Title: Simultaneous change-point and factor analysis for high-dimensional time series
Authors: Matteo Barigozzi - London School of Economics (United Kingdom) [presenting]
Haeran Cho - University of Bristol (United Kingdom)
Piotr Fryzlewicz - London School of Economics (United Kingdom)
Abstract: A method is proposed for simultaneously analysing the factor structure of the data and detecting (possibly) multiple change-points in high-dimensional time series. Firstly, we introduce a piecewise stationary factor model that enables introducing and, consequently, detecting changes not only in loadings but also in factors and idiosyncratic component, which has not been explored in the existing literature. Next, it is shown that the common component estimated with an over-estimated factor number achieves consistency, which motivates our change-point detection methodology. Then, we propose to transform the data so that an existing panel data segmentation method is applicable to the problem of detecting multiple change-points in the factor structure, and consistency of such an approach is established in terms of the total number and locations of estimated change-points. Empirical performance of the proposed method is investigated on simulated datasets as well as macroeconomic and financial time series.