Can FX Conversion Create a Fake Supply-Chain Factor?
September 14, 2026
Altsets
Research by Altsets Research
Yes. Using the wrong exchange-rate date can create artificial historical changes in cross-currency relationship-size ranks even when the underlying commercial relationships did not change.
Data used:Altsets Supply Chain Intelligence: 90k+ entities, 400k+ relationships, 20+ years of history.
Key findings
- The supplied network contains a 4.7T KRW LG Energy Solution-Tesla relationship and a 561M USD HPE-Microsoft relationship, making historical FX normalization necessary before absolute sizes can be compared.
- Supplier revenue share and customer cost share avoid currency translation but use different denominators, so they should remain separate directional feature families rather than being treated as one exposure score.
Yes. Converting historical relationship sizes with the wrong exchange-rate date can create artificial rank changes and make a global supply-chain factor appear to move when the underlying commercial relationships did not. Cross-currency relationship-size features therefore need point-in-time FX normalization, while directional revenue and cost percentages should remain separate feature families rather than being blended into one universal exposure number.
Absolute relationship sizes need a common historical currency
The supplied Altsets data includes a 4.7T KRW relationship between LG Energy Solution and Tesla and a 561M USD relationship between HPE and Microsoft. Those raw magnitudes cannot be ranked against each other because they use different currencies. A historical cross-sectional factor based on absolute relationship size needs a conversion convention tied to the historical snapshot or reporting period.
Official long-run exchange-rate data makes that possible. The Federal Reserve H.10 release provides historical foreign-exchange-rate series, and the BIS maintains broad exchange-rate datasets used in international research. The quant should document whether the factor uses an end-of-month rate, monthly average, or another defensible convention and then apply that convention consistently through the point-in-time history.
Today's FX can rewrite yesterday's factor ranking
Suppose a Korean-won relationship from 2018 and a U.S.-dollar relationship from 2018 are converted using a 2026 exchange rate. The result measures what the old KRW amount would translate to at today's currency value, not the relationship's comparable dollar scale in 2018. A material FX move between the dates can change the rank even though neither commercial relationship changed.
That creates exactly the kind of preprocessing artifact that can become a fake signal. Historical relationship-size ranks, z-scores, and threshold crossings can move because the researcher changed the translation rate rather than because the network changed. Point-in-time FX prevents the model from introducing a modern currency view into an old cross-section.
Directional percentages avoid FX but answer different questions
Supplier revenue percentage and customer cost percentage do not need exchange-rate conversion because each is a ratio inside one company's financials. That makes them useful cross-country features, but they are not interchangeable. The supplied LG Energy Solution-Tesla relationship is associated with 19.03% of LG Energy Solution revenue and 3.41% of Tesla's cost base in the displayed data. The two percentages describe different sides of the relationship and have different denominators.
A quant can use supplier revenue share as a customer-dependence feature and customer cost share as a supplier-materiality feature. They should remain separate columns or feature families. Avoiding FX normalization is not a reason to add them together or rank them as if 19.03 and 3.41 were measurements of the same economic object.
FX can become an economic interaction after normalization is solved
Cross-border relationships create another research possibility. Currency moves can matter to supplier margins, pricing, or translated revenue when the commercial arrangement actually carries currency exposure. The supply-chain graph can identify candidate cross-border company pairs, while public disclosures are needed to determine whether invoice currency, production geography, hedging, or contract terms make the FX channel plausible.
That is different from the preprocessing step. First, normalize absolute relationship sizes correctly so the historical feature is not distorted. Then, if evidence supports it, test an interaction between relationship exposure and currency moves as an economic hypothesis. Company domicile alone should not be used to invent invoice currency.
The conclusion is to keep translation from becoming alpha
Yes, FX conversion can create a fake supply-chain factor if historical amounts are normalized with the wrong rates. Use point-in-time currency conversion for absolute relationship sizes, preserve supplier revenue and customer cost percentages as distinct directional features, and test FX as an economic interaction only when the actual commercial relationship supports that mechanism.
The relationship-size versus relative-exposure guide explains why absolute size and directional percentages answer different questions. The foreign-stock exposure guide explains why international securities can retain cross-border economic dependencies even when the tradable wrapper changes.
For relationship definitions and evidence limits, read the Altsets methodology.