“markets priced at p percent delivered p percent, across a million questions and four platforms”
Alex Tabarrok flags a PNAS study scoring roughly one million resolved questions across Polymarket, Kalshi, Manifold and Metaculus, spanning twelve domains and horizons from hours to years. Calibration held: events priced at p% happened about p% of the time, on every platform. The paper's own caveat is that fine-grained cuts still show systematic over- and underestimation in the most consequential areas, and individual events retain no certainty. Tabarrok's verdict: can't beat that.
No technical background needed
Platforms mentioned: Polymarket, Kalshi, Manifold, Metaculus