a16z highlights research finding no systematic partisan tilt in prediction markets

Andreessen Horowitz has highlighted research finding no evidence that election prediction markets systematically price candidates differently according to partisan affiliation. Across the full sample, the estimated partisan return coefficient was 0.03 percentage points, with a standard error of 2.66, a result consistent with no persistent advantage for either political side.

The analysis tested whether trading returns were associated with candidates’ political affiliations and demographic attributes, rather than asking simply whether markets correctly forecast election outcomes. The study found no political, gender, racial or age coefficient that differed from zero by more than its standard error, apart from a smaller non-U.S. election sample.

That exception suggested an apparent tendency to overprice right-leaning candidates in non-U.S. contests. But the estimated coefficient was -1.42, with a standard error of 1.30 and a 95% confidence interval running from -3.97 to 1.13. Because the interval included zero, the result was not statistically significant.

In the study’s framework, a negative coefficient indicates that markets overprice the relevant candidate attribute, while a positive figure indicates underpricing. The research also identified small, statistically insignificant biases in favour of women and non-white candidates and against older candidates.

The broader sample covered prediction-market returns from 1880 to 2025, combining informal election wagers from the late 19th and early 20th centuries with modern commercial platforms. Although it spans more than a century, much of the data came from the more recent period.

The finding on partisanship does not mean that markets were free from all pricing errors. The NBER abstract, as described by PPC Land, found that most political markets exhibited a favourite-longshot bias, in which long shots were overpriced and favourites underpriced. It also identified very small overpricing of left-leaning outcomes in markets on near-term polling averages.

Those findings measure different effects. A market may make frequent forecasting errors yet still lack systematic partisan bias if its mistakes do not consistently favour one side. a16z therefore characterised the results as evidence that prediction markets generally leave partisan preferences out of election pricing, while the separate favourite-longshot pattern remains a broader market-efficiency issue.