset, oldn);
if (dataset_is_time_series(dset)) {
maybe_extend_lags(dset, oldn, dset->n - 1);
}
at line 1121 by moving the call to "maybe_extend_dummies(dset, oldn)"
inside the if block: I don't think this should be done for a
cross-sectional dataset (maybe for panel datasets...).
Yes, agreed. That's now done in git.
Allin
11:16 a.m.
Am 02.01.2025 um 22:43 schrieb Cottrell, Allin:
On Thu, Jan 2, 2025 at 12:40 PM Riccardo (Jack) Lucchetti
<p002264(a)staff.univpm.it> wrote:
> One of the things I'd do (but wouldn't fix the problem you spotted) is
> to change the code block
>
>
> maybe_extend_dummies(dset, oldn);
> if (dataset_is_time_series(dset)) {
> maybe_extend_lags(dset, oldn, dset->n - 1);
> }
>
>
> at line 1121 by moving the call to "maybe_extend_dummies(dset, oldn)"
> inside the if block: I don't think this should be done for a
> cross-sectional dataset (maybe for panel datasets...).
Yes, agreed. That's now done in git.
Allin
With "that" you mean that it's only done for time series, right?
Referring to Jack's remark "wouldn't fix the problem you (= I)
spotted".
thanks
sven
1:48 p.m.
On Fri, Jan 3, 2025 at 6:16 AM Sven Schreiber
<sven.schreiber(a)fu-berlin.de> wrote:
Am 02.01.2025 um 22:43 schrieb Cottrell, Allin:
> On Thu, Jan 2, 2025 at 12:40 PM Riccardo (Jack) Lucchetti
> <p002264(a)staff.univpm.it> wrote:
>> One of the things I'd do (but wouldn't fix the problem you spotted) is
>> to change the code block
>>
>>
>> maybe_extend_dummies(dset, oldn);
>> if (dataset_is_time_series(dset)) {
>> maybe_extend_lags(dset, oldn, dset->n - 1);
>> }
>>
>>
>> at line 1121 by moving the call to "maybe_extend_dummies(dset, oldn)"
>> inside the if block: I don't think this should be done for a
>> cross-sectional dataset (maybe for panel datasets...).
> Yes, agreed. That's now done in git.
>
> Allin
With "that" you mean that it's only done for time series, right?
Referring to Jack's remark "wouldn't fix the problem you (= I)
spotted".
I mean that extending periodic dummies is not done unless the dataset
is time series, but in addition it's not done unless $pd > 1, so the
change would fix the problem you reported.
Allin
3:53 p.m.
Am 03.01.2025 um 14:48 schrieb Cottrell, Allin:
On Fri, Jan 3, 2025 at 6:16 AM Sven Schreiber
<sven.schreiber(a)fu-berlin.de> wrote:
> Am 02.01.2025 um 22:43 schrieb Cottrell, Allin:
>> On Thu, Jan 2, 2025 at 12:40 PM Riccardo (Jack) Lucchetti
>> <p002264(a)staff.univpm.it> wrote:
>>> at line 1121 by moving the call to "maybe_extend_dummies(dset,
oldn)"
>>> inside the if block: I don't think this should be done for a
>>> cross-sectional dataset (maybe for panel datasets...).
>> Yes, agreed. That's now done in git.
>>
>> Allin
> With "that" you mean that it's only done for time series, right?
> Referring to Jack's remark "wouldn't fix the problem you (= I)
spotted".
I mean that extending periodic dummies is not done unless the dataset
is time series, but in addition it's not done unless $pd > 1, so the
change would fix the problem you reported.
Alright, thanks!
-sven
28
days inactive
43
days old
6 comments
3 participants
participants (3)
-
Cottrell, Allin -
Riccardo (Jack) Lucchetti -
Sven Schreiber