4:08 p.m.
Hello!
Please forgive me if the following questions are a little bit "stupid",
but as I mentioned, I'm a newbie...
My aim is to build up a little macro-model like the "klein.inp", using
ols estimations, and to do forecasts with it. Some questions arise
hereby:
1. In the command reference, page 51 (A4-Version) about using
systems it is mentioned that identities are needed if FIML estimation is
used. But also with ols or other estimations the system has to know how
to create the forecast values of derived variables which are not
estimated inside the system. Did I miss anything in the reference? And I
guess that also with ols-estimation I have to declare the endogenous
variables?
2. Which command is necessary if I want the system to do forecasts
for the system as a whole? In the graphic interface there is the
possibility to select only one estimated variable, but as you know the
clue of using systems is all variables are estimated simultaneous in
mutual dependency...
Thank you for your help,
regards, Jan Limbers
Prognos AG - Basel-Stadt HRA CH-270.3.003.262-6
Geschaftsfuhrer: Christian Bollhoff
Prasident des Verwaltungsrates: Gunter Blickle
2:21 p.m.
Hi Jan,
I don't have any experience using gretl's "system" (the
simultaneous-equation, structural thing, as opposed to VARs and VECMs),
so I can only offer some general comments.
First, it seems to me that this is an area that could be better
documented; in the gretl guide there is no chapter about it AFAICS. And
the command reference is as condensed as it probably should be, but then
of course is no substitute for the missing chapter in the guide. Maybe
that should be added as a minor bug to the bug tracker.
On to your questions:
1. In the command reference, page 51 (A4-Version) about using systems
it is mentioned that identities are needed if FIML estimation is
used. But also with ols or other estimations the system has to
know how to create the forecast values of derived variables which
are not estimated inside the system. Did I miss anything in the
reference? And I guess that also with ols-estimation I have to
declare the endogenous variables?
Identitites as such need not be estimated, so for other estimators I'd
say it's really not necessary. Forecasting is the step after estimation
and is a different issue.
With respect to endogenous variables and OLS: with OLS it's really only
the left-hand variables that can be / should be endogenous. If you have
right-hand variables that are known to be endogenous, OLS will be
inconsistent, that's why you have tsls, 3sls, etc.
2. Which command is necessary if I want the system to do
forecasts
for the system as a whole? In the graphic interface there is the
possibility to select only one estimated variable, but as you know
the clue of using systems is all variables are estimated
simultaneous in mutual dependency…
That I don't know at all. Maybe it's really not implemented yet in
gretl? In principle, being only able to pick a single variable for the
forecast graph does not necessarily mean that no system forecast is done
in the background. But I somehow doubt that that's what's really happening.
But as I said, I don't have experience with that area in gretl, I'm only
speculating. Anybody else know the answer?
good luck,
sven
10:06 p.m.
New subject: Doing forecasts with system equation?
Hi!
One last try before I go back to Eviews :-)
Does anyone - maybe the creators of this gretl feature - know if one can do forecasts with
the system equation tool? Defining macro-models with equations and identities for example
and estimate the variables in mutual dependency? And by chance, has anyone done this
before and can show me the way a little bit?
regards, Jan
________________________________
Von: gretl-users-bounces(a)lists.wfu.edu im Auftrag von Sven Schreiber
Gesendet: So 09.11.2008 15:21
An: Gretl list
Betreff: Re: [Gretl-users] Newbie questions about using systems in Gretl
Hi Jan,
I don't have any experience using gretl's "system" (the
simultaneous-equation, structural thing, as opposed to VARs and VECMs),
so I can only offer some general comments.
First, it seems to me that this is an area that could be better
documented; in the gretl guide there is no chapter about it AFAICS. And
the command reference is as condensed as it probably should be, but then
of course is no substitute for the missing chapter in the guide. Maybe
that should be added as a minor bug to the bug tracker.
On to your questions:
1. In the command reference, page 51 (A4-Version) about using systems
it is mentioned that identities are needed if FIML estimation is
used. But also with ols or other estimations the system has to
know how to create the forecast values of derived variables which
are not estimated inside the system. Did I miss anything in the
reference? And I guess that also with ols-estimation I have to
declare the endogenous variables?
Identitites as such need not be estimated, so for other estimators I'd
say it's really not necessary. Forecasting is the step after estimation
and is a different issue.
With respect to endogenous variables and OLS: with OLS it's really only
the left-hand variables that can be / should be endogenous. If you have
right-hand variables that are known to be endogenous, OLS will be
inconsistent, that's why you have tsls, 3sls, etc.
2. Which command is necessary if I want the system to do
forecasts
for the system as a whole? In the graphic interface there is the
possibility to select only one estimated variable, but as you know
the clue of using systems is all variables are estimated
simultaneous in mutual dependency...
That I don't know at all. Maybe it's really not implemented yet in
gretl? In principle, being only able to pick a single variable for the
forecast graph does not necessarily mean that no system forecast is done
in the background. But I somehow doubt that that's what's really happening.
But as I said, I don't have experience with that area in gretl, I'm only
speculating. Anybody else know the answer?
good luck,
sven
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
10:24 p.m.
New subject: Doing forecasts with system equation?
Hi!
One last try before I go back to Eviews :-)
Does anyone - maybe the creators of this gretl feature - know if one can do forecasts with
the system equation tool? Defining macro-models with equations and identities for example
and estimate the forecast variables in mutual dependency? And by chance, has anyone done
this before and can show me the way a little bit?
regards, Jan
________________________________
Von: gretl-users-bounces(a)lists.wfu.edu im Auftrag von Sven Schreiber
Gesendet: So 09.11.2008 15:21
An: Gretl list
Betreff: Re: [Gretl-users] Newbie questions about using systems in Gretl
Hi Jan,
I don't have any experience using gretl's "system" (the
simultaneous-equation, structural thing, as opposed to VARs and VECMs),
so I can only offer some general comments.
First, it seems to me that this is an area that could be better
documented; in the gretl guide there is no chapter about it AFAICS. And
the command reference is as condensed as it probably should be, but then
of course is no substitute for the missing chapter in the guide. Maybe
that should be added as a minor bug to the bug tracker.
On to your questions:
1. In the command reference, page 51 (A4-Version) about using systems
it is mentioned that identities are needed if FIML estimation is
used. But also with ols or other estimations the system has to
know how to create the forecast values of derived variables which
are not estimated inside the system. Did I miss anything in the
reference? And I guess that also with ols-estimation I have to
declare the endogenous variables?
Identitites as such need not be estimated, so for other estimators I'd
say it's really not necessary. Forecasting is the step after estimation
and is a different issue.
With respect to endogenous variables and OLS: with OLS it's really only
the left-hand variables that can be / should be endogenous. If you have
right-hand variables that are known to be endogenous, OLS will be
inconsistent, that's why you have tsls, 3sls, etc.
2. Which command is necessary if I want the system to do
forecasts
for the system as a whole? In the graphic interface there is the
possibility to select only one estimated variable, but as you know
the clue of using systems is all variables are estimated
simultaneous in mutual dependency...
That I don't know at all. Maybe it's really not implemented yet in
gretl? In principle, being only able to pick a single variable for the
forecast graph does not necessarily mean that no system forecast is done
in the background. But I somehow doubt that that's what's really happening.
But as I said, I don't have experience with that area in gretl, I'm only
speculating. Anybody else know the answer?
good luck,
sven
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
Prognos AG - Basel-Stadt HRA CH-270.3.003.262-6
Geschäftsführer: Christian Böllhoff
Präsident des Verwaltungsrates: Gunter Blickle
11 p.m.
New subject: Doing forecasts with system equation?
On Mon, 10 Nov 2008, jan.limbers(a)prognos.com wrote:
Hi!
One last try before I go back to Eviews :-)
Does anyone - maybe the creators of this gretl feature - know if one can do forecasts
with the
system equation tool? Defining macro-models with equations and identities for example and
estimate
the forecast variables in mutual dependency? And by chance, has anyone done this before
and can
show me the way a little bit?
regards, Jan
I apologise for the scarcity of detail, but the following script should
exemplify what I think you need: it's an example on the Klein model, with
the forecast obtained by filling the exogenous variables with the last
observed value.
<script>
set echo off
set messages off
open klein.gdt
genr W = Wp + Wg
genr A = t - 1931
genr KL = K(-1)
rename I Inv
rename T Tax
Klein <- system
equation C 0 P P(-1) W
equation Inv 0 P P(-1) KL
equation Wp 0 X X(-1) A
identity X = C + Inv + G
identity P = X - Tax - Wp
identity K = KL + Inv
identity W = Wp + Wg
endog C Inv Wp X P K W
end system
estimate "Klein" method=ols
dataset addobs 1
list ALL = dataset
loop foreach i ALL
$i[1942] = $i[1941]
end loop
smpl 1920 1941
fcasterr 1920 1942 C --static --quiet
series Chat = $fcast
fcasterr 1920 1942 W --static --quiet
series What = $fcast
</script>
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
2:03 p.m.
New subject: Doing forecasts with system equation?
As always, Jack's example scripts are very helpful and enlightening. But
am I right in interpreting this as a way of saying "you're right, the
feature is not implemented"?
I need to think about it some more (when I have the time), but this may
be a good area where user-defined function packages (GUI) could be used.
(I mean, of course they can always be used, but here it may be really
sensible to use them.)
cheers,
sven
Riccardo (Jack) Lucchetti schrieb:
On Mon, 10 Nov 2008, jan.limbers(a)prognos.com wrote:
> Hi!
>
> One last try before I go back to Eviews :-)
>
> Does anyone - maybe the creators of this gretl feature - know if one
> can do forecasts with the
> system equation tool? Defining macro-models with equations and
> identities for example and estimate
> the forecast variables in mutual dependency? And by chance, has anyone
> done this before and can
> show me the way a little bit?
>
> regards, Jan
>
I apologise for the scarcity of detail, but the following script should
exemplify what I think you need: it's an example on the Klein model,
with the forecast obtained by filling the exogenous variables with the
last observed value.
<script>
set echo off
set messages off
open klein.gdt
genr W = Wp + Wg
genr A = t - 1931
genr KL = K(-1)
rename I Inv
rename T Tax
Klein <- system
equation C 0 P P(-1) W
equation Inv 0 P P(-1) KL
equation Wp 0 X X(-1) A
identity X = C + Inv + G
identity P = X - Tax - Wp
identity K = KL + Inv
identity W = Wp + Wg
endog C Inv Wp X P K W
end system
estimate "Klein" method=ols
dataset addobs 1
list ALL = dataset
loop foreach i ALL
$i[1942] = $i[1941]
end loop
smpl 1920 1941
fcasterr 1920 1942 C --static --quiet
series Chat = $fcast
fcasterr 1920 1942 W --static --quiet
series What = $fcast
</script>
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
------------------------------------------------------------------------
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
2:58 p.m.
New subject: Doing forecasts with system equation?
On Mon, 10 Nov 2008 jan.limbers(a)prognos.com wrote:
One last try before I go back to Eviews :-)
Does anyone - maybe the creators of this gretl feature - know
if one can do forecasts with the system equation tool? Defining
macro-models with equations and identities for example and
estimate the variables in mutual dependency? And by chance, has
anyone done this before and can show me the way a little bit?
Jack has given one example; I'll add another below. (Yes, I agree
with Sven that this is under-documented at present.) But first
let me say that when you apply the "fcast" command to a
simultaneous equations system you do get "proper" system
forecasts. From the help for fcast:
fcast [ startobs endobs ] [ varname ]
"If the last model is a system of equations, varname has [the]
effect [of] selecting a particular endogenous variable for
forecasting (the default being to produce forecasts for all the
endogenous variables). In the system case, or if varname is not
given, the forecast values can be retrieved using the accessor
$fcast, and the standard errors, if available, via $fcerr."
That said, here's an example of using Klein model 1 to produce a
counter-factual forecast for the effects of an increase in
government wage payments from 1933:
<script>
open klein.gdt
genr W = Wp + Wg
genr A = t - 1931
list endo = C I Wp P W X K
Klein1 <- system
equation C 0 P P(-1) W
equation I 0 P P(-1) K(-1)
equation Wp 0 X X(-1) A
identity P = X - T - Wp
identity W = Wp + Wg
identity X = C + I + G
identity K = K(-1) + I
endog endo
end system
# Baseline estimate and forecast
estimate Klein1 method=fiml --quiet
fcast 1932 1939 --dyn --quiet
matrix F0 = $fcast
# Increase government wage payments from 1933
series Wgsave = Wg
genr Wg = Wgsave + ((t>1932)? 0.5 : 0)
# print Wgsave Wg -o
# Generate counter-factual forecast
fcast 1932 1939 --dyn --quiet
matrix F1 = $fcast
matrix Fdiff = F1 - F0
colnames(Fdiff, endo)
printf "\nEstimated effects, over 1932-39, of a "
printf "sustained increase in"
printf "\n Wg of 0.5 units, starting in 1933\n\n"
printf "%7.3f", Fdiff
</script>
Allin Cottrell
3:12 p.m.
New subject: AW: Doing forecasts with system equation?
Thank you Allin and Ricardo, I will stay close to Gretl, ditch EViews and learn more about
Gretl's scripting language...
Jan
-----Ursprüngliche Nachricht-----
Von: gretl-users-bounces(a)lists.wfu.edu [mailto:gretl-users-bounces@lists.wfu.edu] Im
Auftrag von Allin Cottrell
Gesendet: Dienstag, 11. November 2008 15:58
An: Gretl list
Betreff: Re: [Gretl-users] Doing forecasts with system equation?
On Mon, 10 Nov 2008 jan.limbers(a)prognos.com wrote:
One last try before I go back to Eviews :-)
Does anyone - maybe the creators of this gretl feature - know
if one can do forecasts with the system equation tool? Defining
macro-models with equations and identities for example and
estimate the variables in mutual dependency? And by chance, has
anyone done this before and can show me the way a little bit?
Jack has given one example; I'll add another below. (Yes, I agree
with Sven that this is under-documented at present.) But first
let me say that when you apply the "fcast" command to a
simultaneous equations system you do get "proper" system
forecasts. From the help for fcast:
fcast [ startobs endobs ] [ varname ]
"If the last model is a system of equations, varname has [the]
effect [of] selecting a particular endogenous variable for
forecasting (the default being to produce forecasts for all the
endogenous variables). In the system case, or if varname is not
given, the forecast values can be retrieved using the accessor
$fcast, and the standard errors, if available, via $fcerr."
That said, here's an example of using Klein model 1 to produce a
counter-factual forecast for the effects of an increase in
government wage payments from 1933:
<script>
open klein.gdt
genr W = Wp + Wg
genr A = t - 1931
list endo = C I Wp P W X K
Klein1 <- system
equation C 0 P P(-1) W
equation I 0 P P(-1) K(-1)
equation Wp 0 X X(-1) A
identity P = X - T - Wp
identity W = Wp + Wg
identity X = C + I + G
identity K = K(-1) + I
endog endo
end system
# Baseline estimate and forecast
estimate Klein1 method=fiml --quiet
fcast 1932 1939 --dyn --quiet
matrix F0 = $fcast
# Increase government wage payments from 1933
series Wgsave = Wg
genr Wg = Wgsave + ((t>1932)? 0.5 : 0)
# print Wgsave Wg -o
# Generate counter-factual forecast
fcast 1932 1939 --dyn --quiet
matrix F1 = $fcast
matrix Fdiff = F1 - F0
colnames(Fdiff, endo)
printf "\nEstimated effects, over 1932-39, of a "
printf "sustained increase in"
printf "\n Wg of 0.5 units, starting in 1933\n\n"
printf "%7.3f", Fdiff
</script>
Allin Cottrell
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
Prognos AG - Basel-Stadt HRA CH-270.3.003.262-6
Geschäftsführer: Christian Böllhoff
Präsident des Verwaltungsrates: Gunter Blickle
5649
days inactive
5653
days old
7 comments
4 participants
participants (4)
-
Allin Cottrell -
jan.limbers＠prognos.com -
Riccardo (Jack) Lucchetti -
Sven Schreiber