Garch fitting with mean regressors

[Stefano Balietti]

Hi,
I'm looking for R packages able to perform GARCH-like fitting (estimates),
such as fGarch, tseries and FinTS, but that allow me to specify extra
regressors for the mean equation. It doesn't seem to me that the
above-mentioned packages permit that, but maybe I'm wrong. However,  does
anyone have any suggestion?

Cheers,

Stefano Balietti

        [[alternative HTML version deleted]]

_______________________________________________
R-SIG-Finance@... mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only.
-- If you want to post, subscribe first.

[Yohan Chalabi]

>>>> "SB" == "Stefano Balietti" <futur.dorko@...>
>>>> on Tue, 15 Apr 2008 21:27:39 +0200

   SB> Hi,
   SB> I'm looking for R packages able to perform GARCH-like fitting
   SB> (estimates),
   SB> such as fGarch, tseries and FinTS, but that allow me to
   SB> specify extra
   SB> regressors for the mean equation. It doesn't seem to me
   SB> that the
   SB> above-mentioned packages permit that, but maybe I'm
   SB> wrong. However,  does
   SB> anyone have any suggestion?
   SB>
   SB> Cheers,
   SB>
   SB> Stefano Balietti
   SB>
 
As far as fGarch is concerned, you can only specify the arma regressors
for the mean equation. However, the functions in fGarch are quite
modular and it should not be too much of work to add new regressor to
the mean equation.

regards,
Yohan


--


The 2nd International R/Rmetrics User and Developer Workshop ...
[http://www.rmetrics.org]

_______________________________________________
R-SIG-Finance@... mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only.
-- If you want to post, subscribe first.

[Patrick Burns-2]

You can do the regression on the returns and
then fit the garch model on the residuals.  That
will most probably be very close to the result
if you did it "right".


Patrick Burns
patrick@...
[phone]
http://www.burns-stat.com
(home of S Poetry and "A Guide for the Unwilling S User")

Stefano Balietti wrote:

> Hi,
> I'm looking for R packages able to perform GARCH-like fitting (estimates),
> such as fGarch, tseries and FinTS, but that allow me to specify extra
> regressors for the mean equation. It doesn't seem to me that the
> above-mentioned packages permit that, but maybe I'm wrong. However,  does
> anyone have any suggestion?
>
> Cheers,
>
> Stefano Balietti
>
> [[alternative HTML version deleted]]
>
> _______________________________________________
> R-SIG-Finance@... mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-finance
> -- Subscriber-posting only.
> -- If you want to post, subscribe first.
>
>
>

_______________________________________________
R-SIG-Finance@... mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only.
-- If you want to post, subscribe first.

[Zeno Adams]

On Wed, 16 Apr 2008 10:11:27 +0100
 Patrick Burns <patrick@...> wrote:
> You can do the regression on the returns and
> then fit the garch model on the residuals.  That
> will most probably be very close to the result
> if you did it "right".
>
>


I wonder if you could really do that. After all you would do an
estimation ignoring heteroscedasticity in the returns which biases the
parameter estimates. If you include the exogenous in the mean equation
of a garch model then you take conditional heteroscedasticity into
account. This is easy to do in most commercial software (e.g. EViews,
RATS etc.)

Zeno

_______________________________________________
R-SIG-Finance@... mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only.
-- If you want to post, subscribe first.

[Stefano Balietti]

On Wed, Apr 16, 2008 at 2:36 PM, Diethelm Wuertz <wuertz@...>
wrote:

> Yohan Chalabi wrote:
>
> > "SB" == "Stefano Balietti" <futur.dorko@...>
> > > > > > on Tue, 15 Apr 2008 21:27:39 +0200
> > > > > >
> > > > > >
> > > > >
> >   SB> Hi,
> >   SB> I'm looking for R packages able to perform GARCH-like fitting
> >   SB> (estimates),
> >   SB> such as fGarch, tseries and FinTS, but that allow me to
> >   SB> specify extra
> >   SB> regressors for the mean equation. It doesn't seem to me
> >   SB> that the
> >   SB> above-mentioned packages permit that, but maybe I'm
> >   SB> wrong. However,  does
> >   SB> anyone have any suggestion?
> >   SB>   SB> Cheers,
> >   SB>   SB> Stefano Balietti
> >   SB>  As far as fGarch is concerned, you can only specify the arma
> > regressors
> > for the mean equation. However, the functions in fGarch are quite
> > modular and it should not be too much of work to add new regressor to
> > the mean equation.
> > regards,
> > Yohan
> >
> >
> >
> >
> What about a two stage process ?
>
> To fit first an ARMA model with regressors, and then to fit
> the residuals with a ARMA-GARCH model ?
>
> This idea is similar like fitting independently first an ARMA
> and then a pure GARCH process.
>
> Diethelm
>
> I'm sorry for haven't explained it before, but I'm afraid this procedure

does not make my case, because I need to test the accuracy of my code, which
basically is a simple free implementation of a garch-like maximum likelihood
estimation. It will be part of my degree thesis. With any 2 stage process  I
may get similar estimations, but not exactly the same and  therefore I could
not say (almost) anything about the goodness of the code.  I guess there is
no free solution to my problem yet, but please correct me if I'm wrong.

Cheers,

Stefano

        [[alternative HTML version deleted]]

_______________________________________________
R-SIG-Finance@... mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only.
-- If you want to post, subscribe first.

[Patrick Burns-2]

Zeno Adams wrote:

> On Wed, 16 Apr 2008 10:11:27 +0100
>  Patrick Burns <patrick@...> wrote:
>  
>> You can do the regression on the returns and
>> then fit the garch model on the residuals.  That
>> will most probably be very close to the result
>> if you did it "right".
>>
>>
>>    
>
>
> I wonder if you could really do that. After all you would do an
> estimation ignoring heteroscedasticity in the returns which biases the
> parameter estimates. If you include the exogenous in the mean equation
> of a garch model then you take conditional heteroscedasticity into
> account. This is easy to do in most commercial software (e.g. EViews,
> RATS etc.)
>
> Zeno
>  

Of course we can really do that.  The question is
whether or not it is a good idea to do it.

Yes, we are ignoring heteroscedasticity in the regression.
This makes it inefficient, but bias should be minimal.  There
is also the option to iterate the two stages which, under
suitable conditions, will converge to the maximum likelihood
solution.

If we are worried about violating assumptions, then the two
stage estimation is likely to be one of our lesser sins in the
exercise.

Pat

>
>

_______________________________________________
R-SIG-Finance@... mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only.
-- If you want to post, subscribe first.

[Eric Zivot]

Yes, you can do this. Heteroskedasticity does not generally bias the
coefficients from the regression - just invalidates the usual standard
errors. For basic garch models you can estimate them in a two-step fashion.
Engle showed this in his orignal ARCH paper in 1982  

-----Original Message-----
From: r-sig-finance-bounces@...
[mailto:r-sig-finance-bounces@...] On Behalf Of Zeno Adams
Sent: Wednesday, April 16, 2008 6:37 AM
To: Patrick Burns; Stefano Balietti
Cc: r-sig-finance@...
Subject: Re: [R-SIG-Finance] Garch fitting with mean regressors

On Wed, 16 Apr 2008 10:11:27 +0100
 Patrick Burns <patrick@...> wrote:
> You can do the regression on the returns and then fit the garch model
> on the residuals.  That will most probably be very close to the result
> if you did it "right".
>
>


I wonder if you could really do that. After all you would do an estimation
ignoring heteroscedasticity in the returns which biases the parameter
estimates. If you include the exogenous in the mean equation of a garch
model then you take conditional heteroscedasticity into account. This is
easy to do in most commercial software (e.g. EViews, RATS etc.)

Zeno

_______________________________________________
R-SIG-Finance@... mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only.
-- If you want to post, subscribe first.

_______________________________________________
R-SIG-Finance@... mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only.
-- If you want to post, subscribe first.