Revista OEconomica

Tudor Constantin BĂLAN - Cycles of Freedom: Are Our Rights Today Permanently Secured, or Are We in a Transitional Phase? The Cases of Great Britain, the United States, and Russia

Marius-Cristian PANĂ, Adela BĂLAN - Economic Freedom and Prosperity: A Comparative Analysis of Welfare Determinants at the International Level

Ana Octavia ALBU - When Accurate Forecasts Still Produce Waste: Demand Shocks and Over-Forecasting Risk in Food Retail

Adrian-Ioan DAMOC - The Persian Pivot and the Reconfiguration of Geopolitical Realism: Iran and Venezuela in Strategic Competition among the United States, China, and Russia

Emmanuel Olusegun STOBER - Coercive Diplomacy in the Gulf and the Limits of the Strike-Forward Strategy: Toward a Middle-Power Coalition for Regional Stability

Sorin-Nicolae CURCĂ - On the Informational Entropy of Economic Shock

Florin DĂNESCU - The Banking Sector as Critical Infrastructure in the Logic of Hybrid Conflict: Post-2014 Legislative Convergence in Central and Eastern Europe and the Case of Romania

Mihai LĂCĂTUŞ - GDP as a Means, Not an End: Inequality as the Missing Variable in Fiscal Policy Design – A 60-Country Analysis of Economic Growth and Subjective Well-Being

Articles on Issue Theme

Ana Octavia ALBU

Academia de Studii Economice din Bucureşti

This paper examines how promotions and holidays affect forecasting risk in perishable food retail. Using daily sales data from the Corporación Favorita dataset, the study analyses three stores and ten perishable product families during 2016. The focus is on over-forecast errors, defined as cases in which predicted demand exceeds actual sales and may therefore create excess inventory. A seasonal naïve model is compared with Random Forest and XGBoost forecasts. The results show that promotions and holidays both increase over-forecast errors, with holidays producing the greatest instability. Machine learning models reduce the magnitude of these errors compared with the seasonal naïve benchmark, but remain vulnerable during event-driven demand shifts. The findings also show that conventional accuracy metrics can lead to different conclusions from waste-oriented indicators. While the seasonal naïve model performs best according to MAPE, XGBoost produces the lowest estimated waste. The paper argues that forecasting systems for perishable goods should be evaluated not only by average prediction accuracy, but also by their behaviour during high-risk retail events. This perspective is relevant for retailers seeking to align demand forecasting, inventory control, and food waste reduction.

ŒCONOMICA no. 1-2/2026

Keywords: perishable food retail, demand forecasting, machine learning, over-forecasting, food waste

JEL: C53, L81, Q18, Q56

When Accurate Forecasts Still Produce Waste: Demand Shocks and Over-Forecasting Risk in Food Retail

Select Issue: