The Exchange Loophole

Introduction: The Room Next Door

Water does not stop at the breach. It finds the next low ground, and it does not much care whose property that turns out to be.

In the eight years after the Supreme Court struck down the federal ban on state sports wagering, the industry that grew in the breach built itself a house with rules. Thirty-nine states wrote licensing statutes. Regulators set minimum ages, funded problem gambling programs from a slice of the take, and required operators to answer for their conduct or lose the right to operate. The house was not a fair house, as the articles ahead will document at length, but it was a house with a floor plan and an inspector.

Then a company most bettors had never heard of walked around the side of the building and opened a door that had been sitting there since 1974.

On January 22, 2025, KalshiEX LLC filed an eleven-page certification with the Commodity Futures Trading Commission. The filing did not ask permission. Under the self-certification procedure in the Commission’s rules, an exchange holding a federal license may list a new product simply by certifying that the product complies with federal commodities law, and may begin trading it the following day. The product Kalshi certified was titled, in the flat language of derivatives paperwork, “Will <team> win <title>?” The placeholders were to be filled with the names of NFL, NHL, NBA and NCAA championships. Trading opened on January 23 (KalshiEX, “Championship Winner Contracts”; Zochodne, “Kalshi Preparing to Launch”).

Two weeks later the same mechanism produced contracts on individual games. By the following autumn there were point spreads, totals, touchdown scorers and multi-leg combinations that behaved, to anyone holding the app, exactly like parlays. By February 2026 there were contracts on individual quarters and halves (KalshiEX, “Football Period Contracts”; Willis, “Kalshi Found a Backdoor”). None of this required a single state license, because in the theory of the case none of it was gambling. It was the trading of binary swaps on a federally designated contract market, and state gaming boards had no more business regulating it than they had regulating soybean futures.

This article is about that door, who walked through it, what the courts have said, and the question this series exists to ask. The exchange model genuinely changes the cost structure of a wager. Fees replace the vig. The platform does not profit when you lose. In principle nobody closes your account for winning. Those are real changes, and they deserve to be taken seriously rather than dismissed as marketing.

But a better price is not the same thing as an edge, and the difference between the two is the most useful lesson available this early in the book. A market that charges you less to be wrong has not made you right.

A Contract, Not a Wager

The instrument at the center of all this is simple enough to explain in a sentence. A binary event contract settles at $1 if a stated event occurs and at zero if it does not. It trades between one cent and ninety-nine cents, and its price is therefore readable as the market’s estimate of the probability that the event happens. A contract on the Seahawks changing hands at seventy-three cents says the market thinks Seattle wins about 73 percent of the time (KalshiEX, “Football Period Contracts”; Bloom, “What Is Kalshi”).

A bettor who has ever seen a moneyline will recognize what that is. Seventy-three cents to win a dollar is roughly minus 270 in American odds. The screen is different, the vocabulary is different, and the legal theory is entirely different, but the transaction is a directional position on the outcome of a football game.

The legal theory is worth understanding precisely, because everything else follows from it. When Congress passed the Dodd-Frank Act in 2010, it wrote an expansive definition of a “swap” into the Commodity Exchange Act. A swap includes any transaction providing for a payment “dependent on the occurrence, nonoccurrence, or the extent of the occurrence of an event or contingency associated with a potential financial, economic, or commercial consequence” (Justia, “Kalshiex LLC v. Flaherty”). Kalshi’s argument reads that clause literally. A football game has a financial consequence for somebody. The contract pays on the occurrence of an event. Therefore it is a swap.

The second step does the real work. The Commodity Exchange Act gives the CFTC exclusive jurisdiction over swaps traded on a designated contract market, and preempts state law that interferes with them. If the contract is a swap and the venue is federally designated, then New Jersey’s Sports Wagering Act has nothing to say about it, and neither does anyone else’s.

There is a provision that appears to cut the other way, and it is the hinge on which the entire fight turns. The Commodity Exchange Act contains what practitioners call the Special Rule, implemented as Regulation 40.11, which bars a registered exchange from listing any contract that “involves, relates to, or references terrorism, assassination, war, gaming, or an activity that is unlawful under any State or Federal law” (17 C.F.R. § 40.11). Gaming is on the list. If sports event contracts involve gaming, the rule’s own text bars them, and no state law would be needed to stop them.

The scale that grew on that foundation was extraordinary. Combined monthly trading volume across the two largest venues went from under $5 billion in September 2025 to roughly $24 billion in April 2026, and past $50 billion in July (Radde, “Trading Volume”; Park, “All-Time High”). Kalshi raised $1 billion in May 2026 at a $22 billion valuation, having raised $1 billion at $11 billion in December 2025 (Glasner, “Kalshi Secures $1B”; Kalshi, “Kalshi Raises $1 Billion”). Polymarket, banned from the United States by a 2022 CFTC enforcement order, bought its way back in by acquiring a licensed exchange for $112 million and took a $2 billion investment from the parent company of the New York Stock Exchange (CFTC, “Release 8478-22”; “Polymarket Acquires QCEX”; Sobrado, “Polymarket Secured”). Robinhood earned more from event contracts in the second quarter of 2026 than it did from cryptocurrency (Roberts, “Robinhood CFO”).

And it was sports carrying almost all of it. Kalshi’s 2025 fee revenue was $263.5 million, of which $234.6 million, or 89 percent, came from sports markets (O’Boyle, “Kalshi Fee Revenue”). The elections contracts that won the company its founding legal battle were a rounding error. Whatever the filings say, this is a sports betting business.

The Second Mirror: Volume Is Not Handle

Article 1.3 opened on a number that everybody used and nobody had calculated, the $150 billion illegal market that turned out to trace back to one commissioner’s off-the-cuff remark in 1999. The prediction market era produced its own version of that number within eighteen months, and it is worth naming early, because the habit of accepting a large figure without asking what it measures is the single most reliable way to be misled by this industry.

The figure is trading volume, and the claim built on it is that prediction markets have overtaken sportsbooks.

In February 2026, Kalshi posted the biggest week in its history around Super Bowl LX, roughly $2.8 billion in notional volume against a projected $1.76 billion in legal sportsbook handle nationwide (Cross, “Super Bowl Fuels Kalshi Volume”; American Gaming Association, “Americans to Legally Wager”). In July, after the World Cup, a widely circulated headline announced that prediction markets had “just crushed traditional sportsbooks” in a $50 billion month (Knight, “Prediction Markets Just Crushed Traditional Sportsbooks”). The comparison is arithmetically valid and analytically worthless.

Handle counts the money a bettor puts at risk, once. Notional volume counts every contract that changes hands, on both sides of the trade, valued at its $1 settlement rather than its purchase price, including positions bought and resold a dozen times before the game ends and including the churn of professional market makers who are not betting on anything. Ed Birkin, managing director of the gambling data firm H2 Gambling Capital, put it about as plainly as it can be put. Comparing the two is “akin to measuring apples by the kilo and oranges by the dozen” (Birkin, “Overhyped and Underperforming”).

Birkin’s firm did the conversion. Using ratios derived from exchange research, roughly 45 to 55 percent of volume converts to handle in single-game markets and only five to 20 percent in futures markets, H2 reconstructed the Super Bowl comparison on a like-for-like basis.

Table 1.4.1Super Bowl LX, February 8, 2026: the headline comparison and the reconstruction

Measure	Prediction markets	Licensed sportsbooks
Reported notional volume, Kalshi, all markets, Super Bowl week (Feb 2 to 8)	$2.80 billion	not applicable
Comparable betting volume on the game	$633 million	not applicable
Handle equivalent, after conversion	about $317 million	$1.412 billion
Share of national handle	18 percent	82 percent
Share in states with a legal sportsbook	7 percent	93 percent
Estimated gross revenue	about $2.4 million	about $141 million
Share of revenue	about 2 percent	about 98 percent

Sources: H2 Gambling Capital estimates reported in Birkin, “Overhyped and Underperforming”; the $2.80 billion is volume across all Kalshi markets for the week of February 2 to 8, 2026, from Cross, “Super Bowl Fuels Kalshi Volume.”

Two percent of revenue is not a takeover. It is an entrant. The same reconstruction applied to the World Cup, the tournament that produced the loudest headlines, put prediction markets at roughly 27 percent of all legal United States sports betting activity in the tournament’s first month, up from about nine percent at the start of the year (Popper et al., “Prediction Markets Swell”). That is a serious number and a fast-moving one. It is also less than a third of what the raw volume figures implied.

March Madness is the cleanest case, because there the raw numbers do not even support the claim. Kalshi took $1.8 billion in tournament volume, a genuine record that exceeded what the platform had done on the 2024 presidential election three times over. Projected legal sportsbook handle on the same tournaments was $3.3 billion (McQuillan, “March Madness Volume”; Oke, “March Madness Betting”). The exchange lost that comparison before anyone adjusted anything.

One more figure deserves attention because it cuts against the story everyone is telling. Eilers & Krejcik figures published with RotoWire’s 2026 season projections found that 69 percent of prediction market sports volume originates in states without a legal sportsbook, and that California and Texas alone account for 44 percent (RotoWire, “NFL Betting Handle”). DraftKings told investors it sees roughly one percent customer overlap with its largest prediction market competitor and “no discernible impact” on sportsbook revenue (Motley Fool, “DraftKings Q2 2026”). The exchanges are not, for the most part, taking the sportsbooks’ customers. They are reaching the eleven states the sportsbooks cannot legally touch, plus the eighteen to twenty year olds the sportsbooks are not allowed to serve.

The War Over Who Regulates a Bet

By the summer of 2026 the question of whether a sports event contract is a bet had produced one of the strangest legal maps in modern American regulation, and the strangest feature of it is that a federal financial regulator has been suing states to stop them enforcing their own criminal laws.

The Federal Answer

The high-water mark for the industry came on April 6, 2026, when a divided panel of the Third Circuit affirmed an injunction barring New Jersey from enforcing its gambling laws against Kalshi. Judge David Porter, writing for himself and Chief Judge Michael Chagares, held that sports event contracts are swaps, that the CFTC’s jurisdiction over swaps on a designated contract market is exclusive, and that both field and conflict preemption apply. Allowing New Jersey to enforce its own law, the majority wrote, “would create an obstacle to executing the Act” (KalshiEX v. Flaherty).

Two features of that opinion get lost in the celebration. It resolved a preliminary injunction, meaning it decided only that Kalshi was likely to win, not that Kalshi had won. And the majority did not decide whether the contracts constitute gaming, noting that the Commission “has not yet acted to review or prohibit any sports-related event contracts” and had “chosen not to enforce its regulation against” them. The court did not hold that the contracts are lawful. It held that the only body entitled to say otherwise has declined to speak.

Judge Jane Roth dissented, and her opening observation is the one a bettor recognizes. “These offerings,” she wrote, “are virtually indistinguishable from the betting products available on online sportsbooks, such as DraftKings and FanDuel” (Justia, “Kalshiex LLC v. Flaherty”).

The State Answer

Outside the Third Circuit, Kalshi has mostly been losing, and losing in language that borrows from Roth.

Chief Judge Andrew Gordon of the District of Nevada dissolved his own earlier injunction in November 2025, concluding that the contracts are “sports wagers and everyone who sees them knows it.” Chief Judge Sarah Morrison in the Southern District of Ohio denied relief in March 2026, quoting the Supreme Court’s observation in Epic Systems v. Lewis that “a statute’s meaning does not always turn solely on the broadest imaginable definitions of its component words,” and warning that under Kalshi’s construction all sports bets “would be forced onto DCMs like Kalshi and every sportsbook in the country would be put out of business” (KalshiEX v. Schuler). Judge Robert Shelby in Utah ruled against the company in early August 2026, reasoning that Congress would not have silently reversed two centuries of state authority over gambling through a derivatives statute. Judge Vernon Oliver in Connecticut denied relief on August 10 with a sentence that could serve as the chapter’s epigraph. “Kalshi characterizes its sports-related event contracts in various ways, but at bottom, they are sports wagers” (Nightingale, “Kalshi Rejected in CT Court”).

The enforcement side escalated faster than the doctrine. Arizona’s attorney general filed a twenty-count criminal information against Kalshi entities in March 2026, including four counts of wagering on elections. Minnesota enacted the first outright statutory ban in May. New York’s attorney general sued in July seeking compensatory damages that court filings put at no less than $36 billion, according to The Block (Shen, “New York Sues Kalshi”), alleging among other things that the platform let eighteen to twenty year olds trade sports outcomes in a state where mobile wagering requires twenty-one. A Washington judge ordered Kalshi to geofence the state out of its sports, election and entertainment markets by September 2, 2026, on penalty of $120,000 a day. Baltimore filed suit on August 13 (Arizona Attorney General, “Mayes Charges Kalshi”; Chen, “Minnesota Becomes First State”; New York Attorney General, “Hochul and James Announce”; Bishop, “Kalshi Ordered to Shut Down”; Mitchell, “Baltimore Sues”).

The tribes have brought the argument the industry finds hardest to answer. Under the Indian Gaming Regulatory Act, Class III gaming on tribal land requires a compact. Tribal plaintiffs in California, Wisconsin and New Mexico contend that a contract on a basketball game, traded by a customer standing on tribal land, is exactly that, and that no derivatives statute repealed IGRA by implication. Sixty federally recognized tribes and nine intertribal organizations filed in support of New Jersey in the Third Circuit (Brief of the Indian Gaming Association). A Wisconsin court has already found the Ho-Chunk Nation likely to succeed on the merits of its IGRA claim.

The Regulator With One Commissioner

Standing at the center of this, holding the only gavel that matters under the Third Circuit’s reasoning, is an agency with four of its five seats vacant. Chairman Michael Selig, confirmed in December 2025, is the sole confirmed commissioner. The Commodity Exchange Act sets no quorum, so he can adopt rules alone, a situation the bipartisan leadership of the House Agriculture Committee raised in May 2026 when it urged the White House to fill the vacant seats. As The Crypto Times summarized the concern, it concentrates enormous regulatory discretion in one official, with no internal debate, no minority-party input and no institutional check on rulemaking (Mistry, “House Ag to Trump”).

That agency’s record on the central question is a study in deferral. A prediction markets roundtable was announced in February 2025, rescheduled, and then cancelled without explanation or a reason given to the invited participants (Gouker, “CFTC Cancels”). A 2024 proposed rule on event contracts was withdrawn in February 2026. An advance notice went out in March, and a proposed rule, published in the Federal Register at 91 FR 35806, followed in June, which would finally define what “gaming” means and set out when a contract may be prohibited (CFTC, “Release 9249-26”; “Public Interest Determinations”). Comments closed in the summer of 2026, and the rule has not been finalized. In October the Commission separately issued an interim final rule excluding casino-style gambling products from the definition of “swap,” which took effect on October 19, and proposed another rule that would add event contracts to that definition, open for comment until November 18 (CFTC, “Release 9309-26”; CFTC, “Release 9310-26”).

In the meantime the Commission went on offense. Beginning in April 2026 it sued Arizona, Connecticut, Illinois, New York, Wisconsin, Minnesota, New Mexico, Kentucky and Rhode Island, nine states by the most recent counts, seeking orders that state gambling enforcement is preempted (CFTC, “Release 9206-26”; CFTC, “Release 9260-26”; “44 States Say CFTC Has No Authority”; Karpchuk and Wingfield, “The States Are Coming”). Forty-four state attorneys general responded with a comment letter, filed July 27, arguing that the rule exceeds the Commission’s authority and urging it to “reconsider and draft a new rule consistent with the Commodity Exchange Act and the Constitution” (Ohio Attorney General et al., “Comment on Prediction Markets,” as quoted in Shen, “Coalition of 44 State Attorneys General”). Congress produced at least ten bills on prediction markets in 2026, including a bipartisan Senate measure titled, with no ambiguity at all, the Prediction Markets Are Gambling Act. None has been enacted (Desai et al., “Congress Predictably Turns”; Schiff, “Schiff, Curtis Introduce”).

Updated October 2026. New Jersey asked the Supreme Court to hear the case on September 2, 2026 (Flaherty v. KalshiEX, No. 26-299). Thirty-nine states and the District of Columbia (together, in a single joint brief), gaming regulators, tribes, the American Gaming Association and the NFL have filed briefs supporting New Jersey, and Kalshi’s response is due November 9 (Supreme Court of the United States, “Docket for No. 26-299”). The posture the Court usually waits for has also arrived. The Ninth Circuit ruled for Nevada on August 28 (KalshiEX v. Assad), and on September 25 a unanimous Sixth Circuit panel ruled for Ohio and Tennessee, finding that the contracts had not been shown to be swaps and that the states’ laws were not preempted (Shepstone, “Sixth Circuit Rules Against Kalshi”); Kalshi has asked the full Sixth Circuit to rehear the case (Helling, “Kalshi Asks Full Sixth Circuit”). Like the Third Circuit’s, these are rulings on preliminary injunctions, not final judgments. There is now a circuit split.

The House Follows the Water

The most telling evidence about what these products are does not come from a courtroom. It comes from the balance sheets of the companies that spent eight years and enormous sums building the licensed system.

On November 12, 2025, DraftKings and Flutter, the parent of FanDuel, entered a stipulated order with the Nevada Gaming Control Board voluntarily surrendering their licensing and withdrawing pending applications in the state that invented legal American sports betting. The Board’s chairman stated the reason on the record. The two companies “intend to engage in unlawful activities related to sports event contracts.” A FanDuel spokesman put it more diplomatically and no less clearly, saying the company’s view of the prediction market opportunity outside regulated states was “in direct opposition to Nevada’s priorities for its licensed operators” (Stutz, “DraftKings, FanDuel Agree to Abandon Nevada”).

Read that again. Two operators gave up Nevada in order to keep offering a product Nevada considers illegal, in states where they hold no gaming license at all.

The rest followed quickly. DraftKings bought a licensed exchange in October 2025, launched a predictions app in thirty-eight states in December without using it, and eventually launched its own exchange in June 2026, giving it the brokerage, the exchange and the market maker in one company (O’Boyle, “DraftKings Launches In-House”). FanDuel launched with CME Group in five states and then, in August 2026, moved its sports contracts off CME entirely (Kelemen, “CME Keeps 51%”). Fanatics launched in December 2025 and agreed to buy an exchange and a clearinghouse of its own in July. DraftKings, FanDuel, Fanatics and bet365 all quit the American Gaming Association, the trade body that had spent the year calling sports event contracts illegal gambling (Yogonet, “bet365 Exits AGA”).

Meanwhile Sporttrade, which had built a genuinely peer-to-peer betting exchange under state licenses in five states, shut the whole thing down in May 2026 and applied for a federal designation instead. Its founder’s argument was purely economic. The state-by-state licensing system is a poor fit for a low-margin exchange business (Edelstein, “Sporttrade To Shutter”).

The companies that did not follow the water are showing the marks. BetMGM’s chief executive told investors in July 2026 that “the primary macro impact or primary macro impacts are prediction markets,” as the company pushed its long-term earnings target past 2027 with no replacement date (McFarlane, “BetMGM Q2 2026”). Flutter cut its United States revenue guidance by $395 million and reported a 70 percent decline in second-quarter American adjusted EBITDA (Kelemen, “CME Keeps 51%”), having generated $6 million of prediction market revenue against a category expense line above $200 million (Rybaltowski, “FanDuel Predicts Trails Rivals”).

The pattern here is one this series will keep returning to. The industry’s response to every constraint is to find the room next door. That is not a moral observation, it is a structural one, and it is the reason a bettor should never mistake a company’s regulatory posture for a statement about the product’s value.

The Analyst’s Question: Is the Price Actually Better?

Now to the part that matters to anyone deciding where to put money.

Start with what a sportsbook charges. A standard two-way market at minus 110 requires a bettor to risk $110 to win $100. The implied probabilities on the two sides sum to about 104.8 percent, and the excess is the book’s margin. On balanced action the book keeps about 4.55 percent of everything staked, and a bettor must win 52.381 percent of such wagers simply to break even.

That is the theoretical number, and it is the number every exchange marketing deck compares itself to. It is also nowhere near what the average American bettor actually pays. The national hold rate on legal sports betting was 6.9 percent in 2019 and reached about 10 percent in 2025 (10.2 percent by one count, 10.1 by another), where it has stayed through August 2026 (McQuillan, “Sportsbooks Make More Money”; Casino Reports). The reason is product mix. Casino Reports’ sample covers four states (New Jersey, Colorado, Illinois and Maryland), and the latest months draw on fewer of them, but the split is stark. In August 2026, parlays in that sample held 15.8 percent against 4.6 percent for other bets, and made up about 63 percent of operator revenue on about 33 percent of handle (Casino Reports). The house’s real edge is not the vig on a spread. It is the compounding of vig across correlated legs, which article 5.2 takes apart in detail.

Against that, the exchange fee schedules are genuinely cheaper, and it is worth doing the arithmetic rather than accepting either side’s summary.

Table 1.4.2The cost of a position on a fifty-cent contract, by venue

Venue	Cost per contract	Break-even win rate	Cost as share of amount risked
Sportsbook at -110	embedded in the price	52.38 percent	4.55 percent
DraftKings Predictions	$0.0200	52.00 percent	3.85 percent
FanDuel Predicts	$0.0200	52.00 percent	3.85 percent
Kalshi, taker, 100-lot	$0.0175	51.75 percent	3.38 percent
Polymarket sports, taker	$0.0125	51.25 percent	2.44 percent
Robinhood, non-Gold	$0.0100 plus exchange fee	51.00 percent to 52.00 percent	1.96 percent to 3.85 percent
Kalshi, maker	$0.0044	50.44 percent	0.87 percent
Polymarket, maker	none	50.00 percent	0.00 percent

Sources: published fee schedules at Kalshi, Polymarket, Robinhood, DraftKings and FanDuel; arithmetic by the author. Figures assume execution at the midpoint and exclude the bid-ask spread.

The Kalshi line is the one to sit with. A hundred contracts at fifty cents cost $50 plus $1.75 in taker fees. Risking $51.75 to win $48.25 is equivalent to a moneyline of about minus 107, against minus 110 at the book. On a fair coin flip the exchange trader loses 3.38 percent of stake and the sportsbook customer loses 4.55 percent. That is a real improvement of roughly a point and a fifth, and on a straight main-line bet it is the best publicly available price in American sports wagering.

And then the caveats begin, and they are large enough to eat the entire advantage.

The first is the last line of that table’s footnote. Every figure above assumes you transact at the midpoint, and nobody transacts at the midpoint. On liquid Kalshi markets the spread runs a penny or two with hundreds of contracts resting on each side. On thin ones it runs five to ten cents with a handful. A taker who crosses a one-cent spread on a fifty-cent contract pays half a cent above the midpoint, about one percent of stake, on top of a fee that already costs 3.38 percent; a round trip, in and out, costs the full cent, or two percent. A five-cent spread costs about five percent to cross, more than the entire taker fee, and with the fee the position costs roughly 8.4 percent of stake, close to the hold on a standard two-leg parlay. One comparison of the two venues makes the same point with a thin market quoted at forty-one cents bid and forty-five cents offered, a spread of about four percent that can cost a taker more than a sportsbook’s vig (Shander, “Kalshi Vs A Sportsbook”).

The second is depth. Matt Kalish, a DraftKings co-founder and hardly a neutral witness, asked the most useful question of 2026 about these platforms. “Why is it 93-1 if you bet $10 and 38-1 if you bet $1000 on something?” (Chepkova, “Kalish vs. Kalshi”). A price you cannot get filled at in size is not a price. Sportsbooks guarantee the fill and charge for the privilege; exchanges advertise the price and leave you to find the liquidity.

The third is that the fee structure inverts at the extremes. The fee formula peaks at fifty cents and collapses toward the ends, which sounds favorable until you notice what it means in percentage terms. A one-cent fee on a two-cent longshot contract is a 50 percent tax on the position (Edelstein, “Prediction Market vs Sportsbook”). Bettors drawn to longshots, which is most bettors, pay the worst effective rate on the exchange, just as they do at the book.

The fourth is combinations. Kalshi’s combos and DraftKings’ equivalents look like the great structural victory of the exchange model. In Casino Reports’ four-state sample, sportsbook parlays held 15.8 percent in August 2026 (Casino Reports). But combos are priced through a request-for-quote system in which professional market makers quote you a number. The disclosed fee is not the cost. The quoted spread is, and no one has published a measurement of it. That gap is the most important unmeasured number in this market, and until someone measures it the low headline fee should be read as an absence of evidence rather than evidence of cheapness.

The fifth caveat is the empirical one, and it is the one that should give any reader pause. Citizens examined user-level data drawn from Juice Reel, a bet-tracking app, covering October 2025 to May 2026, which makes it a single, self-selected sample. Prediction market parlays ran lower margins than sportsbook parlays, about 13 percent by InGame’s own research on Kalshi against an 18.5 percent sportsbook parlay hold in the Citizens data, yet prediction market parlay traders returned negative 18 percent against negative 12 percent for sportsbook parlay bettors. Across all users, prediction market customers lost eight percent of bankroll against six percent at the books. The difference that stands out is bet size, though the note says it does not capture other factors. The average prediction market combo was $66, 92 percent larger than the average sportsbook parlay (O’Boyle, “Prediction Market Parlays”). Kalshi disputed the study as based on a flawed subset of data that excluded its market makers, which is a fair objection and also rather the point, since the market makers are not the retail customer.

A cheaper price consumed by a larger bet is not a saving. It is a discount that funded an escalation.

Finally, the question of whether the exchange is a better forecaster than the book, which is different from whether it is cheaper. The only direct study located compared Kalshi and DraftKings win probabilities across 285 NFL games in the 2025 season using Brier scores. Kalshi was marginally more accurate, the difference was barely significant, both platforms became less accurate as kickoff approached, and, most tellingly, in the author’s words, “[o]nce I control for differences in how each platform distributes its quoted probabilities, the accuracy gap largely disappears” (Cotti). The economist Victor Matheson put the same conclusion in one sentence. “From a bettor’s standpoint, Kalshi and sportsbooks are basically identical” (Allyn, “Kalshi Says It’s Not a Sportsbook”).

Kalshi’s own accuracy research, covering 2.3 million bets since 2021, reaches a finding the company reports honestly and its marketing does not emphasize. The platform’s predictions are most accurate in markets that are hard to manipulate, such as elections and economic indicators, and weakest in the easy ones. Sports is in the second category (Hoffman, “Kalshi Banks on Its Predictions’ Accuracy”).

The Claim That the House Is Gone

The strongest thing the exchanges say about themselves is structural rather than numerical. Tarek Mansour, Kalshi’s chief executive, argues that sportsbooks are “designed for customers to lose,” that “when a customer wins on a traditional sportsbook, they block that customer,” and that an exchange has no such incentive because it collects fees regardless of who wins (Florio, “Kalshi CEO Outlines the Difference”).

The first half of that is demonstrably true, and the evidence arrived in 2026 from an unlikely source. Massachusetts became the first state to require sportsbooks to justify account limits, and the operator data submitted under that rule showed limitations affecting 0.64 percent of accounts overall with, in the regulator’s phrasing, “a clear pattern” in which above-average winnings are associated with a higher likelihood of restriction (Dawson, “Massachusetts First to Require”). Books limit winners. That has always been true and is now documented.

The second half is contested. Kalish alleges that Kalshi’s request-for-quote infrastructure shares user identifiers with the Wall Street firms making its markets, allowing them to identify winning accounts and simply decline to quote them. His framing is that the practice is the same and the honesty is not. Kalshi has responded that it is an exchange with no house, and has not addressed the specific mechanism. The allegation is uncorroborated and should be treated as such, but the underlying economics do not require a conspiracy. As one analyst of the sector observes, the asymmetric taker and maker fee structure sorts participants into roles by itself. Sharp traders rest orders and collect the maker rate or a rebate; ordinary users cross the spread because they want the position now. Event contracts generate no cash flows, so the platform’s volume is only worth something when someone on the losing side keeps trading. That is a poker room’s economics, not a stock exchange’s (Rose-Berman, “Kalshi’s Favorite Lie”).

What is unambiguously true, and genuinely valuable, is that a bettor holding a contract can sell it at any time at the prevailing market price rather than accepting a sportsbook’s cash-out offer, which is priced to the book’s advantage. That is a real feature. It is also a feature that mainly benefits someone with a reason to change their mind, which is to say someone doing analytical work rather than someone chasing.

The Gap Between the Rooms

There is one place where the arrival of a second venue creates something a careful reader should file away, and it is not a strategy so much as a structural condition.

For the first time since PASPA fell, the same game is priced simultaneously by two systems with different mechanics, different participants and different cost structures. A sportsbook price is set by a trading desk balancing risk and customer profile. An exchange price is set by whoever is willing to rest an order. There is no reason for those two prices to agree at every moment, and every reason to expect them to diverge in the thin corners, the small conferences, the overnight hours, the minutes after an injury report drops.

Whether those gaps are exploitable is an empirical question that has not been answered, and the honest state of the evidence is worth stating plainly. There is no published study measuring whether exchange prices lead or lag sportsbook line moves, and none measuring which venue offers the better price rather than the better forecast. Aggregators market against that gap without filling it.

What can be said is that finding and acting on a two-cent discrepancy across two venues, hundreds of times a day, in markets that move in seconds, is not human work. It is exactly the sort of high-frequency, low-margin, pattern-recognition problem that machines are built for and people are not. It is one of the few places where a machine edge is plausibly real. For now, note only that the exchange loophole did not just create a cheaper venue. It created a second price, and a second price is the precondition for every arbitrage that has ever existed.

The Bill Nobody Is Paying

The cost comparison in the preceding section covers what a bettor pays the venue. It says nothing about what the venue does not pay, and that ledger is where the exchange model’s advantage is largest and least defensible.

A mobile sportsbook licensed in New York surrenders approximately 51 percent of gross gaming revenue in state tax (New York Attorney General, Verified Petition, para. 29). A prediction market operating in the same state is not a licensed sportsbook and pays none of it. The American Gaming Association estimates that states and tribes have foregone about $1 billion, a figure it has not shown its work on and which should be read as advocacy, though the direction is not in dispute. North Carolina has now written prediction markets into its tax code at six percent, against 23 percent for its sportsbooks (Karpchuk and Wingfield, “The States Are Coming”). Holland & Knight presents that as a cooperative model. Read less charitably, it is a tacit legislative concession that the federal preemption argument has won on the ground even where it has not won in court.

The age floor is the clearest gap. Kalshi permits eighteen year olds nationwide; state sportsbooks generally require twenty-one. Mansour’s position is that the two should be treated the same, meaning at the equities age of eighteen. The NBA and the NFL have asked the Commission for a twenty-one floor, the NFL in a July 27 letter reported by Casino.org (St. Clair, “Kalshi Allowing 18-year-olds”; Breen, “Sports Leagues And Player Unions”; Keast, “NFL Urges CFTC”).

Responsible gambling infrastructure is voluntary. Prediction markets are not subject to state responsible gambling rules and do not connect to state self-exclusion lists, which means a bettor who has excluded himself in New Jersey can trade the same game on an exchange the same afternoon. Kalshi built a separate cross-platform exclusion list with an integrity vendor in April 2026 (O’Malley, “Prediction Markets Move Toward”) and pledged $2 million to the National Council on Problem Gambling. Those are real steps taken voluntarily, by a company with every incentive to be seen taking them, in place of a statutory levy it does not owe. The pledge drew a backlash. By Fortune’s count, four of the council’s state partners have cut ties with it this year, the Michigan Gaming Control Board, the Nevada Council on Problem Gambling, Washington’s Evergreen Council on Problem Gambling and the Ohio Casino Control Commission, the first two reportedly, and its executive director resigned in late September, Fortune reported, citing Barron’s (Grigera Naón, “Head of Prominent Gambling Addiction Nonprofit Resigns”).

Advertising is the same story at larger scale. The American Gaming Association’s tracking found that in 2025 nearly one in five sports betting advertisements consumers saw came from outside state responsible-gaming regulation. By 2026 that share was more than half (American Gaming Association, “Prediction Market Advertising Trends”). Whatever one thinks of state advertising rules, an industry in which more than half the messaging sits outside them is not a regulated industry.

On integrity the record is more mixed than either side admits. The platforms have built real machinery, blocking athletes, referees and politically exposed persons from markets touching their own affairs, and they have referred cases. Kalshi and Polymarket collectively flagged 140 suspicious trades in 2026 (Johnson, “Kalshi and Polymarket Flag”). Actual enforcement has landed on a congressional candidate who bet on himself, an editor who traded on an employer’s unreleased information, an Army sergeant charged over classified information, and a former congressman fined over a market on his own attendance at the State of the Union (Bomey, “Kalshi Suspends Users”; Johnson, “Kalshi and Polymarket Flag”; U.S. Department of Justice, “U.S. Soldier Charged”). What has not happened, and it should be said plainly rather than implied away, is any documented case of a prediction market contract tied to actual on-field manipulation of a game. The integrity failures so far have been political and informational, not athletic.

The leagues’ posture is the most revealing detail in this section. Major League Baseball signed an exclusive partnership with Polymarket in March 2026 and, the same day, a memorandum of understanding with the CFTC on market integrity. In July it told the Commission, in a comment on the proposed rule, that a league’s “affirmative opposition to a particular event contract should weigh heavily” in the public-interest review, and that the framework should apply to sports-related event contracts even when they do not settle on what happens in a game (MLB, “MLB Names Polymarket”; Breen, “Sports Leagues And Player Unions”). The NFL has neither a deal nor a softened position (Keast, “NFL Urges CFTC”). Article 1.3 traced the leagues’ reversal on integrity. This is the same reversal in a new suit, taking the purse with one hand and writing to the regulator with the other.

The Mirror Test

The Prediction-Market Exchange

VerdictConditional

The Mirror Test is this series’ standing verdict: every strategy, product and claim is judged on whether it creates a real edge or only the appearance of one. The verdicts are EDGE, ILLUSION and CONDITIONAL.

The exchange model is a real improvement in price and structure, not an illusion. A taker on a liquid Kalshi main line pays about 3.4 percent of stake against 4.55 percent on a standard sportsbook line, and against roughly ten percent for what the average American bettor actually realizes across all products. He can exit a position at market rather than at the book’s cash-out price, and is not trading against a counterparty that profits from his losses. Those are genuine, verifiable, survivable advantages.

They are also conditional on behavior almost nobody exhibits. The advantage exists only on deep markets, only for a disciplined order-placer, and it vanishes entirely into the bid-ask spread on thin markets, into a punitive effective rate on longshots, and into an unmeasured quoted spread on combos. The only user-level data available, from a single self-selected sample, shows exchange customers losing money faster than sportsbook customers, with bigger bets the likeliest reason.

Why the verdict holds

A better price is not an edge. It is a smaller tax on a wager that still has to be right.

CONDITIONAL The Prediction-Market Exchange

Conclusion: A Better Price Is Not an Edge

Strip away the litigation and the valuations and what remains is a product that is, on its best day, about one percentage point cheaper than the thing it replaced, and only on the narrow subset of bets where its liquidity is deep enough to fill you at the advertised price.

That improvement is real. It should not be dismissed. A bettor who confines himself to main lines on the highest-volume games, who rests limit orders rather than crossing the spread, who never touches a combo, and who reads the order book before clicking, is getting the best price available in the American market. That person also describes perhaps one bettor in a thousand.

For everyone else, the exchange delivers a slightly cheaper version of a negative expectation. The spread eats the fee advantage in thin markets. The longshot fee structure punishes the bets people actually like. The combo product hides its cost inside a quoted spread nobody has measured. And the one piece of user-level evidence available, from a single self-selected sample, says that people trading on exchanges lost money faster than people betting at books, apparently because the cheaper product invited larger positions. That is the oldest result in the study of prices, and it should not surprise anyone. Lower the unit cost of a thing and people buy more of it.

The deeper lesson is the one this article was placed in Part 1 to teach. The exchange changed the venue, the regulator, the tax rate, the minimum age, the fee schedule and the corporate structure of American sports betting. It did not change the skill requirement by a single percentage point. Every question about whether you can beat a market survives the move intact, because those questions were never about who was holding the money.

The house did not disappear. It reorganized as a limited liability company, filed a certification, and opened for business the next morning.

Sources

This is 1.4 of House of Mirrors: A Bettor’s Guide to the Illusions of Modern Sports Gambling, a serialized investigation of the American sports betting industry. Next up is article 2.1, The Dopamine Loop.