Data-Driven Solutions for the Newsvendor Problem: A Systematic Literature Review

Abstract

The newsvendor problem captures the trade-off between ordering decisions, stocking costs and customer service level when the demand distribution is known. Nonetheless, in real case scenarios, it is unlikely that the decision maker knows the true demand distribution and its parameters, encouraging the use of datasets for empirical solutions that will achieve more precise results and reduce misleading decisions. Motivated by the availability of large amount of quality datasets, advances in machine learning algorithms and enhancement of computational power, the development of data-driven approaches has been emerging over the recent years. However, it is still unclear in which settings these data-driven solutions outperform the traditional model-based methods. In this paper, a systematic literature review is conducted for the descriptive analysis and classification of the most relevant studies that addressed the newsvendor problem and its variations under the data-driven approaches. The methods developed to solve the problems with unknown demand distribution are categorized and assessed. For each category, our paper discusses the relevant publications in detail and how they evidence the data-driven performance better. By identifying the gaps in the available literature, the future research directions are suggested.