The Variance Implied Conditional Correlation

Abstract

We use univariate GARCH estimations to construct a computationally simple and flexible filter to predict the time-varying correlation matrix of asset returns. The proposed Variance Implied Conditional Correlation (VICC) filter exploits the polarization result that links the correlation between two variables with variances of linear combinations of their standardized values. In a Monte Carlo study, we show that it yields accurate correlation estimates for common choices of correlation dynamics. We further find in an empirical cross hedging application that the use of the VICC leads to substantially more stable hedge ratios which perform at least as well as the standard benchmark models in terms of variance reduction and achieved decorrelation.

Keywords: Conditional Correlation, GARCH, DCC, Cross Hedging, Hedge Ratio, Regularization

JEL Classification: C10, G11