Learning Nash Equilibrium

We study the exchange economies of Kiyotaki and Wright [1988] in which agents must use a commodity or fiat money as a medium of exchange if trade is to occur. Instead of assuming rational agents as do Kiyotaki and Wright, our agents are artificially intelligent and learn their trading and consumption strategies adaptively. Artificially intelligent agents are modeled as classifier systems (Holland [1986]). In the assignment of credit within the classifier systems, we introduce some innovations designed to study multiagent problems. For most economies, trading and consumption patterns converge to a stationary Nash equilibrium even if agents start with random rules. In economics with multiple equilibria, the only equilibrium that emerges is the one with low storage goods playing the role of media of exchange (i.e., the “fundamental equilibrium” of Kiyotaki and Wright).