Who Pays for the Probability?

Combined monthly trading volume on prediction markets rose from under $5 billion in September 2025 to about $24 billion in April 2026.¹ Most of that is sports. Since July 2024, sports contracts have accounted for roughly 80% of Kalshi’s volume and 39% of Polymarket’s, on platforms whose reputation was built on elections, where political contracts made up 90% and 65% of volume respectively in October and November 2024.¹

A prediction market hands out two different things. One is a number that anybody can read without opening an account, and that number has a respectable forecasting record. The other is a contract, and somebody has to be on each side of it. Across 1.6 million Polymarket accounts that had traded since November 2022, the Wall Street Journal found that 0.1% of accounts captured 67% of all profits, while more than 70% of users lost money.²

Those two facts sit together comfortably once you see what connects them. The forecast is worth reading partly because informed traders are paid to produce it, and they are paid by the people on the other side. This guide works through what the price is worth, what it does not tell you, who funds it, and which of the costs stay with the trader and which land on everyone else.

What the price is doing

The case for prediction markets starts with information rather than with gambling. Hayek’s argument in 1945 was that the knowledge an economy runs on never exists in one place. It exists as “the dispersed bits of incomplete and frequently contradictory knowledge which all the separate individuals possess,” and the price system works as “a mechanism for communicating information” among them.³ Summitward’s guide to why macroeconomic forecasts are not an investment strategy develops that argument and its consequences for forecasting.

A binary event contract is close to a laboratory version of the mechanism. Thousands of people hold fragments of information about whether something will happen. A contract paying $1 on occurrence gives anyone who thinks the current price is wrong a reason to trade against it, and their trades move the price.

The empirical record supports taking this seriously. Berg, Nelson and Rietz compared Iowa Electronic Markets prices against 964 national polls across the 1988 to 2004 presidential elections and found the market closer to the eventual outcome 74% of the time, and significantly more accurate than polls in every election when forecasting more than 100 days out.⁴ Reviewing the field, Wolfers and Zitzewitz concluded that market-generated forecasts “are typically fairly accurate, and that they outperform most moderately sophisticated benchmarks.”⁵

It works away from elections too. Cowgill and Zitzewitz studied internal markets at Google, Ford and a large private materials company and found them “relatively efficient” despite thin trading, weak incentives and traders with reasons to push a result, improving on the firms’ own expert forecasts “by as much as a 25% reduction in mean squared error.”⁶ That 25% is the best of the three firms rather than a typical result, and the comparison against expert forecasters at Ford carried a p-value of 0.104, so it is a direction of travel rather than a precise estimate. The markets also showed an optimism bias and overpriced extreme outcomes relative to a naive prior, and those distortions shrank as traders gained experience.

In 2008 a group including Kenneth Arrow, Paul Milgrom, Robert Shiller, Vernon Smith, Philip Tetlock and Justin Wolfers argued in a Science policy forum that group prediction is “a potent research tool that should be freed of unnecessary government restrictions,” written as the Commodity Futures Trading Commission was considering how to treat event contracts.⁷ This is an old mechanism as well as a new industry. Confederate gold bonds trading in Amsterdam implied roughly a 42% probability of Confederate victory before Gettysburg and Vicksburg, falling to about 15% by the end of 1863, which Summitward covers in what happens to stocks during wars.

What the price leaves out

A contract at 63¢ gets reported as a 63% probability. That is a serviceable approximation and it is not something the theory guarantees.

Manski worked out what an equilibrium price pins down when traders are risk-neutral and disagree with each other. The answer is less than the round number suggests. The price reveals nothing at all about how dispersed beliefs are, and it confines the mean belief only to an interval:

The interval is centred on the price $p$ and has width $2p(1-p)$.⁸ At a price of 50¢, the average belief among traders could be anywhere from 25% to 75%. At 90¢ the interval tightens to 81% and 99%. Confident prices say much more about what traders think than uncertain ones do, which is the reverse of how a market sitting at “too close to call” usually gets reported.

Wolfers and Zitzewitz supplied the counterweight. Under log utility, or under a weaker symmetry condition on demand and beliefs, and provided beliefs are uncorrelated with wealth, the price equals the mean belief exactly. Away from those conditions it stays close as long as beliefs are not too dispersed, and when wealth and belief are correlated the price becomes a wealth-weighted average, which is a different object from one trader one vote. Their conclusion was that prices “typically provide useful (albeit sometimes biased) estimates of average beliefs.”⁹

Two further results bound the optimism. Betting markets show a persistent favorite-longshot bias, with long shots overbet and favorites underbet, and Snowberg and Wolfers found the evidence favours traders misperceiving probabilities over traders simply enjoying risk.¹⁰ And aggregation can fail by design. Galanis, Ioannou and Kotronis show that the securities known to aggregate information under expected utility “no longer aggregate information if some traders have imprecise beliefs and are ambiguity averse,” confirm it in a laboratory prediction market, and prove that “there does not exist a security that is strongly separable for all information structures.”¹¹ No contract design survives every information structure.

The practical reading is that a market price is a market-implied probability. Liquidity, fees, the wording of the contract, the wealth of the people trading it and their appetite for risk all sit inside the number, and a headline that reports it to one decimal place is claiming precision the mechanism does not deliver.

Try it: the price-to-probability explorer

The first panel below draws the Manski interval across every price from 1¢ to 99¢, so you can see how much the price constrains average belief. The second panel takes the costs you face and returns the two prices at which trading stops being worthwhile. On a contract that settles at 100¢ or nothing, a price in cents and a probability in percent are the same unit, so a two-cent spread puts the price at which buying makes sense and the price at which selling makes sense two percentage points apart.

The numbers the tool returns are arithmetic on the inputs you give it, not estimates of any venue’s actual costs. Spread and fee vary by platform and by contract, and several venues price fees as a function of the contract price rather than as a flat amount, so the figure to use is the one from the fee schedule your venue publishes. The point the panel makes is structural. Costs enter twice, on the way in and on the way out, and they define a band of beliefs inside which the correct action is to read the price and do nothing.

Who pays for price discovery

Grossman and Stiglitz gave the answer in 1980, in the paper that established why a perfectly informative price cannot exist. If prices already reflected everything, gathering information would earn nothing and nobody would gather any. What exists instead is “an equilibrium degree of disequilibrium,” in which prices reflect informed traders’ information “only partially, so that those who expend resources to obtain information do receive compensation.”¹² The macro forecasting guide works through what that implies for anyone trying to beat a price.

Applied to an event contract, it says something specific about who is funding the forecast everyone else reads at no charge. The trader with better data, faster execution, a sharper model or genuine private knowledge earns a return, and that return has to come from a counterparty. Recreational traders supply a good deal of it. Two distinct products come out of the same venue, and they go to different people.

The Journal’s account distribution is what those informational rents look like in a retail venue. Fewer than 2,000 Polymarket accounts, out of 1.6 million, took roughly half a billion dollars in aggregate profit, while the typical user was down somewhere between one dollar and a hundred.² On Kalshi, a company spokeswoman told the Journal there were 2.9 unprofitable users for every profitable one, based on the prior month’s data. That concentration is close to what the theory predicts a market producing good forecasts will look like from the inside, and it implicates nobody in wrongdoing.

Why the contract is not an investment

Summitward’s guide to why every portfolio is long-short states the house position on this: relative performance among investors is close to zero-sum before costs, because your overweight is somebody else’s underweight, while the underlying enterprise is not, because companies earn profits and pay dividends and all investors can do well at once.

An event contract has only the first half. A fully collateralised binary pays out exactly what was paid in. Aggregate trading gains equal aggregate losses before costs, and after spreads and fees the participants as a group finish behind. There is no business underneath generating cash flow that everyone can share in, which is what a stock or a bond offers and what produces a return you can expect without being better than your counterparty.

Three reasons to trade one survive that arithmetic. The first is entertainment, where a $10 expected loss buys something worth more than $10 to you. The consumption test Summitward applies to collectibles and to a fantasy football buy-in applies here without modification: would you place the trade knowing with certainty that you would win exactly $0? The second is hedging, where a contract correlated with a real personal or business exposure earns its place the way insurance does, by paying off in the state that hurts. The third is a genuine edge, which is rarer and much harder to establish than it looks; sports betting versus investing works through how many observations it takes to distinguish a real edge from luck, and the answer runs to years.

None of the three is a case for an allocation. Summitward’s framework for compensated versus uncompensated risk puts event contracts firmly in the second category for anyone without a demonstrated advantage.

Private costs, internalities and externalities

Debates about gambling harm tend to collapse several different economic objects into one word. Separating them makes the argument both more defensible and more useful, because the policy implications differ.

The internality row deserves its own note, because it is the one most often mistaken for an externality. A cost you impose on your own future self through present-biased choices is not borne by anyone else, and the economics of it were formalised by Herrnstein, Loewenstein, Prelec and Vaughan, who coined the term, and developed for policy by O’Donoghue and Rabin.¹³ It can justify intervention on paternalistic grounds. It does not justify the claim that taxpayers are picking up the bill.

The rows that genuinely reach third parties have real evidence behind them. The World Health Organization reports that “for every person who gambles at high-risk levels, an average of six others (usually non-gamblers) are affected,” and lists harms including “relationship breakdown, family violence, financial distress, stigma, income-generating crimes (theft, fraud), neglect of children.”¹⁴ A 2025 review of 121 studies found that between 4.5% and 21.2% of adults in general-population surveys report being affected by someone else’s gambling, with harms concentrated in emotional and relationship domains before financial and health ones, and many persisting after the gambling itself stops.¹⁵

There is also a revenue fact that shapes the whole industry. The WHO, citing Gambling Research Exchange Ontario, reports that “people gambling at harmful levels generate around 60% of losses.”¹⁴ The Lancet Public Health Commission’s 2024 synthesis puts past-year gambling at 46.2% of adults worldwide, estimates about 80 million adults with gambling disorder or problematic gambling, and finds that among people using online casino and slot products 15.8% of adults meet criteria for gambling disorder, against 8.9% for sports betting products.¹⁶

Totalling all of this into one dollar figure is where the evidence thins out. A 2025 scoping review of social-cost estimates found seven different methodologies across 26 studies, with anywhere from 5 to 32 cost items counted, producing a mean estimate of 3,980 international dollars per adult against a median of 449.¹⁷ A ninefold gap between mean and median is a statement about method rather than about the world. The existence of these externalities is far better established than any particular estimate of their size.

One fiscal spillover has been measured cleanly. Studying counties near state borders, Goss and Mangrum at the New York Fed found spillover betting spending in non-legal counties within 15 miles of a legal state equal to roughly 14% of the direct effect, and concluded that “spillovers create a fiscal asymmetry: states that have not legalized bear costs from cross-border betting without capturing tax revenue.”¹⁸

What the household evidence covers

The causal research on financial harm has strengthened considerably, and it is worth being precise about what it studies. Sports betting versus investing goes through this literature in detail, including where the studies disagree. In brief: Baker, Balthrop, Johnson, Kotter and Pisciotta, using transaction data on 183,821 households, find that legalization “does not displace other gambling or consumption but significantly reduces savings, as risky bets crowd out positive expected value investments,” concentrated among frequent bettors and low-savings households.¹⁹ Hollenbeck, Larsen and Proserpio, using credit files for about 7 million consumers, find credit scores falling 0.7 points under general legalization and about 12 points where online betting is added to an existing retail market, alongside increases in bankruptcy filings, collections and delinquencies.²⁰ Goss and Mangrum find delinquency rising 0.3 percentage points from a 10.7% base.¹⁸

The studies do not agree on the channel. Caliskan, using the Survey of Income and Program Participation, estimates that online gambling legalization raises credit-card debt by 11.9%, driven by households already carrying a balance, and reports that there are “no effects on income, savings, or stock holdings.”²¹ That is the opposite of Baker and colleagues on the savings question. The datasets differ, transaction records against self-reported survey panels, and so does the treatment, online gambling broadly against sports betting specifically. Taken together they establish that household balance sheets move. They do not yet establish which side of the balance sheet moves most.

A working paper adds a further margin, estimating that legalization “reduces household food sufficiency by 2.1 percent among working-age adults without a college degree,” with effects that cycle with the NFL season.²² It has not been through peer review.

Here is the limitation that matters most for this guide. Every one of those studies is about sports betting or online gambling. None of them measures prediction markets. There is no comparable causal evidence that access to Kalshi or Polymarket reduces saving or raises delinquency, and the platforms differ from a sportsbook in ways that could cut either direction: prices are set by traders rather than posted by a house, some contracts support real hedging, and the user base may not resemble a sportsbook’s.

The distinction is also narrowing. The same research group has a forthcoming review examining “the rapid expansion and convergence of retail betting markets,” treating sports betting, prediction markets and retail options trading as one phenomenon driven by shared technology, behavioural drivers and regulatory arbitrage.²³ With sports contracts at about 80% of Kalshi’s volume,¹ the economic difference between a sports event contract and a sports bet is getting harder to state. The honest position is that the harm evidence comes from adjacent markets, the generalisation is untested, and the untested gap is closing.

Manipulation and inside information

Worries about manipulation are usually aimed at the wrong target. The experimental evidence on price manipulation is consistently reassuring. Hanson, Oprea and Porter ran markets containing traders paid to distort the price and found accuracy held up, because other participants “compensated for the bias in offers from manipulators” by shifting the terms on which they would trade.²⁴ Camerer placed and cancelled large bets at racetracks and moved the posted odds visibly, but the net effect on other bettors came out close to zero.²⁵ Rhode and Strumpf examined the Wall Street political betting markets of 1880 to 1944, an era rich in accusations of party operatives rigging prices, and found manipulation attempts produced brief effects, in one case reverting within an hour.²⁶

Hanson and Oprea went further and showed that a manipulator can raise accuracy. Uncertainty about what price a manipulator is targeting increases the returns to trading on real information, which draws in more informed trading than the manipulation drives out.²⁷

Enforcement has instead landed on traders who knew something, or who could influence the result. In February 2026 the CFTC’s Division of Enforcement issued an advisory following two cases on Kalshi. In one, a political candidate traded contracts on his own candidacy, violating rules that “prohibit trading in a contract over which the trader has direct or indirect influence over the outcome.” In the other, someone with an employment relationship to the subject of a contract traded it while, in the advisory’s words, they “likely had access to material non-public information.” The advisory noted that the Commission “has full authority to police illegal trading practices occurring on any” designated contract market.²⁸ The sanctions were small, in the thousands of dollars with multi-year suspensions, and the principle behind them is the one worth carrying. The counterparty to worry about is the one who already knows the answer.

Design, regulation and the alternatives

If some of the costs land outside the trader, design and regulation become legitimate questions. The evidence on the standard interventions is mixed enough to be worth stating carefully.

Pop-up warnings work better than their reputation suggests. A meta-analysis of 18 studies found moderate effects on both cognition and behaviour, with a standardised mean difference of 0.505 on behavioural measures. The weakness is the setting: only two of those studies observed real gamblers risking real money, and only two collected follow-up data, so durability and real-world transfer are unestablished.²⁹

Voluntary limits look weaker. A randomised trial of 4,328 online slots customers found that prompting people to set a deposit limit raised the share who set one, and left net losses statistically unchanged against control.³⁰ The limits were self-chosen, adjustable and uncapped, which is a fair description of most of what platforms currently offer. Self-exclusion has the opposite problem: it helps people who use it, and almost nobody does. A meta-analysis puts lifetime self-exclusion at 0.26% of the general adult population and 15.2% among people in the problem-gambling category.³¹

Whether the operator wants to find struggling customers is a separate question from whether it can. A New York Times investigation published in September 2026 reported that DraftKings built a model scoring online casino players by how much they were likely to lose after receiving a promotion, directing a large promotional budget toward the highest scorers, while several internal efforts to build models flagging customers sliding toward crisis were shelved. The Times interviewed more than 40 former DraftKings employees. In a statement, the company said its promotions are “directed toward customers who demonstrate sustained, engaged use of our platform, not toward customers based on their losses,” and no regulator has accused it of illegality.³² The mechanism here does not require machine learning. A segmentation rule or a competent analyst could pursue the same objective more slowly. What the technology changes is cost and precision; what determines the outcome is which objective it is pointed at.

The jurisdictional picture is unsettled in a way that will shape the next few years. The CFTC sued Arizona, Connecticut and Illinois in April 2026 to defend exclusive federal jurisdiction over event contracts, and in late July 2026 forty-four state attorneys general wrote to the Commission pressing the opposite view. A federal court in Nevada enjoined sports contracts on both major platforms, with the appeal pending in the Ninth Circuit, Maryland litigation was argued in the Fourth Circuit in May, and tribal suits are live in California, New Mexico and Wisconsin. Polymarket returned to the United States by buying a CFTC-licensed exchange for $112 million in July 2025 and relaunching a US product that November.³³

There is one more consideration that any design debate should include. The claim that restricting wagering destroys the forecast turns out to be too strong, because markets are one of several mechanisms for eliciting good probabilities. Comparing crowd prediction systems on the same questions, Atanasov, Witkowski, Mellers and Tetlock found prediction markets and prediction polls statistically equivalent in accuracy, and small elite crowds outperforming larger ones.³⁴ In a multi-year geopolitical forecasting tournament, trained superforecaster teams reached a Brier score of 0.07 in the final week of the second year, against 0.26 for untrained individual forecasters at the same horizon.³⁵ Real money has real advantages, including a standing incentive to show up and correct a wrong price. It is not the only way to get a calibrated number.

So the design question is genuinely open rather than settled. Which frictions reduce harm without degrading the aggregation that makes the price worth reading? Mandatory caps, cooling-off periods, restrictions on promotional targeting and limits on contract frequency all trade against liquidity to different degrees, and the research base for choosing among them is thin.

What this means for a DIY investor

Three questions are worth keeping apart, because they have different answers.

Is the price useful information? Often, yes. For a liquid, well-designed contract it is one reasonable input among several, and it has the rare virtue of being scoreable against reality later. Read it with the Manski caveat attached, and with more confidence when it sits near the ends than when it sits near the middle.

Is the contract a useful hedge? Occasionally. If an outcome genuinely moves your income or your costs, a contract that pays off in that state can be worth a slightly negative expected return, which is the same logic that justifies insurance. That is a narrow case and it requires the correlation to be real rather than thematic.

Is repeated speculation a way to build wealth? No. There is no underlying enterprise producing a return you can share in, so your gains have to come from being better than your counterparties after costs, from providing liquidity you are compensated for, or from a mispricing that somebody else has not yet corrected. The profit concentration in the Journal’s data is what that contest looks like when 1.6 million people enter it.² The appropriate portfolio allocation is zero, and money that goes into contracts should come from an entertainment budget rather than from savings.

The opportunity cost deserves its own line, because it is the part that compounds. Baker and colleagues find betting money coming out of investment deposits rather than out of other spending.¹⁹ A dollar that never reaches a brokerage account costs more over thirty years than the expected loss on the dollar that was wagered.

What this means for everyone else

Most people who benefit from prediction markets will never open an account, and that is the interesting part. They get a continuously updated probability, attached to a falsifiable claim, that can be scored after the fact. That is a genuine contribution to public argument, and it is structurally similar to reading a price you did not help set.

Two habits make that free information more useful. Treat a number near 50¢ as a statement of uncertainty rather than a precise estimate, since that is exactly where the price tells you least about what traders believe. And keep track of the difference between what the market implies and what it reports: a contract at 70¢ means the market prices roughly a seven-in-ten chance under the assumptions above, which is not the same as the market saying an event will happen.

Bottom Line

Prediction markets convert scattered private information into a public probability, and the evidence that they do it reasonably well goes back decades. Producing that number requires traders willing to fund it, and the returns concentrate heavily among a small group of sophisticated participants. Some of the resulting costs stay with the trader, some fall on their future self, and some land on families, creditors and public programs. A DIY investor can read the price, and should keep any wager inside an entertainment budget.

Key Takeaways

• The forecast and the wager are separate products. The probability is a positive informational externality available to anyone. The contract is a transfer between two traders, minus costs.
• A price is not literally a probability. Under Manski’s assumptions a price of 50¢ confines the average trader’s belief only to somewhere between 25% and 75%, and prices near the middle are the least informative.
• Profits concentrate the way theory predicts. Across 1.6 million Polymarket accounts, 0.1% captured 67% of profits and more than 70% of users lost money. Informed traders earning rents is the mechanism that makes the price worth reading.
• Event contracts have no underlying enterprise. A diversified stock portfolio pays you for bearing risk in a business. A binary contract pays you only if you were better than the person on the other side.
• The harm evidence is about sports betting. Large causal studies find reduced saving, higher credit-card debt and higher delinquency after legalization, and none of them measure prediction markets. With sports at about 80% of one major platform’s volume, that gap is closing.
• Externalities exist; their dollar value is unsettled. The WHO counts about six affected others per high-risk gambler, while social-cost estimates vary ninefold between mean and median depending on method.

Frequently Asked Questions

Are prediction markets gambling or investing?

Economically the contract behaves like gambling, in that returns come from a transfer between counterparties rather than from an underlying business. The market as an institution does something gambling usually does not, which is produce a public forecast with a measurable track record. Both statements are true at once, and which one matters depends on whether you are reading the price or trading it.

Does a 70% contract mean the event is 70% likely?

It means the market prices roughly a seven-in-ten chance, subject to liquidity, fees, contract wording and the risk preferences of whoever is trading. Theory does not guarantee the price equals average belief except under specific conditions, and prices near the middle of the range are the least informative about what traders believe.

If professional traders take most of the profits, is the market rigged?

No, and the concentration is closer to a feature than a flaw. Grossman and Stiglitz showed that prices only carry information if the people who gathered it are compensated. The uncomfortable part is not that informed traders earn money, it is that the compensation comes from a large population of recreational traders who mostly lose.

Can someone manipulate a prediction market?

Moving a price by throwing money at it works poorly. Laboratory experiments, racetrack field experiments and a century of political betting data all find manipulation effects that are small or short-lived, and one model finds that manipulation can improve accuracy by attracting informed traders. The real vulnerability is a counterparty trading on nonpublic information or with influence over the outcome, which is what the CFTC advisory of February 2026 addressed.

Is there any case for holding event contracts in a portfolio?

As a strategic allocation, no. As a hedge, occasionally: if an outcome genuinely moves your income or expenses, paying a small expected loss for protection in that state is the same trade as buying insurance. The correlation has to be real, and the position should be sized against the exposure it offsets rather than against your portfolio.

Would restricting these markets destroy the forecasts?

Not necessarily. Prediction polls with good aggregation perform about as well as markets on the same questions, and small teams of trained forecasters outperform large crowds. Real money brings a standing incentive to correct wrong prices, which is a real advantage, but it is not the only mechanism that produces calibrated probabilities.

What should I do if my trading has stopped feeling optional?

Treat it as a health question rather than a financial one. In the United States the National Council on Problem Gambling helpline is 1-800-522-4700, and it operates 24 hours a day. Self-exclusion tools exist on most platforms and help the people who use them, though Bijker and colleagues find that very few people ever do.

Related Guides

• Sports Betting vs. Investing for the household research in detail, the arithmetic of the house edge, and how long it takes to prove an edge is real.
• Why Macroeconomic Forecasts Are Not an Investment Strategy for Hayek, Grossman and Stiglitz, and what a price already contains.
• Every Portfolio Is Long-Short for why relative performance is zero-sum while owning businesses is not.
• Compensated vs. Uncompensated Risk for the test that decides whether a risk deserves portfolio space.
• What Should a Fantasy Football Buy-In Be? for the consumption test and what changes when there is no house taking a cut.
• What Happens to Stocks During Wars for Confederate gold bonds as a working prediction market in 1863.

Sources

1. Pew Research Center, “Trading volume on prediction markets has soared in recent months,” May 27, 2026. Analysis of data from The Block covering July 2024 to early May 2026. pewresearch.org
2. Neil Mehta, Katherine Long and Caitlin Ostroff, “Why Almost Everyone Loses, Except a Few Sharks, on Prediction Markets,” The Wall Street Journal, May 4, 2026. The Polymarket concentration figures are the Journal’s analysis of 1.6 million accounts trading since November 2022; the Kalshi ratio was supplied by a company spokeswoman. wsj.com
3. F. A. Hayek, “The Use of Knowledge in Society,” American Economic Review 35(4), 1945, 519–530. econlib.org
4. Joyce E. Berg, Forrest D. Nelson and Thomas A. Rietz, “Prediction market accuracy in the long run,” International Journal of Forecasting 24(2), 2008, 285–300. Iowa Electronic Markets against 964 national polls, 1988–2004. doi.org
5. Justin Wolfers and Eric Zitzewitz, “Prediction Markets,” Journal of Economic Perspectives 18(2), 2004, 107–126. doi.org
6. Bo Cowgill and Eric Zitzewitz, “Corporate Prediction Markets: Evidence from Google, Ford, and Firm X,” Review of Economic Studies 82(4), 2015, 1309–1341. The 25% figure is the largest of the three firms; the Ford comparison against expert forecasters carried p = 0.104. doi.org
7. Kenneth J. Arrow, Robert Forsythe, Michael Gorham, Robert Hahn, Robin Hanson, John O. Ledyard, Saul Levmore, Robert Litan, Paul Milgrom, Forrest D. Nelson, George R. Neumann, Marco Ottaviani, Thomas C. Schelling, Robert J. Shiller, Vernon L. Smith, Erik Snowberg, Cass R. Sunstein, Paul C. Tetlock, Philip E. Tetlock, Hal R. Varian, Justin Wolfers and Eric Zitzewitz, “The Promise of Prediction Markets,” Science 320(5878), 2008, 877–878. A policy forum rather than a research article. doi.org
8. Charles F. Manski, “Interpreting the predictions of prediction markets,” Economics Letters 91(3), 2006, 425–429. The bound assumes risk-neutral price-taking traders with heterogeneous beliefs. doi.org
9. Justin Wolfers and Eric Zitzewitz, “Interpreting Prediction Market Prices as Probabilities,” NBER Working Paper 12200, 2006. A working paper; it was not published in a journal. nber.org
10. Erik Snowberg and Justin Wolfers, “Explaining the Favorite–Long Shot Bias: Is it Risk-Love or Misperceptions?” Journal of Political Economy 118(4), 2010, 723–746. doi.org
11. Spyros Galanis, Christos A. Ioannou and Stelios Kotronis, “Information Aggregation Under Ambiguity: Theory and Experimental Evidence,” Review of Economic Studies 91(6), 2024, 3423–3467. doi.org
12. Sanford J. Grossman and Joseph E. Stiglitz, “On the Impossibility of Informationally Efficient Markets,” American Economic Review 70(3), 1980, 393–408. aeaweb.org
13. Richard J. Herrnstein, George F. Loewenstein, Drazen Prelec and William Vaughan Jr., “Utility Maximization and Melioration: Internalities in Individual Choice,” Journal of Behavioral Decision Making 6(3), 1993, 149–185, which introduces the term; and Ted O’Donoghue and Matthew Rabin, “Optimal Sin Taxes,” Journal of Public Economics 90(10–11), 2006, 1825–1849, which develops the policy treatment. doi.org
14. World Health Organization, “Gambling” fact sheet, December 2, 2024. The 60% revenue figure is WHO citing Gambling Research Exchange Ontario (2019). who.int
15. Nicki A. Dowling, Chloe O. Hawker, Stephanie S. Merkouris, Simone N. Rodda and David C. Hodgins, “Addressing gambling harm to affected others: A scoping review (Part I),” Clinical Psychology Review 116, 2025, article 102542. 121 studies since 2000. doi.org
16. Heather Wardle, Louisa Degenhardt, Virve Marionneau, Gerda Reith, Charles Livingstone, Malcolm Sparrow et al., “The Lancet Public Health Commission on gambling,” The Lancet Public Health 9(11), 2024, e950–e994. doi.org
17. Sari Hautamäki, Virve Marionneau, Sari Castrén, Jussi Palomäki, Susanna Raisamo, Tomi Lintonen, Pia Pörtfors and Tiina Latvala, “Methodologies and estimates of social costs of gambling: A scoping review,” Social Science & Medicine 371, 2025, article 117940. 26 studies, seven methodologies, 5 to 32 cost items. doi.org
18. Jacob Goss and Daniel Mangrum, “Sports Betting Across Borders: Spatial Spillovers, Credit Distress, and Fiscal Externalities,” Federal Reserve Bank of New York Staff Reports no. 1184, February 2026, revised March 2026. A staff report rather than a peer-reviewed article. newyorkfed.org
19. Scott R. Baker, Justin Balthrop, Mark J. Johnson, Jason Kotter and Kevin Pisciotta, “Gambling away stability: Sports betting’s impact on vulnerable households,” Journal of Financial Economics 183, 2026, article 104330. 183,821 households, 2014–2023. doi.org
20. Brett Hollenbeck, Poet Larsen and Davide Proserpio, “The Financial Consequences of Legalized Sports Gambling,” Management Science, published online July 21, 2026. Earlier drafts reported larger bankruptcy effects than the published version; the figures quoted here are from the published abstract. doi.org
21. Mustafa Enes Caliskan, “Online Gambling Legalization and Household Credit Card Debt,” Journal of Gambling Studies, published online August 13, 2026. doi.org
22. Xiaohui Guo, Lizhong Peng and Chad Meyerhoefer, “Wagering the Bread Money: Sports Betting Legalization and Food Sufficiency,” NBER Working Paper 35305, June 2026. A working paper awaiting peer review. nber.org
23. Scott R. Baker, Justin Balthrop, Mark J. Johnson, Jason D. Kotter and Kevin Pisciotta, “Retail Betting Markets,” NBER Working Paper 35520, July 2026, forthcoming in the Annual Review of Financial Economics. nber.org
24. Robin Hanson, Ryan Oprea and David Porter, “Information aggregation and manipulation in an experimental market,” Journal of Economic Behavior & Organization 60(4), 2006, 449–459. doi.org
25. Colin F. Camerer, “Can Asset Markets Be Manipulated? A Field Experiment with Racetrack Betting,” Journal of Political Economy 106(3), 1998, 457–482. doi.org
26. Paul W. Rhode and Koleman S. Strumpf, “Manipulating Political Stock Markets: A Field Experiment and a Century of Observational Data,” working paper, 2008. Circulated as a working paper; no journal publication. wfu.edu
27. Robin Hanson and Ryan Oprea, “A Manipulator Can Aid Prediction Market Accuracy,” Economica 76(302), 2009, 304–314. doi.org
28. Commodity Futures Trading Commission, Division of Enforcement, “Advisory on Enforcement Authority over Event Contracts,” February 25, 2026, announced in Release 9185-26. cftc.gov
29. B. Bjørseth, J. O. Simensen, A. Bjørnethun, M. D. Griffiths, E. K. Erevik, T. Leino and S. Pallesen, “The Effects of Responsible Gambling Pop-Up Messages on Gambling Behaviors and Cognitions: A Systematic Review and Meta-Analysis,” Frontiers in Psychiatry 11, 2021, article 601800. Behavioural effect g = 0.505 (95% CI 0.256–0.746) across 18 studies, of which only two observed real gamblers risking real money. doi.org
30. Ekaterina Ivanova, Kristoffer Magnusson and Per Carlbring, “Deposit Limit Prompt in Online Gambling for Reducing Gambling Intensity: A Randomized Controlled Trial,” Frontiers in Psychology 10, 2019, article 639. N = 4,328; pooled intervention versus control, OR = 1.0 (p = 0.921) on the proportion with positive net loss and B = −0.1 (p = 0.291) on its size. doi.org
31. Rimke Bijker, Natalia Booth, Stephanie S. Merkouris, Nicki A. Dowling and Simone N. Rodda, “International Prevalence of Self-exclusion From Gambling: a Systematic Review and Meta-analysis,” Current Addiction Reports 10(4), 2023, 844–859. 0.26% of general adults (95% CI 0.16–0.43) and 15.20% of the problem-gambling category (95% CI 11.00–19.39). doi.org
32. Walt Bogdanich and Jenny Vrentas, “At DraftKings, AI Targets the Gamblers Likeliest to Lose,” The New York Times, September 19, 2026. Reporting based on internal documents and interviews with more than 40 former employees; the company’s response is quoted as printed there. nytimes.com
33. Litigation and regulatory status as of September 2026, compiled from CFTC filings, court records and DLA Piper, “Legal status at odds: tracking developments in prediction markets and sports betting,” September 2026. dlapiper.com
34. Pavel Atanasov, Jens Witkowski, Barbara Mellers and Philip Tetlock, “Crowd prediction systems: Markets, polls, and elite forecasters,” International Journal of Forecasting 41(2), 2025, 580–595. doi.org
35. Barbara Mellers, Lyle Ungar, Jonathan Baron, Jaime Ramos, Burcu Gurcay, Katrina Fincher, Sydney E. Scott, Don Moore, Pavel Atanasov, Samuel A. Swift, Terry Murray, Eric Stone and Philip E. Tetlock, “Psychological Strategies for Winning a Geopolitical Forecasting Tournament,” Psychological Science 25(5), 2014, 1106–1115. A 2024 re-analysis by Kim and colleagues in the same journal contests the statistical conclusions about teams and training. doi.org

Method. The explorer computes Manski’s bound in closed form and derives break-even prices from the spread and fee the reader enters. It assumes a contract settling at 100¢ or 0¢, ignores compounding and position sizing, and does not model any venue’s actual fee schedule. Platform volume and sports-share figures are as of the Pew analysis in May 2026; legal status is as of September 2026 and is changing quickly.

Author disclosure

Summitward has no business relationship with Kalshi, Polymarket, DraftKings or any other firm named here, and receives no compensation from them. This guide is educational and is not investment advice. Several sources are working papers or staff reports rather than peer-reviewed publications and are labelled as such where they appear. Figures from newspaper investigations are reported as those outlets published them. Anyone whose gambling has stopped feeling optional can reach the National Council on Problem Gambling helpline at 1-800-522-4700, 24 hours a day.