Title: A flexible regime-switching model for asset returns
Authors: Patrick Walker - University of Zurich (Switzerland) [presenting]
Marc Paolella - University of Zurich (Switzerland)
Pawel Polak - Swiss Finance Institute (Switzerland)
Abstract: A Markov regime-switching correlation model for a multivariate set of asset returns is proposed. The univariate series are endowed with the usual GARCH structure, and the underlying innovations are multivariate generalized hyperbolic distribution (MGHyp). The multivariate conditional predictive distribution is MGHyp, hence weighted sums of marginals are themselves GHyp and thus tractable, enabling, e.g., portfolio optimization. To accomplish joint likelihood estimation of all the model parameters, a new, fast and efficient two-stage EM-algorithm is developed for estimation. This is coupled with shrinkage estimation of the correlation matrices via a quasi-Bayesian prior, enhancing both estimation ease and forecast quality. Based on Dow Jones 30 data from 1999 to 2014, the new model is demonstrated to outperform all special cases in terms of in-sample fit and out-of-sample density forecasts. An application to portfolio optimization shows the importance of dynamical correlations for optimal asset allocation by providing consistently higher Sharpe ratios for all RSDC models, compared to their CCC counterparts.