The Emergence of Economic Classes in an Agent-based Bargaining Model

Robert Axtell, Joshua M. Epstein, and H. Peyton Young - Brookings Institution

We use an agent-based computational model to study the emergence of equity norms. Agents with finite memory are repeatedly paired at random to play the Nash demand game. Each agent uses a best reply strategy based on the strategies of its most recent opponents. Occasionally each agent plays a random strategy. This gives rise to multi-agent dynamics that, over time, converge to a Nash equilibrium. These results are in accord with theoretical results of Young [1993]. In general, multiple equilibria will exist, and the equilibrium actually obtained may not be the equity norm. However, if the population is permitted to interact for a sufficiently long time then it can switch to a different Nash equilibrium. The dependence of these switching dynamics on the number of agents and the size of agent memory will be presented. Next, the population is separated into two types by giving half the agents an initially meaningless binary "tag." Each agent can noiselessly discern its opponent's tag before play, and conditions its best reply computation based on this tag. Over time these agent tags endogenously acquire organizing salience. For example, it can turn out that agents of each type arrive at an equity norm when playing agents of their own type, but that an inequitable equilibrium emerges when agents of opposite type interact. We interpret this as the emergence of economic classes and study the frequency of the several distinct types of outcomes.

Scheduled for Session 3.5 Simulation Models Of Behavior