Abstract
Combinatorial auctions (CAs) are multi-unit auction mechanisms in which bidders can bid on combinations of items or packages. CAs have received considerable attention from academia, and their practical use has increased significantly over recent years especially in procurement projects. Units in procurement projects often exhibit cost synergies, and using a CA can be particularly advantageous to the auctioneer because it enables bidders to express their cost synergies in the bidding process. If the units were auctioned separately, bidders face a risk of winning only some of the units, known as exposure risk, which prevents them from expressing the synergies as much as they could have in a CA. Indeed, under suitable conditions, allowing combination bids is necessary for an efficient and optimal mechanism in the presence of synergies (Levin 1997).
Allowing package bids, on the contrary, can also hurt the efficiency and optimality of the allocation. Bidders may place discounted package bids even when they do not have cost synergies among the units that form the combination. Indeed, firms may have incentives to bundle units where they have a cost advantage with other units where they do not have an advantage. This practice, which we refer to as strategic bundling, may result in a CA that offers a less efficient and more expensive allocation than a set of separate auctions for each unit (Cantillon and Pesendorfer 2006). This raises an important design issue: how to allow combination bids to allow bidders' to express their synergies while suppressing the negative effects of strategic bundling. In general, the auctioneer would like to permit package bids among units if and only if those units exhibit cost synergies. However, in most practical settings the precise nature of firms' cost synergies is their private information.
With this motivation, in this work we propose an estimation method that identifies important properties of the bidders' cost structure using bidding data. An important challenge is identifying costs separately from mark-ups, which also determine bids. To do so, we use a structural estimation method in which we pose a model of bidders' behavior. In a nutshell, the structural model imposes restrictions on how mark-ups are determined, and thereby enabling us to identify the cost structure of firms. This can then be used to inform auction design decisions such as which package bids to allow. We believe our method and results can bring substantial impacts on enhancing efficiency of CAs.