Bündelung und Auktionsdesign im Revenue Management
Zusammenfassung der Projektergebnisse
In this project we aimed to design highly efficient combinatorial auctions. We have addressed two related questions in the design of combinatorial auction mechanisms. The first three work packages focused on combinatorial auctions with allocation constraints, which are present in almost all applications but have received little attention in the literature. Work packages 4-6 deal with online auctions where no reliable prior distributions are available for all possible packages as is often the case in the field. Here, the key strategic problem is the coordination of bidders and ascending auctions have substantial advantages. Finally, in the last work package 7, we focused on different theoretical questions in the pricing of multi-object auctions. A key contribution in the first set of work packages focusing on allocation constraints was the theoretical analysis of deadness level (DL) auctions, which allow for arbitrary allocation constraints. First, we showed that in these auctions truthful bidding is an ex-post equilibrium if bidder valuations satisfy a form of substitutability. An ex-post equilibrium is a strong solution concept, where bidders do not need to know the distribution of valuations of other bidders. Second, we showed that the computation of DLs is computationally a very hard problem (Pi P2-complete). Yet, exact computations for smaller auctions are feasible in seconds as we show experimentally. Also, such prices can be approximated for larger auctions. In a follow-up paper, we introduced coalitional winning levels, which circumvent some of the problems of DLs. In particular, if bidders bid straightforward on DLs, an auction can take a very large number of auction rounds, many more than what would be acceptable in the field. Coalitional winning levels are studied in an online environment where bidders do not have sufficient information to shade their bids strategically. However, they do face a coordination problem and they need to agree on a partitioning of the item set. Coalitional winning levels are a possibility for the auctioneer to leverage the information he gets in earlier rounds about the preferences of bidders and to actively help bidders coordinate on a set of non-overlapping bids. Results of extensive lab experiments support these findings. Finally, we contributed a number of other papers focusing on different questions about equilibrium strategies and pricing in ascending combinatorial auctions. These papers complement the results of our key work packages and yield insights in various solution concepts. We could generate a number of new results in this area. A part of the results has already been published in conferences and journals. Four papers funded by the project have appeared in A+-ranked journals (according to VHB Jourqual 3), while two other papers have been published in journals ranked A and B. In addition, a number of presentations have been given on our research in conferences and workshops. Importantly, our work on combinatorial auctions has influenced our parallel work in spectrum auction design and industrial procurement which had direct industry impact.
