The Performance of Core-Selecting Auctions: An Experiment

Combinatorial auctions, in particular core-selecting auctions, have increasingly attracted the attention of academics and practitioners. We experimentally analyze core-selecting auctions under incomplete information and find that they perform better than the Vickrey auction. The proportions of efficient allocations are similar in both types of auctions, but the proportions of stable (core) allocations and the revenue are higher in the core-selecting auctions. This is in particular true for an independent private values setting in which theory does not predict this better performance of the core-selecting auction. We trace the causes of the performance differences back to patterns in bids. The core-selecting auctions provide incentives for overbidding the own valuation and - under certain conditions - also for bid-shading, which can hamper performance. In the experiment, bidders react in the predicted direction to these incentives, though less pronouncedly than predicted.