Work

Composite likelihood inference in state space models

Year: 2010
Type: dissertation
Abstract: In general state space models, where the computational effort required in the evaluation of the full likelihood function is infeasible, we analyze the problem of static parameter estimation based on composite likelihood functions, in particular pairwise and split data likelihood functions. We discuss consistency and efficiency properties of these estimators (related to the characteristics of the model) and the bias in stationary models where the invariant distribution is unknown. We focus on numerical methods to compute estimates of the parameter describing a general state space model. We develop an on line Expectation- Maximization algorithm in order to obtain the maximum pairwise likelihood estimate in a general state space framework. We illustrate this method for a linear gaussian model and we extend it to make inference also in jump Markov linear systems. In this framework, some sampling procedures need to be developed to estimate the parameters of the model. In particular, we present an algorithm to sample from the latent discrete state Markov chain given the pairs of observations.
Author: Nadia Frigo
Language: English
Open Access status: green
OpenAlex ID: W2276430096
Cites: 25
Cited by: 1
Related to: 10