We Analyzed 2.3 Million Kalshi Trades. Here's What We Found.

We analyzed every trade on Kalshi since inception. The results challenge what most people believe about prediction markets.

Prediction markets are supposed to be efficient. The crowd is supposed to be wise. And in the aggregate, that is mostly true. But buried in 67,761,406 matched trades and $172,293,209 in volume, we found systematic patterns that tilt the odds in favor of patient, disciplined traders -- and against everyone else.

Here are the four biggest findings.

Kalshi is the first CFTC-regulated prediction market exchange in the United States. Since launching, it has grown into the largest legal event-trading platform in the US, covering categories from politics and sports to weather, crypto, and entertainment.

We pulled every historical trade from the Kalshi API -- 67,761,406 matched trades across 11 categories. Each trade records the price (1-99 cents), the side (YES or NO), and whether the buyer was a maker (limit order) or taker (market order). We then linked each trade to its market's final resolution to compute actual win rates, expected value, and excess returns.

The dataset spans the full history of the platform through early 2026.

The first thing we checked: are the prices actually right?

If a contract trades at 50 cents, it implies a 50% probability. Does the underlying event actually happen 50% of the time?

Yes -- almost exactly. Contracts priced at 50c resolve YES 50.0% of the time. That is as close to perfect calibration as you can get.

[INSERT IMAGE: Kalshi Calibration Curve]

The calibration curve above plots contract price (x-axis) against actual win rate (y-axis). The dashed line represents perfect calibration -- where a contract's price exactly equals its probability. The solid blue line is the actual win rate from 67,761,406 trades.

Through the middle of the price range (roughly 30-70c), the two lines track each other closely. The crowd is indeed wise in this zone.

But at the extremes, things get interesting.

The calibration curve above is not perfectly straight. At low prices (cheap longshots) and high prices (heavy favorites), the actual outcomes systematically diverge from what the prices imply.

This is the longshot bias -- one of the most well-documented phenomena in betting markets, and it is alive and well on Kalshi.

Contracts priced at 1-15 cents win less often than their price suggests. On average, they underperform by 0.8 percentage points.

Here is the full breakdown:

[INSERT IMAGE: Longshot Bias Detail]

Look at the 9-cent contracts: they imply a 9% chance, but they actually win only 7.8% of the time. That is a 1.2 percentage point gap -- which translates to a meaningful edge for anyone selling into that demand.

The pattern is consistent: across the entire 1-15c range, actual win rates fall below implied probabilities. Retail buyers are systematically overpaying for longshots.

Meanwhile, at the other end of the spectrum, heavy favorites (85-99c) outperform their implied probabilities by an average of 0.8 percentage points. This is the mirror image of the longshot bias: favorites are slightly underpriced.

Why does this happen? The same psychological forces at work in horse racing and sports betting. Longshots are exciting. They offer asymmetric payouts -- risk a few cents to win a dollar. Retail traders are drawn to the lottery-ticket appeal, pushing prices above fair value. This creates a systematic and exploitable edge for the other side.

The longshot bias is not evenly distributed. Some categories have far more mispricing than others.

We measured this by computing the maker excess return -- the volume-weighted outperformance of limit-order placers over random -- for each category.

[INSERT IMAGE: Maker Edge by Category]

The standout categories are World Events (+3.7% maker edge) and Media (+3.6% maker edge). These are markets where retail interest is high but informed liquidity is thin -- creating wide mispricings that patient limit-order traders can capture.

At the other extreme, Finance markets show a maker edge of just +0.08%. These markets are dominated by sophisticated traders, and prices reflect true probabilities more accurately.

Esports is a notable outlier with a negative taker excess but positive maker excess, suggesting an unusual market structure where both the bid and ask sides carry different information content.

This may be the most actionable finding of all.

Across every single category, traders who place limit orders (makers) outperform traders who take market orders (takers). No exceptions.

[INSERT IMAGE: Makers vs Takers by Category]

In total, makers earned $+1,437,855 across the dataset, while takers lost $1,424,988.

This is the "smart money" signal hiding in plain sight. Makers -- the traders willing to post a price and wait for someone to fill it -- are systematically on the right side of trades. Takers -- the traders who see a price and hit "buy" immediately -- are systematically on the wrong side.

The gap is not always large. In Finance, it is a razor-thin 0.17 percentage points. But in Media, it balloons to 7.3 percentage points. In World Events, 7.3 percentage points as well.

The implication is clear: how you trade matters as much as what you trade. Even if you have no informational edge, simply placing limit orders instead of market orders shifts you from the losing side to the winning side.

Four practical takeaways:

This is the core thesis behind our daily newsletter: we scan every open Kalshi market daily, identify the contracts where historical calibration data shows the biggest edge, and surface the top picks with full analysis. The longshot bias has persisted for years across every betting market ever studied. As long as retail traders love lottery tickets, the edge is there for the taking.

Data collection. All trade data was pulled from the Kalshi public API. We collected historical trades for every market that has resolved since the platform's launch through early 2026, giving us 67,761,406 matched trades (each with a maker and taker side).

Win rate calculation. For each price point (1-99 cents), we aggregated all trades executed at that price and computed the fraction where the contract resolved YES. This gives the actual empirical win rate at each price level.

Maker/taker classification. Kalshi's API identifies whether each side of a trade was the maker (limit order already resting in the book) or the taker (incoming order that crossed the spread). We computed volume-weighted excess returns for each group.

Category analysis. Markets were classified into 11 categories based on Kalshi's event taxonomy. We computed maker and taker excess returns within each category to identify where mispricing is concentrated.

Expected value. For each price point, we computed the EV of buying YES and buying NO based on empirical win rates. At 69 out of 99 price points, buying NO (the favorite side) had higher expected value than buying YES.

Statistical note. With over a million trades at most price points, standard errors are extremely small (typically <0.05 percentage points). All reported differences are highly statistically significant.

Analysis by the prediction-market-analysis project. Data updated through early 2026. Not financial advice.