Kalshi Research Finds 14,694 Traders Beat Its Prices, and the Rest of the Crowd Covers the Cost

A working paper from Kalshi’s research team identifies 14,694 skilled traders, accounts that forecast better than the prices they trade against and, the authors say, pull those prices toward the right answer. They’re 0.15% of the 9.66 million accounts that traded on the exchange from Oct. 1, 2025, to Sept. 2, 2026, in a sample of 1.08 billion trades across 2.35 million markets, according to the paper by Nicole Kagan and Rubens Baiocchi.

The study looks only at takers, the customers who accept a price already on the board instead of posting their own. The 14,694 made up 22.8% of taker volume, counted in contracts, about 150 times their share of accounts. The side they bought had been underpriced by 3.8 cents on average, measured against the midpoint between the best bid and the best offer. After one of them trades, that midpoint moves 1.7 cents their way by the next order and 3.3 cents by the close.

How the paper sorts skill from luck

Each taker’s price is scored against the midpoint just before the trade, and both are checked against how the contract settled. A trader with no skill who only buys 97-cent favorites goes 20 events without a loss about 54% of the time, the authors note. So they scored only the 571,621 accounts active in at least 100 events, reran each record against 1,000 randomized sets of outcomes and corrected for testing hundreds of thousands of accounts at once, which left 14,694.

Of the traders tagged as skilled in one random half of the events, 44.5% were tagged again in the other half, against 3.0% of all the traders tested. NextPredict’s write-up adds that the skilled set turned up in 92.8% of markets.

The authors concede the score also rewards bots that arbitrage between a contract and the price it tracks, and that part of the price move may be skilled traders reaching news first rather than moving the market.

Who pays

Skilled takers pay 1.5 to 1.8 cents over the midpoint. Within a minute the price has moved past what they paid, so the market maker on the other side, the firm keeping orders posted on both sides of the market, loses 1.3 to 1.5 cents a contract.

The other takers buy sides the midpoint already prices correctly. They pay about 1.2 cents over it, and the makers keep most of that. They also lean one way, buying YES on 67% of their trades against 53% for the skilled set, a tilt the paper links, through outside research, to a premium paid on YES contracts.

The paper’s conclusion says the crowd “still matters, partly because it gives skilled traders a reason to trade,” and that its noisy orders compensate “the makers facing skilled takers,” a mechanism it credits to a Stanford working paper by Robert Bartlett and Maureen O’Hara.

The sportsbook version

In Massachusetts, sportsbooks had cut the bet limits on 0.64% of player accounts as of Dec. 10, 2024, a state Gaming Commission staff memo found. Players who consistently beat the closing line, the final odds before a game starts, were more likely to be the ones limited. Players who didn’t were more likely to have their limits raised, and those accounts were far more likely to be classified as VIPs.

Commission chair Jordan Maynard said winning players are “often limited for justifiable reasons” but get “little to no justification or notification,” the Boston Herald reported. A sportsbook is the other side of every bet, so a customer who keeps beating the closing line costs it money directly. On an exchange the other side is whoever posted the order, and Kalshi’s paper shows those makers losing to its skilled 0.15% and making it back from the other takers.

Kalshi’s case for that arrangement is real, since markets with one skilled trader closed more accurately than markets with similar volume and none. The paper is Kalshi’s argument that its prices deserve trust, and by its own numbers the bill for that trust averages about 1.2 cents a contract, paid by the takers outside the 14,694.

Sources

• Kalshi Research: Who Calibrates Markets? Evidence from Trader Micro-Data (Nicole Kagan and Rubens Baiocchi, working paper, Oct. 2026)
• NextPredict: Kalshi Research links small group of skilled traders to market accuracy (Erik Gibbs, Oct. 20, 2026)
• Massachusetts Gaming Commission: Player Limiting Data Analysis Results and Possible Next Steps (Sept. 30, 2025)
• Boston Herald: Winning gamblers face betting limits, losers made VIPs, gaming commission learns (Matthew Medsger, Sept. 30, 2025)