Time varying covariance structures in financial markets

Reliable estimates of variances and covariances are crucial for portfolio management and risk controlling. This paper investigates alternative methods to estimate time varying variance-covariance matrices: ordinary estimates and exponentially weighted moving averages in comparison to Markov switching models. Different criteria are used to assess the quality of estimators. Bootstrapped confidence intervals for variances and correlations allow to compare the reliability of these estimates. Most importantly, we investigate the quality of these estimators directly in the context of portfolio applications and risk measurement. Therefore, we evaluate the performance of these estimation concepts on the basis of their suitability to select portfolios according to the minimum-variance criterion. This benchmark allows to infer the estimators' ability to capture diversification benefits without diluting this evaluation by any concept aiming to forecast returns. As an additional benchmark we evaluate the validity of forecasted confidence intervals.