报告题目：Rank based quasi-likelihood ratio scan estimate for multiple change-points

报告人：陈占寿（东北师范大学数学与统计学院，教授）

报告时间：2022年10月6日 14:00-15:30

报告地点：腾讯会议ID：274-885-744

校内联系人：李天然

报告摘要：In this talk, we show a rank based quasi-likelihoood ratio scan method to estimate multiple change-points in time series. The new method allows the inference sequences are heavy-tailed. We first applied a rank based quasi-likelihood ratio statistic searching potential change-points, and then estimate true change-points by optimize a loss function. We prove that the new method is consistency for multiple mean change-points in heavy-tailed independent sequences, and multiple structural change-points in heavy-tailed short memory time series. Simulations indicate that the new proposed method performs significantly better than many available methods in the literature when the inference sequences are heavy-tailed. Finally, we illustrate the proposed method by a set of Shanghai Securities Composite Index Yield and a set of Shenzhen Securities Component Index Yield.