A Nonparametric Procedure for Change Point Detection in Linear Regression

• Activity Number: 124
• Type: Contributed
• Date/Time: Monday, August 1, 2016 : 8:30 AM to 10:20 AM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Sunil Mathur and Jing Sun* and Deepak Sakate
• Companies: Georgia Regents University and Augusta University and Augusta University
• Keywords: Least-square estimators ; rank regression ; two-phase linear regression ; F-test ; nonparametric ; non-normal errors

Abstract

Change point detection in linear regression has many applications in climatology, bioinformatics, finance, oceanography and medical imaging. There are some parametric methods to locate change point in linear regression in the literature. Lund and Reeves (2002) proposed a procedure based on F-test for detecting change points in two phase linear regression model assuming normal distribution for error. The F-test is based on the least squares estimator which is optimal under the normality of errors. In this article, we develop a procedure to detect change point in linear regression based on a non- parametric test in McKean and Hettmansperger (1976). The proposed procedure is intended to perform well for non-normal error distribution and is non parametric in nature. A simulation study to compare the performance of the proposed procedure with the procedure in Lund and Reeves (2002) is conducted for Laplace, Student's t, Slash and Cauchy distributions for error. The results of the simulation study reveal that the proposed procedure outperforms its competitor.