A0622
Title: High-dimensional change-point detection using generalized homogeneity metrics
Authors: Runmin Wang - Texas A&M University (United States) [presenting]
Xianyang Zhang - Texas A\&M University (United States)
Shubhadeep Chakraborty - Bristol Myers Squibb Company (United States)
Abstract: The purpose is to study the problem of detecting abrupt changes in the data-generating distributions of a sequence of high-dimensional observations beyond the first two moments. This problem has remained substantially less explored in the existing literature, especially in the high-dimensional context, compared to detecting changes in the mean or the covariance structure. A nonparametric methodology is developed to (i) test the existence of a change-point and (ii) identify the change-point locations in an independent sequence of high-dimensional observations. The approach rests upon recent nonparametric tests for the homogeneity of two high-dimensional distributions. A single change-point test statistic is constructed based on a cumulative sum process in an embedded Hilbert space. Its limiting null distribution is derived, and the asymptotic consistency is presented under the high dimension medium sample size framework. The statistics are also combined with wild binary segmentation to recursively estimate and test for multiple change-point locations. The superior performance of the methodology compared to other existing procedures is illustrated via extensive simulation studies and the application to the stock return data observed during the period of the global financial crisis in the United States.
