Out-of-Sample estimation Garch(1,1) issues

[magnusmj]

Hi everyone,

I would be very grateful for any help on this matter. My dataset is 2514 observations on the s&p500 and basically I want to estimate an out of sample conditional variance for a Garch (1,1) model. I have estimated the model based on the first 1999 observations and want the conditional variance for the remaining 513 observations (I want to compare the forecasted set to the observed values for value at risk implications).

From what I have understood, I can go on the forecast tab (static, and put in the last 513 observations for range) to get the forecasted conditional volatility based on the first 1999 observations. I am interested to get these forecasted values into an excel spreadsheet, however, when I go on proc -- "make Garch variance series" i get "NA" for the values i want to forecast....

Can someone please help me/point me in the right direction? Any help would be very much appreciated.

I have attached the workfile if that helps.

[EViews Gareth]

To get the forecasted values you have to use Proc->Forecast.

[magnusmj]

To get the forecasted values you have to use Proc->Forecast.

Hi Gareth, thanks for helping me out.
This is what I have done though - after estimating the garch(1,1) model I go on proc - forecast and select all observations such that I get estimated alues for the ones I did not estimate the model on. However, I still get NA on the out of sample estimations.
This is what I did:
1. First estimated the Garch(1,1) based on the first 2000 observations.
2. When the model was estimated and I had the overview over the estimates etc I went on Proc then on forecast.

In the forecast:
Forecast name: returnf
Do I need to insert anything for "Garch (optional)"?
I select all observations (1-2514) and choose static.

I get the graph - however, I want the actual values. I then go on "make GARCH variance series" in the proc tab right?
However, here I only get the values for the estimated sample, not the out of sample estimations (just says NA for the last 514 variables).


Any idea??
I really appreciate your help!

Cheers!

I re

[EViews Gareth]

Yes, you need to enter something in the GARCH (optional) bit.

[magnusmj]

Yes, you need to enter something in the GARCH (optional) bit.

You said I must enter something for the Garch space, would this be the returns (residuals) of the data?
I have tried entering the returns data there, but still get NA for the out of sample observations... Why is this?
Am I using the "Garch (Optional)" wrong?

[EViews Gareth]

Just enter the name you want the forecast of the GARCH series to have

[magnusmj]

Just enter the name you want the forecast of the GARCH series to have

Thanks Garth - Works now. When Eviews is forecasting like this, is it using a 1-day rolling forecast based on the same initial parameters?

Thank you!