Polymarket’s Predictive Accuracy

Polymarket serves as a crystal ball, predicting events with nearly 90% accuracy, as shown by a Dune dashboard from New York City-based data scientist Alex McCullough.

McCullough analyzed Polymarket’s historical data, excluding markets with probabilities greater than 90% or less than 10% once outcomes were known but not settled. This ensured the analysis remained accurate.

His research found that Polymarket tends to slightly overestimate event probabilities across most ranges, likely due to biases such as acquiescence bias, herd mentality, low liquidity, and participants favoring high-risk bets.

In his interview with Polymarket’s The Oracle blog, McCullough noted that longer-term markets, which consider events far in advance, tend to be more accurate. They include outcomes that are evidently unlikely, which facilitates easier predictions.

For example, he cited the significant market volume of $54 million regarding Gavin Newsom’s presidential bid during the last election. The extended timeline often results in markets including clearer predictable outcomes, thus enhancing accuracy.

In contrast, head-to-head sports markets present a clearer depiction of predictive accuracy, showing notable improvements as events unfold. With nearly $4.5 billion wagered collectively on various sports outcomes, Polymarket’s growth in this sector is evident.

Interestingly, McCullough’s findings are particularly relevant in Ottawa, where Polymarket indicates that new Liberal Party leader Mark Carney holds a considerable lead over Conservative rival Pierre Poilievre, surpassing predictions from traditional poll aggregators.