the gretl RNG
by Allin Cottrell
For some years now the random number generator (RNG) used in gretl
has been the Mersenne Twister as implemented in the GLib library.
We've recently been investigating an alternative, namely the
SIMD-oriented Fast Mersenne Twister (SFMT); see
http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/SFMT/
We find that this new variant is faster, has better
equidistributional properties, and exhibits quicker recovery from
poor initialization. We're therefore going to make SFMT the
default RNG in gretl: this will be in the next release (gretl
1.9.4) and will land in CVS and the snapshots for Windows and Mac
shortly.
This change is (necessarily) backward incompatible, in the sense
that a gretl script that calls for random values using a specified
seed will now produce a different series of random values from
before. However, to preserve compatibility for anyone who wants to
replicate previous studies exactly, you will be able to call for
use of the old gretl RNG. The command to do this will be
set RNG MT
Having done that, you can switch back to the new default via
set RNG SFMT
Allin Cottrell
12 years, 4 months
GARCH, Forecasting
by Alejandro Mosiño
Hello,
How can we get the out-of-sample volatility / variance forecasts of a
GARCH model?
Thank you,
Alejandro
12 years, 4 months
Residual correlation matrix
by Olle Olsson
Hi all,
Is there a command available to retrieve the residual correlation matrix
returned after performing a normality test (following a VAR estimation)?
/Olle
12 years, 4 months
Re: [Gretl-users] (no subject)
by Allin Cottrell
On Thu, 13 Jan 2011, [big5] ������ ������ wrote:
> 1. If I use data from 1980M01 to 2000M12 to estimate the
> model(in-sample forecasting). It can calculate the MAPE of
> 1980M01-2000M12.
>
> Can it just calculate the MAPE of within-sample like
> 1999M1-2000M12�B1998M01-2000M12 or so on in gretl?
You can calculate MAPE for any series over any range by setting
the sample range and applying the simple formula:
MAPE = sum(abs(y)) / $nobs
> 2. Why does the test of AR and MA show t-statistics in
> gretl1.8.4, but z-statistics in gretl1.9.1?
We decided it was best to display these as z-statistics for
estimators whose validity is based on asymptotic theory and where
standard errors are based on ML estimation. That may be dabatable,
but it's a legitimate choice.
Allin Cottrell
12 years, 4 months
(no subject)
by 不提供 不提供
Dear all:
I have two questions:
1. If I use data from 1980M01 to 2000M12 to estimate the model(in-sample forecasting). It can calculate the MAPE of 1980M01-2000M12.
Can it just calculate the MAPE of within-sample like 1999M1-2000M12、1998M01-2000M12 or so on in gretl?
2. Why does the test of AR and MA show t-statistics in gretl1.8.4, but z-statistics in gretl1.9.1?
Their values in my sample are the same. But degree of freedom of t distribution in my sample is not large,
so the values of t-statistics (gretl 1.8.4) and z-statistics (gretl 1.0.1) should not be the same.
It seems that they are both z-statistics, but it shows t –statistics in gretl1.8.4.
Doesn't it test with t-statistics in general? Why does gretl use z-statistics?
Thanks a lot
12 years, 4 months
Re: [Gretl-users] Correlogram and Forecasting
by Allin Cottrell
On Wed, 12 Jan 2011, [ISO-8859-1] Alejandro Mosi�o wrote:
> I have this question: Is there an gnuplot option to correlograms and
> forecasting?
No, but you can make your own gnuplot graph from corrgm output
quite easily using the corrgm() function. For example
open data9-7
matrix c = corrgm(QNC, 12)
gnuplot 1 --time-series --with-impulses \
--matrix=c --output=display
If you click on the resulting plot and choose "Edit" you can
fine-tune its appearance (e.g. adjust the y-axis range and the
line-width for the impulses).
Allin Cottrell
12 years, 4 months
Re: [Gretl-users] How to generate forecasts
by Allin Cottrell
On Wed, 12 Jan 2011, [big5] ������ ������ wrote:
> I set sample range from 1990M01-2008M12 to estimate the model
> and click on Analysis -> Forecast. Forecast range has been
> automatically set from 2009M01 to 2010M12 in the dialog box and
> I click on OK to have predictions for 2009M01-2010M12
> (out-of-sample forecasting). There is no option can choose to
> generate the forecasts of 2011M01-2013M12.
To get that option, go to the Data menu and choose "Add
observations". Select the number of observations you want (in this
case 36, I think). Then set the sample and estimate your model.
Allin Cottrell
12 years, 4 months
Re: [Gretl-users] Creating X-12-ARIMA spec file
by Allin Cottrell
On Wed, 12 Jan 2011, [big5] ������ ������ wrote:
> I have a time series data in gretl. Can I convert it to
> x-12-arima spec file?
When you choose "/Variable/X-12-ARIMA analysis" in gretl, an
X-12-ARIMA spec file is created for the variable in question.
It's possible to grab and modify this file but it may not be easy
to find, on MS Windows in particular.
However, in current CVS gretl and snapshots (see
http://gretl.sourceforge.net/win32/ ) there's an easier mechanism.
Now if you do "/Variable/X-12-ARIMA analysis" you get the option
of having gretl open the spec file in an editor window. (The exact
option text is "Make X-12-ARIMA command file".) From there you
can modify it and save it to whatever location you like.
Allin Cottrell
12 years, 4 months
Correlogram and Forecasting
by Alejandro Mosiño
Hello,
I have this question: Is there an gnuplot option to correlograms and
forecasting?
Specifically, i have a variable named "l_yde", so i get the correlogram:
corrgm l_yde 12
However, the output is really different from that obtained from the
Gretl toolbar. And the same is for forecasting. For example:
arima 2 0 0 ; l_yde time const
fcasterr 1995:01 1996:07
gives the values, but not the beautiful plot we get from the gretl menus.
Is there an option that i'm not using, or the option does not exist?
Thank you all,
Alejandro
12 years, 4 months
How to generate forecasts
by 不提供 不提供
Dear all:
Giuseppe Vittucci, thank you.
But when I do it this way, I just obtain out-of-sample forecasts (2009M01-2010M12), not forecasts of 2011M01-2013M12. The details are below:
Data: 1990M01-2010M12
In-sample forecasting: 1990M01-2008M12
Out-of-sample forecasting: 2009M01-2010M12
Generate forecasts: 2011M01-2013M12
I set sample range from 1990M01-2008M12 to estimate the model and click on Analysis -> Forecast.
Forecast range has been automatically set from 2009M01 to 2010M12
in the dialog box and I click on OK to have predictions for 2009M01-2010M12
(out-of-sample forecasting).
There is no option can choose to generate the forecasts of 2011M01-2013M12.
Thanks a lot
12 years, 4 months
