Multi-step sales forecasting in automotive industry based on structural relationship identification

Forecasting sales and demand over 6-24 month horizon is crucial for planning the production processes of automotive and other complex product industries (e.g., electronics and heavy equipment) where typical concept-to-release times are 12-60 month long. However, nonlinear and nonstationary evolution and dependencies with diverse macroeconomic variables hinder accurate long-term prediction of the future of automotive sales. In this paper, a structural relationship identification methodology that uses a battery of statistical unit root, weakly exogeneity, Granger-causality and cointegration tests, is presented to identify the dynamic couplings among automobile sales and economic indicators. Our empirical analysis indicates that automobile sales at segment levels have a long-run equilibrium relationship (cointegration) with identified economic indicators. A vector error correction model (VECM) of multi-segment automobile sales was estimated based on impulse response functions to quantify long-term impact of these economic indicators on sales. Comparisons of prediction accuracy demonstrate that VECM model outperforms other classical and advanced time-series techniques. The empirical results suggest that VECM can significantly improve prediction accuracy of automotive sales for 12-month ahead prediction in terms of RMSE (42.73%) and MAPE (42.25%), compared to the classical time series techniques.