Econometric Measures of Connectedness and Systemic Risk in the Finance and Insurance Sectors
Source: Billio, Getmansky, Lo & Pelizzon (2012) · Journal of Financial Economics 104(3), 535–559 · doi:10.1016/j.jfineco.2011.12.010
TL;DR
Proposes return-based econometric measures of connectedness among hedge funds, banks, broker/dealers, and insurers, built from principal-component analysis and Granger-causality networks. All four sectors became highly interrelated over the prior decade, raising systemic risk; the measures identify crisis periods, show out-of-sample predictive power, and reveal an asymmetry in which banks transmit shocks more than the other three sectors.
Problem it solves
Systemic risk is hard to define and balance-sheet/exposure data are often unavailable or lagged. The paper supplies model-light, market-data-only gauges of how tightly the financial system is linked and which institutions transmit shocks — usable for surveillance even without proprietary regulatory data.
The method
Principal-components analysis (PCA). The number and importance of common factors driving the institutions' returns measures commonality; rising variance explained by the top components signals higher connectedness. A derived statistic PCAS measures an institution's contribution to system risk conditional on the system being under stress.
Granger-causality networks. Pairwise linear Granger-causality tests (and, separately, nonlinear Granger-causality tests using a Markov-switching/regime approach to capture tail and nonlinear effects) determine directed links; aggregate network statistics (number of connections, degree of Granger causality, centrality) summarize systemic linkage and direction of transmission.
Assumptions & inputs
Monthly returns, 36-month rolling estimation windows. Hedge fund returns from the CS/Tremont database, January 1994 – December 2008 (asset-weighted indices and individual funds); banks, broker/dealers, and insurers from CRSP via SIC codes (6000–6199 banks, 6200–6299 broker/dealers, insurers).
The 25 largest institutions per sector are used for the individual-firm network analysis (largest by AUM for hedge funds, market cap otherwise).
