Title: Bayesian nonparametric Bernstein copula for uncorrelated dependent MGARCH errors
Authors: Martina Danielova Zaharieva - Erasmus University Rotterdam (Netherlands) [presenting]
Concepcion Ausin - Universidad Carlos III de Madrid (Spain)
Abstract: The proposed model is a Copula-MGARCH, in which the dependency structure of the uncorrelated error term is modeled by a nonparametric copula. The idea of the dependent but uncorrelated MGARCH error term is extended by introducing a Bayesian Bernstein copula based on a Dirichlet process mixture. Hence, the time-varying dependence structure is modeled trough the dynamic linear correlation captured by the MGARCH model and the remaining (nonlinear) dependence trough the flexible nonparametric Bernstein copula, for which we introduce a stick-breaking representation model. We design a full Bayesian MCMC algorithm and provide a simulation study. Finally, we propose an application to portfolio optimization.