Fundamental aspects of privacy and deception in electronic auctions

Auctions have become a major phenomenon of electronic commerce. Governments use auctions to sell spectrum licenses, millions of users trade goods in Internet auctions, and task assignment in multiagent systems as well as procurement in the “real world” is handled via auctions. The contribution of this dissertation is two-fold. The first part addresses strategic bidding whereas the second part deals with privacy issues. There are auction mechanisms that have been proven to lead to allocations of goods that maximize utility among participants (bidders and sellers) and to remove counter-speculation in the bid-preparation process. However, the theoretical model makes several restrictive assumptions to achieve those properties. For example, agreements between bidders (“bidder collusion”) or untruthful auctioneers are not considered. This thesis extends this list by adding the notion of “antisocial” agents, i.e., agents that intend to maximize their own utility while minimizing their competitors’ utilities. We present optimal antisocial bidding strategies and show that it is impossible to construct an auction mechanism that provides the properties mentioned above in the presence of at least one antisocial agent. In the second part of this thesis, the privacy of sealed-bid auctions is investigated. Traditionally, privacy is obtained by consulting a trusted third-party, the auctioneer. However, the correctness of the outcome and the privacy of bids completely depend on the trustworthiness of the auctioneer. The major contribution of this dissertation is the incremental development of distributed protocols that emulate auction mechanisms without relying on any trusted party. This is achieved by applying recently developed cryptographic techniques of secure multiparty computation and distributing the determination of the auction outcome on bidders. The proposed protocols are secure despite any collusion of bidders. Some of them even provide partial privacy against computationally unbounded adversaries.