Model-Based Clustering of Time Series Based on State Space Generative Models

Abstract

Recently refined Markov Chain Monte Carlo techniques for Bayesian inference are combined with the elegant and computationally advantageous specification of state space models to develop and evaluate an approach for the clustering of time series of fixed-income financial instruments. This approach is based upon the specification and estimation of a finite mixture model where each mixture component is represented by a time series generative model that is specified in linear state-space form.