The goal of agent based modelin in economics is to investigate, how a decentralized economy coordinates economic activities. This question is basically the same as the one Keynes (1934) once posed - to what extent, and under what circumstances, particularly under what policy regimes, is the economy self-adjusting? Agent-based computational economics as an approach to this problem, which is a set of techniques for studying a complex adaptive system involving many interacting agents with exogenously given behavioral rules. The fundamental idea behind this approach is that the economic system can exhibit behavioral patterns that are to complex for the individual agent to comprehend. Therefore the approach assumes simple behavioral rules and allows a coordinated equilibrium to be a possibly emergent property of the system itself.
An agent in a rational-expectations-equilibrium model has a behavioral rule that is not independent of what everyone else is doing. In contrast, autonomous agents are endowed with behavioral rules that can tell them what to do in any given situation, independently of each others’ rules, even when no one has access to a correct model of the economy.
The modeling is about autonomous agents and the way they interact. An autonomous agent has simple rules for ﬁnding other people, sending communications to them, and responding to their communications. The interesting questions are what sort of exchange patterns emerge from the interaction between these rules, how orderly and stable are they, and how do they evolve over time.
Modern economic life is largely a sequence of exchanges with other people. Such exchange involves a specialized trader. Traders are the agents that undertake to match buyers and sellers, arrange terms of exchange and bear the costs of adjustment in the face of day-to-day imbalances between spending plans and productive capacity.
It can be modeled, how such a network of coordinating traders might emerge spontaneously, from elementary interactions between people following simple opportunistic rules of behavior. This is a self-organizing model of economy: If the organizational structure was not there, it would quite likely arise from the interaction of intelligent autonomous agents. This organizational structure exhibits the monetary structure of the real-world exchange mechanisms.
Trade is a useful but costly activity. People are motivated by the goal of maximizing consumption. They wander randomly and encounter each other. From time to time people turn into traders.
The economy has N perishable commodities and a discrete number of transactors, each one being identical except for type. A type (i, j) transactor is endowed with commodity i and can consume only commodity j. There is the same number of each type for each ordered pair of distinct commodities. The model focuses on ﬁve activities: entrepreneurship, search, exchange, business failure and price setting. The model goes through a sequence of periods, in each of which these activities take place in the following sequence.
It can be shown that the model has several absorbing states: Arrays of shops, prices, trading relationships and shopkeeper expectations persist indeﬁnitely once established. Different absorbing state can be observed: barter steady state and “monetary steady states, which is much like the monetary equilibrium in the trading post model. It constitutes a Pareto eﬃcient allocation of resources. Aggregate GDP in the economy is total consumption. Capacity GDP is the sum of all endowments minus the operating costs of the shops.
The emergence of this monetary structure is based on the network externality created by the ﬁxed costs of shops.
The ultimate goal of this research is to understand how large complex economic systems are capable of
exhibiting such a high degree of coordination most of the time, and yet from time to time are capable of departing drastically from such a coordinated state. They are many free parameters in agent based models, which needs to be calibrated to actual data. So far first results indicate that agent-based computational economics is capable of addressing some of the questions which a rational-expectations equilibrium approach is incapable of addressing, such as how a multiplier process in real estate lending can cause cumulative deviations from equilibrium and how the degree of intrest ﬂexibility aﬀects the ability of an economy to coordinate activities.