William E. Walsh, David C. Parkes, Tuomas Sandholm, Craig Boutilier
Single-good ascending auctions, including the English Auction and its close variants (e.g., eBay), are the most widely used type of auction. Effective strategies for buyers and sellers in these auctions can have an enormous economic impact. We present a system that, relying on minimal assumptions, computes reserve prices for real-world ascending auctions by inferring the value distribution from past auction data. Our contributions include improvements on previous methods for estimating the bidder value distribution, and novel Bayesian methods for adapting to disruptions in market conditions. We demonstrate the effectiveness of our system in both simulated auctions and in real Internet auctions that we conducted in the domain of selling off returned merchandise.
Subjects: 7.1 Multi-Agent Systems; 1. Applications
Submitted: Apr 15, 2008