Obtaining tstats and forecasting with old GARCH instruction

[jimdimit]

Dear all,

I run a multivariate GARCH-X and I use the old GARCH code for a BEKK GARCH with nonlin, and frml commands for the parameters.

The output of the GARCH estimation looks like this
Variable Coeff Std Error T-Stat Signif
************************************************************************************
1. C11 -0.000000937 0.000000291 -3.21775 0.00129200
2. C12 0.483887134 0.099075640 4.88402 0.00000104
3. C13 0.413337955 0.146033298 2.83044 0.00464846
...
12. VB11 -1.797865109 0.375255458 -4.79104 0.00000166

etc.

The issue is that I want to obtain the t-statistic for say variable VB11 and when I give
compute ts = %TSTATS(12)
disp ts

I get ##undefined##.

Is there any way to 'call' the t-statistic when using the old GARCH code?

Additionally, is there any documentation-files on the old GARCH codes so that can I see how to forecast the variances?

Thank you,

Dimitris

[TomDoan]

That's how you would do it. You need to use MAXIMIZE with either METHOD=BFGS or METHOD=BHHH; METHOD=SIMPLEX doesn't compute standard errors (PMETHOD=SIMPLEX is OK, as long as you also use one of the others as the final method). And you need to get that right after the MAXIMIZE. If, for instance, you do a LINREG or do @MVARCHTEST or some other procedure that includes a LINREG after the MAXIMIZE, you'll re-define the %TSTATS vector.

[jimdimit]

Thank you very much for your prompt reply, I was using Simplex which I shouldn't as I final method.
Do you have reading suggestions on forecasting with the old code?

[TomDoan]

I don't have any examples of that for multivariate models. However, the forecasts can be done by using the same recursion as is used in sample---you just have to feed in the variance in place of the lagged outer product of the U's once the latter are out of sample.

[jimdimit]

Thank you again Tom I will try to do it that way