A leading macroeconomic indicators' based framework to automatically generate tactical sales forecasts. (January 2020)

• Record Type: Journal Article
• Authors: Verstraete, Gylian; Aghezzaf, El-Houssaine; Desmet, Bram
• Is Part Of: Computers & industrial engineering. Volume 139(2020)
• Publication Date: 2020-01
• Subjects: Forecasting -- Sales forecasting -- Tactical sales forecasting -- Macroeconomic indicators -- Decomposition -- LASSO regression; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2019.106169
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: Automatic tactical sales forecasting. Decomposition is used to split sales into independent components. Trend is predicted using macroeconomic indicators. Abstract: Tactical sales forecasting is fundamental to production, transportation and personnel decisions at all levels of a supply chain. Traditional forecasting methods extrapolate historical sales information to predict future sales. As a result, these methods are not capable of anticipating macroeconomic changes in the business environment that often have a significant impact on the demand. To account for these macroeconomic changes, companies adjust either their statistical forecast manually or rely on an expert forecast. However, both approaches are notoriously biased and expensive. This paper investigates the use of leading macroeconomic indicators in the tactical sales forecasting process. A forecasting framework is established that automatically selects the relevant variables and predicts future sales. Next, the seasonal component is predicted by the seasonal naive method and the long-term trend using a LASSO regression method with macroeconomic indicators, while keeping the size of the indicator's set as small as possible. Finally, the accuracy of the proposed framework is evaluated by quantifying the impact of each individual component. The carried out analysis has shown that the proposed framework achieves a reduction of 54.5% in mean absolute percentage error when compared to the naive forecasting method. Moreover, compared to the best performing conventional methods, a reduction of 25.6% is achieved in the tactical time window over three different real-life case studies from different geographical areas.