Ricardo hello,

I just used here the the Gretl supplied NYSE1.gdt data set for exposition only so that I get this difference in the two software packages.
I am not working on this so this is also why no ln first differences were used.

My suspicion was that Eviews just spits out whatever sub-optimal result instead of a no convergence one.

On Thu, Jun 21, 2018 at 3:43 PM Riccardo (Jack) Lucchetti <r.lucchetti@univpm.it> wrote:
On Thu, 21 Jun 2018, Sven Schreiber wrote:

> Am 21.06.2018 um 12:52 schrieb Periklis Gogas:
>> And using Eviews with exactly the same specification and data I get this:
> I don't have an explanation (not knowing very well what happens inside the
> Garch or gig/GARCH routines), but it is not strictly true that it is the same
> specification: In gretl you used the default "sandwich" for the VCV, and
> Eviews says "outer product of gradients". You could try to use OPG in gretl
> as well and tell us what happens.
> It would certainly be interesting whether Eviews gets it right, or just gets
> "something".

Also: it looks as if you're using a series in levels, as opposed to log
differences, which is what you normally feed into garch models. Could you
send me your dataset? I'll have a look at it as soon as I get the chance.

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

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

Gretl-users mailing list