What the Crowd Doesn't Know It Doesn't Know

There is a particular kind of authority that attaches itself to a number. Not the authority of an argument, which can be rebutted, or the authority of an expert, who can be discredited, but something more ambient and harder to dislodge—the authority of a price. Prices feel like verdicts. They carry the weight of aggregated judgment, the implicit testimony of everyone who looked at something and decided what it was worth. When a market says 64 cents, it feels like the world saying 64 percent. It feels, in other words, like the truth.

This is the feeling that prediction markets have been selling for the better part of a decade. And it is, according to a study now sitting in a federal regulatory docket in Washington, not quite right.

The Agency Nobody Watches

The Commodity Futures Trading Commission is not, by temperament, a dramatic institution. It was created in 1974 to oversee futures trading in agricultural commodities—corn, wheat, cattle—and has spent most of its existence in the unglamorous middle distance between Wall Street and Main Street, writing rules that almost nobody reads about instruments that almost nobody fully understands. Its headquarters, a limestone building on Twenty-First Street Northwest, is the kind of place where significant things happen very quietly.

In March of this year, the CFTC published what it called an Advance Notice of Proposed Rulemaking on prediction markets. This is the bureaucratic equivalent of clearing one's throat. It is the agency saying, formally, that it is thinking about thinking about a rule. The document runs to nine pages in the Federal Register. It asks forty questions. It invites public comment through April 30th. It will be read, in its entirety, by perhaps several hundred people in the United States.

One of the comments it received contains a dataset of 291,309 resolved prediction-market contracts, drawn from six platforms including Kalshi—the New York-based exchange that fought all the way to the D.C. Circuit Court of Appeals for the right to let Americans bet on elections. The comment is dense with econometrics. It is not written for general audiences. But its central finding translates with uncomfortable clarity: the prices that prediction markets publish, and that journalists quote, and that campaign managers refresh obsessively on their phones at midnight, are not probabilities. They have never been probabilities. They are something related to probabilities, but bent—systematically, measurably, and by design—in a direction that nobody has been adequately disclosing.

The number that does the bending is called lambda. It is approximately 0.183. It is, the study's author argues, the most important number in the prediction-market industry that the prediction-market industry has never mentioned to its customers.

The Puzzle That Started at the Racetrack

To understand what lambda does, it helps to start not with finance but with a simpler kind of wager—the kind that horse-racing economists have been puzzling over since the 1940s.

The puzzle is this: if you look at decades of parimutuel horse-race results and compare the odds posted on each horse to the frequency with which horses at those odds actually win, you find something odd. Heavy favorites—horses listed at short odds, horses the crowd believes are nearly certain to win—win slightly less often than their odds imply. Long shots—horses the crowd has written off—win slightly more often than their odds imply. The crowd, in other words, is not quite right about either end of the spectrum. It is too confident about the favorite and too dismissive of the underdog.

Economists named this the favorite-longshot bias and spent the better part of fifty years arguing about what caused it. Was it irrational gambling psychology? A love of the underdog? The thrill of the long shot? Probably some of all of that, at the racetrack.

But prediction markets, their proponents have always insisted, are different. They are not casinos. They are not governed by sentiment or superstition. They are governed by money, which is a powerful corrective to irrationality. Rational people, putting real dollars at stake, will price events correctly, because they have every financial incentive to do so. The wisdom of crowds is not a metaphor; it is a mechanism. The market will find the truth.

This is a genuinely compelling argument. It is also, the 291,309-contract study suggests, incomplete in a way that turns out to matter enormously.

What the Argument Leaves Out

Here is what the argument leaves out.

When you buy a prediction-market contract—when you put sixty cents on a contract that pays a dollar if a particular candidate wins an election—you are not simply expressing a belief about probability. You are also bearing risk. Your sixty cents is real, and it is at risk right now, today, regardless of what happens in November. The dollar you might receive is a promise about the future. And promises about the future, as any trader will tell you, are worth less than cash in hand, because the future is uncertain, and uncertainty has a price.

That price is what finance calls a risk premium. It is the extra return that investors demand in exchange for accepting uncertainty. It is why bonds pay interest instead of just returning the principal. It is why insurance companies charge more than the actuarial value of your house burning down. It is, in short, one of the most fundamental concepts in all of finance—and it is why, in every derivatives market that has ever existed, the price of a contract and the true probability of its outcome are related but not identical.

The study filed with the CFTC finds that prediction markets are no exception. The risk premium it measures—lambda—is approximately 0.183, which sounds abstract until you apply it to a concrete case. A contract trading at sixty cents does not imply a sixty percent probability. It implies something closer to forty-nine percent. A contract trading at eighty cents implies something closer to seventy-one percent. Favorites, the ones the market believes in most strongly, are the most systematically underpriced relative to their true odds. Longshots are overpriced. The favorite-longshot bias, it turns out, is not a psychological artifact of the racetrack. It is a structural feature of any market where real money meets genuine uncertainty.

The proof, elegant in its simplicity, comes from the play-money platforms—sites where you can trade prediction contracts using fake currency with no financial stakes at all. On those platforms, the bias reverses. Prices on play-money sites track realized frequencies with reasonable fidelity. Remove the financial risk, and the distortion disappears. This is not what you would expect if the bias were cognitive—if it were a feature of how human minds misperceive probability. It is exactly what you would expect if the bias were financial—if it were a feature of how markets price the bearing of risk.

The market, in other words, is not wrong. It is answering a different question than everyone thinks it is answering.

Two Questions, One Number

The question it is actually answering is this: at what price does a rational, risk-averse person break even on accepting this uncertainty?

The question everyone thinks it is answering is: how likely is this event to occur?

Those are related questions. But they are not the same question. And the gap between them—lambda, 0.183—has been, for the past several years, invisible to essentially everyone who has quoted a prediction-market price as though it were a probability.

Consider what that means in practice. During the 2024 election cycle, a major prediction market briefly showed one of the presidential candidates trading above sixty-four cents on the dollar. This was reported, in serious newspapers and on serious television programs, as evidence of strong market confidence—as a signal that informed, financially incentivized traders believed this outcome was substantially more likely than not. The framing was everywhere: the markets say. As though the markets were oracles rather than exchanges.

Adjusted for lambda, that sixty-four-cent contract implied a probability closer to fifty-three percent. A coin flip, essentially. The dramatic market signal was, in a more precise reading, a shrug.

This does not mean the market was wrong about the election. It means the number was being used to say something it was not, technically, saying. The precision—sixty-four percent, not fifty-three, not fifty-fifty—was doing enormous rhetorical work in newsrooms and campaign war rooms, and that precision was, to a measurable degree, illusory.

The Promise and the Overclaim

Prediction-market companies, for their part, have built entire brands around the claim that prices are probabilities. Kalshi's marketing materials have described its contracts as yielding "real probability estimates." Polymarket has been cited by major publications as offering "the odds" on everything from elections to central-bank decisions to geopolitical crises. The implicit promise—sometimes explicit—is that these prices represent the aggregated, incentivized, wisdom-of-crowds best guess at the truth.

That promise is not entirely false. Prediction markets do aggregate information. They do give financially incentivized participants a reason to trade on genuine beliefs. The prices are not random. They are correlated with outcomes in meaningful ways. The literature on their forecasting value is real, and serious, and worth taking seriously.

But correlation is not calibration. A thermometer that reads five degrees too high is still a thermometer—it still tells you something about temperature—but you would not want a doctor using it to make a diagnosis without knowing about the offset. Lambda is the offset. And the offset has never appeared, as far as anyone can determine, in any disclosure document, marketing material, or editorial note accompanying a prediction-market price.

Neither Kalshi nor Polymarket responded to questions about whether they intend to revise how they present accuracy metrics in light of this analysis. The CFTC declined to comment on specific submissions to the rulemaking.

What the Regulator Can Now Do

What makes this moment genuinely significant—and not merely a technical dispute among economists—is that the CFTC is now, for the first time, in a position to do something about it.

The agency's March notice asked, with some delicacy, whether prediction-market prices are reliable indicators of probability, and what role "informed participants" with "asymmetric information advantages" play in setting those prices. It asked about manipulation. It asked about disclosure. It asked, in the careful language of administrative law, whether the public interest was being served.

The 291,309-contract study answers some of those questions more directly than the agency may have anticipated. It provides, for the first time, a large-sample, platform-agnostic empirical foundation for treating prediction markets as what they legally are—derivative instruments—rather than what they have been culturally marketed as, which is something between a poll and a prophecy.

If the CFTC accepts that framing, the regulatory implications are substantial. Disclosure requirements. Marketing restrictions. Standards around how platforms describe the relationship between prices and probabilities. These would not be radical interventions. They would, in fact, bring prediction markets into alignment with the standards applied to every other derivative instrument sold to the American public—which is to say, standards that require telling customers, with some specificity, what they are actually buying.

The Cost of a Simpler Story

There is something almost poignant about the position prediction markets now find themselves in. They arrived on the scene bearing a genuinely interesting idea: that decentralized, financially incentivized markets might aggregate information more accurately than polls, pundits, or prognosticators. That the crowd, properly incentivized, might know things that experts did not. It was a democratic idea, in its way. It had intellectual heft. The academic literature behind it spans decades and several Nobel-adjacent careers.

And then the industry, in the process of commercializing that idea, overclaimed it. The leap from markets aggregate information usefully to market prices are probabilities is not a small one. It elides an entire field of financial economics—the theory of risk premia, the distinction between risk-neutral and real-world probability measures, the century of scholarship on what prices in financial markets actually represent. The industry made that leap anyway, because the simpler story was more marketable. Sixty-four percent is a headline. A risk-adjusted derivative price implying something in the neighborhood of fifty-three percent, subject to liquidity conditions and the structure of the underlying market is not.

The CFTC's comment docket closes April 30th. Whether the agency will act on what it finds there—whether it will, in effect, require prediction markets to tell a more complicated and more accurate story about themselves—is genuinely uncertain.

The irony is that no market is currently pricing that probability. Or if one is, you should probably adjust the number by about eighteen points before you believe it.

The CFTC's advance notice of proposed rulemaking on prediction markets, docket RIN 3038-AF65, is open for public comment through April 30, 2026, at comments.cftc.gov.

Thank you for reading Low Form News. We're literally just getting started, but let me know what you think so far. Send me a note at editor@lowform.news. Have a great weekend!