Page 1 of 1

Regarding multivariate VAR-GARCH model

Posted: Mon Aug 19, 2019 10:41 am
by Owen Wang
Dear Sir,

I’m a beginner. Could you help me out with the questions as follows:
(1) How to estimate multivariate VAR-GARCH model (generalized) and multivariate VAR-GARCH model (Cholesky)? May I have the code?

(2) Is it possible to obtain impulse response function (in both table and figure) and forecast error variance decomposition after estimation of multivariate VAR-GARCH model? If yes, may I have the code?

Any reply will be appreciated.

Have a nice day.

Owen Wang

Re: Regarding multivariate VAR-GARCH model

Posted: Mon Aug 19, 2019 11:27 am
by TomDoan
I'm not sure what you mean by "generalized" and "Cholesky". A VAR for a mean model is handled by defining a VAR using the SYSTEM commands as described in the User's Guide. The multivariate GARCH model can take pretty much any form---there's no real connection between the mean and variance models.

In general, there is no FEVD for a VAR-GARCH model because the GARCH model breaks the assumption used in computing an FEVD that the shocks are the same sizes. IMPULSE responses, however, can be computed using the IMPULSE instruction, since they depend only upon a one-off set of shocks, however, there is no standard set of shocks since the covariance matrix changes.