A0817
Title: Adaptive detection of change-points for high-dimensional covariance matrices
Authors: Xiaoyi Wang - Beijing Normal University at Zhuhai (China) [presenting]
Abstract: A series of tests are built based on U-statistics to test the high-dimensional covariance matrix change points within the temporal independence assumption. The asymptotic distributions of the constructed U-statistics are derived under the null and local alternative hypotheses. Then, a family of maximum-type statistics is proposed, after which two test methods are developed based on the combination of the p-values of these maximum-type statistics. Some methods are also proposed to estimate the location of the change-point and obtain their corresponding convergence rates. In addition, three new adaptive estimations are built. Finally, the binary segmentation method is proposed to be combined with the three adaptive estimators to detect multiple change-points. The simulation study shows that the proposed test methods can maintain high power under alternatives with different sparsity levels. The proposed adaptive estimators perform well under different alternatives with both single and multiple change-points.
