Forecasting Automotive Production Volume Using Regression and Time Series Modelling

Abstract

Accurate forecasting of automotive production volumes is a critical capability for suppliers navigating an increasingly volatile industry. Overly optimistic forecasts, particularly from Original Equipment Manufacturers (OEMs), lead to misallocated capacity and lost opportunities across the supply chain. This thesis investigates whether advanced statistical models can improve upon benchmark industry forecasts and provide automotive suppliers with more reliable, practical tools for demand planning. Several forecasting methodologies are evaluated, including ARIMA, standard linear regression, Lasso regression, Theta model, and a hybrid Boosted Theta model. Models are tested across North America, Europe, and Greater China using 2000-2024 vehicle production and macroeconomic data. Results show that Theta model outperforms industry forecasts across both 1-year and 5-year horizons in North America and Europe. Its simplicity, low data requirements, and robustness to market volatility make it suitable for industrial use. The model was successfully implemented at Commonwealth Rolled Products, an aluminum rolling mill in Kentucky, portfolio company of American Industrial Partners (AIP), where it was adopted for 2025 planning and drove a shift towards data-centric forecasting practices. This research presents a real-world example of applying academic techniques to solving actual business problems, serving as a valuable reference for suppliers seeking to improve forecast accuracy and operational planning in the evolving automotive landscape.