How sellers choose mechanisms: Information matters

Abstract

Dynamic mechanisms are quite complex, and few experiments have studied how sellers choose them. Here, we propose an experimental design to investigate how human sellers choose between two easily-conducted dynamic mechanisms: the optimal non-clairvoyant dynamic mechanism (NC) (Mirrokni et al., 2020) and the optimal repeated static mechanism (RS) (Myerson, 1981). Our results indicate that human sellers can harness their experience in an environment to choose the optimal mechanism later in the experiment. In addition, sellers tend to adjust heir choice of mechanism based on past revenue. We further find that: (i) sellers generally overprice; and (ii) buyers participate less in NC mechanism environments due to the greater-than-suggested upfront fee, leading to the theoretical-experimental revenue gap. Our results shed light on how sellers choose dynamic mechanisms and can potentially help improve mechanism design.

Keywords: Dynamic, Auction, Experiment, Non-clairvoyant, Learning