On the Impact of Cognitive Limits in Combinatorial Auctions: An Experimental Study in the Context of Spectrum Auction Design

Combinatorial auctions have been studied analytically for several years, but only limited experimental results are available, in particular for auctions with more than 10 items. We investigate the Hierarchical Package Bidding auction (HPB), the Combinatorial Clock auction (CC), and one pseudo-dual price auction (PDP) experimentally, as all these formats were used or suggested for high-stakes spectrum auctions. We want to understand the impact that different auction formats have on bidder behavior and allocative efficiency. Interestingly, we find that the main source of inefficiency in all formats is the bidders’ preselection of packages, rather than their strategies or auction rules; bidders mostly preselect a small number of packages of interest early in the auction. CC achieves high efficiency and revenue in all experiments, but HPB yields similar results even in value models, where hierarchical pre-packaging is difficult. Due to their influence on the decision of the US Federal Communication Commission, we intentionally repeated a set of experiments conducted by Goeree and Holt (GH) [J.K. Goeree, C.A. Holt, Hierarchical package bidding: A paper & pencil combinatorial auction, Games Econ. Behav. (2008)]. We get similar aggregate results but in contrast to GH, we find that small bidders have more difficulties in coordinating their bids to outbid a national bidder in a threshold problem with HPB. In addition to the value models with global synergies in GH, we used value models where bidders have local synergies, which play a significant role in spectrum auctions in the field and lead to different results.