An Agent Model for First Price and Second Price Private Value Auctions

The aim of this research is to develop an adaptive agent based model of auction scenarios commonly used in auction theory to help understand how competitors in auctions reach equilibria strategies through the process of learning from experience. This paper describes the private value model of auctions commonly used in auction theory and experimentation and the initial reinforcement learning architecture of the adaptive agent competing in auctions against opponents following a known optimal strategy. Three sets of experiments are conducted: the first establishes the learning scheme can learn optimal behaviour in ideal conditions; the second shows that the simplest approach to dealing with situations of uncertainty does not lead to optimal behaviour; the third demonstrates that using the information assumed common to all in private value model allows the agent to learn the optimal strategy.