A0507
Title: Efficient particle hybrid sampler for state-space models
Authors: David Gunawan - University of Wollongong (Australia) [presenting]
Robert Kohn - University of New South Wales (Australia)
Christopher K Carter - UNSW (Australia)
Abstract: Particle Markov Chain Monte Carlo (PMCMC) is a general approach to carry out Bayesian inference in non-linear and non-Gaussian state-space models. Our article shows how to scale up PMCMC in terms of the number of observations and parameters by expressing the target density of the PMCMCin terms of the basic uniform or standard normal random numbers, instead of the particles, used in the sequential Monte Carlo algorithm. Parameters that can be drawn efficiently conditional on the particles are generated by particle Gibbs. All the other parameters are drawn by conditioning on the basic uniform or standard normal random variables; e.g. parameters that are highly correlated with the states, or parameters whose generation is expensive when conditioning on the states. The performance of this hybrid sampler is investigated empirically by applying it to univariate and multivariate stochastic volatility models having both a large number of parameters and a large number of latent states and shows that it is much more efficient than competing PMCMC methods. We also show that the proposed hybrid sampler is ergodic.
