Hedging against demand ambiguity in new product development: a two-stage distributionally robust approach

In the globalization era, many manufacturing companies face great uncertainties, as most components in new product development (NPD) are outsourced to external and internal suppliers worldwide. More seriously, component supply chain disruptions have been seen in the recent global outbreak of COVID-19. Hence, some key components must be reserved in advance to control risks considering the suppliers' production plans and uncertain lead time. We propose a key components reservation model for NPD concerning component commonality and substitution. Demand is characterized by a scenario-wise ambiguity set consisting of mean, support, and mean absolute deviation information. Based on a demand unsatisfied index (DUI), we establish a two-stage distributionally robust optimization (DRO) model, which is reformulated to a linear programming (LP) model by duality analysis and solved by a proposed column and constraints generation (CCG) algorithm. We examine the performance of the DRO model over other benchmark models. The proposed DRO model with DUI has a lower probability and magnitude for demand shortage. The scenario-wise ambiguity set also outperforms the single-scenario and deterministic models, especially when product demand varies inconsistently in different scenarios. The proposed CCG algorithm can significantly improve the solution procedure for large-scale problems.