An Experimental Comparison of Combinatorial Procurement Auctions

The need for new procurement auction mechanisms that allow for rich bid types such as bundle bids on multiple items has been raised in many situations in industrial procurement. In addition to strategic problems, the design of these combinatorial auctions exhibits hard computational problems. For example, the winner determination typically leads to NP-hard allocation problems in combinatorial auctions. Recently, researchers have focused on pricing and information feedback in iterative combinatorial auctions. Several auction designs have been proposed in the literature using different types of ask prices, such as linear (i.e. item-level) or non-linear (i.e. bundle) ask prices. Although, there have already been a number of successful applications in the field, little is known about the bidding behavior in combinatorial procurement auctions. Lab experiments are an excellent method to observe the human behavior in a controlled economic environment and they are an important complement to theoretical and computational models as well as field studies. In a set of lab experiments at the TU München in Spring 2007, we examined four different auction formats: the sealed-bid Vickrey-Clarke-Groves (VCG) auction and iterative Combinatorial Auctions (ICAs) such as iBundle, the Combinatorial Clock auction, and ALPSm. iBundle (Parkes and Ungar, 2000) uses non-anonymous and non-linear prices, calculates a provisional revenue maximizing allocation at the end of every round and increases the prices based on the bids of non-winning bidders. The Combinatorial Clock auction is based on linear and anonymous ask prices (Porter et al. 2003, Ausubel et al. 2006). In each round, the ask prices on all over demanded items were increased by a fixed minimum increment and the auction closes when no item is overdemanded. The fourth auction format was the ALPSm auction format, which also uses anonymous linear ask prices. However, instead of increasing the prices directly, the auction lets the bidders submit priced bids and calculates so called pseudo-dual ask prices (Bichler et al. 2007). The termination rule and the eligibility rules have been adapted. Experimental analysis In our experimental setting, we have used four different value models, two with three, one with six and one with nine items. Bidders were interested in up to 27 bundles depending on the number of auctioned items. We provided some level of bidder decision support. The bidders did not have to calculate their payoffs manually as the MarketDesigner platform allowed them to enter their valuations privately in the user interface and the system would calculates …