Break Detection for a Class of Nonlinear Time Series Models

• Activity Number: 8
• Type: Invited
• Date/Time: Sunday, July 29, 2007 : 2:00 PM to 3:50 PM
• Sponsor: International Indian Statistical Association
• Author(s): Richard A. Davis*+ and Thomas C.M. Lee and Gabriel Rodriguez-Yam
• Companies: Colorado State University and The Chinese University of Hong Kong and Universidad Autonoma Chapingo
• Keywords: nonlinear time series ; structural break ; MDL ; genetic algorithm

Abstract

We consider the problem of detecting break points for a broad class of nonstationary time series models. In this formulation, the number and locations of the break points are assumed to be unknown. Each segment of the data is assumed to follow a model from a family of nonlinear time series models. The minimum description length is used as a criterion for estimating the number of break points, the location of break points, and the parameters of the model in each segment. The best segmentation, found by minimizing MDL, is obtained using a genetic algorithm. The implementation of this approach is illustrated using a GARCH model, a stochastic volatility model, and a generalized state-space model as the base parametric model for the segments. Empirical results show the good performance of the estimates of the number of breaks and their locations for these various models.