The probability layer: when a price becomes data

An exchange sells two things. The first is obvious: a place to trade. The second sneaks up on the industry about once a generation — the record of everything that traded, and at what price.

Floor traders had a word for that second product: the tape. The tape is the running print of every trade a venue ever matched, and it ages differently than the venue does.

Take the group built around the London Stock Exchange, an institution that has been matching buyers and sellers since 1801. In 2024, its data-and-analytics division alone brought in £4.0 billion of revenue (excluding recoveries) — more than twice the £1.8 billion of its capital-markets division, the division that contains the actual trading venues (LSEG 2025). The exchange is the landmark. The data is the business.

And the group did not grow that data business out of its own tape — it mostly bought it, at a price that made headlines. That is not a caveat to the lesson; it is the lesson’s sharpest form: when an exchange group works out where the durable value sits, it pays up for the archive.

The shape of the pattern is simple: volume is cyclical, but the archive only accumulates. Trading activity rises and falls with the market’s mood. The tape compounds.

In late 2025, that lesson arrived in this category.

The purchase

In October 2025, Intercontinental Exchange — ICE, the owner of the New York Stock Exchange — announced a strategic investment of up to $2 billion in what its release called “the world’s largest prediction market.” The most interesting sentence in the announcement was not about trading: alongside the investment, ICE said, it would become a global distributor of the venue’s event-driven data — “sentiment indicators,” in its words, “on topics of market relevance” (ICE 2025).

By February 2026, the sentence was a product. Event probabilities, normalized against ICE’s entity and reference databases and mapped to specific securities and companies, delivered through the same consolidated feed that carries securities pricing, fundamental data and corporate actions — with the historical series available alongside, for backtesting (ICE 2026).

Read the delivery mechanism closely, because it is this piece’s thesis in a single fact: probabilities from a prediction market now stream beside the prices of stocks and bonds, as institutional market data — and the archive is part of the offer.

A securities-data company did not pay for entertainment. It bought distribution of a data species its own catalog could not produce.

What a settled probability is

That prices carry information is one of the oldest results in economics. Hayek’s 1945 essay made the canonical argument: a price compresses knowledge scattered across thousands of people — knowledge no single mind holds and no survey can collect — into a single number anyone can act on (Hayek 1945).

Markets have been caught doing this in the wild for decades. In a study now four decades old, the daily move in orange-juice futures — a contract on concentrate from a crop grown almost entirely in one region of Florida — predicted errors in the National Weather Service’s own temperature forecast for that region (Roll 1984). The traders were not meteorologists — some, Roll noted, likely leaned on private forecasts of their own — and that is the point: the price was where everything anyone knew about tomorrow’s weather, including what the official forecast had not yet absorbed, got netted into a number.

A prediction market is that machinery pointed at an explicit question — and the pointing changes what the output is. A stock price says something diffuse about a company’s future. A prediction-market price answers a stated question, by a stated deadline, against a settlement source named in advance — and the answer becomes public either way.

So a settled probability is a dated, priced, accountable forecast. Dated, because the number verifiably existed before the outcome did. Priced, because moving it cost somebody something — exposure to being publicly wrong, in whatever unit the venue keeps score — and nobody gets to restate it for free (where a market price comes from). Accountable, because a source named in advance graded it, under rules written before anyone held a position (how an honest venue settles a question).

Set the familiar public records of belief beside that triad and each drops at least one property. A poll is dated, but it is not a forecast — it records preferences on the day it was taken (Erikson and Wlezien 2008). Commentary is free to state and carries no grade of its own — where pundit predictions have been scored, the scoring was a researcher’s reconstruction, years later, on someone else’s terms. A model’s forecast can be rigorous, but its record is kept at its author’s discretion, on its author’s terms — published, revised, or withdrawn as the author chooses.

The field’s scientists made the institutional case early. In 2008, twenty-two scholars — Nobel laureates among them — argued in Science that prediction-market forecasts carry lower prediction error than conventional forecasting methods, and urged regulators to clear a path for them (Arrow et al. 2008). Seventeen years later, a securities-data company cleared a commercial one.

And the species has documented defects. The same record that shows these prices informative shows them biased at the extremes and least trustworthy on thin, long-dated books (what the accuracy record actually says). That does not disqualify the data. It defines its spec sheet — a clean archive ships with its own error bars.

What a clean archive requires

Not every venue’s history can become this product. Data of this species inherits every property of the venue that produced it, and three of those properties decide whether an archive is an instrument or a curiosity.

Honest books. The number is only worth archiving if it was made the honest way — by real orders, from participants free to disagree, each order carrying that same exposure, whatever the unit (where a market price comes from). A price series produced by anything else — quotes nobody could fill, flow nobody will explain — is noise wearing the costume of belief, and downstream processing can strip out only the noise it can identify. It cannot recover a belief that was never committed.

Named settlement. The outcome column is the archive’s most valuable column, and it is only trustworthy if the settlement rules were fixed before the outcomes arrived (how an honest venue settles a question). Settlement discipline is not compliance hygiene. It is data quality.

Unbroken provenance. The exact wording of a question is part of the datum. A cross-venue map of aligned questions — more than 100,000 listed events across ten venues — had to match language, resolution rules and time scope before it could even call two markets the same question; and between markets judged equivalent, gaps of 2–4% persist after execution costs, held open by structural frictions rather than disagreement (Gebele and Matthes 2026, preprint). Price gaps of that kind have a long pedigree in securities markets too (Lamont and Thaler 2003). A price is a venue-local fact — and an archive that preserves “roughly what was asked” has discarded the part of the record that decides what its numbers meant.

And one property arrives only with time. A probability cannot be scored one question at a time — a 70 that missed was not necessarily wrong — so accuracy exists only in aggregate (what the record shows). Accuracy is a property of tapes, not of trades. Which is why this asset cannot be rushed: the parts of the record that decide between instrument and curiosity — the exact wording as it stood, the settlement rules as written, the book as it filled — are captured at the time or not at all.

The reference layer

Assemble the pieces. Institutional demand for event probabilities is now the stated strategy of an exchange group. The supply is a data species with three hard requirements, and a venue either meets them from the start or never will. And the value sits in the aggregate, which means it compounds with every settled question and belongs to whoever kept the record from the beginning.

Note what the industry’s purchase does and does not settle. Distribution can be bought — that is what just happened. The three requirements cannot be manufactured after the fact; they are properties of how a venue ran on every day of its history, and an archive either has them or it never will.

Securities markets have a name for where this ends: the reference layer — the data that everyone else’s models, products and reporting treat as ground truth. Reference data is unglamorous, decades-deep, and quietly among the most durable businesses in finance. In this category, that seat is the probability layer — and it is being contested now, while public attention stays on the live numbers.

Whoever keeps the cleanest tape holds the strongest claim on the category’s reference layer.

The crowd watches the screen. The industry that prices information for a living has now told you, in its own press releases, where it looks: at the tape. Look there too.

Sources

1. Hayek, F. A. (1945). “The Use of Knowledge in Society.” The American Economic Review 35(4): 519–530. jstor.org
2. Roll, R. (1984). “Orange Juice and Weather.” The American Economic Review 74(5): 861–880. library.caltech.edu
3. Erikson, R. S., Wlezien, C. (2008). “Are Political Markets Really Superior to Polls as Election Predictors?” Public Opinion Quarterly 72(2): 190–215. academic.oup.com
4. Arrow, K. J., et al. (2008). “The Promise of Prediction Markets.” Science 320(5878): 877–878. science.org
5. London Stock Exchange Group (2025). “LSEG Annual Report 2024.” lseg.com
6. Intercontinental Exchange (2025). “ICE Announces Strategic Investment in Polymarket.” Press release, October 7, 2025. ir.theice.com
7. Intercontinental Exchange (2026). “ICE Launches Polymarket Signals and Sentiment Tool Turning Crowd-Sourced Dynamic Views into Market Opportunities.” Press release, February 11, 2026. ir.theice.com
8. Gebele, J., Matthes, F. (2026). “Semantic Non-Fungibility and Violations of the Law of One Price in Prediction Markets.” Preprint, arXiv:2601.01706. arxiv.org
9. Lamont, O. A., Thaler, R. H. (2003). “Anomalies: The Law of One Price in Financial Markets.” Journal of Economic Perspectives 17(4): 191–202. aeaweb.org