Kalshi’s Favorite Lie

They may not be your counterparty, but they still need you to lose

Unlike traditional sportsbooks, prediction markets like Kalshi are not betting against their users. They pair bettors against each other and take a fee, which means they do not care who wins. Sportsbooks take the other side of users’ bets and profit when those users lose; in gambling jargon, they are their users’ counterparty. Prediction markets profit when users trade. The incentives are fundamentally different. Sportsbooks, like casinos, are built to take users’ money. In prediction markets, users compete against each other, not against the house.¹

If you’ve ever read or listened to a prediction market evangelist, you’ve likely heard some version of this argument. On the Axios Show a couple weeks ago, Kalshi CEO Tarek Mansour called it a “fundamental difference in structure.”² Sportsbooks, he said, are “designed for customers to lose”; that model “does not exist in prediction markets.”

A poster from a recent Kalshi ad campaign

This pitch is effective because everything it points at is true. Sportsbooks really do limit and ban winners. Casinos really are extractive by design. Kalshi really does operate as an exchange rather than a bookmaker, profiting through transaction fees instead of directly from user losses.

But Kalshi uses those truths to sell a much bigger fiction: that because it profits from trading volume rather than directly from user losses, its business does not depend on users losing money.

The Exchange Model

Every trade on Kalshi has a maker and a taker. The maker offers a price—say, $0.50 for a contract that pays out $1.00 if the Yankees win—and the taker can choose to accept it.

Engineering a market in which these trades are possible—i.e. “making” the market—is harder than it sounds. Bettors don’t like to wait. They want the Lakers now—not in an hour, not in five minutes, not even in 30 seconds. Sportsbooks handle this by taking the other side of every bet themselves; wagers clear the moment you click place.³ Exchanges can’t do that.⁴ Someone has to have posted a price you’re willing to take, or your order just sits and waits. Kalshi’s central challenge, along with every other prediction market and exchange, is ensuring there is enough liquidity so that whenever a user places a trade, someone is there to take the other side.

Posting an order is riskier than accepting one. Imagine you offer to sell a “Lakers lose” contract, and it sits in the order book for a few minutes. Then news breaks that LeBron James is out with an injury. The Lakers’ chance of winning drops, so the value of a “Lakers lose” contract rises. Your offer is now below the fair price, meaning anyone who sees the news first can snap the contract up at a price that’s favorable to them and unfavorable to you. This is called adverse selection: when you leave an order up, you’re vulnerable to new information moving the market before you can adjust.

Traditional sportsbooks aren’t immune to adverse selection. When a line is mispriced or new information comes out, sharp bettors pile in before the book can adjust. Sportsbooks respond by moving lines fast, limiting or banning bettors who consistently beat them, cultivating a customer base filled with recreational bettors, and by charging a vig: a spread built into the odds on both sides of a bet. It’s why if you want to bet on a 50/50 event they’ll charge you -110, meaning you need to risk $110 to win $100. The vig helps insulate them. Even though some bettors sneak in good bets, the house’s overall portfolio stays profitable.⁵

Market makers on Kalshi and other exchanges face adverse selection without sportsbooks’ tools. They can’t ban bettors who pick off their stale offers.⁶ And the spreads they can charge are constrained by competition with other market makers.

So Kalshi has to make market making worth it. It can do this in a few ways: direct liquidity programs (paying users to market make), fee rebates,⁷ better technical/backend access, dedicated support, and more. But Kalshi’s main tool is the fee structure itself. It makes market making cheaper by making market taking expensive: The user who wants the trade now pays most of the fee, while the trader offering a price pays much less.

There is nothing unusual about an exchange treating makers and takers differently. Market makers get better terms because they supply the liquidity that makes the product usable and helps generate fees for the platform. The key question is how much the taker pays for immediacy.

On Kalshi, the answer is: a lot.

A trade from one user on Coinbase, which routes prediction market trades through Kalshi and applies an additional fee

Take the simplest example: a coin flip. A contract pays $1 if the coin lands heads. The true odds are 50/50, so the fair price is $0.50. With no fees, buying 100 contracts at 50 cents would be a fair bet: you risk $50 to win $50.

But a taker on Kalshi does not pay $50. Kalshi charges $1.75 in taker fees per 100 contracts; the maker on the other side pays about $0.44.⁸ So the taker isn’t risking $50 to win $50. They’re risking $51.75 to win $48.25. That is an expected loss of about 3.4%, meaning for every $100 a user trades, in the long run they’re going to lose $3.40.⁹

Kalshi maintains that this fee is wholly different from the vig. In legal and mechanical terms, that might be true. But for the ordinary user taking a price, the distinction barely matters: either way, a fair price becomes a bad bet.¹⁰

Kalshi does not offer literal coin flips.¹¹ But prediction markets’ own pitch is that prices are incredibly accurate, honed by the wisdom of crowds. They’re correct. Which means the vast majority of users might as well be betting on a coin flip, losing the fee each time.¹²

The asymmetric fees also help lock users into roles. It’s incredibly difficult to win long-term as a market taker on Kalshi, so sharp traders mostly sit on the maker side, posting prices and managing risk.¹³ Ordinary users (the “dumb money” who power exchanges) mostly do the opposite: taking the price in front of them, regardless of fees, because they want the trade now.

The Rake

Exchanges aren’t inherently zero-sum. Stocks and bonds trade on exchanges, and participants can collectively win because the underlying assets can create value. A company can grow. A bond can pay interest. An index fund can rise because the economy becomes more productive.

Prediction market wagers, whether they’re on wars, weather, or sports, do not work this way. A contract on an invasion, a hurricane, or the Lakers can go up in price, but that gain comes from the other side going down in price. It does not generate cash flows or represent ownership of anything productive.¹⁴ It merely resolves.

This distinction matters because it changes what Kalshi needs from its users. A stock exchange can thrive while users collectively build wealth. A prediction market can only thrive when users keep cycling money through zero-sum contracts. A small number of sophisticated users will win, but the system only works if others lose enough to fund them and the fees.

This gives Kalshi the same core dependency as a casino or sportsbook: a steady flow of users who lose enough money to keep the business running.

The main difference is where the money goes first. Sportsbooks collect from losing bettors directly. Kalshi collects from trading volume. But on a zero-sum exchange, volume is not some morally neutral good; it is the process by which ordinary users lose money to sharper ones.

Finding Nemo

The same dynamic exists in poker rooms. The house doesn’t play; it takes a rake from every pot and has no stake in any particular hand. The pros help keep the tables active, but they are not there out of generosity. Like market makers on prediction markets, they show up because they can take money from the losers. Without losing players, they leave—the competition is too good and the fees are too high—and there are no games left to rake.¹⁵

Kalshi doesn’t kick out winners for the same reason poker rooms don’t kick out the pros: the winners are the mechanism that makes the rake work. The liquidity, the tight spreads, the seamless trading—all of it depends on funneling losers to the users who profit from them.

None of this means prediction markets cannot be useful, that they are worse than sportsbooks, or that they should be outlawed.¹⁶ They can aggregate information, generate accurate forecasts, and in some cases let people hedge real-world risks. Even when it comes to sports, prediction markets can offer better pricing and a fairer ecosystem than traditional bookmakers.¹⁷ Skilled users are not punished for winning, which matters.

But prediction markets still need losers.

Kalshi pitches itself as morally superior to casinos because it isn’t on the other side of users’ trades, so it has no stake in whether they lose. That is a lie. Kalshi does not need to beat users directly. But it does need enough regular users to keep losing to sharper ones, while charging a fee on every trade. Those fees are now generating roughly $100 million a month.¹⁸

Thirty years ago, Ace Rothstein explained that in the gambling business, “The cardinal rule is to keep them playing and to keep them coming back. The longer they play, the more they lose, and in the end, we get it all.”

Kalshi might not call itself a casino. But the rule still applies.

1. Most prediction markets, including Kalshi, actually do have in-house trading teams that trade against their users. Kalshi’s affiliate, Kalshi Trading, is the subject of a class-action lawsuit alleging it misled customers about who they were actually betting against. The deeper point, made eloquently by Alfonso Straffon, is that even when an exchange’s affiliate isn’t taking the other side, the designated market makers who are—Susquehanna, Jump, others—take principal risk and profit when users lose, meaning there’s no functional difference between them and a bookmaker.

2. Dan Primack, Axios’s Business editor who conducted the interview, barely pushed back or asked for clarification on this and other claims which were at best misleading. Martin Shkreli (yes, that Martin Shkreli, who has become a major critic of prediction markets) had the best reply to the interview, plainly asking if it was actually an advertisement.

3. At least, they’re supposed to. Related: the popular framing of sportsbook economics—that books set lines to balance action and pocket the vig as a riskless margin—is mostly a myth. Modern operators don’t try to even out exposure on every market. They have the bankroll to absorb variance and generally seek out positive expected value, not balanced books.

4. Well, Kalshi Trading sort of can, but whatever.

5. Actual hold is almost always less than theoretical hold.

6. Market makers for parlays on Kalshi can sort of do this; Kalshi’s RFQ system (Request for Quote), which it uses for combos and parlays, lets market makers see who is requesting a quote and choose whether to respond. In practice, this is a soft form of limiting: a market maker can simply ignore RFQs from sharp users, increasing the likelihood their counterparty is a recreational user.

7. This is the same dynamic as rakeback in poker: the house gives back a portion of the fees it would have charged in order to keep sharp players seated and the action flowing. Polymarket has historically been more aggressive on this front, explicitly paying market makers daily USDC rebates funded out of taker fees. (This is part of why headline volume figures from prediction markets are murky—a meaningful slice of that volume is self-financed liquidity rather than organic user activity.)

8. Fees vary slightly by price and trade size, but these numbers are accurate at 50¢ (see Kalshi’s fee schedule). Other prediction market exchanges charge significantly less, although more (including Kalshi’s main competitor, Polymarket) are increasing fees. Still, Kalshi is by far the most expensive option in the prediction market space, and other platforms like Robinhood and Coinbase, which route through Kalshi, charge additional fees, resulting in comically bad pricing.

9. They also pay the spread, if the market isn’t very liquid. As others and I have written, this strictly -EV (negative expected value) nature of prediction markets is what makes them fundamentally more akin to gambling than traditional investing, and why conflating them with investments and integrating these products into brokerage accounts is so dangerous. As Vanguard CEO Salim Ramji recently described, prediction markets are “a form of financial exploitation,” because the platforms are “representing speculation as a form of empowerment” which has led to too many people viewing them, incorrectly, as “a fast path to financial security.”

10. This is true both in theory and in practice. An analysis in February from Citizens Bank found that the lowest performing prediction market traders at Kalshi lost at a higher rate compared to sportsbooks. A report this week by Bloomberg found that the vast majority of Polymarket traders lose while just a tiny sliver of accounts are net winners. (Though it seems to have some major flaws.)

11. They’re close, though: they self-certified contracts on the NBA draft lottery, which is about as close to a coin flip or roulette spin as a regulated event contract is likely to get.

12. To be clear, this is true on all exchanges; the difference is merely that on Kalshi the fees are way higher, which is why an average user placing a bet on Kalshi mostly gets about the same odds as if they placed the bet on DraftKings or FanDuel.

13. I am by no means the definitive expert on prediction markets and/or market making, but have spoken to many very successful Kalshi traders, and most of them rely heavily on market making. I’ve also traded myself, and the logic behind why this is the case is fairly simple: the large taker fees act as a barrier, preventing takers from taking advantage of “good” prices (and thus making it very difficult for them to win). Even if a maker is off by a cent or two, the taker fee can erase the edge. Note: since publishing I’ve spoken to others who have suggested the maker/taker split by sharp participants is less extreme than I initially posited. Would love to see more data on this, and will happily change my stance here if shown to be wrong, however I don’t believe it meaningfully impacts the main point (that prediction markets like Kalshi rely on traders losing money to fuel the ecosystem).

14. One partial exception: Kalshi pays 3.25% interest on cash and open positions.

15. To use a more vivid metaphor that might make more sense to those unfamiliar with poker: Kalshi is basically operating a shark tank. The sharks (market makers) don’t eat each other, because the fee structure won’t let them. They wait for Kalshi to send in the fish—market takers—then feast, while Kalshi gets a cut of every meal.

16. All businesses require their users to lose money; theorized as payment for entertainment, sports betting and prediction markets are not meaningfully different than spending money at a movie theater or restaurant, with the good the user receives being the chance of winning more money. Problems arise when there is a mismatch between the nature of products and the marketing promoting them.

17. As someone who has been kicked off of all traditional sportsbooks and enjoys using/trading/betting on prediction markets, where I am not punished for making good bets, this argument holds a decent amount of weight.

18. Actual retail losses are much higher than this, because the ~$100 million is just Kalshi’s cut; on top of that, retail users are also funding the profits of the market makers on the other side of their trades.