LEARNING AND IMITATION: TRANSITIONAL DYNAMICS IN VARIANTS OF THE BAM

We study the dynamics of self-organized systems when disturbed by shocks. For this purpose, we consider extensions of the "Bar Attendance Model" [1] (BAM), which provides a stylized setting for the analysis of the emergence of coordination in the behavior of a large collection of agents. We represent the learning process of the agents through genetic algorithms, which respond to global (publicly available) information. In addition, we allow the actions of agents to be influenced by local information, as expressed in the behavior and performance of neighboring individuals. In the context of the BAM, we show that, in the event of a shock, the imitation behavior may become widespread and generate a contagion cascade which mimics a collective panic. We use this framework to represent features of the dynamics of an actual bank run.