Adaptive agents in a persistent shout double auction

Abstract

CIiff [7] has demonstrated that simple, adaptive agents are able to trade in a form of double auction marketplace, in such a way that trade prices converge towards the equilibrium price of the marketplace. However, the marketplace within which the agents trade is unrealistic. In this paper, we consider a more realistic form of double auction market the persistent shout double auction. We present agents based on the ZIP agents of (Xi [7], but with an alternative set of heuristics for use within this auction. We demonstrate that the resulting agents achieve equilibrium significantly faster than ZIP agents do, maintain a more stable equilibrium, and are more robust to changes in learning rate.