We review heterogeneous agent models of financial stability and their application in stress tests. In contrast to the mainstream approach, which relies heavily on the rational expectations assumption and focuses on situations where it is possible to compute an equilibrium, this approach typically uses stylized behavioral assumptions and relies more on simulation. This makes it possible to include more actors and more realistic institutional constraints, and to explain phenomena that are driven by out of equilibrium behavior, such as clustered volatility and fat tails. We argue that traditional equilibrium models and agent-based models are complements rather than substitutes, and review how the interaction between these two approaches has enriched our understanding of systemic financial risk. After presenting a brief summary of key terminology, we review models for leverage and endogenous risk dynamics. We then review the network aspects of systemic risk, including models for the three main channels of contagion: counterparty loss, overlapping portfolios, and funding liquidity. We give an overview of applications to stress testing, including both microprudential and macroprudential stress tests. Finally, we discuss future directions. These include a better understanding of dynamics on networks and interacting channels of contagion, models with learning and limited deductive reasoning that can survive the Lucas critique, and practical applications to risk monitoring using models estimated with the massive data bases currently being assembled by the leading central banks.
Aymanns, C. Farmer, J.D., Kleinnijenhuis, A.M & Wetzer, T. (2018). 'Models of Financial Stability and Their Application in Stress Tests'. in: Hommes, C. & LeBaron, B. (eds.) "Handbook of Computational Economics: Volume 4". Elsevier. pp. 329-391