4:25 a.m.
Hello All,
I have time-series data with a large number of samples (anywhere from 180 to 360 daily
samples). I would like to use GRETL to automatically pick a ARIMA model to forecast values
for as many as 30-60 days into future. I would like to also model for seasonality in the
data.
I have been playing with gretl tool to explore my data with various ARMIA models - like
ARIMA (1,0,1), ARIMA (1,1,1) and it has been a great experience so far.
Now, here is my issue. I would like to write a library to basically take time series data
as input, try various ARIMA models and pick the most appropriate one
"automatically" without human input or intervention. I understand this is fairly
novice way to looking at the forecasting, but my application's use-case is simple
enough to do it this way.
Now, has anyone tried this using GRETL where multiple models are tried and results (fstats
output) are compared to pick one over the other? If so is there a GRETL script out there
that can be shared?
Thanks in advance,
Jeevan
12:06 a.m.
On Mon, 26 Sep 2011, Jeevan Tambuluri wrote:
I have time-series data with a large number of samples (anywhere
from 180 to 360 daily samples). I would like to use GRETL to
automatically pick a ARIMA model to forecast values for as many as
30-60 days into future. I would like to also model for seasonality
in the data.
I have been playing with gretl tool to explore my data with
various ARMIA models - like ARIMA (1,0,1), ARIMA (1,1,1) and it
has been a great experience so far.
Now, here is my issue. I would like to write a library to
basically take time series data as input, try various ARIMA models
and pick the most appropriate one "automatically" without human
input or intervention. I understand this is fairly novice way to
looking at the forecasting, but my application's use-case is
simple enough to do it this way.
Now, has anyone tried this using GRETL where multiple models are
tried and results (fstats output) are compared to pick one over
the other? If so is there a GRETL script out there that can be
shared?
I'm not aware of any such script, although that doesn't mean that
such a script doesn't exist somewhere.
However, my understanding is that TRAMO -- which is available in a
form compatible with gretl, see for example
http://gretl.sourceforge.net/win32/ -- does this sort of thing
(automatic ARIMA model selection). But (I may be wrong) TRAMO
probably does not handle intra-day "seasonality"; I think it's
mostly oriented to monthly or quarterly data.
Maybe Ignacio Diaz-Emparanza, our resident expert on ARIMA models,
could comment?
Allin Cottrell
5:42 a.m.
R has such a function (called auto.arima in forecast package):
auto.arima Fit best ARIMA model to univariate time series
Rem
On 2011.09.28 03:06, Allin Cottrell wrote:
On Mon, 26 Sep 2011, Jeevan Tambuluri wrote:
> I have time-series data with a large number of samples (anywhere
> from 180 to 360 daily samples). I would like to use GRETL to
> automatically pick a ARIMA model to forecast values for as many as
> 30-60 days into future. I would like to also model for seasonality
> in the data.
> I have been playing with gretl tool to explore my data with
> various ARMIA models - like ARIMA (1,0,1), ARIMA (1,1,1) and it
> has been a great experience so far.
>
> Now, here is my issue. I would like to write a library to
> basically take time series data as input, try various ARIMA models
> and pick the most appropriate one "automatically" without human
> input or intervention. I understand this is fairly novice way to
> looking at the forecasting, but my application's use-case is
> simple enough to do it this way.
>
> Now, has anyone tried this using GRETL where multiple models are
> tried and results (fstats output) are compared to pick one over
> the other? If so is there a GRETL script out there that can be
> shared?
I'm not aware of any such script, although that doesn't mean that
such a script doesn't exist somewhere.
However, my understanding is that TRAMO -- which is available in a
form compatible with gretl, see for example
http://gretl.sourceforge.net/win32/ -- does this sort of thing
(automatic ARIMA model selection). But (I may be wrong) TRAMO
probably does not handle intra-day "seasonality"; I think it's
mostly oriented to monthly or quarterly data.
Maybe Ignacio Diaz-Emparanza, our resident expert on ARIMA models,
could comment?
Allin Cottrell
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
11:17 a.m.
El 28/09/11 02:06, Allin Cottrell escribió:
>
> Now, here is my issue. I would like to write a library to
> basically take time series data as input, try various ARIMA models
> and pick the most appropriate one "automatically" without human
> input or intervention. I understand this is fairly novice way to
> looking at the forecasting, but my application's use-case is
> simple enough to do it this way.
>
> Now, has anyone tried this using GRETL where multiple models are
> tried and results (fstats output) are compared to pick one over
> the other? If so is there a GRETL script out there that can be
> shared?
I'm not aware of any such script, although that doesn't mean that
such a script doesn't exist somewhere.
However, my understanding is that TRAMO -- which is available in a
form compatible with gretl, see for example
http://gretl.sourceforge.net/win32/ -- does this sort of thing
(automatic ARIMA model selection). But (I may be wrong) TRAMO
probably does not handle intra-day "seasonality"; I think it's
mostly oriented to monthly or quarterly data.
Maybe Ignacio Diaz-Emparanza, our resident expert on ARIMA models,
could comment?
Allin Cottrell
I have worked on replicating the automatic modelling method of TRAMO
natively in gretl(hansl) and also on the procedure of Mélard and
Pasteels [International journal of Forecasting 16 (2000) 497-508] but
they are not mature enough for publishing yet.
As already mentioned by Remigius, in R you may use the package
'forecast'. And remember that most of the R code may be run from inside
a hansl script. In this case this could be: (fist of all you have to
install.packages("forecast") from inside R)
<hansl>
#This function calls to the function auto.arima()
#in the library "forecast" of R, to automatically identify
#an arima model for the indicated series.
#
function matrix Rautoarima(series myseries, series *uhat[null])
catch smpl myseries --contiguous
if $error
print "Error: May be NAs inside the series"
endif
sx = argname(myseries)
##### Begining of R code
foreign language=R --send-data --quiet
library(forecast)
y <- gretldata[, "myseries"]
arimafit <- auto.arima(as.ts(y))
summary(arimafit)
orders <- as.matrix(arimafit$arma)
mbic<-as.matrix(arimafit$bic)
uhat <- as.ts(arimafit$residuals)
gretl.export(orders)
gretl.export(uhat)
gretl.export(mbic)
end foreign
###### End of R code
outfile --write null
append @dotdir/uhat.csv
orders = mread("(a)dotdir/orders.mat")
mmbic = mread("(a)dotdir/mbic.mat")
scalar p = orders[1]
scalar d = orders[6]
scalar q = orders[2]
scalar P = orders[3]
scalar D = orders[7]
scalar Q = orders[4]
matrix morder = { p, d, q, P, D, Q, mmbic }
outfile --close
print "+++++++++++++++++++++++++++++++++++++++++++++"
printf "The model identified for the series %s is:\n", sx
printf "ARIMA(%1.0f, %1.0f, %1.0f)X(%1.0f, %1.0f, %1.0f) with
BIC=%f\n", p, d, q, P, D, Q, mmbic
print "+++++++++++++++++++++++++++++++++++++++++++++"
return morder
end function
# ------------ main -------------------------
open bjg.gdt
Y=diff(lg)
Y=sdiff(Y)
series ruhat
matrix X = Rautoarima(Y, &ruhat)
</hansl>
--
Ignacio Diaz-Emparanza
DEPARTAMENTO DE ECONOMÍA APLICADA III (ECONOMETRÍA Y ESTADÍSTICA)
UPV/EHU Avda. Lehendakari Aguirre, 83 | 48015 BILBAO
T.: +34 946013732 | F.: +34 946013754
www.ea3.ehu.es
11:48 a.m.
El 28/09/11 13:17, Ignacio Diaz-Emparanza escribió:
printf "ARIMA(%1.0f, %1.0f, %1.0f)X(%1.0f, %1.0f, %1.0f)
with
BIC=%f\n", p, d, q, P, D, Q, mmbic
print "+++++++++++++++++++++++++++++++++++++++++++++"
return morder
end function
# ------------ main -------------------------
open bjg.gdt
Y=diff(lg)
Y=sdiff(Y)
series ruhat
matrix X = Rautoarima(Y,&ruhat)
</hansl>
Sorry, the line beginning with BIC=% correspond to the end of the
previous one and the diff and sdiff commands are not needed, you may run
matrix X = Rautoarima(lg,&ruhat) and the procedure determines the
differences.
--
Ignacio Diaz-Emparanza
DEPARTAMENTO DE ECONOMÍA APLICADA III (ECONOMETRÍA Y ESTADÍSTICA)
UPV/EHU Avda. Lehendakari Aguirre, 83 | 48015 BILBAO
T.: +34 946013732 | F.: +34 946013754
www.ea3.ehu.es
12:41 a.m.
Thank you all for providing excellent pointers!
I will be trying out some of the techniques and will surely post anything I learn from
it.
Jeevan
-----Original Message-----
From: gretl-users-bounces(a)lists.wfu.edu [mailto:gretl-users-bounces@lists.wfu.edu] On
Behalf Of Ignacio Diaz-Emparanza
Sent: Wednesday, September 28, 2011 4:48 AM
To: Gretl list
Subject: Re: [Gretl-users] Help with time-series analysis using ARIMA models in GRETL
El 28/09/11 13:17, Ignacio Diaz-Emparanza escribió:
printf "ARIMA(%1.0f, %1.0f, %1.0f)X(%1.0f, %1.0f, %1.0f)
with
BIC=%f\n", p, d, q, P, D, Q, mmbic
print "+++++++++++++++++++++++++++++++++++++++++++++"
return morder
end function
# ------------ main -------------------------
open bjg.gdt
Y=diff(lg)
Y=sdiff(Y)
series ruhat
matrix X = Rautoarima(Y,&ruhat)
</hansl>
Sorry, the line beginning with BIC=% correspond to the end of the
previous one and the diff and sdiff commands are not needed, you may run
matrix X = Rautoarima(lg,&ruhat) and the procedure determines the
differences.
--
Ignacio Diaz-Emparanza
DEPARTAMENTO DE ECONOMÍA APLICADA III (ECONOMETRÍA Y ESTADÍSTICA)
UPV/EHU Avda. Lehendakari Aguirre, 83 | 48015 BILBAO
T.: +34 946013732 | F.: +34 946013754
www.ea3.ehu.es
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
7:14 a.m.
On Mon, 26 Sep 2011, Jeevan Tambuluri wrote:
Now, has anyone tried this using GRETL where multiple models are
tried
and results (fstats output) are compared to pick one over the other? If
so is there a GRETL script out there that can be shared?
The following simple script does something which I think is close to what
you want, the main difference being that it uses the BIC instead of a
forecast-based fit measure. Of course, it's pretty rought, but refining it
to your needs should be relatively easy.
<hansl>
set echo off
set messages off
open np.gdt
y = ldiff(gnpdefl)
scalar bestsofar = NA
scalar bestp = NA
scalar bestq = NA
scalar maxp = 3
scalar maxq = 3
loop i=0..maxp --quiet
loop j=0..maxq --quiet
arima i 0 j ; y --quiet
c = $bic
printf "Arma(%2d,%2d) : %g\n", i, j, c
if (c<bestsofar) || missing(bestsofar)
bestsofar = c
bestp = i
bestq = j
endif
endloop
endloop
arima bestp 0 bestq ; y
</hansl>
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
4841
days inactive
4843
days old
6 comments
5 participants
participants (5)
-
Allin Cottrell -
Ignacio Diaz-Emparanza -
Jeevan Tambuluri -
Remigijus Lapinskas -
Riccardo (Jack) Lucchetti