Economic Measures of Forecast Accuracy for Demand Planning: A Case-Based Discussion

Successful demand planning relies on accurate demand forecasts. Existing demand planning software typically employs (univariate) time series models for this purpose. These methods work well if the demand of a product follows regular patterns. Their power and accuracy are, however, limited if the patterns are disturbed and the demand is driven by irregular external factors such as promotions, events, or weather conditions. Hence, modern machine-learning-based approaches take into account external drivers for improved forecasting and combine various forecasting approaches with situation-dependent strengths. Yet, to substantiate the strength and the impact of single or new methodologies, one is left with the question how to measure and compare the performance or accuracy of different forecasting methods. Standard measures such as root mean square error (RMSE) and mean absolute percentage error (MAPE) may allow for ranking the methods according to their accuracy, but in many cases these measures are difficult to interpret or the rankings are incoherent among different measures. Moreover, the impact of forecasting inaccuracies is usually not reflected by standard measures. In this chapter, we discuss this issue using the example of forecasting the demand of food products. Furthermore, we define alternative measures that provide intuitive guidance for decision makers and users of demand forecasting.