Can Supply-Chain Data Build Better Counterfactual Stocks for Event Studies?
September 14, 2026
Altsets
Research by Altsets Research
Potentially, yes. The graph can help construct controls that resemble the treated company while excluding stocks exposed to the same customer, supplier, or bottleneck event, producing a cleaner economic counterfactual than a broad sector benchmark alone.
Data used:Altsets Supply Chain Intelligence: 90k+ entities, 400k+ relationships, 20+ years of history.
Key findings
- The supplied 561M USD HPE-Microsoft relationship provides a concrete treated edge where a control group should resemble HPE while excluding companies that also carry material Microsoft exposure.
- Recent causal-event-study research and large placebo exercises show that counterfactual quality is context dependent, making network-based matching and contamination checks empirically testable rather than automatically superior.
Potentially, yes. Supply-chain data can improve the counterfactual in an event study by helping the researcher find stocks that resemble the treated company but do not share the same customer, supplier, or bottleneck exposure. That can produce a more economically defensible control than comparing the treated stock with a broad sector index that may contain companies exposed to completely different business drivers.
The HPE-Microsoft relationship shows what the control problem looks like
The supplied Altsets data contains a 561M USD relationship between HPE and Microsoft. Suppose the quant wants to estimate how a Microsoft-specific enterprise event affects HPE. A conventional event study might compare HPE with the market or with a technology-sector benchmark. That removes broad market movement but does not necessarily create a close counterfactual for what HPE would have done without the Microsoft-specific information.
A supply-chain-aware control process can use several dimensions at once. Candidate controls can match HPE on industry, size, volatility, momentum, and pre-event returns while excluding companies with meaningful Microsoft relationships. The graph can also match companies on broader dependency structure, such as customer concentration or supplier-network shape, so the control resembles HPE economically without sharing the treatment path being tested.
Synthetic-control methods make this a portfolio-construction problem
Synthetic control methods construct a weighted combination of untreated units designed to reproduce the treated unit before an event. Recent work on financial event studies argues that synthetic portfolios can be useful when traditional factor models provide a poor counterfactual, especially over volatile or longer event windows. Federal Reserve research using large numbers of placebo studies also emphasizes that counterfactual-method performance is context dependent and that placebo testing is necessary rather than assuming one estimator dominates everywhere.
Supply-chain data adds a new set of matching variables to that framework. A synthetic HPE can be asked to resemble HPE in pre-event returns and conventional characteristics while also resembling HPE's overall network structure, excluding the specific Microsoft link that defines treatment. If the synthetic portfolio tracks HPE closely before the event and diverges only after Microsoft releases the relevant information, the causal interpretation becomes more credible than a raw HPE price move.
The graph can help prevent contaminated controls
A common event-study mistake is allowing the control group to receive the same treatment through another path. If the event is Microsoft-specific, a control stock that also depends materially on Microsoft is not truly untreated. The relationship graph can identify those hidden exposures even when the company belongs to a different sector.
The same logic applies to supplier shocks. A Micron event study around an ASML disruption should not use another semiconductor company as a clean control without checking whether that company also depends on ASML. Industry matching alone can accidentally place the same upstream shock on both sides of the comparison. Network-aware exclusion can make the control group cleaner before any return data is examined.
Placebo tests can determine whether the network actually improves the counterfactual
A relationship-aware synthetic control should be evaluated against simpler alternatives. Run placebo events on dates with no relevant customer news, compare pre-event tracking error with and without network features, and test whether the method produces false treatment effects when the treated company is replaced by an untreated company.
If network matching improves pre-event fit but also creates unstable post-event estimates, the added complexity may not help. If it consistently creates better pre-event replication and cleaner placebo behavior, the supply-chain variables are contributing something that sector and price-history matching missed. The graph should earn its place through counterfactual quality, not because it sounds economically sophisticated.
The conclusion is that the network can improve who counts as untreated
Potentially, yes. Supply-chain data can improve event-study counterfactuals by helping a quant build controls that look like the treated company while excluding companies exposed to the same customer, supplier, or bottleneck event. That is a distinct use of the graph: not forecasting the event response, but improving the benchmark against which the response is measured.
The event-study guide explains how relationship data can define treatment before returns are observed. The causal-spillover guide explains the placebo and falsification tests needed before calling a connected-stock response a genuine information channel.
For relationship definitions and evidence limits, read the Altsets methodology.