Forecast accuracy and inventory performance: Insights on their relationship from the M5 competition data

Although it is generally accepted that more accurate forecasts contribute towards better inventory performance, this relationship may often be weak, also depending on the structural characteristics of the products being forecast, the inventory policy considered, and the underlying expenses, among others. To empirically explore the connection between forecast accuracy and key costs associated with inventory control, namely holding, ordering, and lost sales costs, we consider the data set of the M5 competition and conduct detailed simulations using popular methods to generate quantile forecasts. Our results are analyzed for various setups of the order-up-to policy and for series of different demand patterns. We find that forecast accuracy is more relevant when holding cost is similar or larger than that associated with lost sales. Therefore, in applications where the latter cost exceeds the former, the preferable forecasting method may not be the most accurate one, especially for relatively short review periods and lead times, as well as products characterized by intermittency. Based on our results we discuss some practical concerns for decision making.