Determining the number of breaks in a piecewise linear regression model

In this paper we propose a sequential method for determining the number of breaks in piecewise linear structural break models. An advantage of the method is that it is based on standard statistical inference. Tests available for testing linearity against switching regression type nonlinearity are applied sequentially to determine the number of regimes in the structural break model. A simulation study is performed in order to investigate the finite-sample behaviour of the procedure and to compare it with other alternatives. We find that our method works well in comparison for both single and multiple break cases.

Introduction: Models with structural breaks (SB) have been of interest to many researchers for at least the last four decades. Most of the work in this area of research has been related to the case of detecting and estimating a single break. See Chow (1960), Andrews (1993), and Bai, Lumsdaine, and Stock (1998), among others. The questions related to multiple structural changes have received somewhat less attention. Early works include Yao (1988) and Liu, Wu, and Zidek (1997) who advocated the use of the (modified) Bayesian Information Criterion and showed that the number of breaks can be estimated consistently (at least for a normal sequence of random variables with shifts in mean).

Author: Birgit Strikholm

Source: The Economic Research Institute

Download URL 2: Visit Now