The field of computational economics is a young and growing field. While this paper presents the first agent-based model of transition economies, many others researchers have begun to draw on ideas from the field of artificial life to model economic phenomena. Critically, the emergent behavior exhibited by many artificial life systems has the potential to explain how large-scale economic behavior arises from local interactions among a multitude of heterogeneous agents (Arthur De Vany, personal communication).
One of the most complete artificial life models of an economy is the Aspen system developed by Richard Pryor and colleagues at Sandia National Labs [Basu et al.1996]. Aspen has a rich set of agents, including households, firms in four different sectors, a realtor, a capital goods producer, banks, and a government. The firms use a genetic algorithm to select a price that maximizes profits. The Aspen model duplicates several results first described by Modigliani in his research on the FMP model.
Sugarscape, a versatile system that has become a testbed for studying problems in the social sciences, has been used to examine simple trading models. In these trading models, agents in a two-dimensional landscape trade two goods (sugar and spice) that the agents then metabolize. The agents endogenously determine the prices of sugar and spice, and the quantities exchanged. This simulation replicates several findings described in standard economic literature. For example, the number of agents that can exist on a landscape increases when trading is added to the simulation. However, the Sugarscape simulations also demonstrate that, under certain conditions, prices do not converge to the general equilibrium price, a result that differs from standard economic theory.
Work by Youssefmir, Huberman, and Hogg provides a possible explanation for why markets crash, a particularly timely topic given the market gyrations of 1997 [Youssefmir, Huberman, & Hogg1996]. In their model, a set of heterogeneous agents trade in an asset market based on the agents' expectations of what future prices will be. These expectations are based on two components: the fundamental price and a trend. Each agent has a slightly different perception of the fundamental price, and a trend gets established based on the past behavior of prices. Agents differ in their belief in how long a trend will last. So a rising trend can lead to speculative bubbles since most trend followers are likely to believe strongly in the trend, and some fundamentalists will believe in the trend for a while. As prices move away from the fundamental price, fundamentalist agents will expect the trend to reverse itself and eventually some trend followers will also lose faith in the trend, and the speculative bubble will deflate. The price response to buy/sell orders and the individual trend horizon are set exogenously, and asset prices are determined endogenously. These results show that asset prices can deviate sharply from their fundamental values.
In Tesfatsion's Trade Network Game (TNG), the player set is a collection of traders consisting of pure buyers, pure sellers, and buyer-sellers[Tesfatsion1997]. Buyers repeatedly submit trade offers to sellers, who either refuse or accept these offers. If a seller accepts a trade offer from a buyer, the seller and buyer engage in a risky trade modeled as a standard prisoner's dilemma game. The iterated prisoner's dilemma strategies used by buyers and sellers to conduct their trades are evolved by means of a genetic algorithm. The fact that traders are able to choose and refuse their trading partners makes this a better model of real-world trading than standard game models in which partners are matched randomly or by round robin assignment. Simulations are run for two types of markets: Endogenous-type markets comprising only buyer-sellers; and two-sided markets comprising equal numbers of pure buyers and pure sellers. The findings illustrate how en ante capacity constraints, in the form of buyer offer quotas and seller acceptance quotas, are a primary driving force determining the evolution of trading behavior. For example, given relatively large seller acceptance quotas and relatively small buyer offer quotas, sellers tend to be parasitized by buyers in the sense that buyers are able to latch on to cooperative sellers and successfully defect against them.