Self-exciting threshold autoregressive models for nonlinear time series

Abstract

This dissertation focuses on nonlinear time series methods. These have become increasingly popular since data in many sectors cannot be suitably modelled using linear time series models. Regime switching models are introduced, highlighting the Threshold Autoregressive Model and the various forms this model can undertake. Moreover, it discusses nonlinearity tests, with special reference to threshold nonlinearity. A simulation study was performed to study the efficiency of the estimation procedure proposed. An illustration of this model is also given using a real time series and the forecasting accuracy of the estimated model was analysed.