1:09 p.m.
Hello,
I own several packages; Xplore, R, Gretl Omatrix etc, I am at the the
very basic level. It seems that the open source tools are quckly gaining
in capability. I stillsee estima and several others doing their thing.
What is the expected level of capabillity of Gretl in the area of times
series analysis lets say in a year??
Allan Bart
7:52 p.m.
On Thu, 20 Jul 2006, Allan_Bart(a)ibi.com wrote:
What is the expected level of capabillity of Gretl in the area
of times series analysis lets say in a year??
What we have already:
* Proper handling for numerous data frequencies: decennial,
annual, quarterly, monthly, weekly, daily (5-, 6- or 7-day) and
hourly).
* Easy and high-quality time series graphs, individual or
multiple.
* Various GLS methods for models with autoregresive errors
(Cochrane-Orcutt, generalized Cochrane-Orcutt, Hildreth-Lu,
Prais-Winsten).
* ARCH and GARCH.
* VARs and VECMs.
* ARIMA, including seasonal ARIMA, with a choice of estimation
method.
* Filters: Hodrick-Prescott and Baxter-King (command mode only,
no GUI yet).
* Correlogram and spectrum.
* Breusch-Godfrey autocorrelation test.
* Unit root tests: Dickey-Fuller and KPSS.
What might be added in the next year, depending on how
priorities work out:
* More advanced hypothesis testing for VECMs.
* Structural VARs
* GUI interface for time-series filters
* Your requests here
Allin Cottrell
3:06 a.m.
On Fri, July 21, 2006 01:52, Allin Cottrell wrote:
On Thu, 20 Jul 2006, Allan_Bart(a)ibi.com wrote:
> What is the expected level of capabillity of Gretl in the area
> of times series analysis lets say in a year??
What we have already:
...
* ARCH and GARCH.
At the moment we are only providing "plain vanilla" GARCH. Its implementation
is excellent, being based on the code written by the three famous
computational econometrics gurus Fiorentini, Calzolari & Panattoni. However,
we lack an extension to the veritable zoo of *ARCH models that have been
invented in the past 20 years, such as EGARCH, APARCH, GARCH-M and so on.
Gretl *can*, at present, estimate these models, but only through the mle
command: the user has to write down the log-likelihood manually, and there's
no GUI interface.
* ARIMA, including seasonal ARIMA, with a choice of estimation
method.
I'll venture to say that the ARMA routines in the upcoming 1.6.0 version are
among the best around. Possible improvements are: analytical gradients,
constraints on coefficients, fractional models. But ordinary (S)ARIMA
estimation is fully implemented.
* Unit root tests: Dickey-Fuller and KPSS.
And it's rather easy to write more using scripts. For example, DF-GLS takes
only a few lines.
What might be added in the next year, depending on how
priorities work out:
* More advanced hypothesis testing for VECMs.
* Structural VARs
* GUI interface for time-series filters
* Your requests here
* a few more tools for long-memory models
Riccardo "Jack" Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"
Ancona
9:40 a.m.
Allin Cottrell schrieb:
What might be added in the next year, depending on how priorities work out:
* More advanced hypothesis testing for VECMs.
That one is the only reason why I am still regularly (and legally, mind
you) using PcGive. If gretl could do that, it would be great. While I
greatly admire Jurgen Doornik for his work there, I think ten years or
so is enough in terms of licence fees for his proprietary program. (It
would also make it easier to leave windows behind, but that's another
story, and PcGive apparently is being ported to Linux this year.)
However, I guess that implementing the relevant switching algorithms is
not trivial, or other packages would already have followed the lead of
PcGive. Afaics, the best reference is Boswijk and Doornik (2004),
Identifying, estimating and testing restricted cointegrated systems: An
overview. In: Statistica Neerlandica 58(4), pp.440-465. See especially
section 3 (most general and most difficult part I guess).
So what do you think? Is it doable? In what way could I help (given that
I still don't know or want to learn C)?
This is clearly a more longer-term thing. That's why I'm sending it just
before I go offline for a while :-)
cheers,
Sven
9:51 a.m.
On Fri, 28 Jul 2006, Sven Schreiber wrote:
> What might be added in the next year, depending on how priorities
work out:
>
> * More advanced hypothesis testing for VECMs.
That one is the only reason why I am still regularly (and legally, mind
you) using PcGive. If gretl could do that, it would be great. While I
greatly admire Jurgen Doornik for his work there, I think ten years or
so is enough in terms of licence fees for his proprietary program. (It
would also make it easier to leave windows behind, but that's another
story, and PcGive apparently is being ported to Linux this year.)
However, I guess that implementing the relevant switching algorithms is
not trivial, or other packages would already have followed the lead of
PcGive. Afaics, the best reference is Boswijk and Doornik (2004),
Identifying, estimating and testing restricted cointegrated systems: An
overview. In: Statistica Neerlandica 58(4), pp.440-465. See especially
section 3 (most general and most difficult part I guess).
Thanks for the reference. Do you have a PDF of that article by
any chance? I'd like to take a look and see how difficult it
seems.
Allin.
6683
days inactive
6691
days old
4 comments
4 participants
participants (4)
-
Allan_Bart＠ibi.com -
Allin Cottrell -
Riccardo Jack Lucchetti -
Sven Schreiber