Particle methods are popular computational tools for Bayesian inference in nonlinear non-Gaussian state space models. For this class of models, we present two particle algorithms to compute the score vector and observed information matrix recursively. The first algorithm is implemented with computational complexity 𝒪(N) and the second with complexity 𝒪(N²), where N is the number of particles. Although cheaper, the performance of the 𝒪(N) method degrades quickly, as it relies on the approximation of a sequence of probability distributions whose dimension increases linearly with time. In particular, even under strong mixing assumptions, the variance of the estimates computed with the 𝒪(N) method increases at least quadratically in time. The more expensive 𝒪(N²) method relies on a nonstandard particle implementation and does not suffer from this rapid degradation. It is shown how both methods can be used to perform batch and recursive parameter estimation.
Biometrika is primarily a journal of statistics in which emphasis is placed on papers containing original theoretical contributions of direct or potential value in applications. From time to time, papers in bordering fields are published.
Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide. OUP is the world's largest university press with the widest global presence. It currently publishes more than 6,000 new publications a year, has offices in around fifty countries, and employs more than 5,500 people worldwide. It has become familiar to millions through a diverse publishing program that includes scholarly works in all academic disciplines, bibles, music, school and college textbooks, business books, dictionaries and reference books, and academic journals.
This item is part of JSTOR collection
For terms and use, please refer to our Terms and Conditions
Biometrika
© 2011 Biometrika Trust
Request Permissions
