Polymarket and Kalshi sold themselves as "truth machines" — prices that aggregate information better than pundits, polls, or the Bureau of Labor Statistics. The data on who trades, what they trade, and how often they are wrong tells a different story.
Polymarket and Kalshi sold themselves as "truth machines" — prices that aggregate information better than pundits, polls, or the Bureau of Labor Statistics. The data on who trades, what they trade, and how often they are wrong tells a different story.
Prediction markets are pitched with the language of science — wisdom of crowds, aggregated priors, efficient information. The accounts tell a different story: a few sharks, a sea of retail, and a set of platforms quietly booking the spread.
The question is not whether markets can be informative. It is whether these markets are.
The claim of a "truth machine" is empirical. It is testable. And when tested, it doesn't hold up.
A Vanderbilt study of election outcomes found Polymarket calling races correctly only 67% of the time — and Kalshi 78%. Against a baseline of polls, pundits, and forecasters, the markets are not systematically ahead. In several categories, they are behind.
Two-thirds right is worse than most political forecasters. Four-in-five is the baseline of a competent poll aggregator. Neither is the performance of an oracle.
What the chart below shows is a market that is informative in the weak sense — prices move when news arrives — and unreliable in the strong sense — the price you pay rarely matches the outcome you get.
Defenders of these markets say insider trading is a feature, not a bug. Non-public information, they argue, gets priced in faster. Accuracy improves.
They are not wrong about the mechanism. They are wrong about who benefits.
If insiders always arrive first, the public is always arriving second. The accuracy gain — such as it is — is paid for by the retail trader on the other side of every informed bet.
"Faster price discovery" is another way of saying the uninformed lose faster. That is not a bug on the road to efficiency. It is the business model.
Every market is a negotiation between someone who knows and someone who doesn't. In equities, the SEC draws a line. In sports betting, leagues police player contact. In prediction markets, the position is that there should be no line at all — that information asymmetry is the product.
The 0.04% is not a mystery. It is what the rules were designed to produce.
The marketing is oracles and elections. The order book tells a different story.
Eighty-five to ninety percent of activity on Kalshi is sports betting. The "information aggregation platform" is, by volume, a sportsbook that happens to list a few political contracts on the side.
A regulated sportsbook has problem-gambling disclosures, state licensing fees, age verification required by each state, and mandatory self-exclusion programs. A CFTC-regulated event contract platform has none of these by default.
Same activity. Different shelf.
If a state-licensed sportsbook produced these outcomes, the regulator would intervene. The prediction-market platforms are not state-licensed. The regulator of first resort is a federal commodities agency that was not designed to police consumer gambling harms.
The result: a worse deal, sold under a better label.
Why fight so hard for federal CFTC oversight? The answer is on the tax return and the license agreement.
State gambling regulators require: licensing fees paid per state, revenue taxes that frequently exceed 10%, mandatory problem-gambling funding, state-by-state KYC, and operational audits. Federal commodity oversight requires none of these by default.
Kalshi's legal position — upheld by a federal court in 2024 — is that its event contracts are commodities, not bets. This puts them beyond the reach of the state gambling commissions that would otherwise tax and regulate them as the sportsbooks they functionally are.
Polymarket, re-entering the US market under similar logic, is following the same playbook.
If 85% of Kalshi's activity is sports wagering and most of it is happening with US users, state gambling authorities are watching a large and growing chunk of gambling revenue walk past their borders — toward an exchange that owes them nothing.
The problem-gambling hotlines, the treatment funding, the self-exclusion registries: all of it is funded by state gambling taxes these platforms do not pay.
The phrase "truth machine" was doing a lot of work. It implied neutrality, accuracy, public good. The data on this page — concentration of winnings, missed forecasts, composition of volume, user outcomes, and regulatory route — suggests something narrower.
These are markets. They aggregate something. What they aggregate most reliably is losses from a retail base toward a 0.04% — across a regulatory seam built to keep states out.
This story is adapted from reporting and on-air analysis summarized from a video investigation of prediction-market platforms, including specific figures cited on Polymarket and Kalshi.
Concentration of profits (0.04% / 70%) and the February 2026 BLS payrolls miss (150,000 jobs) are drawn from that investigation. The election-accuracy comparison (Polymarket 67% / Kalshi 78%) is attributed to a Vanderbilt University study cited in the source material.
The composition of Kalshi volume (85–90% sports) and the 3-month user-loss comparison to traditional sportsbooks are attributed to a Citizens Bank analyst report cited in the source material.
Regulatory framing — CFTC jurisdiction vs. state gambling regulators, and the state-level tax, licensing, and problem-gambling obligations bypassed by federal commodity treatment — reflects the current posture of Kalshi and Polymarket as described in the source material.
This is an opinion piece. Numbers are reproduced as cited; the framing is editorial.