Forecasting Steel Demand: Comparative Analysis of Predictability across diverse Countries and Regions

Abstract

Steel demand forecasting is crucial for steel industries, enabling effective production planning, inventory management, resource optimization, and strategic decision-making. This also holds true for steel plant manufacturers, who require insights into future demand trends to assess market potential, evaluate project feasibility, and identify growth regions. This paper addresses the predictability of steel demand across countries and regions worldwide, employing a comparative analysis of linear and non-linear modeling approaches to forecast steel demand per capita. A global model is trained in order to integrate information from diverse countries at varying stages of development, allowing the model to learn from comparable conditions and historical patterns. For each modeling approach hyperparameter tuning and selection of lagged input variables was performed. The performance of different model configurations was evaluated and compared for individual countries. This evaluation reveals that the predictability varies across countries and regions and shows for which countries a simple regression model based on past data is sufficient and for which countries more sophisticated models are needed. The developed models are useful in providing insights into consumption patterns, comparative risk assessment, and development of steel consumption.