Automated Design of Near Optimal Auctions for Realistic Scenarios

Automated construction of revenue maximizing auctions poses many challenges which are not addressed by classic auction theory. In this paper we describe a system that facilitates generic automatic construction of near optimal auctions for realistic scenarios. In order to test our system we execute it on several benchmark distributions. On all of these benchmarks, our system yields revenue higher than the English auction. On some of them, the gap is significant. The system gives initial insights into several issues that have not been explored so far, such as the price of fairness in single item auctions, the power of ascending auctions, etc. Many open questions are presented as well.