Towards A New Experimental Economics: Complex Behaviour In Bargaining

In this paper we show that simple cognitive learning models, in a Computational Organization set-up can predict very well the bargaining behaviour observed in laboratory experiments with human players. Economist are interested in bargaining because transactions are not determined by market forces, and the because conceptually is the opposite to the idealized perfect competition equilibrium models. In fact in bargaining the complexity of the "institutional" setting depends upon the way the strategies of the negotiation is being unfolded all along the process. And this process is essential to export economic theory to managerial economics. Game theory has provided a rigorous conceptual frame to analyse strategic decisions and negotiations. But while game theory departs in this way from the competitive equilibrium both share three basic assumptions. The players are fully rational decision makers; they comprehend the faced situation and they know all the relevant institutional parameters.Experimental economics today has became a well accepted tool of economic research. To gain credit in managerial and applied economics many laboratory experiments on bargaining were design. Although some of the qualitative predictions of game theoretic models of bargaining have received some support, the existing models have performed poorly as point predictors. And experimental economics as conducted so far have faced the basic resistence: "I'm an economist; I don't accept anecdotal evidence". Most recently, this led to interesting research, Erev and Roth (1998), on how well simple but robust learning and coordination models can predict the observed experimental outcomes. The main findings being that institutional complexity can be induced by simple learning rules. And that more complex learning will deteriorate the predictive capacity of the model.We totally agree with their claim that "approximating the strategies used by players...will be the area of future research in which low-rationality adaptive game theory will need to interact most closely with cognitive theory". But as our paper demonstrates, bargaining and negotiation problems can be engineered even better by instruments developed in multiagent system theory. The agents are endorsed with behavioural rules empirically based, since they come from a previous bargaining experiment with human agents. Learning procedures as used in adaptive game theory, based on mechanical and optimizing players, will be replaced by cognitive learning and agenda based agents. And by so doing we argue computational organization theory will become a useful part of both economic theory and applied economics.