Date: Fri, 1 Dec 2006 20:59:36 -0500
Reply-To: "Atamer, Murat" <Murat_Atamer@UNC.EDU>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: "Atamer, Murat" <Murat_Atamer@UNC.EDU>
Subject: Re: ARCH-Correction
Content-Type: text/plain; charset="iso-8859-1"
I got close enough results with your suggestion.
Many thanks & Regards,
Murat
"If everything seems under control, you're not fast enough"
Ayrton SENNA DA SILVA
-----Original Message-----
From: rdporto1 [mailto:rdporto1@terra.com.br]
Sent: Friday, December 01, 2006 8:28 PM
To: Atamer, Murat
Cc: SAS-L
Subject: Re: ARCH-Correction
Murat,
you'll get 22 parameters + intercept but using their
mean as initial values in a iterated way, it is supposed
that they will converge for the true values. Of course,
supposing that your model is right. See the macro part.
Rogerio.
---------- Cabeçalho original -----------
De: "SAS(r) Discussion" SAS-L@LISTSERV.UGA.EDU
Para: SAS-L@LISTSERV.UGA.EDU
Cópia:
Data: Thu, 30 Nov 2006 20:50:20 -0500
Assunto: Re: ARCH-Correction
> Hi Rogerio:
>
> Thank you so much for your answer. nlag covers the theta part. Is there a way to fix the parameters to be the same before running the model maybe using proc model? Because p=22 gives me 22 parameters. All I need to estimate is intercept, theta, a and b. Thank you for the warning, there are squares on error terms I am sorry about the typo. The model is:
>
> R= intercept + error + theta*lag1(error); * Error is distributed normally with mean zero, variance sigma^2;
>
> Sigma^2=a+b(lag1(error^2)+....+lag22(error^2))/22;
>
> Many thanks & Regards,
>
> Murat
>
> -----Original Message-----
> From: rdporto1 [mailto:rdporto1@terra.com.br]
> Sent: Thursday, November 30, 2006 7:46 PM
> To: Atamer, Murat
> Cc: SAS-L
> Subject: Re: ARCH-Correction
>
> Murat,
>
> I don't know what you wanna do with your estimates
> but you could try
>
> proc autoreg;
> model r= /nlag=1 garch=(p=22);
> run;
>
> Note I'm suposing there are *squares* on the
> lagged errors terms at the variance equation.
>
> If you are lucky to get a convergence, you could
> test if all the garch parameters are equal and give
> a reasonable value for a start up using the *initial*
> option.
>
> If your model is correct, after some try and error,
> you should get the desired estimates. By the way,
> you could also make a macro to do this.
>
> HTH,
>
> Rogerio.
> ---------- Cabeçalho original -----------
>
> De: "SAS(r) Discussion" SAS-L@LISTSERV.UGA.EDU
> Para: SAS-L@LISTSERV.UGA.EDU
> Cópia:
> Data: Thu, 30 Nov 2006 18:50:00 -0500
> Assunto: Re: ARCH-Correction
>
> > Hi again:
> >
> > I made a correction and explained with capital letters. Instead of
> > (lag(error)+ lag22(error))/22 I need all lags in between, as well.
> >
> > I am trying to run a model of the following form:
> >
> > Arch model
> > R= intercept + error + theta*lag1(error); * Error is distributed
> > normally with mean zero, variance sigma^2;
> >
> > Sigma^2=a+b(lag1(error)+....+lag22(error))/22; *I NEED ALL 22 LAGS HERE;
> >
> > Because we have the same parameter in front of all lagged errors,
> > divided by 22, and since we have a lagged error term multiplied by theta
> > in the model itself, I couldn't figure out how to set this up using proc
> > autoreg. The examples are all garch (1,1).
> >
> >
> > I would greatly appreciate any comments you may have.
> >
> > Thanks & Regards,
> >
> > Murat
> >
>
| 
