Prediction Market Accuracy Is Not "Wisdom of the Crowd" – Polymarket Full Transaction Data Reveals Price Discovery by an Informed 3% Minority

The Limits of the Conventional "Wisdom of Crowds" Narrative

CEOs of Polymarket, Kalshi, and Robinhood have consistently attributed prediction market accuracy to the "Wisdom of Crowds." The authors formalize this view in a Kyle (1985)-style model and test it using the complete on-chain trading record from Polygon, where Polymarket settles its trades.

The team developed a novel "sign-randomization test," resampling each account's actual trade directions (buy/sell) 10,000 times at the event level to separate skill from luck. The 1.72 million accounts fall into five groups: skilled winners (3.14%), lucky winners (29.0%), unlucky losers (61.4%), unskilled losers (6.4%), and market makers (0.1%).

The Informed 3% Drives Price Discovery

Skilled winners and market makers together represent only 3.5% of accounts but capture over 30% of total profits. Average per-account profit reaches USD 3,255 for skilled winners and USD 11,832 for market makers, while 67% of losing accounts absorb the entirety of aggregate losses – a clear zero-sum structure.

Most notable is skill persistence. Of accounts classified as skilled in the training set, 44% remain skilled in the test set – more than four times the 10% persistence rate observed for active mutual funds, suggesting structural information advantage rather than luck. Around FOMC announcements and corporate earnings releases, only skilled winners' order flow consistently moved in the direction of the news surprise.

Insider Trading Risk Materializes

The paper documents a case in which three Polymarket accounts opened just before the January 2026 U.S. operation to capture Venezuelan President Maduro took large positions and collectively realized over USD 630,000 in profits. In April 2026, the CFTC filed a complaint against an active-duty U.S. Army service member for insider trading on the platform. With light regulatory oversight and pseudonymous participation, integrity controls are emerging as an industry-wide priority. Polymarket's monthly volume grew roughly 600x, from USD 3.3 million in December 2023 to USD 1.98 billion in December 2025, making this issue increasingly material.

[Business Development Insights]

1. Market opportunity for informed-signal data products: The order flow of the skilled 3% is a far higher-value information source than aggregate "crowd" sentiment. With energy traders reportedly already using Polymarket signals in global oil markets, there is structural demand for B2B information products that extract and distribute skilled-trader order-flow indicators to institutional investors and corporates.
2. Skill-validated copy trading and investment products: Naive PnL-based rankings misclassify "lucky winners" as truly skilled (60% of whom flip to losses in the next period). Copy-trading platforms and skilled-trader-linked index products built on sign-randomization-based skill scores represent a clear differentiator in the Web3 fund space.
3. Expanding demand for compliance technology: A two-factor anomaly-detection logic combining "timing" (accounts opened just before an event) and "conviction" (concentrated single-event positions) will likely be required as U.S. regulation of prediction markets advances. For on-chain analytics firms (TRM Labs, Chainalysis, etc.), insider detection is shaping up as a next major revenue stream.

[Sources]

• Gómez-Cram, Guo, Jensen, Kung (2026) "Prediction Market Accuracy: Crowd Wisdom or Informed Minority?", SSRN Working Paper ID 6617059
• TRM Labs "How Prediction Markets Scaled to USD 21B in Monthly Volume in 2026" (March 2026)
• Cointelegraph "Prediction markets reflect 'wisdom of an informed minority,' not crowd: Study" (April 2026)
• The Block "Skilled Polymarket traders are a 3% minority" (April 2026)
• Polymarket official statistics page