[Gretl-users] Re: State space modelling
Am 12.12.2020 um 17:22 schrieb Artur Tarassow:
Am 12.12.20 um 17:09 schrieb Brian Revell:
> Thanks this has been helpful thus far. Next question, having run just
> the basic Structural TS model, how do generate the forecasts as the
> the GUI only produces the graph of the fitted model vs actual. There
> is no link to Analysis. So do I need to save the model as a bundle and
> feed into another separate downloadable function to produce the
> forecast variable plus CI? Is there yet another function I need to
> pass the parameters to to see the smoothed series
Brian did you see the sample script for StrucTiSM? It includes an
example on how to compute the forecasts.
Click File -> Function packages -> On local machine and the then right
click on StrucTiSM -> Sample script.
That's a useful pointer, but it is also true that we should probably try
to enable forecasting via the package's output window. (I'm a co-author
of the package.)
(Internal question: the packaging guide says in section 4.3 that a
"bundle-fcast" designation is part of the spec, but that it's "unused
at
present". Does that mean it doesn't work yet?)
In the meantime what you can do is indeed to save the bundle from the
package's output window, say under the name "bstrucTS" or whatever. Then
in a script window or gretl console you'd need to do (without the hansl
tags):
<hansl>
include StrucTiSM.gfn
STSM_fcast( &bstrucTS, 12)
series depvarfcast = bstrucTS.fcast
series depvarfcvar = bstrucTS.fcastvar
</hansl>
Here the number 12 is the arbitrary forecast horizon. You need to make
sure you have those observations in the dataset beyond the estimation
sample, possibly by adding observations before you do the forecast.
BTW, good to have someone testing the package who is able to report any
problems on the list!
thanks
sven