3% of Traders Drive Prediction Market Accuracy
Only 3% of traders drive prediction markets’ accuracy, not the crowd, study finds
A new study has found that prediction markets’ forecasting performance is largely driven by a small minority of participants, rather than by broad “wisdom of the crowd” dynamics.
According to the research, roughly 3% of traders account for most of the accuracy observed in prediction markets. The finding challenges a common assumption that these markets are reliable primarily because they aggregate many independent opinions into a single price.
The result matters because prediction markets are often cited as a way to produce probabilistic forecasts on real-world outcomes, from elections and policy decisions to business events. If accuracy depends heavily on a narrow slice of participants, the market’s reliability may hinge on whether those informed traders are present, active, and able to trade freely.
In practical terms, the study suggests that market accuracy may be less about the sheer number of participants and more about who is participating. That has implications for market design and oversight, including questions about liquidity, incentives, and how platforms handle fees, position limits, and access restrictions that could influence whether high-skill traders stay engaged.
The finding also adds context to ongoing debates in crypto-adjacent prediction markets, where onchain platforms and traditional venues alike often argue that market prices provide a neutral, data-driven signal. The study indicates that the signal may reflect the actions of a small group with superior information or forecasting ability, rather than a broad consensus.
More broadly, the research contributes to a growing body of work examining where prediction markets get their informational edge—and under what conditions that edge can weaken. If only a small fraction of users meaningfully improves accuracy, platform operators and policymakers may need to think differently about how participation, market structure, and data quality interact.
