Systemic risk in interbank networks: disentangling balance sheets and network effects

TL;DR

This paper compares empirically observed systemic risk in interbank networks with risk inferred from banks' balance sheets, highlighting the importance of network structure and events like crises for accurate risk assessment.

Contribution

It introduces a maximum-entropy network model constrained by balance sheet variables to estimate systemic risk and compares it with empirical data, emphasizing network effects.

Findings

Aggregate systemic risk estimates align well with empirical data during stable periods.

Individual bank risk levels are significantly influenced by their network positions.

Network details are crucial for precise stress testing, especially during systemic crises.

Abstract

We study the difference between the level of systemic risk that is empirically measured on an interbank network and the risk that can be deduced from the balance sheets composition of the participating banks. Using generalised DebtRank dynamics, we measure observed systemic risk on e-MID network data (augmented by BankFocus information) and compare it with the expected systemic risk of a null model network, obtained through an appropriate maximum-entropy approach constraining relevant balance sheet variables. We show that the aggregate levels of observed and expected systemic risks are usually compatible but differ significantly during turbulent times (in our case, after the default of Lehman Brothers and the VLTRO implementation by the ECB). At the individual level instead, banks are typically more or less risky than what their balance sheet prescribes due to their position in the network. Our results confirm on one hand that balance sheet information used within a proper maximum-entropy network model provides good aggregate estimates of systemic risk, and on the other hand the importance of knowing the empirical details of the network for conducting precise stress tests on individual banks, especially after systemic events.