Multi-Item and Multi-Attribute Auctions

While the theory of single-item auctions is mature, several unresolved issues remain in the more complex setting of auctions with several items and/or several attributes. Typically, in an electronic market setting, these are the auctions of relevance. Also, the electronic market setting itself gives rise to new, unusual auction types such as a digital goods auctions where items can be replicated at no cost.

The mechanism designer faces several decisions:

• Should items be sold simultaneously or in a sequence
• Should combinatorial bids for bundles of items be allowed or not?

Due to the complexity of these auction types, these questions are typically not analytically tractable and hence lend themselves naturally to a computational approach. In addition, it is necessary to insist on computational efficiency of the mechanism itself. In practice, this means that requirements of maximum economic efficiency or optimality must sometimes be relaxed. If an item for sale (or in procurement, a service demanded) has multiple attributes, such as price, quality, and delivery date, not only is the resulting model not analytically tractable, it is not fully developed. Meanwhile, systems for automating negotiations approach multi-attribute auctions but are not fully understood theoretically.

Finally, from the anti-trust literature it is well-known that a lot of alleged collusion occurs in auction settings. The scope for collusion as well general misuse of data is closely tied to the availability of information to the players and to the auction house. One can therefore expect tools from cryptography to assist the tools of economics in designing mechanisms more robust towards misuse of information by bidders or by the auction house. The latter can sometimes be the most important issue, namely if the auction house can misuse the information it learns in a context that is larger than the auction itself.

In general, we aim to improve the analytical tools for dealing with multi-item and multi-attribute auctions as well as the efficiency and applicability of computational methods. We also aim to make the resulting systems secure against collusion and misuse of information in general.