Title: Dynamic change-detection with application to mutual fund selection
Authors: Lilun Du - HKUST (Hong Kong) [presenting]
Changliang Zou - Nankai University (China)
Abstract: In the era of big data, it has become particularly important to rapidly and sequentially identify individuals whose behavior deviates from the norm. In such applications, the state of a stream can alternate, possibly multiple times, between a null and an alternative state. Aiming to balance the ability to detect two types of changes, i.e., a change from the null to the alternative and back to the null, we develop a large-scale dynamic testing system in the framework of false discovery rate (FDR) control. By fully exploiting the sequential feature of datastreams, we propose a new procedure based on a penalized version of the generalized likelihood ratio test statistics for change-detection. The FDR at each time point is shown to be controlled under some mild conditions on the dependence structure of datastreams. A data-driven approach for choosing the penalization parameter is developed, giving the new method an edge over existing methods in terms of FDR control and detection delay. Its advantage is demonstrated using a real data example in Chinese financial market.