Am 08.11.2014 um 15:48 schrieb Karthik Raju:
Hello Sven,
Please send these things to the gretl mailing list, not by private
email. (This now goes to gretl-users.)
I want to do 'out of sample' forecasting using GARCH variants like TARCH
or EGARCH. For example, after estimating a EGARCH model and saving the
bundle contents I use the following script,
series forecast = omega + alpha * (ok(uhat(-1)) ? uhat(-1)^2 :
forecast(-1)) / forecast(-1) + beta * forecast(-1) + gamma *
(ok(uhat(-1)) ? uhat(-1)^2 : forecast(-1))/(forecast(-1))
But the forecasted values has higher magnitude of deviations from
realized values.
I have also seen negative R2s with ARCH models which is even worse.
Could you please suggest for a solution ?
Well, you have to find a better model. That's what research is about,
it's nothing gretl-specific.
good luck,
sven