Discrete Double Auctions with Artificial Adaptive Agents: A Case Study of an Electricity Market Using a Double Auction Simulator

A key issue raised by previous researchers is the extent to which learning versus market structure is responsible for the high efficiency regularly observed for the double auction in human-subject experiments. In this study, a computational discrete double auction with discriminatory pricing is tested regarding the importance of learning agents for ensuring market efficiency. Agents use a Roth-Erev reinforcement learning algorithm to determine their bid and ask prices. The experimental design focuses on two treatment factors: market capacity; and a key Roth?Erev learning parameter that controls that degree of agent experimentation. For each capacity setting, it is shown that changes in the learning parameter have a substantial systematic effect on market efficiency.