A data augmentation scheme embedding a sequential Monte Carlo method for Bayesian parameter inference in state space models

Irene Votsi (1) , Paul-Henry Cournède (1)

1 MICS - Mathématiques et Informatique pour la Complexité et les Systèmes

State space models (SSMs) are successfully used in many areas of science to describe time series and/or dynamical systems. In this work, we revisit the data augmentation algorithm introduced by Tanner and Wong (1987) for bayesian parameter estimation in SSMs. We propose to employ sequential Monte-Carlo and adaptive Monte-Carlo Markov chain methods to improve the performance of the algorithm. We provide a first numerical example that allows us to evaluate the convergence of the posterior estimate to the true posterior distribution. Our objective is to evaluate the performance of the proposed method to nonlinear/non-Gaussian models, with a special interest to plant growth models.

Mots clés: Data augmentation; state-space models; sequential Monte Carlo methods

Domaines: Statistiques [stat]

Irene Votsi, Paul-Henry Cournède. A data augmentation scheme embedding a sequential Monte Carlo method for Bayesian parameter inference in state space models. 48èmes Journées de Statistique de la SFdS, May 2016, Montpellier, France. ⟨hal-01355334⟩