Learning Algorithms in a Decentralized General Equilibrium Model

A model is developed in which economic agents learn to make price-setting, price-response, and resource allocation decisions in decentralized markets where all information and interaction is local. Computer simulation shows that it is possible for agents to act almost as if they had the additional necessary information to define and solve a standard optimization problem. Their behaviour gives rise endogenously to phenomena resembling Adam Smith's invisible hand. The results also indicate that agents must engage in some form of price comparison for decentralized markets to clear--otherwise there is no incentive for firms to respond to excess supply by lowering prices. This suggests that agent-based models with decentralized interaction risk untenable results if price-response decisions are made without being first directed toward the most favourable local price.