A score-driven model of short-term demand forecasting for retail distribution centers

• Published in: Journal of retailing Vol. 97; no. 4; pp. 715 - 725
• Main Authors: Hoeltgebaum, Henrique, Borenstein, Denis, Fernandes, Cristiano, Veiga, Álvaro
• Format: Journal Article
• Language: English
• Published: Greenwich Elsevier Inc 01-12-2021; Elsevier Limited
• Subjects: Artificial intelligence; Datasets; Decision making; Distribution center aggregate data; Fuzzy logic; Inventory; Inventory control; Inventory management; Literature reviews; Log-normal predictive density; Neural networks; Retail demand forecasting; Retail sales; Retail stores; Sales forecasting; Score-driven models; Seasonal variations; Time series; Trends; Variables; Log-normal predictive density; Retail demand forecasting; Score-driven models; Distribution center aggregate data
• ISSN: 0022-4359; 1873-3271
• DOI: 10.1016/j.jretai.2021.05.003

•GAS models are competitive to forecast daily retail demand.•GAS models are suitable to calculate daily retail demand predictive density.•Time varying variance GAS did not perform well with DC-level aggregate demand.•Price must be used as an exogenous variable, regardless of the adopted model.
[Display omitted]
Forecasting is one of the fundamental inputs to support planning decisions in retail chains. Frequently, forecasting systems in retail are based on Gaussian models, which may be highly unrealistic when considering daily retail data. In addition, the majority of these systems rely on point forecasts, limiting their practical use in retailing decisions, which often requires the full predictive density for decision making. The main contribution of this paper is the modeling of daily distribution centers (DCs) level aggregate demand forecasting using a recently proposed framework for non-Gaussian time series called score-driven models or Generalized Autoregressive Score (GAS) models. An experimental study was carried out using real data from a large retail chain in Brazil. A log-normal GAS model is compared to usual benchmarks, namely neural networks, linear regression, and exponential smoothing. The results show the GAS model is a competitive alternative to retail demand forecasting in daily frequency, with the advantage of producing a closed form predictive density by construction.