Robust Financial Networks

We study networks of financial institutions where only aggregate information on liabilities is available. We introduce the robust liability network, that is, the network with the worst expected losses among all networks with the same aggregate liabilities and assets. We provide an algorithm to identify the robust liability network and, using aggregate data provided by bank holding companies to the Federal Reserve in form FR Y-9C, determine robust liability networks for U.S. banks under various network configurations. We find that the robust liability network is sparse, with links between institutions that hold highly correlated portfolios. We illustrate the potential of our approach in two ways: We study the evolution of robust liability networks around the onset of the COVID-19 pandemic and evaluate the importance of network structure for financial institutions that are subject to a regulation that limits risk-taking based on each institution’s conditional value-at-risk.

Supplemental Material: The online appendices and data files are available at https://doi.org/10.1287/opre.2022.0272.