Evolution of sunspot like behavior in the agent based economies of bank runs

摘要

This paper describes evolution and adaptation of an agent-based model in a version of the Diamond-Dybvig, bank-run economy. In addition to pure strategy, run and no-run equilibria, the theoretical model also has a multiplicity of sunspot equilibria where agents' expectations and subsequent actions are driven by coordination on an extrinsic random variable. In this paper, agents that adapt using the individual evolutionary (IEL) algorithm make decisions on whether or not to follow extrinsic random variable announcement on whether to 'wait' (not run) or 'withdraw' (run). Their strategies also advise them as what to do if they do not follow the announcements. The results of simulations, that are robust for a number of different parameterizations, show 'no-run' and 'run' outcomes for low levels of strategic uncertainty. However, for high level of strategic uncertainty, the IEL agents evolve strategies that follow sunspot announcements. These results correspond to those obtained in the experiments with human subjects. Moreover, the results scale-up when the number of agents increases ten-fold and twenty-fold.