A1096
Title: Detecting breaks in the group-structure of high-dimensional data
Authors: Stefano Soccorsi - Department of Economics, Lancaster University Management School (United Kingdom) [presenting]
Haeran Cho - University of Bristol (United Kingdom)
Abstract: The focus is on large panels of time series with common factors pervasive to all cross sectional units and group factors pervasive only within an unknown cluster of time series according to a group structure which is subject to multiple change points. We study the problem of estimating group memberships; extending previous results, we do so while allowing them to change over time in a piecewise-stationary framework. Consistent change-point detection, factor estimation and clustering are established, while in an empirical application on stock market data we show the usefulness of allowing for time variation in second order property of the data and their cluster memberships.
