Title: Detection performance of likelihood ratio test for change-points based on bootstrap for AR(1) models
Authors: Ceyda Yazici - Middle East Technical University (Turkey) [presenting]
Ceylan Yozgatligil - Middle East Technical University (Turkey)
Inci Batmaz - Middle East Technical University (Turkey)
Abstract: The detection of change-points in time series is an important issue especially in economics, finance, meteorology and energy. Change in mean, change in variance or any sudden increase or decrease in the series can cause breakpoints. In AR(1) models, the likelihood ratio test is conducted to test for a single breakpoint. However, if the sample size is small or the location of the breakpoint is close to the end or the beginning of the series, the detection performance becomes worse. In order to increase the correct detection percentage of the likelihood ratio test in these cases, a bootstrap method for dependent data is applied and its performance is investigated when the change is only in the mean under several breakpoint scenarios. The test is applied to simulated data and the results are compared with the results obtained from tests in the literature.