Variance Clustering Improved Dynamic Conditional Correlation MGARCH Estimators

Abstract

It is well-known that the estimated GARCH dynamics exhibit common patterns. Starting from this fact we extend the Dynamic Conditional Correlation (DCC) model by allowing for a clustering structure of the univariate GARCH parameters. The model can be estimated in two steps, the first devoted to the clustering structure, and the second focusing on correlation parameters. Differently from the traditional two-step DCC estimation, we get large system feasibility of the joint estimation of the whole set of model's parameters. We also present a new approach to the clustering of GARCH processes, which embeds the asymptotic properties of the univariate quasi-maximum-likelihood GARCH estimators into a Gaussian mixture clustering algorithm. Unlike other GARCH clustering techniques, our method logically leads to the selection of the optimal number of clusters.

Keywords: dynamic conditional correlations, time series clustering, multivariate GARCH, composite likelihood

JEL Classification: C32, C38, C53, C51, C52, C58