Nearly optimal multi attribute auctions

Abstract

In almost every procurement situation, non-price attributes of the items to be purchased play a crucial role. Procurement protocols which take these attributes into account are called multi-attribute auctions.We study the following problem called optimal multi-attribute auction design: A buyer wants to procure an item which can be supplied in many possible configurations. The buyer has a value v(x) for each possible configuration x. Every seller i has a privately known cost ci(x) of supplying each possible configuration. Given a probability distribution on the cost functions, our goal is to design an auction which maximizes the expected utility of the buyer.This paper offers a generic method for the construction of nearly optimal multi-attribute auctions. The computational time of our mechanisms equals the time required for computing (or approximating) the optimal mechanism on a small number of agents. Our method can be successfully applied to many variants of multi-attribute auction design.