Title: Advances in Bayesian computations with application to stochastic volatility models
Authors: Yuliya Shapovalova - Maastricht University (Netherlands) [presenting]
Abstract: Stochastic volatility models (SVM) belong to the class of state-space models. Volatility in this framework is an independent latent process as opposed to GARCH-type models. The presence of the latent structure makes the estimation challenging. It has been documented in the literature that standard techniques such as Quasi-Maximum Likelihood and General Method of Moments do not perform well. Particle filters have been considered as the most promising solution for quite long time, intuitively clear implementation and good performance of simulated likelihood makes this procedure appealing. However, it is computationally very demanding. For practitioners speed of estimation might be crucial and hence, methods that might work faster are of interest. We consider a few other methods in Bayesian framework, namely Approximate Bayesian Computation (ABC), Variational Bayes, Expectation Propagation algorithm and particle MCMC. We discuss advantages and disadvantages of these methods in terms of the quality of estimation and computational speed in univariate stochastic volatility models. Further, we discuss possibilities for multivariate extensions.