Feature-Rich, Data-Private: A Sparse Learning Framework for the High-Dimensional Newsvendor

Modern firms increasingly rely on high-dimensional data to improve inventory decisions, but extracting useful signals from large numbers of features while protecting sensitive data remains challenging. In their paper, “Feature-Rich, Data-Private: A Sparse Learning Framework for the High-Dimensional Newsvendor,” Chang et al. develop a framework for the feature-based newsvendor problem that addresses both challenges. Their approach combines sparse learning with differential privacy, enabling firms to focus on the demand drivers that matter while safeguarding sensitive information. The authors establish theoretical guarantees showing that the method’s performance depends primarily on the number of relevant features, with only logarithmic dependence on the overall feature dimension. This result supports the use of rich contextual information while sharply limiting the usual dimensionality penalty. Simulation studies and an application to a real-world grocery retail data set further demonstrate the practical value of combining rigorous privacy protection with data-driven inventory optimization.