ISBIS4 Abstract

Contact Author's Name: Christian M. Hafner
Title of Abstract: Semi-Parametric Modelling of Correlation Dynamics
Author(s): Christian M. Hafner, Dick van Dijk and Philip Hans Franses
Affiliation: Erasmus University Rotterdam

Multivariate volatility models for financial asset returns usually contain many unknown parameters, which hampers their application to empirically relevant settings. The literature puts forward various models that somehow impose structure on these models in order to make estimation feasible. In this paper we argue that these structures often are too restrictive, and therefore we propose a new flexible semi-parametric variant. Our model uses parametric univariate GARCH-type specifications for the individual conditional volatilities, while the conditional correlation matrix is estimated nonparametrically by means of kernel regression. Application of our model to the daily returns of the 30 stocks included in the Dow Jones index demonstrates that our model improves upon many rival specifications, in particular in terms of out-of-sample forecasting performance.