EcoSta 2017: Start Registration
View Submission - EcoSta2017
A0806
Title: MCD meets MCD: Minimum Covariance Determinant Methods with Mixed Correlation Dynamics Authors:  Marco Gambacciani - University of Zurich and Swiss Financial Institute (Switzerland) [presenting]
Marc Paolella - University of Zurich (Switzerland)
Abstract: A new model class for large-scale multivariate financial asset returns is proposed and studied. It expands upon recent work using the multivariate normal mixture model, estimated using the minimum covariance determinant (MCD), in two ways. First, the multivariate Laplace is used in place of the normal, and its benefit is shown both in terms of density forecasting and portfolio performance. Second, the model is extended to support a DCC-GARCH type of structure, whereby both components in the mixture are endowed with a DCC law of motion for the two time-varying covariance matrices. This is accomplished by using the MCD fit to generate posterior probabilities associated with each component, at each point in time.