CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Nuisance parameter free changepoint detection in non-stationary time series Authors:  Michal Pesta - Charles University, Faculty of Mathematics and Physics (Czech Republic)
Martin Wendler - University of Greifswald (Germany) [presenting]
Abstract: For short range dependent time series, a new ratio type changepoint test is proposed which does involve any estimation of the long run covariance nor the choice of any tuning parameters. The new test is not affected by heteroskedasticity. The asymptotic distribution is derived with the help of the continuous mapping theorem. We will also present a robustified version of our test statistic and compare it to previous proposals: the ratio test and the self-normalized test. Some simulation results on the finite sample performance will illustrate our findings.