It has been hard to find an example of a bootstrapped forecast, using the ugarchboot() function in the Rugarch package for a full ARFIMA/ARFIMA-GARCH model. i.e: The model which has a Fractional Integral component different from zero.
The regular examples supplied with the most common references for the Rugarch package seems not to be working with full ARFIMA-GARCH models.
Mr. Walther
Thank you very much for your advice.
I was looking foward to perform a ARFIMA(3,d,3)-GARCH(1,1) bootstrapped forecast. However, as I can see, the Rugarch package allows only for mutually exclusive AR or MA mean models when using a Fractional Integral component. Did I got this right ?!
Best Regards.
Jorcy Neto
Currently I'm tryng to previously estimate the Fraction Integral component using the fracDiff package, and then feeding the fractionally differentiated series for estimating a ARFIMA(3,0,3)-GARCH(1,1) model.
O = fdGPH(log.data)
K = fdSperio(log.data)
diff.log.data = diffseries(log.data,O$d)
...
I'm not sure whether this will be so accurate as the d parameter is stand-alone estimated, however now it is possible to do a bootstrapped forecast.
Sometimes you need to do it step by step. First simulate the paths using garchsim; then resample a number of time to get the stable estimated parameters. With these parameters you can plug them in the original AFRIMA - GARCH model to generate forecast starting from the first observation in the outsample space.
Dears
Thanks for the advice !
I'm now looking for which would be the most suitable methodology (like Box-Jenkins) for determining the number of parameters in a given GARCH model (Standard GARCH, GJR-GARCH, E-GARCH, etc ... ) .
Best Regards
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R