Harvard researchers found $143 million in anomalous Polymarket profit. The mechanism that makes prediction markets accurate is the same mechanism that makes insider trading invisible.
Researchers at Harvard screened every trade on Polymarket from February 2024 through February 2026. Ninety-three thousand markets. Nearly fifty thousand unique wallet addresses. More than two hundred thousand suspicious wallet-market pairs. They found one hundred and forty-three million dollars in anomalous profit. Flagged traders achieved a sixty-nine point nine percent win rate, a result exceeding the null distribution of random chance by more than sixty standard deviations.
One account placed its first trade seventy-one minutes before news of the U.S.-Israeli strike on Iran broke. Markets implied a seventeen percent probability. The account collected five hundred and fifty-three thousand dollars. On April 7, at least fifty newly created accounts placed bets on a U.S.-Iran ceasefire hours before Trump announced one. Three accounts collected over six hundred thousand dollars.
The study does not call this insider trading. It says the statistical anomalies are large enough that the word lucky requires a sixty-standard-deviation coincidence.
The Mechanism
Prediction markets work by aggregation. Thousands of individual estimates collide in a continuous auction. The price that emerges is more accurate than any individual participant because it incorporates information no single participant possesses. This is Hayek's insight from 1945: the price system communicates dispersed knowledge that exists nowhere in totality.
But aggregation has a second function that operates simultaneously with discovery: concealment. When a market converts thousands of individual bets into a single probability, it strips the provenance from every signal. An intelligence analyst's fifty-thousand-dollar position and a college student's fifty-dollar bet both enter the price. The probability that emerges is correct precisely because it launders the source.
This is not a bug in the mechanism. It is the mechanism.
The accuracy and the opacity are the same function. A market that revealed which signals were informed would allow everyone to free-ride on those signals, destroying the incentive to bring information in the first place. Grossman and Stiglitz proved in 1980 that perfectly informationally efficient markets are impossible for exactly this reason. The market needs insiders to be accurate. It needs their identity concealed to maintain the incentive. The concealment that enables insider profit is structurally identical to the concealment that enables price discovery.
The Harvard study's blockchain analysis attempts to reverse the laundering, tracing aggregated price signals back to individual wallets using five forensic dimensions: bet size, profitability, timing, concentration, and cross-sectional anomaly. But even this analysis is probabilistic. It identified anomalous trading, not insider trading. The laundering is definitional: once information enters the price, the price does not remember where it came from.
The Response
The policy apparatus is mobilizing as if the laundering were incidental rather than constitutive.
The White House warned staff that using nonpublic information for prediction market bets violates federal ethics regulations. On April 9, Senator Blumenthal sent a letter to Polymarket calling it an illicit market to sell and exploit national security secrets unlike any in history. Representative Torres introduced the Public Integrity in Financial Prediction Markets Act. Senators Curtis and Schiff introduced the Prediction Markets Are Gambling Act. At least a dozen bills now target prediction markets from different angles.
Each intervention assumes the problem can be isolated from the product. Ban government officials. Restrict military-adjacent contracts. Increase surveillance. Prosecute the anomalies. The logic is borrowed from securities regulation, where insider trading is a deviation from a market that functions perfectly well without insiders.
But prediction markets are not securities markets. In equities, fundamental value exists independently of who trades. A company has earnings, assets, cash flows. The price should reflect those fundamentals regardless of participants. In prediction markets, there is no fundamental value independent of the information participants bring. The probability is the aggregate of private estimates. Remove the most informed estimates and the probability degrades. The insider is not parasitic on the market. The insider is the market's substrate.
The Precedent
Credit rating agencies ran the same mechanism for decades. Thousands of individual mortgage assessments were aggregated into a single letter grade. The rating was useful precisely because it concealed the underlying complexity. An investor did not need to evaluate six thousand individual mortgages. The rating did the laundering.
The laundering worked until it didn't. The same concealment that made ratings useful made fraud invisible. Subprime mortgages entered collateralized debt obligations, the CDO entered the rating model, the model produced AAA, and the AAA entered pension portfolios. At every step, the aggregation stripped provenance. No single participant could trace the chain from a NINJA loan in Phoenix to a retirement fund in Oslo. The information was simultaneously aggregated and laundered.
Dodd-Frank imposed disclosure requirements, separated rating from consulting, and mandated skin in the game. But the structural tension was never resolved. Ratings still aggregate. Aggregation still conceals. The regulation manages the tension. It does not dissolve it.
The Impossibility
The Harvard study quantified what prediction markets would prefer to leave unnamed: the mechanism that makes them accurate is the same mechanism that makes them exploitable. Every information aggregation system faces this tradeoff. Peer review aggregates expert judgment and conceals the politics. Polling aggregates voter intent and conceals the methodology. Intelligence estimates aggregate classified sources and conceal the sourcing. Each is valuable because of the aggregation and dangerous because of the concealment.
One hundred and forty-three million dollars in anomalous profit is not a bug to be patched. It is the market's own accuracy, measured in the currency of those who brought the information.
The question is not whether prediction markets can eliminate insider profit. The question is whether the public is willing to pay for accuracy with opacity, whether the same mechanism that launders national security secrets into probabilities is worth preserving because it launders them.
Sixty standard deviations say the market knows more than it should. The entire regulatory apparatus now mobilizing says that knowing more than you should is the problem. Neither has reckoned with the possibility that these are the same statement.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)