Distributionally Robust Newsvendor Under Stochastic Dominance with a Feature-Based Application

• 成果类型: Article
• 署名作者: Fu, Mingyang; Li, Xiaobo; Zhang, Lianmin
• 署名单位: National University of Singapore; Shenzhen Research Institute of Big Data; The Chinese University of Hong Kong, Shenzhen
• 刊物名称: M&SOM-MANUFACTURING & SERVICE OPERATIONS MANAGEMENT
• ISSN/ISSBN: 1523-4614
• DOI: 10.1287/msom.2023.0159
• 发表日期: 2024
• 页码: 1962-1977
• 关键词: distributionally robust optimization; Newsvendor Problem; demand covariates; kernel estimation; Wasserstein Distance; Asymptotic convergence; stochastic dominance; Kullback-Leibler divergence

摘要

Problem definition: : In this paper, we study the newsvendor problem under some distributional ambiguity sets and explore their relations. Additionally, we explore the benefits of implementing this robust solution in the feature-based newsvendor problem. Methodology and results: : We propose a new type of discrepancy-based ambiguity set, the JW ambiguity set, and analyze it within the framework of first-order stochastic dominance. We show that the distributionally robust optimization (DRO) problem with this ambiguity set admits a closed-form solution for the newsvendor loss. This result also implies that the newsvendor problem under the well-known infinity-Wasserstein ambiguity set and Levy vy ball ambiguity set admit closed-form inventory levels as a by-product. In the application of feature-based newsvendor, we adopt general kernel methods to estimate the conditional demand distribution and apply our proposed DRO solutions to account for the estimation error. Managerial implications: : The closed-form solutions enable an efficient computation of optimal inventory levels. In addition, we explore the property of optimal robust inventory levels with respect to the nonrobust version via concepts of perceived critical ratio and mean repulsion. The results of numerical experiments and the case study indicate that the proposed model outperforms other state-of-the-art approaches, particularly in environments where demand is influenced by covariates and difficult to estimate.