A multivariate approach for multi-step demand forecasting in assembly industries: Empirical evidence from an automotive supply chain

Demand forecasting works as a basis for operating, business and production planning decisions in many supply chain contexts. Yet, how to accurately predict the manufacturer's demand components presence of end-customer uncertainty remains poorly understood. Assigning proper order quantities suppliers thus becomes nontrivial task, with significant impact on planning, capacity inventory-related costs. This paper introduces multivariate approach throughout multiple forecast horizons using different leading indicators shifts. We compare autoregressive integrated moving average model exogenous inputs (ARIMAX) Machine Learning (ML) models. Using real case study, we empirically evaluate performance regression models over component's life-cycle. The experiments show that proposed provides superior inventory compared traditional univariate benchmarks. Moreover, it reveals applicable life-cycle, not just single stage. Particularly, found signals at beginning life-cycle are predicted better by ARIMAX model, but is outperformed ML-based later stages.