Polymarket reaches 94% accuracy in event predictions

Research by data scientist Alex McCullough shows that Polymarket, a blockchain-based prediction market, has demonstrated up to 94% accuracy in forecasting events hours before they occur.

The findings, based on historical data, indicate that the platform maintains around 90% accuracy even one month in advance. 

McCullough analysed Polymarket’s past performance using a Dune dashboard, filtering out extreme probabilities—those above 90% or below 10%—to refine the accuracy assessment.

According to his study, the platform tends to slightly overestimate probabilities, which may be influenced by factors such as acquiescence bias, herd mentality, low liquidity, and a preference for high-risk bets.  

“Longer-term markets tend to be more accurate because they include many outcomes that are clearly unlikely, making predictions easier,” McCullough said in an interview with Polymarket’s The Oracle blog.

He cited the example of Gavin Newsom’s presidential chances in the last election, where a $54 million market showed an outcome that was easily predictable, helping boost the platform’s overall accuracy.  

Sports betting on Polymarket has also shown significant growth, with nearly $4.5 billion in total volume wagered on major events, including the NBA, MLB, Champions League, and Premier League finals.

Unlike political betting markets that often feature extreme long-shot candidates, sports markets have a more balanced distribution of outcomes, which contributes to their increasing accuracy over time.  

McCullough’s findings suggest that Polymarket’s predictive power could attract more attention, including from political analysts.

In Canada, for example, Polymarket currently indicates that new Liberal Party leader Mark Carney has a stronger lead over his Conservative opponent Pierre Poilievre than traditional poll aggregators suggest.  

Polymarket’s performance highlights the potential of decentralised prediction platforms in forecasting real-world events with high accuracy.

However, factors such as biases and liquidity constraints remain challenges in ensuring precise probability assessments across different market types.