Page 1 of 1
multivariate garch
Posted: Thu Jun 16, 2011 3:26 am
by mmkubilay
is it possible to estimate multivariate garch or egarch for volatility spillover by "make system"?
i have VAR equations as mean equations. microsoft ibm apple are the variables. i wanna estimate microsoft volatility by its own arch-garch terms and garch terms of ibm and apple. unfortunately i cannot find any solution to insert garch terms of apple and ibm. eviews restricts me to add these variables only in mean equation. thats also the case when the ibm and apple are dependent variable.
thanks for ur help;)
Re: multivariate garch
Posted: Thu Jun 16, 2011 8:14 am
by EViews Gareth
No, it is not possible to specify them in the variance equation.
Re: multivariate garch
Posted: Fri Jun 17, 2011 12:36 pm
by SnakeCharmerII
What about (1) estimating single GARCH models for each variable in the VAR vector, (2) generate a sigma variance (var, se OR log-var) series with the garch models, (3) add these sigmas to the VAR, as exogenous variables.
This would be like a VAR-with-GARCH-in-Mean.
Re: multivariate garch
Posted: Sat Jun 18, 2011 8:02 am
by mmkubilay
but i wanna see the spillover effect of these variances. so i need to insert them into conditional variance equation. for eg:
var1=c+var1(-1)+res1(-1)+var2(-1)+var3(-1)
Re: multivariate garch
Posted: Wed Jun 22, 2011 1:26 pm
by SnakeCharmerII
It seems that you can do things like that in RATS (I've also found some code in GAUSS), with VAR-GARCH models.
Re: multivariate garch
Posted: Wed Jun 22, 2011 11:27 pm
by trubador
It seems that you can do things like that in RATS (I've also found some code in GAUSS), with VAR-GARCH models.
You can do such things in EViews as well, but not as a built-in procedure. You have to use the LogL object along with some programming features of EViews. There are sample programs on similar issues in the "Program Repository" forum.
Re: multivariate garch
Posted: Fri Jun 24, 2011 2:52 pm
by SnakeCharmerII
It seems that you can do things like that in RATS (I've also found some code in GAUSS), with VAR-GARCH models.
You can do such things in EViews as well, but not as a built-in procedure. You have to use the LogL object along with some programming features of EViews. There are sample programs on similar issues in the "Program Repository" forum.
True. If you know the form of the (log) likelihood of your model, then you can do it in Eviews.