Skip to content
Prev 2384 / 15274 Next

Garch and multivariate garch

The Bauwens survey paper in the Journal of Applied Econometrics is very good, but it lacks any practical examples of actually estimating multivariate garch models. In fact, I don't know of any survey paper that discusses the real life practical issues of estimating multivariate garch models. A good intuitive guide to forecasting correlation is in Carol alexander's book Market Models. Unfortunately, she spends a lot of time plugging her orthogonal garch model which is rather problematic in practice. A big problem in assessing the accuracy of estimating something like a conditional correlation is that the "true" time varying correlation is not observable so that the output of a multivariate garch model cannot be compared to an "actual" correlation to assess forecasting accuracy. One can try to follow Andersen and Bollerslev and try to compare a garch correlation forecast to a realized correlation 
computed from high frequency data. However, this is also fraught with problems as it is not clear how one is supposed to compute realized correlation. I have a paper under preparation that looks at evaluating multivariate garch models using realized correlation. Unfortunately, none of the typical multivariate garch models work very well - especially the dynamic conditional correlation model of Engle. This is often the worst model! I have a reference list of some empirical multivariate garch papers and I'll post some references later this week.


****************************************************************
*  Eric Zivot                  			               *
*  Professor and Gary Waterman Distinguished Scholar           *
*  Department of Economics                                     *
*  Box 353330                  email:  ezivot at u.washington.edu *
*  University of Washington    phone:  206-543-6715            *
*  Seattle, WA 98195-3330                                      *                                                           *
*  www:  http://faculty.washington.edu/ezivot                  *
****************************************************************
On Sun, 13 Apr 2008, michal miklovic wrote: