Consider a simple economy in which autonomous agents are endowed with two goods, g1 and g2, and with the capability of producing more of each. At regular intervals each agent is allowed to produce one of the goods, at a rate determined by his own unique production function, or to trade with other agents in the economy. The opportunity cost of each activity is the inability to pursue other activities, that is, if an agent decides to produce good 2, he cannot produce good 1 or initiate trade with other agents in that time period. Each good also yields utility for the agents according to a utility function with the standard first and second derivatives. This paper studies how utility maximizing agents optimize in these circumstances, examines the aggregate characteristics of the economy, and investigates the internal organization of production and exchange.The autonomous agents in this study follow a simple procedure. In each period an agent draws a random sample of other agents as potential trade partners. He negotiates an exchange price and quantity with each partner to find which agent offers the best deal. He then compares the utility he would get from such an exchange with the utility gained from production and he selects the best alternative, to produce or trade. A second agent then draws a sample of potential traders and follows the same decision tree. Over time agents learn about the other agents in the economy. Specifically, they remember who offered the better deals in previous periods, and they can revisit those traders. This memory, however, is constrained in space and time. That is, agents remember only a few periods back and a limited number of traders in each of those previous periods. One set of research questions concerns the effects of this memory, that is, how does the size and length of the agents' memory affects the economy's aggregate characteristics and the distribution of goods? Given this simple structure, our agents display an assortment of behavior. A common outcome is for some agents to produce one good and trade for the other. The sequential nature of this economy leads these agents to adopt a cyclical pattern such as production in one period, trade in the next. Other agents, however, specialize further. For example, one agent may be highly proficient in the production of g1. If he is "discovered" by other agents as a ready source of g1, he may be able to produce in almost every period and trade for g2 at someone else's initiative. We also find some agents specialize in trade. These "merchants" may not produce at all and exist by buying in one period and selling in the next. With repeated simulations the initial conditions of the economy can be reset and the same population can be observed under a different sets of rules. For example, in one case trade is completely suppressed so that every agent produces in every period. We then compare aggregate characteristics, such as total output, income distributions, and Pareto utility improvements to economies that allow trade. We also explore the impact of transactions cost by introducing the condition that negotiation and exchange absorb resources. More importantly, these costs change over time. For example, transactions cost fall as agents deal repeatedly with the same trade partners. In essence, these falling costs reflect a world in which agents are initially strangers and not knowing the practices of others makes negotiation and exchange relatively expensive. With experience, agents learn about the trading conventions of other agents and the cost of negotiation declines for those familiar trading partners. Again, we explore the consequences of these changing transaction costs on the aggregate characteristics of the economy and the dynamic behavior of specific agents. We are especially interested in production coalitions and trade networks that emerge and survive over several periods.
