Abstract
The classical approach to the newsvendor problem is to first estimate the demand distribution (or assume it to be given) and then determine the optimal inventory level. Data-driven optimization offers an alternative, where the inventory level is determined directly from the data. In this paper, we consider the data-driven newsvendor problem under a service-level constraint. We show that existing approaches to this problem suffer from overfitting, resulting in service-levels that are below the target service-level. We propose new data-driven approaches and corresponding mathematical optimization models based on methods of distributionally robust chance-constrained optimization—which have not yet been applied and empirically tested in the context of the data-driven newsvendor problem. We assess the effectiveness of our approaches by means of an extensive numerical study. To that end, we conduct structured experiments based on simulation as well as experiments based on a real-life bikesharing system where we consider the daily usage data along with information on weather and seasonal factors. The results demonstrate that our methods achieve on-target service-levels even in absence of large amounts of data. All in all, our study provides ample empirical evidence that distributionally robust chance-constrained optimization is a viable approach for addressing the data-driven newsvendor problem.
Original language | English |
---|---|
Article number | 108509 |
Number of pages | 16 |
Journal | International Journal of Production Economics |
Volume | 249 |
DOIs | |
Publication status | Published - Jul-2022 |
Keywords
- Data-driven decision making
- Distributionally robust optimization
- Newsvendor model
- Service-level