A deep learning-based approach to a newsvendor problem considering uncertainty and time-varying costs

We study a newsvendor problem in which demand is uncertain and newsvendor costs are highly time-varying. In this work, we propose a deep learning based newsvendor model that systematically considers uncertainty in inventory decision-making. Unlike the literature assuming a known demand distribution or considering fixed newsvendor costs, the proposed model is distributional assumption free and makes inventory decisions in response to time-varying costs. By exploiting two sources of uncertainty in decision-making, data and model uncertainty, we construct a forecast distribution and use this distribution for newsvendor inventory decisions. The numerical analysis evaluates the proposed model in the context of energy-commitment decision-making in a day-ahead market. The results demonstrate that the proposed model is promising when newsvendor costs change over time in terms of lowering inventory costs. Moreover, we verify that considering both data and model uncertainty enables us to reduce more inventory costs.