Iterative Combinatorial Auctions with Bidder-Determined Combinations

In combinatorial auctions, multiple distinct items are sold simultaneously and a bidder may place a single bid on a set (package) of distinct items. The determination of packages for bidding is a nontrivial task, and existing efficient formats require that bidders know the set of packages and/or their valuations. In this paper, we extend an efficient ascending combinatorial auction mechanism to use approximate single-item pricing. The single-item prices in each round are derived from a linear program that is constructed to reflect the current allocation of packages. Introduction of approximate single-item prices allows for endogenous bid determination where bidders can discover packages that were not included in the original bid set. Due to nonconvexities, single-item prices may not exist that are exact marginal values. We show that the use of approximate single-item prices with endogenous bidding always produces allocations that are at least as efficient as those from bidding with a fixed set of packages based on package pricing. A network resource allocation example is given that illustrates the benefits of our endogenous bidding mechanism.