A Comparative Study of ETS, ARIMA, and Reconciliation Techniques for Hierarchical Product Forecasting in Retail SMEs

Small and medium-sized enterprises (SMEs) retail supply chains are important to the global economy but are susceptible to inventory issues. Forecasting products is important to SMEs' stock management since it streamlines inventory, minimizes lost sales, and boosts profit. Product forecasting in retail organizations is complex due to the increasing number of stock-keeping units (SKUs) and the varying demand patterns at different hierarchy levels. However, most current studies on demand forecasting for SMEs focus on only one level of the hierarchy-either the bottom, middle, or top-without considering the entire hierarchy. This study addresses the gap by forecasting across all hierarchy levels and evaluating various reconciliation techniques to enhance prediction accuracy and coherence. The baseline models used are the ETS and ARIMA models, widely used benchmarks for forecasting competitions. The reconciliation techniques analyzed include Bottom-Up, TopDown established by historical proportions (Average Proportions, Proportions Average), Top-Down established by forecast proportions, and MinT in various variants (OLS, OLS Non-Negative, WLS, WLS Non-Negative). The results showed that the combination of the base ETS model with the OLS NonNegative reconciliation method produced the most accurate forecast across all hierarchical levels, with average SMAPE of 35.335%, 13.9% better than the base ETS Model (SMAPE 41.085%) and 6.8 % better than base ARIMA (SMAPE $\mathbf{3 8. 2 6 0 \%}$). These findings offer substantial implications for SME supply chain management by demonstrating that the adoption of hierarchical forecasting and reconciliation methods can significantly enhance forecast accuracy to improve inventory management.