Temporal Fairness in Learning and Earning: Price Protection Guarantee and Phase Transitions
研究了价格保护保证对数据驱动动态定价中在线学习的影响，揭示了最优遗憾随价格保护期长度变化的相变行为，并指出在温和条件下设置长保护期无害。
Temporal Fairness in Data-Driven Dynamic Pricing Temporal fairness has long been recognized as a major issue in data-driven dynamic pricing. To address this, price protection guarantee has been proposed to mitigate such concern. Under this widely adopted guarantee, a customer who purchases a product can receive a refund from the seller if the seller lowers the price during the price protection period (defined as a certain time window after the purchase). In this paper, we initiate the study of the impact of this guarantee to online learning for data-driven dynamic pricing with initially unknown customer demand. Our results provide a fundamental characterization on the statistical complexity of this problem. In particular, we reveal a surprising phase transition behavior of the optimal regret with respect to the length of the price protection period. Not only that, our findings also offer practical insights in real-world deployment of price protection guarantees in data-driven dynamic pricing. That is, there is no harm to setting a long price protection period under very mild and realistic conditions.