A new working paper from researchers at London Business School and Yale casts doubt on the familiar pitch that prediction markets succeed because “the crowd” aggregates distributed knowledge. Using every Polymarket trade from 2023–2025, the team finds that a tiny, informed minority — not the mass of participants — drives price discovery.
What the researchers did
- Authors Roberto Gómez-Cram, Yunhan Guo, Theis Ingerslev Jensen and Howard Kung analyzed 1.72 million accounts and $13.76 billion in trading volume on Polymarket.
- To separate skill from luck, they replayed each trader’s bets 10,000 times with everything held constant except the buy/sell direction (a coin-flip benchmark). This produced an expected profit distribution for each trader assuming no edge.
Key findings
- Just 3% of traders account for most price discovery — they consistently move prices toward the correct outcome.
- The remaining 97% mainly provide liquidity and volume but, on aggregate, lose to the informed minority.
- Among the biggest raw winners, only 12% beat the coin-flip benchmark; about 60% of apparent “lucky winners” flip to losers when their performance is evaluated on a separate set of events.
- Skilled traders are faster to incorporate new public information — e.g., Fed announcements or earnings — and their presence measurably improves market accuracy, especially near resolution.
- When traders use non-public information (insider-style trades), price impact per dollar is far larger — roughly 7–12x the effect of ordinary skilled trades — but those cases are rare and concentrated in a few events rather than driving day-to-day price discovery.
A concrete red flag: the Maduro example
The paper highlights a striking episode tied to the U.S. removal of Nicolás Maduro from power in January. Three newly created Polymarket accounts placed unusually large bets on a contract asking whether Maduro would be removed, when the market priced the odds at about 10%. The accounts placed tens of thousands of shares before prices moved; after the operation, they collectively pocketed more than $630,000. Two stopped trading entirely soon after; the third went mostly dormant. The authors note there is no public evidence of wrongdoing, but the episode illustrates how non-public information — if it appears in markets — can produce outsized, concentrated profits.
Implications for crypto prediction markets
Polymarket and Kalshi both ban trading on non-public information, but this research makes the enforcement challenge concrete. More broadly, the study upends the standard “wisdom of crowds” narrative for prediction markets: accuracy appears to hinge less on broad participation and more on a small set of repeat, informed traders. For crypto-native markets and tokenized prediction platforms, that raises questions about market design, transparency and surveillance — and about whether ordinary users are trading against skillful insiders more often than they realize.
Bottom line
Prediction markets do appear to be informative — but not because of a dispersed crowd. They work largely because a small, skilled minority consistently moves prices in the right direction; everyone else tends to subsidize those gains.
Read more AI-generated news on: undefined/news
