On Kalshi’s single-name markets, the ones where traders are betting on whether a specific person or company will do a specific thing, traders buy “yes” about 61% of the time.
And “yes” wins about 32% of the time.
That monster gap is one of the main findings in a new paper from Robert Bartlett of Stanford Law School and Maureen O’Hara of Cornell’s Johnson College of Business. The paper, Adverse Selection in Prediction Markets: Evidence from Kalshi, analyzed 41.6 million trades across 478,167 Kalshi markets.
The authors split the markets into two broad categories: single-name markets and broad-based markets.
Single-name markets include contracts tied to a specific person or company, like whether Jeff Bezos would attend the Super Bowl, whether Tesla would hit a delivery target, whether a company would mention a certain phrase on an earnings call.
Broad-based markets include contracts tied to things like the S&P 500, Bitcoin, Federal Reserve decisions, CPI, unemployment, and other events where information is spread across a wider market.
Both categories lean “yes.” On broad-based markets, traders bought “yes” 56.3% of the time by volume, while those markets settled “yes” 39.1% of the time. On single-name markets, traders bought “yes” 60.9% of the time by volume, while those markets settled “yes” 32.5% of the time.
But the paper’s point is not simply that “yes” loses more often than it wins.
A side can win 32% of the time and still be a good bet if the price is right. A +220 underdog does not need to win half the time. It needs to win often enough to justify the price.
The issue, according to Bartlett and O’Hara, is that on single-name markets, the price often is simply not right.
Improperly priced
Broad-based markets were much closer to properly priced. “Yes” lost more often than it won, but the prices largely accounted for that. Single-name markets were different. The paper found that “yes” prices exceeded actual “yes” settlement rates, and often by eye-popping numbers.
The middle of the board was especially rough. Contracts trading around a 46% implied “yes” probability actually settled “yes” only 21% of the time.
In short: Kalshi traders are not just buying “yes” too often on single-name markets. They are also paying too much for it.
For anyone who has spent time around gambling markets, the idea is not exactly foreign. Longshots get overbet. Parlays get overbet. The more fun side of a bet often comes with a worse price.
“Yes” means the thing happens. Bezos shows up. The company says the phrase. The target gets hit. “No” is usually the less exciting side: Nothing happened. Ho-hum.
But in this data set, much like in life, boring often wins.
That dynamic helps market makers clean up. The paper found that market makers earned 1.91 cents per contract on single-name markets, more than twice the 0.82 cents per contract they earned on broad-based markets. Makers won 63.4% of contracts on single-name markets.
On single-name markets, makers averaged 16.34 cents when they won and lost 23.13 cents when they lost. The losses were bigger than the wins, but they still came out ahead because they won a lot more often than they lost.
That edge has grown along with Kalshi. From early 2024 to early 2026, the gap between winning and losing trades on single-name markets widened sharply. As trading volume increased, the “yes” problem got worse, not better.
Here (doesn’t) come Bezos
The paper also found more evidence that even when there is informed trading (insider trading, a rose by any other name), it still didn’t change the calculus.
The Bezos Super Bowl market is the paper’s clearest example.
Kalshi ran a market on whether Bezos would attend the Super Bowl. According to the paper, Bezos’ stepson, a member of Sigma Alpha Epsilon at the University of Miami, told fraternity brothers that Bezos was not going. Those fraternity brothers then bought “no,” and the “yes” price fell from roughly 70 cents to 30 cents.
That is the risk in single-name markets. Somebody may know something. Or somebody may know somebody who knows something.
But even in that market, the market makers finished ahead.
The paper says makers earned $29,274 on the Bezos market. The reason was the same basic “yes” problem: Uninformed “yes” buyers bought 518,000 contracts at an average price of 19 cents. Bezos did not attend. Those “yes” contracts lost. The makers collected enough from that segment to offset the sharp “no” money.
The authors describe all of this as a kind of “cross-subsidy,” in which losses to informed traders are being covered by traders who overpay for “yes.”
As Bartlett and O’Hara put it, the health of these markets depends on “the continued willingness of uninformed takers to participate.”
