论文部分内容阅读
In this paper,serial autoregression rank score statistics are introduced as a measure of the serial dependence of residuals.Based on these statistics,non-parametric tests are constructed for testing threshold nonlinearity of autoregressive models and these tests can be used in detecting structural change of an underlying process.Compared with existing methods,such as portmanteau tests and other tests based on predictive residuals of least square estimators,such tests are more robust to outliers in time series and they are asymptotically distribution free.Moreover,they do not require estimation of nuisance parameters.Simulation study is given to demonstrate the performance of the tests.