Rolling-VAR-MGARCH-M

[hasanov]

Thank you for your reply!
What if I remove the "M" from the model?
How can I achieve the forecast computations through the loop?
Regards

[TomDoan]

If the mean model is just a VAR, you can forecast the mean using the regular FORECAST instruction.

[hasanov]

Thank you!
I will try with FORECAST instruction then.

[hasanov]

Hi Tom,

What is the purpose of writing the variable names (e.g., rvar1, rvar2, rvar3 in the code chunk below) right after the garch instruction? Does it have any effect on the computations?

garch(model=varmah, mv=bekk, robust, distrib=norm, rvectors=rv, hmatrices=hh, mvhseries=hhs, pmethod=simplex, piters=15, cvcrit=0.0001, iters=1000, presample=cv0) regend-d+1 regend rvar1 rvar2 rvar3

Thank you

[TomDoan]

If you use a MODEL option, the list of series is ignored, so they do nothing.

[hasanov]

Thank you! This is what I was thinking.

[hasanov]

Hi Tom,
Would it be possible to combine threshold VAR (with multiple breaks) with the BEKK variance-covariance? My intention is to capture the different regimes.
Any idea is appreciated.
Thank you

[TomDoan]

Is it possible? Yes. Is it likely to work very well? I doubt it. In a regular threshold VAR, the likelihood is easy to compute given a test value of the threshold, because all that changes are the subsamples for a VAR---changing the regime at entry t has no effect on the likelihood function for t-1 or t+1. All that is required is to run the different subsample regressions and find the break(s) that maximize the likelihood. If you add to that a GARCH process for the likelihood, it's completely different. Change the regime at t, and you change the GARCH process for t+1, t+2, ..., possibly substantially, and each different set of regime breaks will require estimating a full fledged GARCH model (with a complex mean process). The only feasible way I can think for doing that is to estimate the break points using the standard threshold VAR methods and then take the residuals from that as input for the GARCH model. Of course, if what you're really interested in is the regime breaks, that wouldn't really help, since the "GARCH" part of the analysis would have no effect on that.

[hasanov]

Thanks for being a big help!