11:48 p.m.
Greetings,
I am playing with sunspot data on a Sunday night, making periodograms in
various software packages (R, SAS, ITSM, and Gretl). The Gretl
periodogram is quite nice, with the period on the upper x axis and the
"scaled frequency" on the lower x axis. None of the other packages
present scaled frequency as a default (I don't know if they offer it at
all), and I have been unable to nail down exactly what it is. Can anyone
offer a simple explanation of scaled frequency as it is used in Gretl,
or even better, a specific reference? Thanks.
Tom
1:09 a.m.
On Sun, 2 Mar 2008, Thomas La Bone wrote:
I am playing with sunspot data on a Sunday night, making
periodograms in various software packages (R, SAS, ITSM, and
Gretl). The Gretl periodogram is quite nice, with the period on
the upper x axis and the "scaled frequency" on the lower x axis.
None of the other packages present scaled frequency as a default
(I don't know if they offer it at all), and I have been unable
to nail down exactly what it is. Can anyone offer a simple
explanation of scaled frequency as it is used in Gretl, or even
better, a specific reference? Thanks.
W. Greene, Econometric Analysis, 4e, chapter 18, section 18.2.8.
I suppose there must be a counterpart in more recent editions;
I'll see if I can find a newer reference.
Allin Cottrell
9:01 a.m.
El Monday 03 March 2008 02:09:24 Allin Cottrell escribió:
On Sun, 2 Mar 2008, Thomas La Bone wrote:
> I am playing with sunspot data on a Sunday night, making
> periodograms in various software packages (R, SAS, ITSM, and
> Gretl). The Gretl periodogram is quite nice, with the period on
> the upper x axis and the "scaled frequency" on the lower x axis.
> None of the other packages present scaled frequency as a default
> (I don't know if they offer it at all), and I have been unable
> to nail down exactly what it is. Can anyone offer a simple
> explanation of scaled frequency as it is used in Gretl, or even
> better, a specific reference? Thanks.
W. Greene, Econometric Analysis, 4e, chapter 18, section 18.2.8.
I suppose there must be a counterpart in more recent editions;
I'll see if I can find a newer reference.
I don't have this reference, but looking at the numbers I can guess that:
if T is the period, and Nobs the number of observations, the scaled frequency
is
f=Nobs/T,
or said in another way, if w is the angular frequency, (w=2*pi/T)
f=w*(Nobs/2*pi)
--
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.et.bs.ehu.es
10:45 a.m.
Yes, that seems to be what Gretl is doing. Thanks. The next question is
why? What is the advantage of scaling the frequency in this way?
Tom
Ignacio Diaz-Emparanza wrote:
I don't have this reference, but looking at the numbers I can guess that:
if T is the period, and Nobs the number of observations, the scaled frequency
is
f=Nobs/T,
or said in another way, if w is the angular frequency, (w=2*pi/T)
f=w*(Nobs/2*pi)
5:23 p.m.
On Mon, 3 Mar 2008, Thomas La Bone wrote:
Yes, that seems to be what Gretl is doing. Thanks. The next question
is why?
What is the advantage of scaling the frequency in this way?
If you have a time series with a marked cycle, a peak in the spectrum at a
specified "scaled frequency" tells you how many cycles you're likely to
observe in the data. Nice to have. Try this:
nulldata 150
setobs 1 1 --special
set seed 12345
e = normal()
series y = 0
y = 1.5*y(-1) - 0.8*y(-2) + e
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
9:46 a.m.
El Monday 03 March 2008 18:23:37 Riccardo (Jack) Lucchetti escribió:
> Yes, that seems to be what Gretl is doing. Thanks. The next
question is
> why? What is the advantage of scaling the frequency in this way?
If you have a time series with a marked cycle, a peak in the spectrum at a
specified "scaled frequency" tells you how many cycles you're likely to
observe in the data. Nice to have. Try this:
nulldata 150
setobs 1 1 --special
...
What does the option --special mean? It is not documented in the manual
--
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.et.bs.ehu.es
2:08 p.m.
On Wed, 5 Mar 2008, Ignacio Diaz-Emparanza wrote:
El Monday 03 March 2008 18:23:37 Riccardo (Jack) Lucchetti escribió:
> nulldata 150
> setobs 1 1 --special
> ...
What does the option --special mean? It is not documented in
the manual
Hmm, you're right, that needs to be added. It's short for
--special-time-series
which can be used to indicate that a dataset is time series, but
not with any standard, recognized frequency. The difference is
that if you did
nulldata 150
setobs 1 1 --time-series
you would be defined the data as annual, starting in the year 1,
but with --special you're being non-committal about the dates.
Allin.
12:56 a.m.
In this example an AR(2) dataset is simulated. I do not have a lot of
experience with the Gretl language and it is not clear to me how to
generate an MA dataset. Can someone suggest the appropriate commands
needed to generate an MA(2) dataset, for example. Thanks
Tom
Riccardo (Jack) Lucchetti wrote:
On Mon, 3 Mar 2008, Thomas La Bone wrote:
> Yes, that seems to be what Gretl is doing. Thanks. The next question
> is why? What is the advantage of scaling the frequency in this way?
If you have a time series with a marked cycle, a peak in the spectrum at
a specified "scaled frequency" tells you how many cycles you're likely
to observe in the data. Nice to have. Try this:
nulldata 150
setobs 1 1 --special
set seed 12345
e = normal()
series y = 0
y = 1.5*y(-1) - 0.8*y(-2) + e
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Universit` 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
2:34 a.m.
On Wed, 5 Mar 2008, Thomas La Bone wrote:
In this example an AR(2) dataset is simulated. I do not have a
lot of experience with the Gretl language and it is not clear to
me how to generate an MA dataset. Can someone suggest the
appropriate commands needed to generate an MA(2) dataset, for
example.
After the preamble,
nulldata 500
setobs 1 1 --special
The language is pretty much transparent:
series e = normal()
series MA2 = e + .4*e(-1) + .1*e(-2)
Allin Cottrell
6:39 a.m.
On Wed, 5 Mar 2008, Allin Cottrell wrote:
On Wed, 5 Mar 2008, Thomas La Bone wrote:
> In this example an AR(2) dataset is simulated. I do not have a
> lot of experience with the Gretl language and it is not clear to
> me how to generate an MA dataset. Can someone suggest the
> appropriate commands needed to generate an MA(2) dataset, for
> example.
After the preamble,
nulldata 500
setobs 1 1 --special
The language is pretty much transparent:
series e = normal()
series MA2 = e + .4*e(-1) + .1*e(-2)
See also the "Cheat sheet" chapter in the user's guide.
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
10:47 a.m.
Thanks everyone. How could a student miss a chapter called "cheat
sheet"? That chapter has a taste of what I have been looking for --
examples of Gretl code doing a variety of tasks. As a new user, it would
be very helpful if there was a section on the main Gretl web page that
had examples of code submitted by more experienced users (like profs
using Gretl to teach a time series course). Just a suggestion.
Tom
Riccardo (Jack) Lucchetti wrote:
On Wed, 5 Mar 2008, Allin Cottrell wrote:
> On Wed, 5 Mar 2008, Thomas La Bone wrote:
>
>> In this example an AR(2) dataset is simulated. I do not have a
>> lot of experience with the Gretl language and it is not clear to
>> me how to generate an MA dataset. Can someone suggest the
>> appropriate commands needed to generate an MA(2) dataset, for
>> example.
>
> After the preamble,
>
> nulldata 500
> setobs 1 1 --special
>
> The language is pretty much transparent:
>
> series e = normal()
> series MA2 = e + .4*e(-1) + .1*e(-2)
See also the "Cheat sheet" chapter in the user's guide.
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
2:13 p.m.
New subject: LMF serial correlation statistic
Hello,
where can i find the exact formula used for calculation the LMF
statistic in the serial correlation LM test? I checked the Kiviet
(1986) formula (LMF, equation 5, p. 245) but it doesn't produce
the same result and i don't know why.
Thanks in advance.
Best regards,
Arek Wiśniowski
Allin Cottrell pisze:
On Sun, 2 Mar 2008, Thomas La Bone wrote:
> I am playing with sunspot data on a Sunday night, making
> periodograms in various software packages (R, SAS, ITSM, and
> Gretl). The Gretl periodogram is quite nice, with the period on
> the upper x axis and the "scaled frequency" on the lower x axis.
> None of the other packages present scaled frequency as a default
> (I don't know if they offer it at all), and I have been unable
> to nail down exactly what it is. Can anyone offer a simple
> explanation of scaled frequency as it is used in Gretl, or even
> better, a specific reference? Thanks.
W. Greene, Econometric Analysis, 4e, chapter 18, section 18.2.8.
I suppose there must be a counterpart in more recent editions;
I'll see if I can find a newer reference.
Allin Cottrell
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
8:25 p.m.
New subject: LMF serial correlation statistic
On Mon, 3 Mar 2008, rwarekw wrote:
where can i find the exact formula used for calculation the LMF
statistic in the serial correlation LM test? I checked the
Kiviet (1986) formula (LMF, equation 5, p. 245) but it doesn't
produce the same result and i don't know why.
Ah, we've been using Ramanathan's version of the LMF calculation.
See http://econ.ucsd.edu/~rramanat/LMFtest.pdf ; he gives
((T-k-n)/n) * R2 / (1-R2)
where R2 is the unadjusted R-squared from the auxiliary
regression. This is equivalent to an F-test for the null that
all coefficients in the auxiliary regression are zero.
I see this is not identical to Kiviet's formulation, and will
differ numerically if the sample partial correlation (controlling
for the lagged residuals) between X and the current residual is
non-zero. Perhaps we should switch to Kiviet's version.
Allin Cottrell
6452
days inactive
6456
days old
12 comments
5 participants
participants (5)
-
Allin Cottrell -
Ignacio Diaz-Emparanza -
Riccardo (Jack) Lucchetti -
rwarekw -
Thomas La Bone