"ugarchfit" function of "rugarch" package needs at least 100 data points.

classic Classic list List threaded Threaded
5 messages Options
Reply | Threaded
Open this post in threaded view
|

"ugarchfit" function of "rugarch" package needs at least 100 data points.

Tanvir Khan
Right now I'm trying to fit a GJR Garch model on the inflation rate of my
country, Bangladesh, but the problem is the function "ugarchfit" of the
"rugarch" package requires at least 100 data points to run. I have yearly
(12 months average) data and my country is not even 50 years old! So, is
there any way to use this function to fit a data with less than 100
observations? Any type of suggestion will be extremely helpful.

--
**********************************************************
Tanvir Khan
Applied Statistician
Institute Of Statistical Research & Training
University Of Dhaka
http://bd.linkedin.com/pub/tanvir-khan/39/281/536
[hidden email]
[hidden email]
**********************************************************

***THE WORLD IS OPEN SOURCE***

        [[alternative HTML version deleted]]

_______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only. If you want to post, subscribe first.
-- Also note that this is not the r-help list where general R questions should go.
Reply | Threaded
Open this post in threaded view
|

Re: "ugarchfit" function of "rugarch" package needs at least 100 data points.

alexios
Dear Tanvir,

To have an 'accurate' estimate of the persistence of a GARCH process you
really do need more than 100 datapoints, and anything less than this is
not likely to be informative or accurate, and might even fail to
converge. More generally, in the approach adopted by the rugarch package
(frequentist), the question of determining the presence of GARCH
dynamics is difficult to answer with very few datapoints. Instead, if
you insist on using GARCH, with limited history monthly data, my
suggestion is to try one of the Bayesian packages (try the bayesGARCH
package or LaplacesDemon, the latter has extensive examples including
how to fit a number of GARCH models).

Regards,

Alexios


On 09/10/2012 19:58, Tanvir Khan wrote:
> Right now I'm trying to fit a GJR Garch model on the inflation rate of my
> country, Bangladesh, but the problem is the function "ugarchfit" of the
> "rugarch" package requires at least 100 data points to run. I have yearly
> (12 months average) data and my country is not even 50 years old! So, is
> there any way to use this function to fit a data with less than 100
> observations? Any type of suggestion will be extremely helpful.
>

_______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only. If you want to post, subscribe first.
-- Also note that this is not the r-help list where general R questions should go.
Reply | Threaded
Open this post in threaded view
|

Re: "ugarchfit" function of "rugarch" package needs at least 100 data points.

Patrick Burns-2
In reply to this post by Tanvir Khan
The blog post:

http://www.portfolioprobe.com/2012/09/17/variability-of-garch-estimates/

shows how variable garch results are using
2000 daily observations.

Even if there is a way to get your model fit,
I doubt that it would mean very much.

If you do find a way, I would suggest that you
create multiple datasets that are simulations of
the model the same size as your data.  Then estimate
the model on the simulations and see how variable
those model estimates are.  They will be very
variable, I predict.

Pat

On 09/10/2012 19:58, Tanvir Khan wrote:
> Right now I'm trying to fit a GJR Garch model on the inflation rate of my
> country, Bangladesh, but the problem is the function "ugarchfit" of the
> "rugarch" package requires at least 100 data points to run. I have yearly
> (12 months average) data and my country is not even 50 years old! So, is
> there any way to use this function to fit a data with less than 100
> observations? Any type of suggestion will be extremely helpful.
>

--
Patrick Burns
[hidden email]
http://www.burns-stat.com
http://www.portfolioprobe.com/blog
twitter: @portfolioprobe

_______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only. If you want to post, subscribe first.
-- Also note that this is not the r-help list where general R questions should go.
Reply | Threaded
Open this post in threaded view
|

Re: "ugarchfit" function of "rugarch" package needs at least 100 data points.

Eric Zivot
While GARCH parameter estimates can be quite variable (as Pat has nicely
shown), the resulting volatility forecasts tend to be much more stable (at
least in the short term). If the goal is short-term vol forecasting then one
doesn't really care too much about the GARCH point estimates. The important
long-term parameters are the unconditional vol and the persistence of the
GARCH process as these dictate where and how fast the vol forecasts evolve.
For the short term, most GARCH forecasts look a lot like simple EWMAs. So if
you only have 100 obvs and want to get a simple short-term forecast, use an
EWMA. That's the riskMetrics approach.

-----Original Message-----
From: [hidden email]
[mailto:[hidden email]] On Behalf Of Patrick Burns
Sent: Tuesday, October 09, 2012 1:05 PM
To: [hidden email]
Subject: Re: [R-SIG-Finance] "ugarchfit" function of "rugarch" package needs
at least 100 data points.

The blog post:

http://www.portfolioprobe.com/2012/09/17/variability-of-garch-estimates/

shows how variable garch results are using
2000 daily observations.

Even if there is a way to get your model fit, I doubt that it would mean
very much.

If you do find a way, I would suggest that you create multiple datasets that
are simulations of the model the same size as your data.  Then estimate the
model on the simulations and see how variable those model estimates are.
They will be very variable, I predict.

Pat

On 09/10/2012 19:58, Tanvir Khan wrote:
> Right now I'm trying to fit a GJR Garch model on the inflation rate of
> my country, Bangladesh, but the problem is the function "ugarchfit" of
> the "rugarch" package requires at least 100 data points to run. I have
> yearly
> (12 months average) data and my country is not even 50 years old! So,
> is there any way to use this function to fit a data with less than 100
> observations? Any type of suggestion will be extremely helpful.
>

--
Patrick Burns
[hidden email]
http://www.burns-stat.com
http://www.portfolioprobe.com/blog
twitter: @portfolioprobe

_______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only. If you want to post, subscribe first.
-- Also note that this is not the r-help list where general R questions
should go.

_______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only. If you want to post, subscribe first.
-- Also note that this is not the r-help list where general R questions should go.
Reply | Threaded
Open this post in threaded view
|

Re: "ugarchfit" function of "rugarch" package needs at least 100 data points.

Patrick Burns-2
I agree with Eric.  His suggestion is simple
and likely to give about as good of an answer
as possible.

On 09/10/2012 21:22, Eric Zivot wrote:

> While GARCH parameter estimates can be quite variable (as Pat has nicely
> shown), the resulting volatility forecasts tend to be much more stable (at
> least in the short term). If the goal is short-term vol forecasting then one
> doesn't really care too much about the GARCH point estimates. The important
> long-term parameters are the unconditional vol and the persistence of the
> GARCH process as these dictate where and how fast the vol forecasts evolve.
> For the short term, most GARCH forecasts look a lot like simple EWMAs. So if
> you only have 100 obvs and want to get a simple short-term forecast, use an
> EWMA. That's the riskMetrics approach.
>
> -----Original Message-----
> From: [hidden email]
> [mailto:[hidden email]] On Behalf Of Patrick Burns
> Sent: Tuesday, October 09, 2012 1:05 PM
> To: [hidden email]
> Subject: Re: [R-SIG-Finance] "ugarchfit" function of "rugarch" package needs
> at least 100 data points.
>
> The blog post:
>
> http://www.portfolioprobe.com/2012/09/17/variability-of-garch-estimates/
>
> shows how variable garch results are using
> 2000 daily observations.
>
> Even if there is a way to get your model fit, I doubt that it would mean
> very much.
>
> If you do find a way, I would suggest that you create multiple datasets that
> are simulations of the model the same size as your data.  Then estimate the
> model on the simulations and see how variable those model estimates are.
> They will be very variable, I predict.
>
> Pat
>
> On 09/10/2012 19:58, Tanvir Khan wrote:
>> Right now I'm trying to fit a GJR Garch model on the inflation rate of
>> my country, Bangladesh, but the problem is the function "ugarchfit" of
>> the "rugarch" package requires at least 100 data points to run. I have
>> yearly
>> (12 months average) data and my country is not even 50 years old! So,
>> is there any way to use this function to fit a data with less than 100
>> observations? Any type of suggestion will be extremely helpful.
>>
>
> --
> Patrick Burns
> [hidden email]
> http://www.burns-stat.com
> http://www.portfolioprobe.com/blog
> twitter: @portfolioprobe
>
> _______________________________________________
> [hidden email] mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-finance
> -- Subscriber-posting only. If you want to post, subscribe first.
> -- Also note that this is not the r-help list where general R questions
> should go.
>
>

--
Patrick Burns
[hidden email]
http://www.burns-stat.com
http://www.portfolioprobe.com/blog
twitter: @portfolioprobe

_______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-finance
-- Subscriber-posting only. If you want to post, subscribe first.
-- Also note that this is not the r-help list where general R questions should go.