« Return to Thread: Problem with GarchFit [NC]

Re: Problem with GarchFit [NC]

by sylvain.archenault :: Rate this Message:

I am pretty new to garchFit as well.

I noticed that with my time series, it was the mean estimation which makes
the hessian matrix computation failed. When I compare the mean of the
timeseries I tried to fit. I noticed that the mean value of the series
generating by garchSim was around 1e-3 (as the mean value is set to 0). On
the other side, the timeseries that makes garchFit failed has mean of
1e-17.

On top of that, the garch estimates seem coherents.

Thanks.

markleeds@...
08/04/08 01:33 AM

To
Sylvain ARCHENAULT/fr/socgen@socgen
cc

Subject
Re: Re: [R-SIG-Finance] Problem with GarchFit [NC]

>From: sylvain.archenault@...
>Date: 2008/04/07 Mon AM 11:28:35 CDT
>To: R-SIG-Finance@...
>Subject: Re: [R-SIG-Finance] Problem with GarchFit [NC]

good but i've never used GarchFit so make sure it's doing what you think
it's doing. i was just guessing but if i helped, i'm glad to but make
sure i did because i really have no clue/experience about GarchFit.

>Thanks, it works great.
>
>I didn't understand well the "include.mean" parameter, I thought it has
>only effects if a mean formul (ARMA) was specified.
>
>
>
>markleeds@...
>07/04/08 06:21 PM
>
>
>To
>Sylvain ARCHENAULT/fr/socgen@socgen
>cc
>
>Subject
>Re: Re: [R-SIG-Finance] Problem with GarchFit [NC]
>
>
>
>
>
>
>>From: sylvain.archenault@...
>>Date: 2008/04/07 Mon AM 11:14:48 CDT
>>To: R-SIG-Finance@...
>>Subject: Re: [R-SIG-Finance] Problem with GarchFit [NC]
>
>if you're confident that there isn'ta non zero
>mean, then
>is there an option for fixing mu at zero (
>i/e: zero mean ) ? maybe that could
>fix things ? I'm no expert at this as
>you can tell so i'm just throwing things
>out there. good luck.
>
>
>
>>Plotting the data seems correct. The process failed when the gardhFit
>>method try to compute the Hessian matrix. Actually, it seems that the
>>problem comes from the mu parameter which is
>>very close to 0 (mu: 9.714126e-17), and then the solve method on the
>>Hessian matrix failed.
>>
>>"trace=F" is just to prevent printing all informations about the
>>optimisation process.
>>
>>
>>
>>
>>
>>markleeds@...
>>07/04/08 06:01 PM
>>
>>
>>To
>>Sylvain ARCHENAULT/fr/socgen@socgen
>>cc
>>
>>Subject
>>Re: [R-SIG-Finance] Problem with GarchFit [NC]
>>
>>
>>
>>
>>
>>
>>>From: sylvain.archenault@...
>>>Date: 2008/04/07 Mon AM 10:53:28 CDT
>>>To: r-sig-finance@...
>>>Subject: [R-SIG-Finance] Problem with GarchFit [NC]
>>
>>i woud plot the data and see if anything
>>looks strange about it. it sounds
>>like the routine is trying to take
>>an inverse and can't. also, what does trace = F do ?
>>maybe change that ?
>>
>>
>>>Hello,
>>>
>>>I have problems with garchFit. When trying to fit a time series, I got
>>the
>>>following error :
>>>
>>>> garchFit(~garch(1,1),data=x,trace=F)
>>>Error in solve.default(fit$hessian) :
>>> Lapack routine dgesv: system is exactly singular
>>>
>>>The method works fine on series generating by garchSim. What could be
>the
>>
>>>problem with the timeseries I use ?
>>>
>>>Thanks,
>>>Sylvain
>>>*************************************************************************
>>>This message and any attachments (the "message") are

co...{{dropped:22}}
>>
>>_______________________________________________
>>R-SIG-Finance@... mailing list
>>https://stat.ethz.ch/mailman/listinfo/r-sig-finance
>>-- Subscriber-posting only.
>>-- If you want to post, subscribe first.
>
>*************************************************************************
>This message and any attachments (the "message") are co...{{dropped:22}}

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

« Return to Thread: Problem with GarchFit [NC]