GARCH forecasting

[trubador]

If you select "dynamic forecast", then it becomes an out-of-sample regardless of the period of your choice. It treats the specified period as unknown even if it has observations in it. If you select "static forecast" then the forecasted values will be identical to those of full in-sample fit. You can try and see it for yourself as EViews will do the hard work.

[econworker]

If you select "dynamic forecast", then it becomes an out-of-sample regardless of the period of your choice. It treats the specified period as unknown even if it has observations in it. If you select "static forecast" then the forecasted values will be identical to those of full in-sample fit. You can try and see it for yourself as EViews will do the hard work.

Thank you, understood, so the difference arises as: "dynamic forecast" will take previously forecasted values while ""static forecast" will take actual values to make next step forecast, so if I'm forecasting only one period ahead, in this case out of sample forecasting doesn't make sense because the immediate previous value is an actual value and it is automatically an in sample forecast even if i chose dynamic forecast, is it correct?

But still I cant understand how it deals with the estimation sub-sample when I applied the whole sample as my estimation sub-sample!
I only want to know that what I did is it correct or not? I chose the whole sample as my estimation sub-sample and then I perform dynamic forecast just for one period ahead for a specific time within that sample? is it correct?

I appreciate if you reply me, thanks

[trubador]

Yes, in the dynamic case, one-step ahead forecast takes the forecasted value, whereas in the static case it takes the actual (realized) value. So, if you do not have observations for the forecast period you cannot perform static forecast even if you wanted to.

In your specific case, there is nothing technically wrong. From the practical point of view, however, it does not make any sense as you use the whole sample to estimate the parameters. If you want to treat a period (or sub-sample) as unknown and generate (dynamic) forecasts, then the parameter estimates should come from the sample period prior to it. Otherwise, you implicitly assume that new information have no impact on the parameters and that they are identical in both periods.

[ryan2993]

Hi

I was wondering if anyone could help with this problem?

I'm using a VAR model to estimate the relationship between share returns and its volatility that I have estimated using a GARCH(1,1) model. I can generate the return of the shares but I am unsure of how to get the GARCH(1,1) model into a time series of conditional variances to generate the VAR model.

Any advice would be much appreciated.

Thanks

[chigata]

Hi there,

I also need help with value at risk based garch (1,1) model for my dissertation.
I don't how to enter those 95% or 99% confidence level in Eviews!

HELP HELP.