Recursive identification, estimation and forecasting of non- stationary time series. Ph.D. thesis

Abstract

The thesis describes a new, fully recursive method for the identification, estimation and forecasting of non-stationary time series. This new approach is based on a step-wise decomposition of the time series data into its constituent components; and the separate identification and estimation of the signal generating models for these components. The various signal generating models are combined to yield an overall state-space model, which then provides the basis for forecasting using standard Kalman Filter (KF) methods. An "adaptive' forecasting method is subsequently developed based on these recursive estimation and forecasting procedures. -from Author