Thank You. Actually I had a gig user manual downloaded from internet
that does not have the section 3.5 on forecasting.
Now, I retrieved the gig.pdf file file you mentioned. Based on the
example provided I was able to replicate the example for b-g.gdt dataset
but not for my dataset.
My dataset consists of 1489 observations of 6 years daily stock index
log return series. I want to generate 150 days out of sample forecast.
The script I used is :
egarch = gig_setup(Y, 7, const)
series h = egarch.h
series uhat = egarch.uhat
matrix theta = egarch.coeff
scalar omega = theta
scalar alpha = theta
scalar beta = theta
dataset addobs 150
scalar latest = $nobs - 1000
smpl latest ;
series fore = h[latest]
series fore = omega + beta * fore(-1) + alpha * (ok(uhat(-1)) ?
uhat(-1)^2 : fore(-1))
But the forecast generated by the above script was in range from -10 to
negative values in millions.
Please let me know how to correct this and also I want to know how to
measure the forecast performance using loss functions like Mean square
error, Root mean square error etc.
Sven Schreiber wrote on 2014-09-27 01:10 PM +0530:
> No, I'm referring to the "gig" manual, because you
> gig.pdf is a separate file, for example click "help" on the gig dialog
> window when you launch gig.
> Am 27.09.2014 um 07:20 schrieb Karthik Raju:
> Hello Sven,
> Thank you for your response.
> I think you refer to Chapter 29 - Forecasting in "Gnu Regression,
> Econometrics and Time-series Library" by Allin Cottrell. But I cannot
> replicate to what is explained there.
> Gretl-users mailing list