Time Structured Priors
Peter Congdon · 2019
The chapter considers schemes for modelling correlated observations and latent effects in time series. It explores autoregressive and state-space priors for time series analysis, and also considers state-space methods for discrete time series. The chapter discusses the Bayesian approaches to stochastic volatility and explains the models adaptive to temporal discontinuities. In Bayesian analyses, it is common to estimate parameters without presuming stationarity, but instead obtain the posterior probabilities of stationarity via monitoring the sampled parameters. Nonstationary models based on state-space priors are widely used in applications where time series parameters are evolving through time, especially in analysing separate unobserved components representing trend, cyclical, or seasonal effects. Aberrant observations or shifts in a series can bias parameter estimates and other inferences in time series models and a variety of methods exist for modelling shifts or outliers in the observation, state, or error series.