An Experimental Comparison of Linear and Nonlinear Price Combinatorial Auctions

Combinatorial auctions are used for the efficient allocation of heterogeneous goods and services. They require appropriate software platforms that provide automated winner determination and decision support for bidders. Several promising ascending combinatorial auction formats have been developed throughout the past few years based on primal-dual algorithms and linear programming theory. The ascending proxy auction and iBundle result in Vickrey payoffs when the coalitional value function satisfies buyer submodularity conditions and bidders bid their best responses. These auction formats are based on nonlinear and personalized ask prices. In addition, there are a number of designs with linear prices that have performed well in experiments, the approximate linear prices auction, and the combinatorial clock auction. In this paper, we provide the results of lab experiments that tested these different auction formats in the same setting. We analyze aggregate metrics such as efficiency and auctioneer revenue for small-and medium-sized value models. In addition, we provide a detailed analysis not only of aggregate performance metrics but also of individual bidding behaviour under alternative combinatorial auction formats.