Non-Clairvoyant Dynamic Mechanism Design : Experimental Evidence

Dynamic mechanisms provide a powerful means for optimizing repeated auctions. Imple- menting them is complicated, however, due to a number of conditions that are difficult to satisfy in practice. These include the fact that the auction designer must be clairvoyant, in the sense that they must have reliable forecasts of participants’ valuation distributions in all future periods. Recently, Mirrokni et al. (2020) introduced a non-clairvoyant dynamic mech- anism (NC) and showed it is optimal within the class of dynamic mechanisms that do not rely on strong assumptions about the future. We showed, however, that an optimal repeated static mechanism (RS) (a Myerson auction) can sometimes outperform NC. Here, we report data from an experiment to test NC in relation to RS. Our results support the theory: NC outperforms or underperforms RS according to theory predictions. Our results highlight the practical value of non-clairvoyant mechanisms as implementable approaches to dynamic auction design