Multiscale Factor Modelling of High Dimensional Locally Stationary Time Series

Details

Supervisors

Van Bellegem, Sébastien

Faculty

Faculté des sciences économiques, sociales, politiques et de communication

Degree label

Master [120] en sciences économiques, orientation économétrie, à finalité approfondie

Abstract

Extracting relevant information on the immense amount of time series data is paramount in contemporary multivariate analysis. Furthermore, those time series are usually non-stationary, serially and cross-correlated. Economic shocks and earnings announcements generally induce such series behaviours. In this paper, we provide a factor model based on the wavelet spectral representation of time series that allow us to handle both high dimensionality data and nonstationarity. On the one hand, the very essence of factor models is to encapsulate the variability of many variables in just a few common factors. On the other hand, Wavelets have been extensively used to capture abrupt changes and non-stationarities while allowing a sparser representation of data. We first define the model before developing an estimator for the time-varying factor loadings. We also report an empirical application.