ABSTRACT
Dynamic linear models (DLMs) arise via state-space formulation of standard time series models as already illustrated in Chapter 2 and also, as natural structures for modeling time series with nonstationary components. A review of the structure and statistical theory of basic normal DLMs is given here, with various special cases exemplified, followed by development of simulation methods for routine time series analysis within the DLM class. Most of the methods summarized here are based on the theory of West and Harrison (1997). Markov chain Monte Carlo methods for filtering in conditionally Gaussian dynamic linear models are also summarized and illustrated.
