Macroeconomic research conducted at the Central Bank of Armenia in Dilijan, Armenia during summer of 2017. A heterogeneous, reinforcement learning model of bank bailout situations and policy to quantify moral hazard, compare against empirical results, propose optimal central bank bailout policy, and explore the effects of asset volatility.
Abstract: In the wake of the 2008 financial crisis, conventional understanding of bailout policy grew to accept the bailout of insolvent banks due to issues of systemic risk. A model is created to simulate such decision-making on the part of a central bank in order to study commercial bank behavior and adaptation patterns and analyze relevant tradeoffs and factors in policymaking. A field of heterogeneous commercial bank agents are endowed with an endogenous reinforcement learning mechanism through which they make investment decisions and learn from the profitability of their choices. The central bank's bailout policy is represented by an objective function, and its decisions impact future commercial bank behavior. The model quantifies and produces clear visualizations of the tradeoff between moral hazard and aversion of systemic damage, and the model's relationships mirror relationships in empirical data. It also recommends bailout policy that compares favorably to empirical bailout rates. Finally, the model demonstrates how asset volatility across a financial market affects the results of a particular policy.