A1335
Title: A nonparametric method for the detection of changepoints in multivariate time series
Authors: Moumanti Podder - Indian Institute of Science Education and Research Pune (India) [presenting]
Abstract: Changepoint analysis in time series data are focused on detecting distributional changes within the timeline. These are useful in a variety of fields, especially in handling financial data. Detection of changepoints in time series data has a vast existing literature of its own. However, these methods are often limited by strict assumptions made on the underlying distribution or by poor small sample performances in multivariate case. To that end, we propose a simple and computationally efficient test statistic for detecting changepoints, without making any assumptions on the distributional form of the underlying asset characteristics. Our test is based on a nonparametric approach aimed to quantify the difference in the dependence structure between the variables before and after the occurrence of the changepoint. It is easy to implement and has attractive asymptotic properties. We derive relevant theoretical results for the proposed test statistic and subsequently asses its small sample performance for a large class of data generating processes. Our simulation studies demonstrate that the proposed method is generally superior to many existing methodologies. We also present a real life application in this work.
