Testing Supply Chain Network Centrality as a Quant Factor

Key findings

• The supplied Tesla and Shin-Etsu views illustrate why simple degree and economically weighted degree answer different questions, but local screenshots are not enough to claim a global centrality ranking.
• Academic studies have linked supply chain centrality with crash risk and stock-price efficiency, making centrality a legitimate feature family while still requiring replication against simple size, degree, and coverage baselines.

Supply chain network centrality is a candidate quantitative factor, not evidence of alpha by itself. To test it, define the economic mechanism, construct the directed graph at each historical formation date and compare the resulting feature with size, industry and coverage baselines. A central company can have different risk without offering a predictable excess return.

The supplied Altsets views show why the feature definition matters before any return regression. Tesla connects to upstream companies including LG Energy Solution, Samsung Electronics, Lens Technology, Huayu Automotive Systems and Toyota Industries. That local view establishes observed counterparties, not Tesla's rank in a global network. Shin-Etsu Chemical connects to Samsung, TSMC and Intel with unequal directional economic weights. Counting those customers equally and weighting them by their importance to Shin-Etsu answer different questions.

The graph therefore changes the experiment rather than supplying its conclusion. It lets the researcher compare breadth, economic magnitude and position among connected firms while keeping supplier and customer roles explicit. The test must then determine whether network position adds information after ordinary company characteristics and uneven disclosure coverage are accounted for. Calculating a sophisticated statistic on a local screenshot, or on today's graph joined to old returns, cannot answer that question.

Define direction and weights before choosing a centrality measure

Use a commercial adjacency matrix with supplier companies as rows and customer companies as columns. On that convention, outgoing degree counts observed customers and incoming degree counts observed suppliers. A demand shock can travel economically from customer to supplier, opposite the recorded commercial direction; that does not justify silently reversing the graph between calculations.

Weighted degree needs a similarly explicit perspective. Supplier revenue shares on a company's outgoing edges describe measured customer dependence. Customer cost shares on its incoming edges describe measured procurement exposure. Combining the two into one sum mixes revenue and cost denominators. Relationship size offers a monetary weighting, but requires consistent currency and period treatment.

Shin-Etsu's displayed TSMC, Samsung and Intel relationships illustrate the choice. A binary count assigns each one connection; the directional weights preserve TSMC as the largest of the three on both displayed percentage measures. That establishes a local weighting distinction. It does not establish a global eigenvector or betweenness score, and it does not establish substitutability among customers.

Match the statistic to a mechanism that could fail

Betweenness also requires care when weights enter shortest-path calculations. A stronger economic relationship is not automatically a shorter distance. State the transformation and test alternatives rather than passing exposure percentages to a library whose distance convention was designed for another problem.

Predefine the outcome separately. Crash risk, price efficiency, residual volatility and expected return are different targets. Existing studies cited below associate supply chain centrality with crash risk and stock-price efficiency in their own research settings; those associations motivate replication rather than establish a tradable result for Altsets or another market.

Test coverage as a competing explanation

Large firms may appear in more disclosures, receive more analyst attention and operate in industries with more observable relationships. Centrality can capture those conditions along with economic position. Compare against market capitalization, industry, liquidity, analyst coverage and available relationship counts before attributing a result to global network structure.

Track measured and structural-only edges separately. Replacing an unknown economic weight with zero changes weighted centrality; dropping every unmeasured edge changes the graph itself. A stable-coverage subsample and a binary-graph benchmark can help reveal whether a purported factor is largely a quantification-coverage effect. The missing-data guide develops those encoding choices.

Reconstruct the whole admissible graph at each formation date

Global centrality depends on relationships beyond the focal firm. A future edge elsewhere can change a company's centrality through indirect paths, even if its own direct supplier list is unchanged. Using today's graph historically can therefore contaminate several observations at once.

Build each formation-date graph from the accepted historical nodes, edges and metric versions. Preserve entity mappings, coverage state and preprocessing choices. Fit any normalization on training data, then compare the same candidate features across subsequent test intervals. The historical join guide explains admissibility; it does not validate the centrality hypothesis for the researcher.

Require an incremental result worth the graph complexity

Compare the chosen centrality measure with contemporaneous degree, economic edge weights and ordinary firm characteristics on the same sample and outcome. Use network-aware evaluation so related firms and shared shocks do not masquerade as independent evidence. Report whether improvement appears in risk estimation, price efficiency or returns, rather than calling every association alpha.

The test is successful as research even when the simpler representation wins. That result would suggest that local economic dependence is more useful for the specified outcome than global position. The factor-adjusted downside-beta study asks a narrower edge-versus-centrality question. The network-motif guide and GNN guide explain how local patterns and learned graph representations can be compared with centrality.

Sources

• Altsets supply chain dataset
• Supply chain network centrality and stock price crash risk
• Supply chain network structure and stock price efficiency

Related research

• What Supply Chain Network Motifs Capture That Centrality Misses

• Why Treating Missing Supply Chain Data as Zero Biases Quant Models

• Graph Neural Networks for Supply Chain Trading

• Supply Chain Networks Shrink Your Effective Sample Size

Methodology

Read the methodology for this research.

Cite this research

Altsets Research. "Testing Supply Chain Network Centrality as a Quant Factor." Published September 22, 2026. https://www.altsets.com/research/supply-chain-network-centrality-quant-factor

Copy citationCanonical URLMethodology