AFAIK, this is one you’d have to do yourself via MLE (or GMM, maybe?). I think that JMulti has this class of model available from a menu or something similar, but I’ve never used it.




From: [] On Behalf Of Anutechia Asongu
Sent: Sunday, April 22, 2012 3:50 PM
To: Gretl list
Subject: [Gretl-users] STAR Model with Gretl


Hi Gretl Users,

                      Could some gurus here(Allin, Riccardo, Sven, Artur et al.) gist me on how to apply a STAR(Smooth Transition Autoregression) model with Gretl?



From: Allin Cottrell <>
To: Gretl list <>
Sent: Friday, April 20, 2012 11:29 PM
Subject: Re: [Gretl-users] Restrict command after VECM

On Thu, 19 Apr 2012, Sven Schreiber wrote:

> On 04/19/2012 05:55 PM, Allin Cottrell wrote:
>> On Thu, 19 Apr 2012, Artur Tarassow wrote:
>>> Since some while it is possible to do use the name of a variable for the
>>> 'restrict' command after OLS. But this is not the case for the VECM
>>> command. Is there any reason to handle this case differently?
> I thought the only reason is that nobody has implemented it, but Allin
> seems to disagree a little:
>> Yes: typically a restriction on a system such as a VECM will
>> need two "subscripts" on the coefficients (although in the
>> case you give, of a VECM with a single cointegrating relation,
>> that happens not to apply). I guess we could support names in
>> that context but it doesn't seem as "natural" as in the
>> straightforward single-equation case.
> Hm, isn't it the case already that a single index can be used here to
> mean "variable number x in all cointegration relations"?

You're right, I was forgetting that. So I've now enabled in CVS what
Artur wanted.

> I tend to think that Artur's request is perfectly valid, but I don't
> think the example he's giving is the top most typical hypothesis. In
> other words, if some syntactic sugar were to be added to the
> Vecm-restrict apparatus, I think it would be worthwhile to also cover
> some other use cases.
> At the risk of sounding spoiled, some of those could be:
> * test of weak exogeneity of some variable, zero row in alpha: 'a[myvar]=0'
> * test of whether variable number x is stationary, given the
> cointegration rank (one appropriate unit vector in beta, and that
> variable not appearing in the other relations)
> etc.
> I wonder whether there would be some way to implement this as a
> function/add-on/etc. without touching the C source?

That's a good thought. It might need a little help from the C code
level; e.g. the ability to call up the list of endogenous variables
in the VECM, which I don't think is there right now.

Gretl-users mailing list