Polymarket's 78.5% Earnings Accuracy Exposes 35-Point Analyst Flaw

If you trade earnings season, the consensus number on your terminal is not a neutral forecast. New research shows Polymarket traders correctly called quarterly earnings outcomes 78.5% of the time, versus just 43.7% for Wall Street analyst consensus. That 35-point gap is a challenge to how earnings surprises are measured and how investors should read guidance.

The idea that a pool of anonymous online bettors could outpredict Wall Street's best-compensated analysts sounds like an anomaly story, but a new study suggests it happened consistently. According to Beating the Earnings Game, a study by Daniel Rabetti, Jiaqi Shao and Che Zhang, one of whom researches at the National University of Singapore, Polymarket traders correctly called whether companies would beat quarterly earnings 78.5% of the time over a six-month sample. The Wall Street sell-side consensus was right just 43.7% of the time. Larry Swedroe summarized the findings for Financial Advisor Magazine and called the 35-point gap a signal that deserves a moment of attention from anyone who reads earnings headlines through a brokerage app.

“

The contract settles at $1 per share if the company beats estimates and $0 per share if it misses.

The mechanics behind each number are different, and that difference is central to understanding the result. A Polymarket earnings contract is a real-money binary bet on whether a specific company will clear a stated earnings-per-share threshold in a specific fiscal quarter. Traders buy and sell the contract at any time before the report, and the price floats between zero and one dollar based on what buyers are willing to pay. The contract settles at $1 per share if the company beats estimates and $0 per share if it misses. Because real money is on the line, the live price behaves as a probability estimate that updates continuously as new information arrives.

A consensus estimate is something quite different. It is the mean or median of forecasts submitted by sell-side analysts who cover a stock, compiled by data vendors and refreshed when individual analysts publish new numbers. That process is not a market. It is an aggregation of individual, often lagging and strategically managed forecasts. The study points to a structural bias: companies deliberately walk down analyst estimates before the quarter closes, engineering an easy bar to beat. That turns the consensus number into a negotiated target rather than a neutral forecast of economic reality. The result is a high reported beat rate with limited informational value, while prediction market participants have real financial incentives to trade toward truth.

There are important caveats. The sample covers only six months and is limited to firms with active prediction markets, which skews the universe toward larger, more liquid and widely followed names. The accuracy gap may not generalize to smaller or less-covered companies. Still, the size of the gap is hard to explain away. It implies that properly incentivized crowds can extract and aggregate information more efficiently than traditional sell-side research in certain settings.