SMC2: an efficient algorithm for sequential analysis of state space models

• 成果类型: Article
• 署名作者: Chopin, N.; Jacob, P. E.; Papaspiliopoulos, O.
• 署名单位: Institut Polytechnique de Paris; ENSAE Paris; Institut Polytechnique de Paris; ENSAE Paris; Universite PSL; Universite Paris-Dauphine; ICREA; Pompeu Fabra University
• 刊物名称: JOURNAL OF THE ROYAL STATISTICAL SOCIETY SERIES B-STATISTICAL METHODOLOGY
• ISSN/ISSBN: 1369-7412
• DOI: 10.1111/j.1467-9868.2012.01046.x
• 发表日期: 2013
• 页码: 397-426
• 关键词: CENTRAL-LIMIT-THEOREM; MONTE-CARLO METHODS; stochastic volatility; Particle filters; inference; simulation

摘要

. We consider the generic problem of performing sequential Bayesian inference in a state space model with observation process y, state process x and fixed parameter . An idealized approach would be to apply the iterated batch importance sampling algorithm of Chopin. This is a sequential Monte Carlo algorithm in the -dimension, that samples values of , reweights iteratively these values by using the likelihood increments and rejuvenates the -particles through a resampling step and a Markov chain Monte Carlo update step. In state space models these likelihood increments are intractable in most cases, but they may be unbiasedly estimated by a particle filter in the x-dimension, for any fixed . This motivates the SMC2 algorithm that is proposed in the paper: a sequential Monte Carlo algorithm, defined in the -dimension, which propagates and resamples many particle filters in the x-dimension. The filters in the x-dimension are an example of the random weight particle filter. In contrast, the particle Markov chain Monte Carlo framework that has been developed by Andrieu and colleagues allows us to design appropriate Markov chain Monte Carlo rejuvenation steps. Thus, the -particles target the correct posterior distribution at each iteration t, despite the intractability of the likelihood increments. We explore the applicability of our algorithm in both sequential and non-sequential applications and consider various degrees of freedom, as for example increasing dynamically the number of x-particles. We contrast our approach with various competing methods, both conceptually and empirically through a detailed simulation study, and based on particularly challenging examples.