Prediction markets entered the mainstream selling a simple promise: prices are smarter than polls because traders put money behind their beliefs. Polymarket founder Shayne Coplan called his platform "the most accurate thing we have as mankind right now." The US midterms are about to test that claim against a growing record of spectacular misses. The failures are not edge cases. In Brazil's 2022 presidential runoff, Polymarket priced a Bolsonaro victory at 80-84% while polls showed a statistical tie; Lula won 51.8% to 48.2%. In Serbia's 2026 snap election, Polymarket gave Vučić's party a 70% chance; the opposition took 52.6%. In Wisconsin's Democratic gubernatorial primary, both platforms gave Francesca Hong roughly 95%; she lost by half a point. In New York City's mayoral primary, Polymarket gave Andrew Cuomo 66%; Zohran Mamdani beat him by 12 points. Polymarket deleted an X post after the Wisconsin miss. A structural explanation is emerging that undermines the founding mythology. Research shows just 1% of Polymarket accounts captured three-quarters of all profits since 2022, with the top 0.1% taking half. Two-thirds of accounts have lost money. A Yale-affiliated study estimates only 3% of accounts belong to skilled traders whose activity actually drives prices toward accuracy. This is not the wisdom of crowds. It is the wisdom of a tiny informed elite — which works only when the elite has room to operate. The mechanism breaks precisely when it matters most. High-profile elections draw waves of recreational money from bettors treating wagers as expressions of loyalty rather than probabilistic assessments. A Vanderbilt analysis of 2,500 political markets during the 2024 presidential race — involving $2.3bn in transactions — found accuracy dropped to 67% on Polymarket and 78% on Kalshi, with prices diverging sharply across platforms and contract types. The signal-to-noise ratio collapses under volume. There is an important technical distinction that prediction market evangelists exploit and critics overlook. Market prices express win probabilities, not vote shares. A 70% probability assigned to a candidate polling at 51% is not necessarily wrong — it reflects the compounding of small leads through election mechanics. But this gap also provides cover: almost any outcome can be retroactively reconciled with almost any price. The markets get credit for Trump 2024 (where polls showed a toss-up and markets saw a clear lean) while explaining away Brazil, Serbia, and NYC as within their probabilistic range. The regulatory architecture amplifies the stakes. Trump's pick for CFTC chair, Mike Selig, previously lawyered for both Polymarket and Kalshi. The CFTC now states on its website that event contracts can "sometimes forecast outcomes better than polling." Newsrooms including AP, CNN, the Wall Street Journal, and CNBC have integrated live election odds into coverage. A former congressman recently cited a candidate's "99% chance to win on Polymarket" without referencing polls. The platforms have embedded themselves into the information infrastructure before proving their accuracy claims. October 2025 trading volume hit $21bn — a monthly record, up from $90m in the same period before the 2024 presidential election, and $4m before the 2022 midterms. That 5,000x growth in three years means the midterms will test not just accuracy but market integrity: whether prices can remain informative when the ratio of informed traders to recreational bettors has shifted dramatically, and when the financial incentives to manipulate outcomes have grown by orders of magnitude.
