[Gretl-users] Forecasting variance of a GARCH model

Riccardo (Jack) Lucchetti r.lucchetti at univpm.it
Sun Sep 28 03:34:47 EDT 2014


On Sun, 28 Sep 2014, Karthik Raju wrote:

> Hello Sven,

[...]

> But the forecast generated by the above script was in range from -10 to
> negative values in millions.

You can't simply cut and paste the example provided: that applies to a 
model (GARCH) in which the law of motion for the volatility is formulated 
in terms of the conditional variance. Instead, you estimated an EGARCH 
model, which uses its log. It's not at all clear to me (nor, I suspect, in 
the literature) how precisely you ought to forecast volatility in that 
case, as you should deal with non-linearity properly in your loss funcion. 
I guess that for a crude approximation of a 1-step, 2-step, 3-step ahead 
forecast you could exponentiate what you get from the recursive formula, 
but definitely I wouldn't recommend that from a statistical viewpoint.

-------------------------------------------------------
   Riccardo (Jack) Lucchetti
   Dipartimento di Scienze Economiche e Sociali (DiSES)

   Università Politecnica delle Marche
   (formerly known as Università di Ancona)

   r.lucchetti at univpm.it
   http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------


More information about the Gretl-users mailing list