Multivariate stochastic volatility via Wishart processes : a continuation

This paper picks up on a model developed by Philipov and Glickman (2006) for modeling multivariate stochastic volatility via Wishart processes. MCMC simulation from the posterior distribution is employed to fit the model. However, erroneous mathematical transformations in the full conditionals cause false implementation of the approach. We adjust the model, upgrade the analysis and investigate the statistical properties of the estimators using an extensive Monte Carlo study. Employing a Gibbs sampler in combination with a Metropolis Hastings algorithm inference for the time-dependent covariance matrix is feasible with appropriate statistical properties. -- Bayesian time series ; stochastic covariance ; timevarying correlation ; Markov chain Monte Carlo