Prediction markets are widely considered the most accurate forecasting tool we have, routinely outperforming polls, pundits, and expert panels. But they are not oracles. Every so often, the crowd gets it spectacularly wrong, and traders who spot the mispricing walk away with life-changing payouts. Below, we rank the most consequential prediction market upsets of the last decade, the structural reasons the crowd missed each one, and what today’s traders can learn from these historic blowouts.
1. Brexit, June 2016: 89% Remain, 0% Remain
Hours before the United Kingdom voted to leave the European Union, prediction markets on Betfair and PredictIt pegged the probability of “Remain” at roughly 89%. Within six hours of polls closing, that number collapsed to effectively zero. It was one of the largest single-event probability moves ever recorded on a liquid prediction market.
The error was not random. Markets over-weighted urban London polling, discounted the enthusiasm gap in Leave-leaning regions, and reflected the biases of the mostly financial-sector traders who dominated the Betfair book. Sharp traders who had been quietly accumulating “Leave” positions at prices between 11 and 18 cents cashed out at 92 cents or higher. One syndicate of British quants reportedly cleared more than £8 million on the night.
2. Trump Wins the Presidency, November 2016
Four and a half months after Brexit, prediction markets made the same mistake at even larger scale. On the morning of November 8, 2016, Betfair, PredictIt, and the now-defunct IEM all showed Hillary Clinton with an implied probability between 78% and 85%. By 11 p.m. Eastern, Trump contracts had traded from the mid-teens into the 90s in the span of three hours.
The lesson, in hindsight, was that prediction markets inherit the biases of their underlying inputs. Nearly every public poll had Clinton leading in Wisconsin, Michigan, and Pennsylvania. The markets correctly aggregated those polls. The polls themselves were wrong. For the first time, serious traders realized that a prediction market is only as accurate as the information sources feeding it, and that polling-model concentration was a hidden systemic risk.
3. Leicester City Wins the Premier League, 2016
At the start of the 2015 to 2016 English Premier League season, Leicester City was priced at 5,000 to 1 on UK betting exchanges, implying a probability of 0.02%. The club had narrowly avoided relegation the previous year. By the time Leicester clinched the title in May 2016, it had become the single most famous outlier in sports prediction market history.
Only 25 bettors in the United Kingdom reportedly held outright Leicester tickets from the preseason. The total payout exceeded £25 million. The episode is now taught in finance programs as a textbook case of tail-risk mispricing: markets are efficient at the median, but fat-tail events, by definition, cannot be priced by consensus.
4. The 2022 FTX Collapse on Polymarket
In early November 2022, Polymarket contracts on “Will FTX be solvent at year end?” traded above 95% for more than 48 hours after the first CoinDesk report revealed cracks in Alameda’s balance sheet. A small group of on-chain analysts who had actually read the leaked balance sheet took the other side at 3 to 7 cents. When FTX filed Chapter 11 on November 11, those positions paid out between 14x and 33x. One wallet, later identified as an ex-Jump Research trader, cleared $4.2 million from a $180,000 stake.
5. The 2026 Nevada Senate Shocker
The most recent major upset happened just last week. Kalshi and Polymarket both had the Democratic incumbent in Nevada at 73% to 76% to hold the seat through the night of November 3. The Republican challenger won by 2.1 points, driven by a late surge in Clark County suburbs that no major poll had detected. Contracts on the Republican flip, which had traded as low as 18 cents on Election Day morning, settled at $1.00 within hours. Combined volume across the two venues exceeded $41 million, and the move triggered a brief but sharp drop in the overall GOP Senate control contract, which quickly recovered.
What These Upsets Have in Common
Every one of these mispricings shares three structural features. First, the consensus was built on a narrow set of correlated inputs: UK polls, US state polls, football form tables, or a single auditor’s opinion. Second, the losing side of the trade was psychologically uncomfortable, which kept casual money away and let prices drift further from reality. Third, the sharp traders on the winning side had access to primary-source information, whether that was on-the-ground canvassing data, balance-sheet forensics, or simply a willingness to read what the crowd ignored.
The takeaway for today’s traders is not that prediction markets are unreliable. On the overwhelming majority of contracts, they beat every other forecasting method on the planet. The takeaway is that the biggest profits in this space come from identifying the small number of contracts where the crowd is systematically miscalibrated, and having the discipline to size into them.
Where to Trade
If you want to position yourself for the next major upset, the two deepest and most liquid venues in the market remain Kalshi, the CFTC-regulated US exchange with the broadest coverage of political and economic contracts, and Polymarket, the leading crypto-settled venue for global events and tail-risk trades. For a full side-by-side of fees, available markets, and funding methods, see our complete rankings of the best prediction markets.