Network Modeling of Extreme Dependence in High-Dimensional Financial Time Series

Abstract

Modelling the systemic risk of the ever-growing global financial market has garnered significant interest since multiple worldwide crises showed that contagion of financial distress could occur on a system-wide scale. In this thesis, We study the phenomenon of extreme dependence, also known as financial contagion, in which extreme events in financial time series data move together and propagate distress. To model the interconnected relationships of a large market, where the dimension is high, we employ network-based methods. In particular, we construct an original framework based on the theory of generalized linear models to detect contagion by quantifying the amount of co-movement of extreme events. We employ the LASSO to perform a simultaneous estimation of the links of a specific instrument to the rest of the market. Finally, we use the estimates to represent the market as a network and derive summary statistics that illustrate relevant facets of the underlying complex network, such as the mean connectivity of the network. We test the validity of this method by comparing the time series of these summary statistics against known crisis periods.