Prediction Markets More Accurate Than Polls, But Beware Israeli Elections

A new study examining nearly one million contracts on prediction markets like Polymarket found them to be highly accurate in forecasting events. Professor Michael Gilead, a cognitive-social psychologist at Tel Aviv University, co-authored the research published in PNAS. The study analyzed data collected by Justin Dickerson, comparing market predictions to actual outcomes across various domains, including sports, economics, and even pop culture.

While prediction markets demonstrate significant accuracy, Gilead cautions against relying on them for Israeli elections. He notes that Israel's smaller market size leads to lower trading volumes, making it easier to manipulate prices and create a false sense of momentum, especially close to election day. In contrast, US prediction markets, despite the crisis in polling due to low response rates, have shown strong predictive power. Gilead highlights that prediction markets incentivize realistic assessments due to the financial stakes involved, unlike polls which can be influenced by wishful thinking.

The research measured accuracy using the Brier score, which quantifies the difference between predicted probabilities and actual outcomes. The study found that prediction markets consistently outperformed random chance (like a coin toss) and other benchmarks. However, Gilead also pointed out potential biases, particularly in security-related predictions, where markets were less accurate for positive outcomes like peace compared to negative ones like war, suggesting emotional influence. He also noted that in smaller markets, like Israel's, a few million dollars can significantly sway prices, potentially leading to manipulation by political actors.

Gilead expressed surprise at the markets' accuracy, contrasting it with the work of psychologists Daniel Kahneman and Amos Tversky, who emphasized human irrationality. He believes the collective wisdom of crowds, where diverse perspectives average out errors, drives this accuracy. However, he strongly advises against using prediction markets for election insights in Israel, especially when polls are not being published, due to the high risk of manipulation by political players aiming to create a false impression of momentum. He also offered a personal prediction that the Likud and Yisrael Beiteinu parties would gain more mandates than expected, while Bennett and Lapid's parties would weaken, with one potentially falling below the electoral threshold.