Stress Testing the Unknown - The Impact of Network Reconstruction on Systemic Risk Estimates

Abstract

Banks connect through a complex network of interbank credit exposures that determine contagion and hence the systemic stability of the network. Yet, the bilateral linkages of these institutions and thus the true structure of the network are typically unobservable, with leading reconstruction methods, such as maximum entropy or minimum density, potentially underestimating the true contagion risk of the network. Based on publicly available data from large European institutions, this paper analyses different network reconstruction techniques and the contagion risk impact of different network structures in a stress testing context. My findings suggest that uncertainty about the true interbank network leads to substantially varying risk assessments, where a Bayesian reconstruction approach can be used to identify contagion risk boundaries. In addition, network structure metrics that explain the variation in risk assessments are identified.

Keywords: systemic risk, network reconstruction, bank network structure, contagion

JEL Classification: D85, G21, L14