統計学輪講 第01回

An efficient particle Markov chain Monte Carlo methodology is proposed for the rolling-window estimation of state space models. The particles are updated to approximate the long sequence of posterior distributions as we move the estimation window. To overcome the well-known weight degeneracy problem that causes the poor approximation, we introduce a practical double-block samplerwith the conditional sequential Monte Carlo update where we choose one lineage from multiple candidates for the set of current state variables. Our proposed sampler is justified in the augmented space through theoretical discussions. In the illustrative examples, it is shown to be successful to accurately estimate the posterior distributions of the model parameters. This is a joint work with Naoki Awaya.