The Dynamic Linear Model

Abstract

The first-order polynomial and simple regression models of the preceding two chapters illustrate many basic concepts and important features of the general class of Normal Dynamic Linear Models, referred to as Dynamic Linear Models (DLMs) when the normality is understood. This class of models is described and analysed here, providing a basis for the special cases in later chapters and for further generalisations to follow. The principles used by a Bayesian forecaster in structuring forecasting problems, as introduced in Section 1.3 of Chapter 1, are reaffirmed here. The approach of Bayesian forecasting and dynamic modelling comprises, fundamentally, (i) a sequential model definition; (ii) structuring using parametric models with meaningful parametrisation; (iii) probabilistic representation of information about parameters; (iv) forecasts derived as probability distributions.