EcoSta 2017: Start Registration
View Submission - EcoSta2017
A0797
Title: Bayesian semiparametric multivariate stochastic volatility Authors:  Martina Danielova Zaharieva - Erasmus University Rotterdam (Netherlands) [presenting]
Mark Trede - University of Muenster (Germany)
Wilfling Bernd - University of Muenster (Germany)
Abstract: The proposed framework is a Cholesky-type multivariate stochastic volatility model, in which the multivariate distribution of the error term is modeled as an infinite scale mixture of Gaussian distributions. A Bayesian non-parametric approach, in particular a Dirichlet process mixture, is adopted. This allows for high flexibility of the return distribution with respect to the kurtosis. Furthermore, the Cholesky decomposition allows for parallel univariate process modeling, creating potential for the estimation of higher dimensional models. Markov Chain Monte Carlo methods are applied for the posterior simulation and the computation of the predictive density. Finally, a five-dimensional stock market application is presented.