NBER Paper: Prediction Markets Have Shown No Partisan Bias Since 1880

NBER Working Paper No. 35846 found that prediction markets did not systematically overestimate left-wing or right-wing candidates. Across all markets, the partisan coefficient was +0.03 percentage points, with a standard error of 2.66; the confidence intervals for all four political regressions included zero. On October 9, a16z cited the study in “This Week in Charts,” using its Kalshi market chart for Brazil’s presidential election to illustrate how prediction markets reflected the outcome about three weeks in advance. The study shows only that errors did not consistently favor one party; it does not prove that markets make accurate predictions.