|
Abstract:
|
For time series data, the variance may change with time. We consider situations where variance changes with time in a step function fashion. The goal is to detect if there are variance changes and where the change points are. This kind of variance change is studied by numerous authors (Wichern, Miller, and Hsu 1976, Hsu 1977, Tsay 1988, Inclan & Tiao 1994). Among existing detection methods, we did not find a satisfactory method. Some methods do very well if there is only one variance change. But when there are multiple variance changes, they do not perform very well any more due to masking effects. We propose a method that uses a moving window search criterion in an iterative procedure. We hope that this method can greatly reduce the masking effect because it is less likely to have more than one variance change in the same window. Simulation studies are done to compare several methods.
|
