Abstract

We discuss a problem of an online changepoint detection procedure for instabilities within a nonlinear expectile model. The underlying model is firstly trained using a historical dataset and, later, possible structural disorders of different types are detected in terms of a sequential statistical test. Resampling techniques are used to avoid plug-in methods when obtaining the limiting distribution of the test statistic. Theoretical properties are presented and finite sample empirical performance and real data applications are provided for a brier illustration.