Detection of nonlinearity in time series disturbance models.

摘要

There are many diagnostic and statistical tests available in the literature that are used for the detection of nonlinearity in time series models. Many of these tests were developed from an econometrics perspective and are based on idealized time series model structures. The objective of this thesis was to develop a procedure for using these tests to (1) detect the presence of nonlinearity in non-ideal disturbance models, based on physical phenomena, and (2) screen data for candidate nonlinear model structures.; The empirical behaviour of the test statistics indicated that the tests do have value as screening tools for model structure, somewhat analogous to the use of the autocorrelation function in linear time series modelling. When the model structure of the generated data was nested within the alternative model of a given test statistic, the power of that test was comparable to the power of the test for which the alternative model was correctly specified. (Abstract shortened by UMI.)