12:51 a.m.
All,
Please forgive my simple questions. For ARIMA modelling, using the ACF
and PACF to determine the pdq parameters requires subjective judgment
based on experience. Answers are seldom clear-cut. As the subject
header says, can a somewhat more objective path be found in quickly
running through ARIMA models, varying parameters, acquiring AIC, BIC,
HQC, and comparing? I don't think the time-series-->VAR lag selection
option gets at precisely the same thing. Thank you TMN
5:17 a.m.
Of course it is possible. Please consider the following texample with
artificial data:
<hansl>
nulldata 500
setobs 1 1 --special-time-series
genr time
set seed 20140413 # For reproduction
series wn = randgen(n,0,1)
# Stationary ARMA(2,2) model
series z = 4.5
z = 1 + 0.25*z(-1) + 0.55*z(-2) + wn + 0.5*wn(-1) + 0.5*wn(-2)
# Visual eyecandy, can be omitted
# gnuplot z time --with-lines --output=display
#corrgm z 24 --plot=display
matrix AICs = zeros(6,6)
matrix BICs = zeros(6,6)
loop i=1..6 --quiet
loop j=1..6 --quiet
arima $i 0 $j ; z
AICs[$i,$j]=$aic
BICs[$i,$j]=$bic
endloop
endloop
mwrite(AICs,"AICs.csv")
mwrite(BICs,"BICs.csv")
</hansl>
What it does: varies the $p$ and $q$ parameters of ARIMA(p,d,q) estimation
and generates two matrices with the criteria. Just delete the first line
from the .csv files and use your favourite means of visualisation (e.g.
Excel2LaTeX) for comparison. A beautiful way of doing that via a heatmap is
described here:
http://texblog.org/2013/06/13/latex-heatmap-using-tabular/You may want
to transform the table for easier interpretation (e.g.
subtract the second smallest value from all and take logs if the order of
variance is too high, like 371.1 371.2 371.7 370.8 410.4 456.9).
Please see the final result on page 9 of this document:
https://www.dropbox.com/s/g7vf0per3pp45r3/avr-sem.pdf (it is in Russian,
except for the Universal Language of Mathematics).
You may also want to perform Poskitt—Tremayne test in order to determine
whether to include a model or not.
Is that what you wanted?
--
Yours sincerely, | С уважением,
Andreï V. Kostyrka. | Андрей Викторович Костырка.
http://kostyrka.ru, http://kostyrka.ru/blog
2014-04-26 4:51 GMT+04:00 Tim Nall <tnall.ling(a)gmail.com>:
All,
Please forgive my simple questions. For ARIMA modelling, using the ACF
and PACF to determine the pdq parameters requires subjective judgment
based on experience. Answers are seldom clear-cut. As the subject
header says, can a somewhat more objective path be found in quickly
running through ARIMA models, varying parameters, acquiring AIC, BIC,
HQC, and comparing? I don't think the time-series-->VAR lag selection
option gets at precisely the same thing. Thank you TMN
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
9:09 a.m.
On Sat, 26 Apr 2014, Andreï | Андрей Викторович wrote:
# Stationary ARMA(2,2) model
series z = 4.5
z = 1 + 0.25*z(-1) + 0.55*z(-2) + wn + 0.5*wn(-1) + 0.5*wn(-2)
An alternative way to generate an arma process is to use the filter()
function. This leads to a formulation that's probably less intuitive, but
more efficient and compact:
<hansl>
phi = {0.25, 0.55}
theta = {1, 0.5, 0.5}
z = filter(wn, theta, phi)
</hansl>
-------------------------------------------------------
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
-------------------------------------------------------
6:56 a.m.
You can use armax_auto (see time series menu in Gretl GUI). The function has a drawback:
the cycle crushes on encountering a unit root in MA part of the model Also you can use
auto.arima{forecast} in R My experience tells it would be better to determine the order of
integration (use unit root tests) beforehand and use, say "d=1" in auto.arima
(type ?auto.arima in R console)
--- Оригінальне повідомлення ---
Від кого: "Tim Nall" < tnall.ling(a)gmail.com >
Дата: 26 квітня 2014, 03:53:32
All,
Please forgive my simple questions. For ARIMA modelling, using the ACF
and PACF to determine the pdq parameters requires subjective judgment
based on experience. Answers are seldom clear-cut. As the subject
header says, can a somewhat more objective path be found in quickly
running through ARIMA models, varying parameters, acquiring AIC, BIC,
HQC, and comparing? I don't think the time-series-->VAR lag selection
option gets at precisely the same thing. Thank you TMN
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
8:46 a.m.
On Sat, 26 Apr 2014, oleg_komashko(a)ukr.net wrote:
You can use armax_auto (see time series menu in Gretl GUI). The
function
has a drawback: the cycle crushes on encountering a unit root in MA part
of the model Also you can use auto.arima{forecast} in R My experience
tells it would be better to determine the order of integration (use unit
root tests) beforehand and use, say "d=1" in auto.arima (type
?auto.arima in R console)
That's a function package, you may need to install it first: go to
Files>Function Files>On server.
-------------------------------------------------------
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
-------------------------------------------------------
9:23 a.m.
Another option for automatized ARIMA selection is to use TRAMO. You should install
binaries from Gretl site. Use short output option,
--- Оригінальне повідомлення ---
Від кого: "Tim Nall" < tnall.ling(a)gmail.com >
Дата: 26 квітня 2014, 03:53:32
All,
Please forgive my simple questions. For ARIMA modelling, using the ACF
and PACF to determine the pdq parameters requires subjective judgment
based on experience. Answers are seldom clear-cut. As the subject
header says, can a somewhat more objective path be found in quickly
running through ARIMA models, varying parameters, acquiring AIC, BIC,
HQC, and comparing? I don't think the time-series-->VAR lag selection
option gets at precisely the same thing. Thank you TMN
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
2:36 p.m.
Dear Riccardo,
The filter function is quite useful. There is a slight misprint in the PDF
Command Reference: the second sum starts from {i=i}; of course, it should
say {i=1}.
--
Yours sincerely, | С уважением,
Andreï V. Kostyrka. | Андрей Викторович Костырка.
http://kostyrka.ru, http://kostyrka.ru/blog
2014-04-26 13:23 GMT+04:00 <oleg_komashko(a)ukr.net>:
Another option for automatized ARIMA selection is to use TRAMO. You
should install binaries from Gretl site. Use short output option,
--- Оригінальне повідомлення ---
Від кого: "Tim Nall" <tnall.ling(a)gmail.com>
Дата: 26 квітня 2014, 03:53:32
All,
Please forgive my simple questions. For ARIMA modelling, using the ACF
and PACF to determine the pdq parameters requires subjective judgment
based on experience. Answers are seldom clear-cut. As the subject
header says, can a somewhat more objective path be found in quickly
running through ARIMA models, varying parameters, acquiring AIC, BIC,
HQC, and comparing? I don't think the time-series-->VAR lag selection
option gets at precisely the same thing. Thank you TMN
_______________________________________________
Gretl-users mailing
listGretl-users@lists.wfu.eduhttp://lists.wfu.edu/mailman/listinfo/gretl-users
!
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
3:24 p.m.
On Sun, 27 Apr 2014, Andreï | Андрей Викторович wrote:
Dear Riccardo,
The filter function is quite useful. There is a slight misprint in the PDF
Command Reference: the second sum starts from {i=i}; of course, it should
say {i=1}.
Of course. It's now fixed in CVS, thanks.
-------------------------------------------------------
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
-------------------------------------------------------
3897
days inactive
3898
days old
7 comments
4 participants
participants (4)
-
Andreï | Андрей Викторович -
oleg_komashko＠ukr.net -
Riccardo (Jack) Lucchetti -
Tim Nall