The Question
When a major automaker reports strong sales, which semiconductor firms benefit? When a commodity price spike hits chemical manufacturers, which downstream consumer goods companies see margin compression?
Traditional factor models treat these as independent events—different sectors, different risk exposures. But capital flows through supply chains, and returns follow capital.
Can you systematically map these commercial relationships and predict correlated movements before the market recognizes them?
The Approach
We construct supply chain networks using Numinor's Customer & Supplier dataset, which captures disclosed relationships between listed companies and their key trading partners (both listed and unlisted). Each edge in the graph is weighted by transaction amount or period-end balance—reflecting economic importance, not just binary connectivity.
We build two types of networks:
1. Direct networks: Explicit supplier → customer relationships from disclosure data
2. Extended networks: Add indirect connections—supplier A and supplier B both serve customer C, creating a latent "supplier-supplier" edge that captures shared exposure
We then apply Leiden community detection to identify clusters of companies with dense internal supply chain connections. These communities represent economically cohesive groups that share risks and opportunities through the value chain.
The hypothesis: stocks within the same supply chain community should exhibit higher return correlations than randomly selected pairs, providing both risk management insight (hedging) and alpha opportunities (lead-lag relationships).
The Finding
Directly connected stocks showed significantly stronger return correlations than unconnected stocks—for example, a mean correlation of 0.42 versus 0.36 for unconnected pairs in one representative period. This wasn't subtle: the effect persisted across multiple reporting periods, market cap segments, and industry classifications.
Indirect connections also mattered. Stocks connected through shared trading partners (supplier-supplier links via common customers, customer-customer links via common suppliers) exhibited elevated correlations relative to baseline, though weaker than direct relationships. This suggests second-order dependencies flow through the network, not just first-order capital transfers.
Supply chain communities detected via Leiden algorithm provided incremental information beyond traditional industry classifications. Stocks in the same community had higher correlations (e.g., 0.38 vs. 0.36 for out-of-community pairs), and these communities had low overlap (~1-2%) with standard industry groupings. In other words, supply chain structure reveals market segmentation that sector codes miss entirely.
The effect was robust: it held within specific market cap buckets (controlling for size effects) and within industries (controlling for sector co-movement). This is pure relationship alpha, not a proxy for other known factors.
Try It Yourself
Mapping supply chain relationships enables multiple strategies:
- Lead-lag trading: When upstream suppliers move, predict which downstream customers will follow (or vice versa)
- Correlation-aware hedging: Construct sector-neutral portfolios that account for hidden supply chain exposures, reducing unexpected common risk
- Pairs trading: Identify statistically cointegrated pairs grounded in fundamental commercial relationships rather than spurious historical correlations
- Event-driven alpha: When news hits one node (earnings surprise, production disruption), predict which connected stocks will react next—and trade ahead of the crowd
The challenge is data freshness: supply chain relationships change as companies disclose new customers or suppliers drop off. Institutional investors need pipelines that update these networks quarterly and re-run community detection dynamically.
Ready to build supply chain network strategies into your research process? Book a call to discuss graph construction, backtest infrastructure, and portfolio integration.