This paper aims to contribute to the study of auction design within the domain of agent-based computational economics. In particular, we investigate the efficiency of different auction mechanisms in a bounded-rationality setting where heterogeneous artificial agents learn to compete for the supply of a homogeneous good. Two different auction mechanisms are compared: the uniform and the discriminatory pricing rules. Demand is considered constant and inelastic to price. Four learning algorithms representing different models of bounded rationality, are considered for modeling agents' learning capabilities. Results are analyzed according to two game-theoretic solution concepts, i.e., Nash equilibria and Pareto optima, and three performance metrics. Different computational experiments have been performed in different game settings, i.e., self-play and mixed-play competition with two, three and four market participants. This methodological approach permits to highlight properties which are invariant to the different market settings considered. The main economic result is that, irrespective of the learning model considered, the discriminatory pricing rule is a more e±cient market mechanism than the uniform one in the two and three players games, whereas identical outcomes are obtained in four players competitions. Important insights are also given for the use of multi-agent learning as a framework for market design.
