3:37 a.m.
There is an irritation in using the Forecast Function in Analysis in that
it treats Panel data as a continuum between adjacent panel units in
plotting by observation number sequence. So as illustrated below, there are
21 annual observations in Panel 1 ending in 2021. However the forecast
funtion treats this as as a observation continuum from 1:21 to 2:01(year
2001) and plots an irritaing line between1:21 and 2:01......and similarly
between all the adjacent 10 Panel Units. Is there anyway to suppress this
joining line, as the forecast graph plots beautiful CI shaded areas.
The workaround is to transfer the data to Excel -but that does not provided
a shaded 95% CI -which is clearly more elegan -only the ability to plot the
Upper and lower CI bounds as lines..
The Graph function on the other hand will show model fitted and actual
values by Panel Unit number, but no confidence intervals!
Brian
1:20 6.963190 6.860621 0.283066 6.301620 - 7.419622
1*:21 6.729824 6.452257 0.284134 5.891148 - 7.013367*
2:01 6.952729
2:02 5.905362
Perhaps an option in the graphics GUI to plot the forecasts for each panel
by Panel numbered units might remove the problem.?
[image: Capture.JPG]
Brian J Revell
Professor Emeritus (Agricultural Economics)
Harper Adams University UK
Current Chair of Defra Economic ADvisory Panel
Former President Agricultural Economics Society
Tel: home 01952 728153 Mobile 07976 538712
Address: Orchard Croft, Vineyard Rd, Homer, Much Wenlock TF!3 6NG
Alt. Email bjrevell(a)harper-adams.ac.uk
8:49 a.m.
Am 28.02.2024 um 09:37 schrieb Brian Revell:
There is an irritation in using the Forecast Function in Analysis in
that it treats Panel data as a continuum between adjacent panel units
in plotting by observation number sequence. So as illustrated below,
there are 21 annual observations in Panel 1 ending in 2021. However
the forecast funtion treats this as as a observation continuum from
1:21 to 2:01(year 2001) and plots an irritaing line between1:21 and
2:01......and similarly between all the adjacent 10 Panel Units. Is
there anyway to suppress this joining line, as the forecast graph
plots beautiful CI shaded areas.
The workaround is to transfer the data to Excel -but that does not
provided a shaded 95% CI -which is clearly more elegan -only the
ability to plot the Upper and lower CI bounds as lines..
Hi Brian,
I tried to run a forecast on the grunfeld example panel dataset shipped
with gretl. I restricted the sample to leave out the last two time
periods, to save them for forecasting, and then estimated a trivial
dynamic/reduced-form fixed-effects regression with "invest" as the dep.
variable. (Let's ignore the Nickell bias issues for this example.) No
contemporaneous variables, but in the model window under the Analysis
menu the "Forecasts..." entry was still greyed out, so I couldn't
cross-check what you described.
So, it would be helpful if you could describe how exactly and with what
data you produced your forecasts.
thanks
sven
11:27 a.m.
Hi Sven
I am not sure
Brian J Revell
Professor Emeritus (Agricultural Economics)
Harper Adams University UK
Current Chair of Defra Economic ADvisory Panel
Former President Agricultural Economics Society
Tel: home 01952 728153 Mobile 07976 538712
Address: Orchard Croft, Vineyard Rd, Homer, Much Wenlock TF!3 6NG
Alt. Email bjrevell(a)harper-adams.ac.uk
On Wed, 28 Feb 2024 at 13:50, Sven Schreiber <sven.schreiber(a)fu-berlin.de>
wrote:
Am 28.02.2024 um 09:37 schrieb Brian Revell:
There is an irritation in using the Forecast Function in Analysis in that
it treats Panel data as a continuum between adjacent panel units in
plotting by observation number sequence. So as illustrated below, there are
21 annual observations in Panel 1 ending in 2021. However the forecast
funtion treats this as as a observation continuum from 1:21 to 2:01(year
2001) and plots an irritaing line between1:21 and 2:01......and similarly
between all the adjacent 10 Panel Units. Is there anyway to suppress this
joining line, as the forecast graph plots beautiful CI shaded areas.
The workaround is to transfer the data to Excel -but that does not
provided a shaded 95% CI -which is clearly more elegant -only the ability
to plot the Upper and lower CI bounds as lines..
Hi Brian,
I am running the forecast function via the Estimated Model output/
/Analysis/Forecast in the GUI. My aim is not to generate forecasts beyond
time period 2021 but simply to produce a prediction for actual Y values
(not mean) with associated 95% CI spread. This works perfectly (see data
section below). The graph however, as I said , seems to join the datapoints
from adjacent panel data units i.e .it will add a line between observation
1:21 and 2:01 etc. for the dependent variable l_RSE (the number of
returning adult salmon) The graph of the prediction, acutal and CI joins
inserts a line joining the final value of Y (ln RSE) in 1:21 to that of
2:01 , 2:21 to 3:01 et al.. . The graph is generated by selecting 0 for
the number of preforecast obsvns to graph and a tick in checkbox for show
fitted values for preforecast range
* l_RSE prediction std. error 95% interval1:01
*
1:01
1:02
1:03
1:04
1:05 7.184629
1:06 7.034388 6.894189 0.282773 6.335766 - 7.452613
1:07 6.697034 6.633725 0.282741 6.075366 - 7.192085
1:08 6.632002 6.691674 0.282374 6.134040 - 7.249307
1:09 6.411818 6.456549 0.284239 5.895231 - 7.017867
1:10 6.661855 6.971589 0.285004 6.408761 - 7.534417
1:11 6.635947 6.491402 0.283699 5.931152 - 7.051652
1:12 6.670766 6.814620 0.282740 6.256262 - 7.372978
1:13 6.526495 6.500384 0.282566 5.942371 - 7.058397
PANEL UNIT 1 -River No. 1
1:14 7.074117 7.339223 0.289105 6.768296 - 7.910149
1:15 7.193686 7.441791 0.285464 6.878053 - 8.005528
1:16 6.944087 7.205822 0.283033 6.646886 - 7.764759
1:17 6.601230 6.696775 0.283867 6.136192 - 7.257357
1:18 6.556778 6.177856 0.285445 5.614156 - 6.741556
1:19 7.309881 6.955872 0.283390 6.396231 - 7.515514
1:20 6.963190 6.860621 0.283066 6.301620 - 7.419622
1:*21 6.729824 *6.452257 0.284134 5.891148 - 7.013367
* 2:01 6.952729*
2:02 5.905362
2:03 7.029088 7.102234 0.295125 6.519420 - 7.685049
2:04 7.018402 7.091428 0.286250 6.526138 - 7.656717
2:05 6.964136 6.865752 0.286381 6.300206 - 7.431299
2:06 6.498282 6.909504 0.286639 6.343447 - 7.475561
2:07 7.304516 7.068966 0.289634 6.496995 - 7.640937
2:08 6.805723 6.882406 0.288406 6.312860 - 7.451953
2:09 6.725034 6.891609 0.286613 6.325603 - 7.457614
2:10 6.887553 7.073868 0.288169 6.504789 - 7.642946Panel
Unit 2 River No. 2
2:11 6.855409 6.920967 0.288442 6.351349 - 7.490584
2:12 6.927558 7.083822 0.286813 6.517421 - 7.650222
2:13 6.908755 6.889174 0.286116 6.324149 - 7.454199
2:14 7.604396 7.255976 0.287378 6.688459 - 7.823493
2:15 7.282074 7.247741 0.288574 6.677862 - 7.817620
2:16 7.522941 7.159051 0.286828 6.592620 - 7.725482
2:17 6.526495 6.772668 0.290510 6.198966 - 7.346371
2:18 6.891626 6.668582 0.288539 6.098773 - 7.238391
2:19 7.988543 7.155666 0.287025 6.588847 - 7.722484
2:20 6.556778 6.971224 0.293826 6.390974 - 7.551473
2:21 6.226537 6.513208 0.288322 5.943828 - 7.082588
3:01 5.476464
3:02 5.129899
3:03 6.016157 5.892951 0.292152 5.316006 - 6.469896
Hope that clarifies the issue.
Brian
______________________________________________________________________________
I tried to run a forecast on the grunfeld example panel dataset shipped
with gretl. I restricted the sample to leave out the last two time
periods,
to save them for forecasting, and then estimated a trivial
dynamic/reduced-form fixed-effects regression with "invest" as the dep.
variable. (Let's ignore the Nickell bias issues for this example.) No
contemporaneous variables, but in the model window under the Analysis menu
the "Forecasts..." entry was still greyed out, so I couldn't cross-check
what you described.
So, it would be helpful if you could describe how exactly and with what
data you produced your forecasts.
thanks
sven
_______________________________________________
Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it
To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it
Website:
https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
11:33 a.m.
Sven
PS -perhaps it might be possible (if I knew where the forecast data table
was stored, ) to delate the first Y value in each panel for the purpose of
inserting a blank ie discontinuity/break in what the graphics function
interprets as a continuous series instead of 10 time series of the same
same variable relating to different panel units..
Brian
Professor Emeritus (Agricultural Economics)
Harper Adams University UK
Current Chair of Defra Economic ADvisory Panel
Former President Agricultural Economics Society
Tel: home 01952 728153 Mobile 07976 538712
Address: Orchard Croft,
On Wed, 28 Feb 2024 at 13:50, Sven Schreiber <sven.schreiber(a)fu-berlin.de>
wrote:
Am 28.02.2024 um 09:37 schrieb Brian Revell:
There is an irritation in using the Forecast Function in Analysis in that
it treats Panel data as a continuum between adjacent panel units in
plotting by observation number sequence. So as illustrated below, there are
21 annual observations in Panel 1 ending in 2021. However the forecast
funtion treats this as as a observation continuum from 1:21 to 2:01(year
2001) and plots an irritaing line between1:21 and 2:01......and similarly
between all the adjacent 10 Panel Units. Is there anyway to suppress this
joining line, as the forecast graph plots beautiful CI shaded areas.
The workaround is to transfer the data to Excel -but that does not
provided a shaded 95% CI -which is clearly more elegan -only the ability to
plot the Upper and lower CI bounds as lines..
Hi Brian,
I tried to run a forecast on the grunfeld example panel dataset shipped
with gretl. I restricted the sample to leave out the last two time periods,
to save them for forecasting, and then estimated a trivial
dynamic/reduced-form fixed-effects regression with "invest" as the dep.
variable. (Let's ignore the Nickell bias issues for this example.) No
contemporaneous variables, but in the model window under the Analysis menu
the "Forecasts..." entry was still greyed out, so I couldn't cross-check
what you described.
So, it would be helpful if you could describe how exactly and with what
data you produced your forecasts.
thanks
sven
_______________________________________________
Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it
To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it
Website:
https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
1:05 p.m.
On Wed, Feb 28, 2024 at 8:50 AM Sven Schreiber
<sven.schreiber(a)fu-berlin.de> wrote:
Am 28.02.2024 um 09:37 schrieb Brian Revell:
There is an irritation in using the Forecast Function in Analysis in that it treats Panel
data as a continuum between adjacent panel units in plotting by observation number
sequence. So as illustrated below, there are 21 annual observations in Panel 1 ending in
2021. However the forecast funtion treats this as as a observation continuum from 1:21 to
2:01(year 2001) and plots an irritaing line between1:21 and 2:01......and similarly
between all the adjacent 10 Panel Units. Is there anyway to suppress this joining line, as
the forecast graph plots beautiful CI shaded areas.
The workaround is to transfer the data to Excel -but that does not provided a shaded 95%
CI -which is clearly more elegan -only the ability to plot the Upper and lower CI bounds
as lines..
Hi Brian,
I tried to run a forecast on the grunfeld example panel dataset shipped with gretl. I
restricted the sample to leave out the last two time periods, to save them for
forecasting, and then estimated a trivial dynamic/reduced-form fixed-effects regression
with "invest" as the dep. variable. (Let's ignore the Nickell bias issues
for this example.) No contemporaneous variables, but in the model window under the
Analysis menu the "Forecasts..." entry was still greyed out, so I couldn't
cross-check what you described.
So, it would be helpful if you could describe how exactly and with what data you produced
your forecasts.
There's a sort of "loophole" here in that you're offered the Forecast
option under the Analysis menu if you estimate via pooled OLS on panel
data -- and in that case you get what Brian's describing. It seems we
weren't really ready to offer GUI forecasts for panel data, but the
OLS case slipped through.
Allin
2:36 p.m.
Alin\ Sven
You are correct in that the forecasts are not available in the Panel data
FE and RE model options. But the Panel Data series can be modeled via
Pooled OLS with DVs.. It is a workaround to estimate a similarly specified
pooled data model to an FE panel data equivalent. The Pooled OLS model does
provide an Analysis/Forecast option.
Brian
On Wed, 28 Feb 2024 at 18:06, Cottrell, Allin <cottrell(a)wfu.edu> wrote:
On Wed, Feb 28, 2024 at 8:50 AM Sven Schreiber
<sven.schreiber(a)fu-berlin.de> wrote:
>
> Am 28.02.2024 um 09:37 schrieb Brian Revell:
>
> There is an irritation in using the Forecast Function in Analysis in
that it treats Panel data as a continuum between adjacent panel units in
plotting by observation number sequence. So as illustrated below, there are
21 annual observations in Panel 1 ending in 2021. However the forecast
funtion treats this as as a observation continuum from 1:21 to 2:01(year
2001) and plots an irritaing line between1:21 and 2:01......and similarly
between all the adjacent 10 Panel Units. Is there anyway to suppress this
joining line, as the forecast graph plots beautiful CI shaded areas.
> The workaround is to transfer the data to Excel -but that does not
provided a shaded 95% CI -which is clearly more elegan -only the ability to
plot the Upper and lower CI bounds as lines..
>
> Hi Brian,
>
> I tried to run a forecast on the grunfeld example panel dataset shipped
with gretl. I restricted the sample to leave out the last two time periods,
to save them for forecasting, and then estimated a trivial
dynamic/reduced-form fixed-effects regression with "invest" as the dep.
variable. (Let's ignore the Nickell bias issues for this example.) No
contemporaneous variables, but in the model window under the Analysis menu
the "Forecasts..." entry was still greyed out, so I couldn't cross-check
what you described.
>
> So, it would be helpful if you could describe how exactly and with what
data you produced your forecasts.
There's a sort of "loophole" here in that you're offered the Forecast
option under the Analysis menu if you estimate via pooled OLS on panel
data -- and in that case you get what Brian's describing. It seems we
weren't really ready to offer GUI forecasts for panel data, but the
OLS case slipped through.
Allin
_______________________________________________
Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it
To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it
Website:
https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
3:33 p.m.
So back to my original question. How to prevent The Forecast graphic
treating the Panel data series as a continuum between panels, and joining
panel units together.
Brian
On Wed, 28 Feb 2024, 18:06 Cottrell, Allin, <cottrell(a)wfu.edu> wrote:
On Wed, Feb 28, 2024 at 8:50 AM Sven Schreiber
<sven.schreiber(a)fu-berlin.de> wrote:
>
> Am 28.02.2024 um 09:37 schrieb Brian Revell:
>
> There is an irritation in using the Forecast Function in Analysis in
that it treats Panel data as a continuum between adjacent panel units in
plotting by observation number sequence. So as illustrated below, there are
21 annual observations in Panel 1 ending in 2021. However the forecast
funtion treats this as as a observation continuum from 1:21 to 2:01(year
2001) and plots an irritaing line between1:21 and 2:01......and similarly
between all the adjacent 10 Panel Units. Is there anyway to suppress this
joining line, as the forecast graph plots beautiful CI shaded areas.
> The workaround is to transfer the data to Excel -but that does not
provided a shaded 95% CI -which is clearly more elegan -only the ability to
plot the Upper and lower CI bounds as lines..
>
> Hi Brian,
>
> I tried to run a forecast on the grunfeld example panel dataset shipped
with gretl. I restricted the sample to leave out the last two time periods,
to save them for forecasting, and then estimated a trivial
dynamic/reduced-form fixed-effects regression with "invest" as the dep.
variable. (Let's ignore the Nickell bias issues for this example.) No
contemporaneous variables, but in the model window under the Analysis menu
the "Forecasts..." entry was still greyed out, so I couldn't cross-check
what you described.
>
> So, it would be helpful if you could describe how exactly and with what
data you produced your forecasts.
There's a sort of "loophole" here in that you're offered the Forecast
option under the Analysis menu if you estimate via pooled OLS on panel
data -- and in that case you get what Brian's describing. It seems we
weren't really ready to offer GUI forecasts for panel data, but the
OLS case slipped through.
Allin
_______________________________________________
Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it
To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it
Website:
https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
6:18 a.m.
Am 28.02.2024 um 21:33 schrieb Brian Revell:
So back to my original question. How to prevent The Forecast graphic
treating the Panel data series as a continuum between panels, and
joining panel units together.
Brian
OK, so from what I understand you're running a fixed-effects model by
using the explicit LSDV approach, in order to "trick" gretl's pooled OLS
routine to show you something.
Econometrically I'd say that this requires a reasonably large time
dimension; in a short-T panel you cannot consistently estimate the
unit-specific intercepts. It sounded as if for you T=21, so it might
possibly just be enough. I'm also assuming that you only have exogenous
regressors, especially no lagged dependent variables.
Technically I'm not aware of an easy way to obtain this plot, currently.
It is feasible, because you can save the static forecast values and its
standard errors, and then you could construct a plot of the various
units in the time series dimension. In the latest gretl version (2023c)
there is a new feature that allows to plot several bands in a single
plot, which would be needed here. (There have been some bug fixes in
that area since 2023c was released, not sure if those fixes would be
needed as well.) But it would involve some scripting, would admittedly
be a little fiddly and not completely trivial.
I challenge our practical panel guy Artur to present a solution, hehe!
cheers
sven
7:22 a.m.
Hi Sven
understood. The workaround will have to suffice. I insert vertical arrow
separating the Panels on the horiz axis that obscure the offending joining
lines between the Panels
Brian
PS what is the issue with inclusion of a lagged dependent variable -other
than reduction in the sample length?
On Thu, 29 Feb 2024 at 11:18, Sven Schreiber <sven.schreiber(a)fu-berlin.de>
wrote:
Am 28.02.2024 um 21:33 schrieb Brian Revell:
So back to my original question. How to prevent The Forecast graphic
treating the Panel data series as a continuum between panels, and joining
panel units together.
Brian
OK, so from what I understand you're running a fixed-effects model by
using the explicit LSDV approach, in order to "trick" gretl's pooled OLS
routine to show you something.
Econometrically I'd say that this requires a reasonably large time
dimension; in a short-T panel you cannot consistently estimate the
unit-specific intercepts. It sounded as if for you T=21, so it might
possibly just be enough. I'm also assuming that you only have exogenous
regressors, especially no lagged dependent variables.
Technically I'm not aware of an easy way to obtain this plot, currently.
It is feasible, because you can save the static forecast values and its
standard errors, and then you could construct a plot of the various units
in the time series dimension. In the latest gretl version (2023c) there is
a new feature that allows to plot several bands in a single plot, which
would be needed here. (There have been some bug fixes in that area since
2023c was released, not sure if those fixes would be needed as well.) But
it would involve some scripting, would admittedly be a little fiddly and
not completely trivial.
I challenge our practical panel guy Artur to present a solution, hehe!
cheers
sven
_______________________________________________
Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it
To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it
Website:
https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
9 a.m.
Am 29.02.2024 um 13:22 schrieb Brian Revell:
Hi Sven
understood. The workaround will have to suffice. I insert vertical
arrow separating the Panels on the horiz axis that obscure the
offending joining lines between the Panels
Brian
PS what is the issue with inclusion of a lagged dependent variable
-other than reduction in the sample length?
Well, it's the classic problem of the Nickell bias
(https://www.jstor.org/stable/1911408). That's why for short-T panels
the Arellano-Bond type dynamic panel methods were developed.
cheers
sven
191
days inactive
192
days old
9 comments
3 participants
participants (3)
-
Brian Revell -
Cottrell, Allin -
Sven Schreiber