2016/01/06 by Oliver Grothe, Grothe, Oliver, Tore Selland Kleppe +3
Decision Sciences · Economics, Econometrics and Finance · #Forecasting Techniques and Applications #Financial Risk and Volatility Modeling
paper · pdf · doi:10.48550/arxiv.1601.01125
We consider Particle Gibbs (PG) as a tool for Bayesian analysis of non-linear\nnon-Gaussian state-space models. PG is a Monte Carlo (MC) approximation of the\nstandard Gibbs procedure which uses sequential MC (SMC) importance sampling\ninside the Gibbs procedure to update the latent and potentially\nhigh-dimensional state trajectories. We propose to combine PG with a generic\nand easily implementable SMC approach known as Particle Efficient Importance\nSampling (PEIS). By using SMC importance sampling densities which are\napproximately fully globally adapted to the targeted density of the states,\nPEIS can substantially improve the mixing and the efficiency of the PG draws\nfrom the posterior of the states and the parameters relative to existing PG\nimplementations. The efficiency gains achieved by PEIS are illustrated in PG\napplications to a univariate stochastic volatility model for asset returns, a\nnon-Gaussian nonlinear local-level model for interest rates, and a multivariate\nstochastic volatility model for the realized covariance matrix of asset\nreturns.\n