1:23 p.m.
Hi,
I'm observing something strange working with the grunfeld.gdt sample
datafile in the GUI. This is with the latest Windows snapshot. The
dataset has units 1 to 10, and time periods 1935 through 1954, so T = 20
(fully balanced).
I go to Sample / Set range; it shows N=10, T=20, NT=200. OK.
I click on the "down" arrow next to the end period 1954, to leave out
the last year. The number jumps to 1953 alright, but now it says T=18
(two down). When I click on "up" to get back to 1954, then T=19 allegedly.
Next, different but related problem (having freshly reloaded the
dataset): in the sample range setter, I now click "up" next to the
_starting_ period 1935 to get rid of the first time period. It changes
to 1936, but the count T=20 doesn't change. And after clicking "OK" the
sample isn't changed.
Finally, in the console I type "smpl year > 1935 --restrict --time"
without having checked the required syntax, and gretl crashes on me. (I
now know it's not correct; instead "smpl 1936 1954 --time" works OK.)
thanks
sven
6:26 p.m.
On Thu, 27 Apr 2023, Sven Schreiber wrote:
Hi,
I'm observing something strange working with the grunfeld.gdt
sample datafile in the GUI. This is with the latest Windows
snapshot. The dataset has units 1 to 10, and time periods 1935
through 1954, so T = 20 (fully balanced).
I go to Sample / Set range; it shows N=10, T=20, NT=200. OK.
I click on the "down" arrow next to the end period 1954, to leave
out the last year. The number jumps to 1953 alright, but now it
says T=18 (two down). When I click on "up" to get back to 1954,
then T=19 allegedly.
Next, different but related problem (having freshly reloaded the
dataset): in the sample range setter, I now click "up" next to the
_starting_ period 1935 to get rid of the first time period. It
changes to 1936, but the count T=20 doesn't change. And after
clicking "OK" the sample isn't changed.
What you're seeing there is "off by one" stuff, due to an internal
confusion between the case where "panel time" is defined (years,
quarters or whatever, as in grunfeld.gdt) and the case where it's
just a matter of periods 1 to T. That's now fixed in git and
snapshots.
Finally, in the console I type "smpl year > 1935 --restrict
--time" without having checked the required syntax, and gretl
crashes on me. (I now know it's not correct; instead "smpl 1936
1954 --time" works OK.)
Insufficient error-checking in the relatively new panel sampling
code. That's now fixed in git; snapshots will follow.
Allin
2:58 a.m.
New subject: About adding observations in forecasts with an estimated AR model
I just found it when I estimated an AR model (via ARIMA) with sub-sample
observations to generate the out-of-sample forecasts in GUI. A window then
pops up to ask "Add observations."
With the reduced observations in the sub-sample, the forecasting in an
estimated AR (or ARMA) model seems unnecessary to add new observations.
Adding observations seems to only be required for only out of the current
whole sample.
I further test the same forecasts function in an OLS model with the same
lagged variable with the same sub-sample. It does not require me to add
observations.
I tested three versions of gretl - 2022c, 2023a, and 2023b-git (build date
of 2023-04-27) and found that the results were consistent across all three.
Is this strange or intended? As I can recall in earlier version, forecasts
in AR models with sub-sample need not to add new observations.
Thanks.
Yi-Nung Yang
7:23 a.m.
New subject: About adding observations in forecasts with an estimated AR model
Next time, please start a new thread for a new topic instead of replying
to another message. (even if you change the subject line, it's still
treated as a reply.)
Am 01.05.2023 um 08:58 schrieb yinung at Gmail:
I just found it when I estimated an AR model (via ARIMA) with
sub-sample observations to generate the out-of-sample forecasts in
GUI. A window then pops up to ask "Add observations."
I cannot confirm this in a quick test with a recent snapshot, estimating
an ARMA model with bjg.gdt dataset in a subsample. Please be more
specific what you did to provoke such behavior.
thanks
sven
9:29 a.m.
New subject: About adding observations in forecasts with an estimated AR model
Am 01.05.23 um 13:23 schrieb Sven Schreiber:
Next time, please start a new thread for a new topic instead of
replying
to another message. (even if you change the subject line, it's still
treated as a reply.)
Am 01.05.2023 um 08:58 schrieb yinung at Gmail:
> I just found it when I estimated an AR model (via ARIMA) with
> sub-sample observations to generate the out-of-sample forecasts in
> GUI. A window then pops up to ask "Add observations."
I cannot confirm this in a quick test with a recent snapshot, estimating
an ARMA model with bjg.gdt dataset in a subsample. Please be more
specific what you did to provoke such behavior.
I can also not confirm this difference between ARIMA and OLS using the
GUI with recent gretl under Ubuntu.
Artur
12:39 p.m.
New subject: About adding observations in forecasts with an estimated AR model
Thanks for quick responses. I re-testes it and found that "adding
observations" windows occurs if the estimated AR is without a constant. For
example,
open bjg.gdt
smpl ; -10
model1 <- arima 1 0 1; g --nc
Then, turn to GUI of model1 to do the forecasts.
ps: My apology, I will create a new thread next time
Yi-Nung
Artur T. <atecon(a)posteo.de> 於 2023年5月1日 週一 下午9:29寫道：
Am 01.05.23 um 13:23 schrieb Sven Schreiber:
> Next time, please start a new thread for a new topic instead of replying
> to another message. (even if you change the subject line, it's still
> treated as a reply.)
>
> Am 01.05.2023 um 08:58 schrieb yinung at Gmail:
>> I just found it when I estimated an AR model (via ARIMA) with
>> sub-sample observations to generate the out-of-sample forecasts in
>> GUI. A window then pops up to ask "Add observations."
>
> I cannot confirm this in a quick test with a recent snapshot, estimating
> an ARMA model with bjg.gdt dataset in a subsample. Please be more
> specific what you did to provoke such behavior.
I can also not confirm this difference between ARIMA and OLS using the
GUI with recent gretl under Ubuntu.
Artur
_______________________________________________
Gretl-devel mailing list -- gretl-devel(a)gretlml.univpm.it
To unsubscribe send an email to gretl-devel-leave(a)gretlml.univpm.it
Website:
https://gretlml.univpm.it/postorius/lists/gretl-devel.gretlml.univpm.it/
8:47 p.m.
New subject: About adding observations in forecasts with an estimated AR model
On Mon, May 1, 2023 at 12:39 PM yinung at Gmail <yinung.cycu(a)gmail.com> wrote:
Thanks for quick responses. I re-tested it and found that "adding observations"
windows occurs if the estimated AR is without a constant. For example,
open bjg.gdt
smpl ; -10
model1 <- arima 1 0 1; g --nc
Then, turn to GUI of model1 to do the forecasts.
Thanks for the report, Yi-Nung. I see what you mean. We'll try to fix
that as soon as we can.
Allin Cottrell
3:28 p.m.
New subject: About adding observations in forecasts with an estimated AR model
On Mon, May 1, 2023 at 8:47 PM Cottrell, Allin <cottrell(a)wfu.edu> wrote:
On Mon, May 1, 2023 at 12:39 PM yinung at Gmail <yinung.cycu(a)gmail.com> wrote:
>
> Thanks for quick responses. I re-tested it and found that "adding
observations" windows occurs if the estimated AR is without a constant. For example,
>
> open bjg.gdt
> smpl ; -10
> model1 <- arima 1 0 1; g --nc
>
> Then, turn to GUI of model1 to do the forecasts.
Thanks for the report, Yi-Nung. I see what you mean. We'll try to fix
that as soon as we can.
This is now fixed in git and snapshots. For the record, the issue was
that despite reserving 10 observations for forecasting, via
smpl ; -10
Yi-Nung found that after estimating (via the GUI) an ARIMA model
without a constant, gretl reported that there were no observations
available for forecasting.
The problem was specific to the GUI, and to the case where the model
had no regressors as such.
Allin
452
days inactive
457
days old
7 comments
5 participants
participants (5)
-
Allin Cottrell -
Artur T. -
Cottrell, Allin
-
Sven Schreiber -
yinung＠Gmail