Causal Discovery in Multi-Echelon Supply Networks: Leveraging Foundation Models for Demand Propagation Analysis

Abstract

The complexity of modern multi-echelon supply networks presents significant challenges in understanding demand propagation patterns and causal relationships across network tiers. Traditional correlation-based approaches fail to capture the true causal mechanisms underlying supply chain disruptions and demand amplification phenomena. This research proposes a novel framework that integrates causal discovery methodologies with foundation models to analyze demand propagation in multi-echelon supply networks. By leveraging large-scale pre-trained models adapted for supply chain analytics, we develop a system capable of identifying causal relationships between demand signals, inventory decisions, and operational parameters across network echelons. The framework employs graph neural networks combined with causal inference algorithms to construct dynamic causal graphs that represent inter-echelon dependencies. Our approach addresses the limitations of existing methods by explicitly modeling directional causality rather than mere correlation, enabling more accurate root cause attribution and predictive capabilities. Empirical validation using supply chain simulation data demonstrates that network structural parameters significantly impact demand amplification, with high echelon configurations exhibiting peak amplification ratios exceeding fifty times baseline levels. The temporal evolution analysis reveals distinct propagation patterns across different network structures, validating the framework's ability to capture complex spatio-temporal dynamics. This research contributes to the emerging field of foundation models in supply chain management while advancing causal discovery techniques for complex network structures.