Dear all
I have some problem with my code to forecasting based on ARMA-GARCH model and AR(1). Indeed, have 2 points:
1. My data are ranged from 2000:1:1 to 2017:1. I want to do forecasting only during the period 2000:1:1 to 2013:12 and then do forecast during the period 2014:1 to 2017:1. Therefore I have specified the following line in my code SMPL 1985:1 2013:12. When I do this the forecasting based on GARCH did not work and I have the following message:
## REG20. GARCH Cannot Be Used with Gaps/Missing Values
2. My second concern is about how we get the forecasted values of my variance based on GARCH in order to do the test of Diebold and Mariono.
My code is as follows
CALENDAR(M) 2000:1:1
Allocate 2015:12
OPEN DATA
data(format=xls,org=columns)
print
*
SMPL 2000:1:1 2013:12
equation(AR=6, MA=5) M1eq infa
group garch1 M1eq
garch(model=garch1, p=1,q=1,ASYMMETRIC, hseries=h,resids=u) / infa
@garchfore(steps=13) h u
forecast(model=garch1,from=2000:1,to=2017:1,results=rhat)
print / h
thank you a lot
ARMA-GARCH Model
Re: ARMA-GARCH Model
First, why are you choosing something as complicated as an ARMA(6,5)? That's almost certainly (very) overparameterized.
The estimation problem is that you're telling it to start the estimation at the very beginning of your data set, even though you need 6 own lags for the mean. The SMPL instruction exists in RATS but is more of a TSP/EViews idea. Instead, get rid of the SMPL and just use
garch(model=garch1, p=1,q=1,ASYMMETRIC, hseries=h,resids=u) * 2013:12 infa
to restrict the upper end of the estimation range. By the way, you can skip a step and use the EQUATION directly without defining a MODEL:
garch(equation=M1eq, p=1,q=1,ASYMMETRIC, hseries=h,resids=u) * 2013:12 infa
Forecasts of the variance don't depend upon the mean model, so just follow the description in the manual. However, you can't use Diebold-Mariano on the variance because the volatility isn't observable.
The estimation problem is that you're telling it to start the estimation at the very beginning of your data set, even though you need 6 own lags for the mean. The SMPL instruction exists in RATS but is more of a TSP/EViews idea. Instead, get rid of the SMPL and just use
garch(model=garch1, p=1,q=1,ASYMMETRIC, hseries=h,resids=u) * 2013:12 infa
to restrict the upper end of the estimation range. By the way, you can skip a step and use the EQUATION directly without defining a MODEL:
garch(equation=M1eq, p=1,q=1,ASYMMETRIC, hseries=h,resids=u) * 2013:12 infa
Forecasts of the variance don't depend upon the mean model, so just follow the description in the manual. However, you can't use Diebold-Mariano on the variance because the volatility isn't observable.