Why it matters
An event contract that pays $1 if a stated outcome occurs and nothing otherwise trades between 1 and 99 cents. The convenient reading is that the price is the probability: a contract at 62 cents means the market thinks the event 62% likely. Such quotes now appear in newspapers and in policy discussion because they compress dispersed information into one number that updates as facts arrive.
The reading is approximately right and precisely wrong, and the difference matters wherever the number prices something. A contract price is produced by traders with capital constraints, risk preferences, funding costs and views on whether the contract will resolve as they expect. Each introduces a wedge between the quote and the average belief, and what matters in risk management or valuation is the size and direction of that wedge, not that one exists.
How it works
The instrument is a fully collateralised binary. Contracts trade between 1 and 99 cents and pay $1 if the outcome occurs, so the venue holds the whole payout and no participant carries counterparty leverage. Kalshi operates an order book; Polymarket users trade against each other. Both now sit inside the US regulatory perimeter as designated contract markets with affiliated clearing.
Because the two sides must sum to $1, the quotes give a band around the implied probability, not a point. The position economics are plain: a buyer at 62 cents risks 62 to win 38, so the break-even probability is 62%. A risk-neutral trader with no funding cost buys whenever their probability exceeds the price, so the price sits where the marginal trader is indifferent.
Resolution is the other half of the instrument, and not a formality. Exchange-listed contracts resolve against sources named in the contract terms. Onchain venues resolve through an optimistic oracle: an outcome is proposed with a bond, anyone may challenge it, and a contested question escalates to a token holder vote. A contract is a probability of an event only to the extent that resolution tracks the event.
Economic mechanism
The canonical treatment is Wolfers and Zitzewitz, NBER Working Paper 12200 of April 2006, which asks when price equals mean belief. With log utility, "prediction market prices are equal to the mean belief among traders", an identity rather than an approximation. Away from log utility it breaks in a direction set by risk aversion: with low risk aversion "prices will be biased toward 1/2", with higher risk aversion "prices will tend to be biased toward zero or one". The gap widens as beliefs disperse; their calibrations put it "typically within a percentage point" under moderate dispersion, and the approximation is best between $0.20 and $0.80.
Follow the money to see why the tails are worst. Capital in a binary is locked until resolution and earns nothing meanwhile. A trader who believes an event has a 2% chance can sell at 3 cents, tying up 97 cents of collateral to earn one cent over however many months remain: the annualised return is derisory even when the view is right, so the capital does not arrive and the price stays above the belief. That is the funding-cost account of the favourite-longshot bias: longshots are overpriced because arbitraging them is capital-intensive and slow. Where collateral earns a return, through interest on margin or a yield-bearing stablecoin, the wedge narrows, which makes collateral choice a determinant of price accuracy rather than a plumbing detail.
Who is paid for supplying the information is the other half. Market makers earn the spread, informed traders the difference between price and outcome, the venue its fees. The consumer of the price pays nothing: the output is a public good, undersupplied relative to its social value and subsidised by participants trading for entertainment or hedging.
Participants
Designated contract markets list and clear the contracts under CFTC oversight, Kalshi and Polymarket being the two of consequence. Brokers and platforms distribute them, with Robinhood and Coinbase among the retail routes. Market makers supply two-sided quotes. Hedgers are the category the CFTC is most interested in and the thinnest in practice. Data distributors, Intercontinental Exchange the most prominent, sell the prices on to institutions. Resolution sources determine payout, and the CFTC's 2026 rulemakings define which questions may be asked.
Examples
The 2024 US presidential election produced the strongest published claim for these prices and the strongest rebuttal, from the same university. A Vanderbilt team (Cutting, Hughes-Berheim, Johnson, Baroud and Goldstein) published on 7 July 2025 a comparison in which Polymarket outperformed polling averages: its 95% intervals no longer crossed 50% by mid-October 2024, while polling intervals never firmly indicated either candidate even on election day. Their stated limitations are substantial: one market source against aggregated polls, unknown bettor demographics, and Polymarket's exclusion of US users at the time.
On 5 December 2025 Joshua Clinton and TzuFeng Huang, also at Vanderbilt, analysed about 2,500 markets and $2.5bn of volume across Polymarket, Kalshi and PredictIt in the final five weeks of that election, reporting correct resolution rates of 67%, 78% and 93% respectively, negative serial correlation in 58% of Polymarket's presidential markets, and inefficiency increasing in the last two weeks. Kalshi's Jack Such rejected the methodology: accuracy means calibration (whether contracts at 70 cents resolve yes 70% of the time) and on that measure Kalshi is close to accurate. Clinton replied that the issue was not the hit rate but that price changes in similar markets were largely unrelated day to day, implying traders reacting to each other rather than to news. The disagreement is over the scoring rule.
Risks and limitations
Manipulation is the most cited risk and the least settled empirically. A large directional bet moves a quoted probability that journalists and institutions read, and in a thin market the cost is small against the value of influencing a narrative. The counter is that manipulation creates a profitable trade for anyone with a better view, so it self-corrects where there is depth. Depth varies enormously across contracts on one venue.
Resolution risk is underweighted. Where a contract turns on interpretation rather than a published number, the payout depends on an adjudication, and disputes over whether an event occurred inside a stated window are routine. A price is a probability of a payout and only derivatively of an event.
The regulatory perimeter is being drawn now, and it will change what the prices measure. The CFTC's advance notice of 16 March 2026 recorded that designated contract markets listed about five event contracts a year from 2006 to 2020, 131 in 2021 and roughly 1,600 in 2025, and asked what role such contracts play in "managing and assuming price risks, discovering prices, or disseminating pricing information". On 10 June 2026 the Commission proposed a framework for contracts involving the activities enumerated in section 5c(c)(5)(C) of the Commodity Exchange Act (terrorism, assassination, war, gaming and unlawful conduct), applying public interest factors contract by contract. If the categories carrying most of the volume are constrained, the liquidity that makes the remaining prices informative goes with them.
Finally, calibration is a property of a large sample, not of a single quote: a venue can be well calibrated in aggregate and badly wrong on the contract in front of you, and it chooses the sample of resolved contracts on which it is measured.
Key metrics
For a single contract: the bid-ask band rather than the midpoint, open interest, and volume since the last material news, which distinguishes a stale quote from a current one. For a venue: calibration curves over resolved contracts with the sample defined, Brier or logarithmic scores against a benchmark, and the dispute rate on resolution. For the market: open interest and notional by category (the record $1.48bn of mid-June 2026 rested substantially on sports contracts), and the return earned on posted collateral, which sets the floor under the favourite-longshot wedge.
Related research
Wolfers and Zitzewitz remains the reference for when a price is a probability and how far it departs. The two Vanderbilt papers of July and December 2025 reach opposite conclusions on the same election, a fair summary of the evidence. The CFTC's advance notice of 16 March 2026 and its proposal of 10 June 2026 record how a regulator is weighing the informational value of these contracts against their other uses.