Bridging uncertainty: A data-driven DRO approach for correcting censored demand in newsvendor problems

Abstract

When dealing with short-life cycle products, small and medium-sized enterprises (SMEs) commonly confront challenges stemming from limited local censored demand data. This often leads to a lack of comprehensive understanding of demand distribution and can result in suboptimal order decisions. To address this issue, we introduce a data-driven newsvendor framework that combines a novel cost-driven data correction procedure with distributionally robust optimization (CDDC-DRO). With cost minimization objectives, the proposed procedure integrates local censored demand data and external demand information to adaptively generate high-value improved censored datasets, while circumventing reliance on static correlations. Furthermore, we consider the granularities of external demand information and propose three DRO-based data correction strategies to effectively reduce demand censoring. Tests on both simulated and actual data indicate that the CDDC-DRO procedure adaptively corrects censored data based on demand characteristics and cost structures, thereby eliminating significant errors induced by demand censoring and improving the precision and robustness of order decisions. The correction degree of the improved censored datasets dynamically depends on cost structure. A high degree of data correction is employed under high critical ratios, whereas a minimal correction degree is applied under low critical ratios. In response to the significant negative impacts of demand censoring, SMEs prefer to implement the DRO-based data correction strategy with finer-grained external demand information. This strategy enhances correction capabilities while minimizing variations in decision accuracy. Even when finer-grained external demand information is unavailable, SMEs are able to make well-informed order decisions using the DRO-based data correction strategy with local censored demand data.