The estimation of parametric change in time-series models

Abstract

This thesis examines methods for detecting structural change in parametric time-series models. This detection is accomplished through the use of random walk models of the parameter variation. Although the model of main interest is the transfer function models the methods developed are largely adaptations of procedures used for regression models as the exact theory for the time-series case is generally too complex. An instrumental variable smoothing algorithm for estimating parametric change is developed and is shown to provide good estimates of the variation. Other aspects of the procedure are also discussed^including the estimation of the statistics of the parameter variation. Finally some computer simulations and analyses of real data are provided. These illustrate some of the main points discussed in the thesis. Show more