Abstract
Historical data are typically limited. This study explores the robust mechanism design problem, where a seller aims to choose a pricing mechanism that maximizes the revenue from a buyer with uncertain valuation distributions. In this talk, I will present my recent works on robust mechanism design. In the first problem, we consider a seller who optimizes the pricing mechanisms to maximize the competitive ratio relative to an optimal policy with full valuation distribution information. We focus on situations where the seller only knows the exact probability of sales linked to multiple historical price points. We derive the optimal deterministic/randomized pricing mechanisms and analyze the competitive ratio corresponding to the optimal mechanism. Furthermore, we study the problem of the pricing experiment design and fully characterize the robust optimal experiment policy. Our findings illustrate that a limited number of pricing experiments suffice for achieving a strong performance guarantee and provide novel insights on the value of dynamic pricing. In the second problem, we assess the effects of non-linear utility on the robust optimal pricing mechanisms, and offer the corresponding profit guarantee for the seller across all possible valuation distributions within a mean-variance ambiguity set. When the utility function is of a quadratic form and the coefficient of variation is low, we find that a quadratic pricing mechanism yields the optimal revenue guarantee. Conversely, a unit price mechanism becomes optimal when the coefficient of variation is high. We further extend our analysis to general elasticity utility functions and investigate how the optimal worst-case revenue is influenced by elasticity and coefficient of variation. Additionally, recognizing the widespread implementation of two-part tariff mechanisms in practical applications, we also investigate the optimal two-part tariff mechanism and establish performance guarantees for varying elasticity coefficients.
Time
2024年5月15日（周三）10:00 – 12:00
Speaker
王震，香港中文大学（深圳）博士后。主要研究兴趣是不确定优化算法及其在定价和收益管理中应用。
Room
信息管理与工程学院308室
