K Mayo and MP Wellman

2nd ACM International Conference on AI in Finance (ICAIF), Article No.: 20, pages 1–8, November 2021.

Extended abstract appeared in 20th International Conference on Autonomous Agents and Multiagent Systems (AAMAS), pages 1599–1601, May 2021.


Portfolio compression, the elimination of debt cycles in a financial network, is employed in over-the-counter derivative markets as a method for simplifying balance sheets. Canceling debts through compression can sometimes promote financial stability by stopping the spread of contagion from an insolvent firm hit by a negative financial shock. However, previous work has demonstrated that in some cases compression can exacerbate systemic risk by removing paths of shock absorption. We analyze portfolio compression as a strategic decision made by firms within a debt network. We define a network game in which firms represented as nodes have only local information and we ask what criteria the firms should consider in their decision to compress. We propose a variety of heuristic strategies and evaluate them using agent-based simulation and empirical game-theoretic analysis. In our experiments, compression may or may not benefit nodes on a cycle, and our results show that simple strategies based on local information can effectively improve firms’ decision making. We further examine which features are most useful under various conditions, and find that the results depend on the rate at which assets can be recovered from insolvent nodes. When recovery rates are low, firms focus on avoiding contagion, and when they are high they consider the benefits of retaining cycles to cushion shocks. Finally, we then analyze the effects on systemic risk in equilibrium, finding that when nodes strategically choose to compress, the outcome is more likely to be beneficial than harmful.