【報告人簡介】

Dr Melvyn Sim is a Provost's Chair Professor in the Department of Analytics and Operations (DAO) at the National University of Singapore (NUS) Business School. His research interests broadly encompass decision-making and optimisation under uncertainty, with applications in finance, supply chain management, healthcare, and engineered systems. He currently serves as a Department Editor for Manufacturing and Operations Management (MSOM).

【摘要】

We propose a robust benchmark satisficing framework for data-driven decision-making under uncertainty, designed to identify decisions whose expected revenue exceeds that of a comparator by a user-specified surplus—even when the true distribution is unknown. This framework generalizes the robust satisficing model of Long et al. (2023), by accommodating a broader range of benchmark-driven decision criteria as individuals often evaluate their performance relative to others or to reference standards. Built on distributionally robust optimization, our model employs the Wasserstein metric to model distributional ambiguity while ensuring finite-sample performance guarantees. Within this framework, we identify the optimal linear transformation of the uncertain parameters that minimizes conservatism, formulated as a determinant minimization problem with an exponential moment constraint. When estimating the deviation matrix from data, we also introduce a spectral regularization constraint to limit its condition number and prevent its determinant from collapsing to zero. We derive tractable reformulations under various structural assumptions on both the primary and comparator revenue functions, including settings with linear recourse. We validate the framework through two computational studies. In a portfolio optimization problem, our model consistently outperforms an equal weighted benchmark, offering improved risk-return profiles, especially with our proposed deviation matrices. In a multi-product newsvendor setting, where product demands depend on S&P 500 and gold prices, the model ensures revenue superiority over the better-performing benchmark. Together, these results underscore the framework’s flexibility and practical effectiveness in benchmark-driven, uncertain environments.

撰稿：王戈 審核：許建軍 單位：現代供應鏈管理研究院