Coordination in an agent-based division of labor model

This paper compares the performance of coordination by decentralized and centralized mechanisms in a division of labor model, where there are returns to specialization. My primary hypothesis is that central coordination is a superior mechanism for organizing systems that display complex webs of interdependency between participants. The model demonstrates that centralized coordination provides efficiency benefits over decentralized systems when duplication of effort is costly and close synchronization of effort is required; that centralized coordination can be more efficient than decentralized coordination even in the absence of opportunism; and that redundancy is a necessary element of decentralized systems. The comparison of centralized and decentralized systems in economic production has a long history in economics. Coase [1937] first pointed out that economic agents incur an unseen cost when they rely on decentralized markets. Hayek [1944] argued that decentralized systems have information processing advantages since economic agents acting on local information could process more information than a central coordinator. Most recently, Williamson [1975, 1985] proposed that hierarchical organizations arise to solve bargaining problems when production involves investments by one agent that are specialized to another economic agent. This model demonstrates the advantages of central coordination in overcoming synchronization and assignment problems [Milgrom and Roberts, 1992]. Synchronization problems arise when many participants in a joint process must time their contributions precisely; in this situation agents acting on their own behalf may not synchronize their actions properly. Assignment problems, on the other hand, arise when redundancy is particularly costly. For instance, if ambulances were dispatched through a decentralized market, then several emergency vehicles might respond to a call requiring just one team, leaving the system unprepared for the next emergency. These advantages of centralized coordination of production have not been closely explored, probably because they are dynamic and do not lend themselves to analytical modeling. This paper uses an agent-based computer model to compare the advantages of centralized and decentralized coordination in conditions that share many of the characteristics of synchronization and assignment problems. The model requires synchronization, since all the steps of the production process must be performed in every round or the product will not be produced. Thus when an agent exits the simulation, all its trading partners are reset to solo production since they have lost the steps that a partner provided. In addition, the model is an example of the assignment problem in reverse, since it demonstrates the problems of decentralized systems when they lack sufficient redundancy.