报告摘要：
Cross-sectional and temporal heterogeneity among consumers suggests that personalized dynamic price promotions could significantly benefit retailers of storable goods. However, the conventional estimate-then-optimize approach for designing personalized promotions may fail to realize the potential, hindered by two primary challenges: model misspecification and limited training data. In this work, we propose a novel offline reinforcement learning approach designed to overcome these challenges. First, by estimating the state-action value function directly from data, we effectively bypass the need for specifying a consumer choice model. Second, by analyzing the structural properties of the problem, we devise a quadratic functional approximation to the state-action value function, effectively mitigating the issues of limited data and distribution shift. Finally, we apply binning to the state variables, converting each continuous state variable into a binary vector, thereby capturing possible nonlinear dependencies between decision variables and state variables. Our approach has been shown to yield consistently higher revenue in a series of experiments, including simulations with synthetic data, analyses using real-world datasets, and field experiments. We explore the conditions under which our approach is most effective; we also offer managerial insights on selecting the optimal strategy for personalized dynamic pricing promotions in the storable goods market.
嘉宾简介：
Yanzhi Li (David) is Professor of Marketing and Management Sciences and also an Affiliated Professor of School of Data Science. He is Head of Department of Marketing and Director of Fintech and Business Analytics Centre (FBAC) at the College of Business, City University of Hong Kong. He received a bachelor's degree in Computer Science from Tsinghua University and his Ph.D. in Industrial Engineering and Engineering Management from Hong Kong University of Science and Technology. His research interests include (1) Decision analytics for marketing, logistics, and supply chain; and 2) Interface research between Operations and Marketing, such as pricing, assortment planning, channel management, and advertising. His papers have appeared in journals such as Marketing Science, Manufacturing & Service Operations Management, Operations Research, and Production and Operations Management.
