Can Customer and Supplier Relationship Churn Become a Quant Signal?
September 14, 2026
Altsets
Research by Altsets Research
Point-in-time supply-chain history can turn edge formation, persistence, magnitude change, weakening, and dissolution into temporal features, provided the model separates genuine commercial change from disclosure, coverage, and corporate-action changes.
Data used:Altsets Supply Chain Intelligence: 90k+ entities, 400k+ relationships, 20+ years of history.
Key findings
- Historical relationship snapshots can produce edge-age, persistence, churn, magnitude-change, and termination-hazard features that capture how the economic network is changing rather than only its current shape.
- Relationship-dissolution research and evidence on supply-chain shocks show that ties can weaken, terminate, or partially re-form for economically meaningful reasons, but a quant dataset must distinguish those events from observation changes before treating graph churn as a signal.
Potentially. Customer and supplier relationship churn can become a quant signal if genuine commercial changes are separated from disclosure thresholds, backfills, acquisitions, delistings, and coverage changes. Point-in-time history lets a model treat relationship formation, persistence, weakening, and dissolution as events rather than static background information.
Edge birth and edge death are not ordinary binary features
A new customer edge can mean several things. The company may have genuinely won new business, an old relationship may have become large enough to be disclosed, or the dataset may have obtained better evidence for a relationship that existed earlier. A disappearing edge can likewise reflect true termination, a drop below a disclosure threshold, an acquisition, a delisting, or missing coverage. The raw 0-to-1 and 1-to-0 transitions therefore need evidence and observation rules before they become trading signals.
Research on buyer-supplier relationship dissolution makes this distinction important. A 2026 review emphasizes that relationship endings can reflect multiple stressors and that dissolution is a process rather than one simple event. Other empirical research shows that mergers can terminate customer-supplier ties and that supplier incidents can increase the probability of relationship termination or partial trade cuts. These findings create legitimate quant hypotheses, but they also warn against treating every missing historical edge as a confirmed breakup.
Relationship duration can become a feature in its own right
A long-lived relationship may represent deep integration, switching costs, product qualification, or durable demand. A newly formed relationship can contain more uncertainty but also more potential information if the market has not fully incorporated the new commercial path. With monthly point-in-time snapshots, the quant can construct relationship age, persistence, consecutive-observation counts, and time-since-last-change features without reducing the whole history to one current edge.
Duration models are especially natural for this problem. The researcher can estimate the hazard that a relationship disappears as a function of economic importance, customer concentration, industry conditions, company fundamentals, or prior network changes. The predicted termination risk can then be tested as a supplier-risk feature or used to condition event studies. The goal is not to assume old relationships are safe. It is to quantify whether persistence contains information after other factors are controlled.
Magnitude change can be more informative than edge disappearance
Many important relationship changes happen on the intensive margin. A customer can remain present while its share of supplier revenue rises sharply, and a supplier can remain in the network while its importance to customer costs declines. A binary graph misses those changes completely. A quant with directional economic metrics can construct percentage-change features, rank changes, acceleration, or transitions between exposure buckets while preserving the difference between supplier revenue share and customer cost share.
The LG Energy Solution-Tesla relationship provides an intuitive current example of why context can deepen around an existing edge. The supplied data already identifies Tesla as economically important to LG Energy Solution, while public 2026 announcements add Megapack 3 product and Lansing production context. That does not provide a historical relationship-change series by itself, but it illustrates the kind of commercial development that a point-in-time graph can eventually capture as persistence or changing magnitude. The quantitative hypothesis is whether such changes contain information before ordinary financial variables fully adjust.
Churn can be measured at the company level as well as the edge level
A company can have a stable number of relationships while replacing many counterparties underneath the total. Another can have a rapidly expanding network with little attrition. Company-level churn metrics can therefore summarize the fraction of customer or supplier edges that are new, lost, or materially changed over a rolling window. High churn might indicate rapid strategic repositioning, weak relationship stability, acquisition integration, or changing disclosure quality, so the sign of the feature should not be assumed in advance.
The model should separate genuine commercial churn from database churn. Coverage expansion, entity-resolution improvements, and historical backfills can create artificial edge births. Acquisitions can internalize relationships without destroying the underlying economic activity. A defensible historical dataset needs corporate-action mapping and observation-date discipline before churn becomes an investable variable. Once those controls are in place, relationship dynamics offer a feature family that static network centrality cannot capture.
The conclusion is that the graph has a velocity as well as a shape
A current network tells the quant who is connected. Historical snapshots can tell the quant whether those connections are becoming stronger, weaker, newer, older, or less stable. Relationship churn turns supply-chain history into a temporal feature space where formation, persistence, magnitude change, and dissolution can be tested directly against future fundamentals, risk, and returns. The key is to distinguish real economic change from disclosure and coverage change before calling graph movement a signal.
The relationship-history guide explains when historical snapshots add information beyond the current graph. The strategic-repositioning guide explains how an existing relationship can change meaning before segment reporting fully reflects the shift.
For relationship definitions and evidence limits, read the Altsets methodology.