A Computational Approach to Compare Information Revelation Policies

Revelation policies in an electronic marketplace differ in terms of the level of competitive information disseminated to participating sellers. Since sellers who repeatedly compete against one another learn based on the information revealed and alter their future bidding behavior, revelation policies affect welfare parameters—consumer surplus, producer surplus, and social welfare—of the market. Although different revelation policies are adopted in several traditional and Web-based marketplaces, prior work has not studied the implications of these policies on the performance of a market. In this paper, we study and compare a set of revelation policies using a computational marketplace. Specifically, we study this in the context of a reverse-market where each seller’s decision problem of choosing an optimal bid is modeled as an MDP (Markov decision process). Results and analysis presented in this paper are based on market sessions executed using the computational marketplace. The computational model, which employs a machine-learning technique proposed in this paper, ties the simulation results to the model developed using the game-theoretic models. In addition to this, the computational model allows us to relax assumptions of the game-theoretic models and study the problem under a more realistic scenario. Insights gained from this paper will be useful in guiding the buyer in choosing the appropriate policy.