In November 2024, when American polling firms projected a tight race between Donald Trump and Kamala Harris, predictive markets – led by Polymarket – showed a probability of Trump's victory exceeding 60%. The final outcome proved them right. This was not a stroke of luck; it is the logical consequence of a fundamental economic mechanism that political science has been slow to recognize.
Polls rely on a sample of a few hundred or thousand people interviewed at a specific moment, using a methodology that has not fundamentally evolved since the 1960s. They measure declared intentions, subject to multiple biases: social desirability, strategic response, emotional volatility. Predictive markets, on the other hand, aggregate the convictions of thousands of participants who risk their own money on the outcome. Information is not declared – it is revealed by economic behavior.
Hayek's theorem, formulated as early as 1945, already explained why markets outperform experts: no individual, however competent, can hold all the information dispersed throughout society. Market prices aggregate this information in a decentralized, real-time manner, with an efficiency that no pollster can match. Each bettor incorporates data that traditional models ignore: local rumors, ground-level trends, faint signals picked up by networks.
Classic objections – possible manipulation, insufficient liquidity, participant bias – have lost their relevance with the rise of platforms like Polymarket, Kalshi, and Metaculus. Trading volumes on Polymarket during the 2024 US election exceeded $3 billion. At this level of liquidity, manipulation attempts become extremely costly and self-correct rapidly.
Should we conclude that polls are obsolete? Not necessarily. Polls remain useful for understanding motivations, concerns, and sociological divides. But for predicting an outcome – an election, a referendum, the adoption of a policy – predictive markets are, structurally, a superior tool. It is time for the media, analysts, and policymakers to acknowledge this and incorporate them into their analysis frameworks.
