Bayesian Inference in State‐Space Time Series Models
Matt Sekerke · 2015
State-space models are introduced as a coherent, rigorous, and overarching framework for the elements of Bayesian inference discussed in previous chapters. The notion of a latent state space is introduced along with the classical problems of filtering and smoothing state estimates. Sequential estimation is then developed in a concrete way with dynamic linear models. Dynamic linear models offer a flexible strategy for implementing Kalman filtering and smoothing, a tool as fundamental to state-space time series analysis as the normal linear regression model is to cross-sectional data analysis. Component-wise construction of dynamic linear models is discussed, as well as the handling of parameter uncertainty, model uncertainty, and discounting.