New research has confirmed the accuracy of the blockchain-based prediction platform Polymarket. According to the data collated and analyzed, the project is at least 90% accurate.

Alex McCullough, a data scientist based in New York, conducted the research and outlined the results in a dashboard on the market analytics platform Dune. The dashboard tracks Polymarket’s accuracy over a period of one month, one week, a day, 12 hours, and four hours before the markets resolve.

A 90% Success Rate

McCullough’s findings revealed that Polymarket is 90.5% accurate one month before markets resolve, 89.2% correct one week before, and 88.6% accurate a day before. The platform also has a 90.2% accuracy 12 hours before markets resolve, and the figure surges to 94.2% four hours before the bets are over.

During an interview with Polymarket’s blog, The Oracle, McCullough revealed that he picked the time frames because they showed the most interesting data. The data scientist said four hours was the minimum time needed because markets do not resolve immediately. Sometimes, markets even take days between when the predicted event occurs and when the resolve happens.

McCullough measured the accuracy by counting markets above 50% that resolved to “Yes” and “No” as correct. He also studied Polymarket’s historical data and removed any extreme probabilities.

Long-Term Markets Have Higher Accuracy

Further discoveries revealed that prediction markets may get more accurate over time; however, this is not reflected on the platform until four hours before bets are resolved. While Polymarket makes accurate predictions the majority of the time, McCullough found that bias affects the platform’s results.

Causes of the bias on Polymarket include herd mentality, low liquidity, and acquiescence bias. Due to these factors, market participants seem to overestimate the likelihood of most events by a few points. This also results in most markets being overpriced and resolving to “Yes” less frequently than expected.

When asked why markets are more accurate a month out than a day before resolution, McCullough explained that such scenarios occur during markets that stay open for longer periods.

“The longer-term markets tend to have some options that are extreme locks as a “No” for example Gavin Newsom becoming president in the last election. Longer term markets tend to have more of these highly certain outcomes, which explains their higher accuracy as a group,” he added.