Semiparametric Multivariate GARCH Models for Volatility Asymmetries and Dynamic Correlations

We propose a simple class of semiparametric multivariate GARCH models, allowing for asymmetric volatilities and time-varying conditional correlations. Estimates for time-varying conditional correlations are constructed by means of a convex combination of estimates for averaged correlations (across all assets) and dynamic realized (historical) correlations. Our model is very parsimonious. Estimation based on Functional Gradient Descent is computationally feasible also in very large dimensions without resorting to any variance reduction technique. We back-test the model on a six-dimensional exchange-rate time series and collect empirical evidence of its strong predictive power compared to other related existing procedures.