Forecasting GARCH model

[Kraus]

I have weekly time series with 261 observations and I am interested in the predicted covariance matrix for the coming three months. The code looks essentially as follows:

Code: Select all

GARCH(NOPRINT,P=1,Q=1,HMATRICES=EstCovMat,RVECTORS=EstResid) / EUR USA JAP UKI
*
SOURCE mvgarchfore.src
@MVGARCHFORE(STEPS=13) EstCovMat EstResid
*  
DIS EstCovMat(274)

This should give me a prediction for t=274. Correct? How do I get a cov-mat-prediction for the whole period from t=262 to t=274?



I want to have a prediction of the

[TomDoan]

I'm not sure what you mean. EstCovMat(262) would have the forecast for the covariance matrix at 262, and similarly for the other entries. If you want the covariance across time periods, you would need to do simulations.

[Kraus]

Thank you, Tom. I need the covariance matrix for a portfolio optimization with a three months holding period. Since I am working with weekly data, EstCovMat(262) will give me an estimation for the coming week, but probably not a good estimator for the coming quarter.

[TomDoan]

So what do you need? The covariance of the sum of over the next 13 weeks? Wouldn't that be just the sum of the period-by-period covariances (since the deviations from the mean are assumed to be independent)?

[Kraus]

mmmhhh... Guess you're right. Thanks a lot!