Should a Supply-Chain Model Predict Earnings Before It Predicts Stock Returns?
September 14, 2026
Altsets
Research by Altsets Research
Often, yes. Supplier earnings, revenue, margins, and analyst revisions are closer to the customer-supplier mechanism than noisy short-horizon stock returns, making them a cleaner first validation target.
Data used:Altsets Supply Chain Intelligence: 90k+ entities, 400k+ relationships, 20+ years of history.
Key findings
- The supplied Nvidia relationships with Micron and SK Hynix create a direct economic mechanism through which customer demand can affect supplier fundamentals before valuation and market expectations determine the final stock response.
- A two-stage design can first test whether network features improve fundamental forecasts and then test whether any of that information remains unpriced enough to create return predictability.
Often, yes. Earnings, revenue, margins, and analyst revisions are closer to the economic mechanism in customer-supplier data than next-day stock returns, so they can provide a cleaner first test of whether the network contains real information. Return prediction can come second, after the researcher has shown that the relationship features improve an economically adjacent forecast.
Customer demand should reach the business before it reaches an alpha model
The supplied Altsets network maps Micron and SK Hynix to Nvidia, and current public evidence connects both memory suppliers to Nvidia's AI roadmap. If Nvidia demand strengthens, the most direct hypothesis is not automatically that Micron stock rises tomorrow. The business mechanism is that customer demand can change memory shipments, pricing, capacity utilization, supplier revenue expectations, or analyst estimates. The stock response adds another layer because valuation, prior expectations, factor moves, and market efficiency determine how that business information affects the security.
That makes supplier fundamentals a useful first-stage target. A model can test whether customer-side information improves forecasts of supplier revenue growth, earnings changes, guidance, or consensus revisions. If the supply-chain variables fail to improve those outcomes, a strong stock-return backtest deserves extra skepticism because the proposed economic channel is difficult to observe where it should be closest to the business.
A two-stage design separates economics from mispricing
Machine-learning research has shown that financial statement information can improve earnings forecasts. Supply-chain data adds a different source of information: the changing state of economically connected companies. A supplier model can combine its own fundamentals with customer growth, analyst revisions, event variables, and directional relationship metrics, then compare that forecast with one built only from the supplier's own history.
Stage one asks whether the graph improves the fundamental forecast. Stage two asks whether the market has already incorporated that improved forecast. If customer-network features improve Micron revenue or earnings estimates but add no return predictability after momentum and analyst revisions are controlled, the relationship data may still be useful for valuation, risk, or earnings research. If it improves both fundamentals and subsequent return ranking, the researcher has a much more coherent alpha story.
The target should match the cadence of the relationship data
Daily returns are extremely noisy. Many supply-chain relationships and metrics evolve more slowly. A monthly network snapshot can be naturally suited to forecasting next-quarter fundamentals, monthly revisions, or event sensitivity rather than next-day returns. The right target depends on the mechanism.
A customer-demand feature can belong in a quarterly supplier-revenue model. A relationship-churn feature may belong in a multi-month analyst-revision model. A supplier disruption can belong in a short event-window risk model. Choosing the outcome based on the economic path avoids forcing every alternative dataset into the same daily-return template merely because market prices are easy to download.
Fundamental success can still fail as a trading signal
A useful economic forecast does not guarantee a profitable strategy. The market can anticipate the same information, the stock can be too expensive or too cheap for unrelated reasons, and trading costs can consume a small return edge. That distinction is valuable because it tells the researcher where the information stops being incremental.
It also allows better comparisons with other datasets. A sentiment dataset may add short-horizon return information without improving fundamentals. Supply-chain data may improve supplier earnings forecasts but have weaker immediate return predictability. Those datasets are measuring different parts of the information process, and evaluating each only by the same short-horizon Sharpe ratio can obscure where it is actually strongest.
The conclusion is to prove the economic channel before demanding alpha
Often, a supply-chain model should predict the business before it predicts the stock. If customer and supplier features improve forecasts of economically adjacent fundamentals or expectations, the researcher has evidence that the network contains real business information. The next test is whether any of that information remains mispriced enough to create a tradable return signal.
The customer analyst-revisions guide explains how changing customer expectations can become supplier-level features. The incremental-alpha guide explains how return models should test network variables after conventional predictors are already present.
For relationship definitions and evidence limits, read the Altsets methodology.