The Science of the Deal: Optimal Bargaining on eBay Using Deep Reinforcement Learning

Abstract

Bargaining is ubiquitous. How can people bargain better? We train a reinforcement learning agent to bargain optimally in "Best Offer" listings on eBay, and we characterize its behavior in a manner that humans can use. As a buyer, the agent starts lower than human buyers and bargains longer. As the seller, the agent interprets offers as signals---of the buyer's willingness to pay and of the item's desirability---that human sellers ignore. Simple strategies derived from these agents purchase more items for lower prices than human buyers and sell more items for higher prices than human sellers.