On 8 July 2026, Bloomberg Opinion published an essay by Émile Servan-Schreiber, co-founder of The Forecasting Machine and head of Hypermind, under the headline “Prediction Markets Can Work Without Money on the Line”. He shared it the same day. It was syndicated a week later.
The timing was deliberate. Kalshi and Polymarket were handling billions of dollars in World Cup wagers. Servan-Schreiber listed recent troubles in the sector: suspected insider trading on military operations, manipulated temperature readings, and threats against journalists. His target was the claim that such markets only work when real money is at stake, which he called false. He said he has made a living from prediction markets for 25 years “not as a gambler, but as a scientist and entrepreneur”.
A wager over wine
The essay goes back to 2003. After the Pentagon’s research agency DARPA cancelled its policy prediction market, Servan-Schreiber challenged the economist Justin Wolfers, then at Stanford, on a Washington Post online forum. He bet French wine against California wine that play money would forecast as well as real money.
With David Pennock, they compared NewsFutures, the play-money exchange Servan-Schreiber ran, with TradeSports, a real-money betting exchange based in Ireland. They used every game of the 2003 NFL season. The resulting paper, published in Electronic Markets in 2004, covers 208 games between September and December 2003. Each forecast was the last traded price before noon Eastern time on game day.
- Favourites won 65.9% of games on TradeSports and 66.8% on NewsFutures.
- The two markets were indistinguishable on mean absolute error (0.439 against 0.436), on squared error and on logarithmic score. A randomisation test found no significant difference.
- In an online contest against about 2,000 individual forecasters, the two markets finished 6th and 8th at season’s end, with play money ahead. The average of all the human forecasts ranked 39th.
The paper’s conclusion was short: “In this case, (real) money does not matter.” The authors offered a possible reason. Real money may push traders to find information, while play money may combine it more efficiently. In Bloomberg, Servan-Schreiber summed up the result as “Money did not matter at all.” He collected his wine, a “Stanford Edition” chardonnay.
Design, not dollars
The rest of the essay argues that structure matters more than stakes. Hypermind, which he launched in 2014, is a play-money market with small cash prizes. He wrote that it has handled nearly a million trades on more than 1,000 questions and is well calibrated at every probability level. He also cited Hypermind’s record on Brexit, Donald Trump and Emmanuel Macron in 2016 and 2017, where he said it beat PredictIt and Betfair.
scale and liquidity do not drive accuracy. Design does.
Émile Servan-Schreiber, Bloomberg Opinion, 8 July 2026
His examples of design are equal endowments and the aggregation of polled forecasts. With equal endowments, influence follows insight rather than wealth, because, as he put it, “the big money that moves prices is not necessarily the smart money”. He pointed to the Good Judgment Project, which won a four-year US intelligence forecasting tournament in the early 2010s by weighting forecasters’ predictions by track record and recency. He also cited Glimt, whose 20,000 volunteers forecast in support of Ukraine, and a Johns Hopkins public-health forecasting poll launched a year before Covid. In all three cases the forecasters were not staking their own money.
Integrity and AI
Servan-Schreiber argued that markets without wagers are much less exposed to the manipulation and fraud now affecting real-money platforms. In the comments he linked what he called “just the latest scandal”. Trump’s teleprompter operator was reported to have won more than $100,000 betting on Kalshi “mentions” markets about the President’s own speeches. PBS reported that he was placed on unpaid leave. Kalshi said its surveillance team had flagged the trades and referred them to the Commodity Futures Trading Commission.
On 28 August he made the policy distinction more explicit. Responding to a poll about election betting, he called wagers on elections, wars and terrorism ethically and nationally problematic while arguing that forecasting those outcomes remains useful. He pointed to Glimt as the alternative: a free contest, powered by Hypermind for Sweden’s FOI, whose aggregated forecasts are supplied to Ukrainian decision-makers without asking participants to bet. That is his policy argument, not a finding of the 2004 football study.
The essay ends with AI. Servan-Schreiber wrote that forecasting bots have reached the accuracy of wise human crowds. Panels of language models can now produce evidence-backed probabilities in minutes. He quoted Rafał Kierzenkowski of the OECD’s Strategic Foresight Unit, who said forecasting machines can help with horizon scanning at a scale and speed not possible before. In his account, the role of human forecasting tournaments shifts toward making foresight “a civic act”.
