Cause of divergence between Polymarket and polling forecasts in 2024

Determine the underlying factors that caused Polymarket betting-market predictions and aggregated public-opinion polling forecasts to diverge beginning around October 18, 2024 in predicting the outcome of the 2024 U.S. presidential election, including identifying structural, behavioral, or data-generating mechanisms specific to these two information sources.

Background

Using Bayesian Structural Time Series models, the paper finds that market-based probabilities from Polymarket and polling-based estimates began producing distinct, non-overlapping forecasts in mid-October 2024. While Polymarket forecasts dynamically responded to major events and increasingly favored Trump, polling aggregates remained more static and generally favored Harris. The authors explicitly note that the reason for this divergence is unknown and suggest demographic differences between market participants and poll respondents as one possible—but unverified—explanation.