Description:
This paper presents a new approach to the modeling of conditional correlation matrices within the multivariate GARCH framework. The procedure, which consists in breaking the matrix into the product of a sequence of matrices with desirable characteristics, in effect converts a highly dimensional and intractable optimization problem into a series of simple and feasible estimations. This in turn allows for richer parameterizations and complex functional forms for the single components. An empirical application involving the conditional second moments of 69 selected stocks from the NASDAQ100 shows how the new procedure results in strikingly accurate measures of the conditional correlations.