Overview
Aug 20, 2026 - Kalshi Research released a study analyzing over 2.2 million data points, including proprietary data never previously released, examining Kalshi's forecasting ability.
This data provides relevant insights to many of the most popular questions about prediction markets:
How well do prediction markets forecast the future?
What factors affect the markets' forecasting ability?
Does the quality of forecast change depending on the category of market?
Is the 'wisdom of the crowd' theory of why prediction markets work backed by empirical evidence?
Main Findings
The study reaches a few key conclusions:
Overall, Kalshi produces forecasts that are closely correlated with how often events occur.
Forecasting ability improves with time, market depth, and participation.
Many concerns about market data — namely, that the markets aren't useful or accurate because of demographic bias, manipulation, or thin trading volume — do not have empirical support.
With the exception of trader count, all of this data is public. We encourage reproduction.
Calibration
Calibration compares each probability on Kalshi to the amount of times the market resolves yes.
The idea is simple: if a forecaster were perfect, events that they assign a 30% chance of happening would happen 30% of the time, events at 60% would happen 60% of the time, and so on. To measure this, you simply sort the forecasts by likelihood and compare them with how often they occur.
Perfect is represented by the gray dotted line; the closer to this line, the closer Kalshi is to being a perfect forecaster.
Beyond overall calibration, there is a better way to explore Kalshi's predictive power.
Sports, mentions, and intraday markets are unique because there is a rapid influx of price-relevant information as the event unfolds. Examining the sample where information can be aggregated in advance (so, removing those market categories) provides a clearer measure of traders' predictive ability.
This chart below, showing calibration without 'live-event' categories, is an even more high-quality signal for evaluating the predictive power of Kalshi:
Key Takeaways:
Data shows that predicted probabilities on Kalshi match closely to real-world outcomes.
From a policy perspective, this is important context for the wider conversation about prediction markets. Often, concerns are raised about the markets being unreliable due to the demographic makeup of the platform skewing the odds a certain way, or bad actors intentionally manipulating prices to distort perception.
If there was any sort of systemic bias on Kalshi — whether via demographics, intentional market manipulation, or something else — this data would show it. It does not.
In fact, it shows quite the opposite: there is no evidence of significant distortion in any direction, with forecasts increasing in confidence over time.
Calibration: Brier Scores
The study also examines Kalshi's Brier scores, a common metric used in fields like meteorology to quantify the success of forecasts.
You can think of the Brier score as measuring 'the distance away from the reference line' in the calibration charts. In this scoring system, lower is better. 0 is perfect, 1 is as bad as you could possibly be, 0.25 represents random chance.
Because it is entirely dependent on the subject matter, there is no universally agreed upon standard for what a 'good' Brier score is. Important context is that the world's most skilled superforecasters typically have Brier scores of around 0.1 to 0.15. (See recent tournament scoring data from the Forecasting Research Institute here).
All Kalshi markets, sorted by time horizon, have the below Brier scores:
Elections, politics, and finance have the strongest forecasts. The Brier scores of Kalshi markets, sorted by category:
The Brier scores of Kalshi markets, sorted by trader count:
The Brier scores of Kalshi markets, sorted by volume:
Key Takeaways:
Market forecasts improve in quality with time, participation, and volume.
Even markets with as little as ~$10,000 in volume, or as low as 20 traders, provide useful forecasts.
This data suggests it is the nature of the market mechanism itself, rather than the presence of millions of dollars in liquidity, that powers Kalshi's ability to provide useful forecasts. Why? Mis-priced markets are free money lying on the ground; people will pick up free money whether it is $100 or $100 million.
Additional insights in the full study. Except trader count, all data is public and reproducible, available from the Kalshi API.
