[R-SIG-Finance] GARCH/APARCH with fSeries

[Efferz at gmx.de]

Hi,

I try to make VaR predictions with different GARCH setups, and have some questions about the implementation with fSeries.

1) Is the following setup correct to predict the one-step ahead VaR with skewed t GARCH(1,1)?
  
  (retdata contains the daily returns)

   st.fit<-sstdFit(retdata)
   g.fit<-garchFit(formula=~garch(1,1),
                   data=retdata,
                   cond.dist="dsstd",
                   skew=st.fit$est['xi']
                   shape=st.fit$est['nu']
                   include.shape=T,
                   include.skew=T,
                   include.mean=T)	
   param<-predict(g.fit,1)
   g.os.pred<-qsstd(0.05,mean=as.numeric(param['meanForecast']),
                    sd=as.numeric(param['standardDeviation']),
                    nu=as.numeric(g.fit at fit$params$params['shape']),
                    xi=as.numeric(g.fit at fit$params$params['skew']))

2) How to predict the 5day VaR within that setup? (not the daily VaR in 5   
   days!) Are the predicted values in predict(g.fit,5)[5,] those for the  
   daily mean and sd in 5 days?

3) When using the same setup as in 1) but now for skewed t APARCH(1,1) 
   (formula=~APARCH(1,1)) I get extremly large predicted standard  
   deviations of 0.3 and larger (in contrast to 0.02 of the skewed t  
   GARCH(1,1) setup). I think that isn´t reasonable and therefore I have
   doubts if my implementation in 1) is correct. 

Kind regards,

Martin
--