Sequential Monte Carlo methods for filtering and smoothing in hidden Markov models

Summary form only given. In the statistical analysis of time series and stochastic dynamic systems, data often arrive sequentially over time, and sequential importance sampling (SIS) provides a natural framework for performing Monte Carlo computation sequentially to update estimates and posterior distributions. In real applications, there are typically also unknown parameters in the dynamic systems. We explain how SIS with resampling can be used to address these problems, and in particular apply SIS to an important class of hidden Markov models, namely, change-point autoregression models.