Evaluating Non-Clairvoyant Dynamic Mechanisms: Theory and Experiment

Abstract

Dynamic mechanisms provide a powerful means for optimizing revenue in repeated auctions. However, they are complex to implement, as they assume clairvoyance, i.e., that all participants accurately predict future demands. Recently, Mirrokni et al. (2020) introduced an optimal non-clairvoyant dynamic mechanism (NC) with a simple structure that does not rely on future knowledge. Here, we design an experiment to test the performance of NC as compared to another non-clairvoyant mechanism, the optimal repeated static mechanism (RS) (Myerson, 1981). Results show that non-clairvoyant mechanisms work as intended: NC either outperforms or underperforms RS according to theory predictions. However, contrary to theory, the optimal clairvoyant mechanism using full information performs no better than the non-clairvoyant mechanisms. Our results highlight the practical importance of non-clairvoyant mechanisms as implementable approaches to dynamic auction design.

Keywords: Non-clairvoyant, Dynamic, Mechanism, Experiment