ABSTRACT
This chapter concerns the class of autoregressive models with time-varying parameters, or TVAR models, that are widely used in empirical analysis of time series and short-term forecasting. They extend AR models to a broader class useful in describing nonstationary time series. One main focus is on the underlying structure in time series that TVAR models can identify, especially related to quasiperiodic components. Such data arise in many areas of applications, such as biomedical monitoring, speech signal processing, climatological studies, and econometrics. TVAR decompositions allow us to partition a time series into a collection of processes that are often scientifically meaningful, defining a statistically sound approach to time-frequency analysis. TVAR models are special classes of DLMs, so the theory of Chapter 4 applies for estimation, forecasting, smoothing, and inference on latent structure in nonstationary time series.
