Sequential Change-Point Detection in Time Series: An Adjusted-Range-Based Self-Normalization Approach
Date:
This talk discusses Sequential Change-Point Detection in Time Series: An Adjusted-Range-Based Self-Normalization Approach, presented at an academic seminar at Renmin University of China.
Citation: Jiajing Sun. (2025). “Sequential Change-Point Detection in Time Series: An Adjusted-Range-Based Self-Normalization Approach.” Presented at Renmin University of China, Beijing, China.
Abstract:
The ability to update models in real-time to reflect the evolving scope of real-world data is a fundamental task in statistics. Existing cumulative sum (CUSUM)-type procedures need to specify tuning parameters such as kernel, bandwidth, or block size in block bootstrap when estimating the long-run variance (LRV). The weak power of the KS-type statistic using the existing self-normalization (SN) method proposed by Shao (2010) limits its use in sequential change-point detection. This paper proposes a novel adjusted-range self-normalized sequential change-point monitoring scheme. We conduct asymptotic analysis under the null hypothesis and establish the consistency of the proposed sequential change-point scheme under weak regularity conditions. Through Monte Carlo simulations and empirical analysis, we find that our proposed method can sequentially detect structural changes in a timely fashion and is robust to “mild” misspecification in the training sample.