Multivariate stress scenario selection in interbank networks

Stress testing of financial systems has been increasingly important after several financial crises in recent years, thereby drawing keen attention to the choice of appropriate stress scenarios to effectively test the robustness of the systems. To this end, we consider the problem of identifying extreme-yet-plausible stress scenarios for financial systems, particularly the most likely scenarios that lead to adverse impacts on interbank networks, and propose an optimization-based approach to this problem based on the Eisenberg-Noe network model, a widely known risk contagion framework. To be more specific, we formulate tractable mixed-integer programs whose optimal solutions represent the most likely stress scenarios among all extreme scenarios in which the negative impact on the network (e.g., the total shortfall in payments, the total loss in asset values, and the number of defaulting banks) exceeds a certain threshold. We also present an extension of our approach to the case allowing for the disposal of illiquid assets.