The persistence and asymmetry of time-varying correlations

Existing multivariate GARCH models either impose strong restrictions on the parameters or do not guarantee a well-defined (positive definite) covariance matrix. We focus on the multivariate GARCH model of Baba, Engle, Kraft and Kroner (BE=) and show that the covariance and correlation is not adequately specified. This implies that any analysis of the persistence and the asymmetry of the correlation is difficult and potentially biased. We illustrate this by the use of Monte-Carlo simulations for different correlation processes and propose a new Bivariate Dynamic Correlation (BDC) model that parameterizes the conditional correlation directly and eliminates the shortcomings of the BEKK model. Empirical results for correlations of the German stock market index with three international stock market indices reveal that correlations exhibit different degrees of persistence and different asymmetric reactions than variances. In addition, we find that correlations do not necessarily increase with variantes implying a justification for international portfolio diversification.