question about the garchFit

Yohan Chalabi chalabi at phys.ethz.ch
Fri Sep 5 13:02:55 CEST 2008


>>>> "PAD" == "Parra Amado Daniel" <dparraam at banrep.gov.co>
>>>> on Mon, 1 Sep 2008 17:37:12 -0500

   PAD> My name is Daniel Parra, I'm economist and I'm studying time series
   PAD> models for financial variables. I'm working in R with their package
   PAD> (fGarch), which I found very interesting and useful.
   PAD>  
   PAD> I would ask you please help me with the following question.
   PAD>  
   PAD> How do I do to estimate a ARMA(6,1)-GARCH(1,1) without the coefficients
   PAD> AR(1), AR(2), AR(3), AR(4) and AR(5)?
   PAD>  
   PAD> For example, I've tried in this way,
   PAD>  
   PAD> y = garchFit(~arma(6,1)+garch(1,1), r.IGBC, trace=F,cond.dist = "dnorm")
   PAD>  
   PAD> and result is:
   PAD>  
   PAD> Coefficient(s):
   PAD> Mu       0.20950912  
   PAD> ar1       -0.27087286   
   PAD> ar2       0.06539889  
   PAD> ar3       -0.08086325     I don't want estimate these coefficients
   PAD> ar4       0.00391767   
   PAD> ar5        0.05273713  
   PAD> ar6       -0.04344611   
   PAD> ma1     0.49958456   
   PAD> omega  0.44834015   
   PAD> alpha1  0.34024705   
   PAD> beta1    0.46750871  
   PAD>  
   PAD> According to the results above, you can see that the routine estimate
   PAD> the coefficients AR (1) to AR (5), which I don't want in the model. How
   PAD> do I restrict them on the code?.
   PAD>  

Hi Daniel,

in the current implementation you can not fix specific parameters in the optimization, which are in your case the coefficients AR(1)-AR(5).

regards,
Yohan

-- 
PhD student
Swiss Federal Institute of Technology
Zurich

www.ethz.ch



More information about the Rmetrics-core mailing list