Title: Identifying coordinates with change in high-dimensional panel data using tail-summed scores
Authors: Anica Kostic - London School of Economics and Political Science (United Kingdom) [presenting]
Piotr Fryzlewicz - London School of Economics (United Kingdom)
Abstract: Detection of possibly aligned change-points is considered in the high-dimensional mean-shift model. The interest is not only in the detection of the number and locations of change-points, but also in determining which panel components have undergone a change at a given time. As changes in some components can be insignificant when considered individually, we propose a new iterative multiple testing and signal discovery procedure, which considers components in groups and uses tail-summed scores. We show its attractive properties both in theory and in practice, with particular advantages over the state of the art when the mean changes are small but dense across the panel.