Univariate Time Series Modelling and Forecasting using TSMARS

This monograph examines nonlinear threshold time series models using TSMARS, a time series extension of the Multivariate Adaptive Regression Splines (MARS). MARS is model free and can detect and measure linear and curvilinear structure in data. Novel aspects include applications to Ireland's Trade Statistics and the introduction of regime dependent threshold seasonal time series models - the effect of seasonal adjustment in the presenence of a threshold is examined using these models. Two important new advances are incorporated into TSMARS. The first allows TSMARS to automatically treat ordinary and dynamic outliers. The second is a new procedure to estimate treshold moving average models within TSMARS. Both of these advances are described, implemented in SAS/IML, tested and results are reported. Finally, parametric and nonparametric bootstrapped procedures are described and the forecasts investigated.

Autorentext

I am a mathematician/statistician working in asylum and immigration statistics. My research interests include nonlinear time series analysis, probability and statistical models using parametric, bayesian and non-parametric methods.