A Genetic Algorithm Approach to Repeated Bargaining Under Symmetric and Asymmetric Information

Christoph Zott - The University of British Columbia

Asymmetric information can lead to inefficient bargaining outcomes and even market failure. Learning has been suggested as a means to alleviate these problems. This paper investigates how a genetic algorithm affects the outcome of bilateral bargaining in a repeated ultimatum game between two agents, one who offers an investment project and a profit share, and the other who must accept or reject this offer. The quality of the underlying investment project is unknown to either agent, but each receives a potentially different signal regarding this quality. The genetic algorithm is part of a classifier system used to model human learning. Three lines of investigation are pursued within this model: comparisons of learned behavior with analytically determined Nash bargaining solutions; determination of the relative importance of learning by experience (updating rule strengths) versus learning by experimentation (introducing new rules); and potential normative uses of genetic algorithm learning for the design of financial contracts.

Scheduled for Session 4.6 Agent-Based Computational Economics - II