Prediction Markets Explained: How They Work, Pricing, Risks and Accuracy

Prediction markets are exchanges where people buy and sell contracts tied to future outcomes, such as an election result, an inflation release or a sporting event. Many of these contracts are binary, paying $1 if a defined outcome occurs and $0 if it does not, so their prices can be read as market-implied probabilities: a contract at 65 cents suggests roughly a 65% chance. Other structures, including ranges and numerical contracts, use different payout rules.

That number is a live price produced by whoever is currently willing to trade, and its usefulness depends on liquidity, fees, who can participate, how the question is worded and who decides what happened. Reading the price without reading those things is where most people go wrong.

Key Takeaways

• Prediction markets let people trade contracts on future outcomes. Binary contracts commonly settle at $1 if the outcome occurs and $0 if it does not.
• A binary contract at 65 cents is usually read as about 65% implied probability. The price is a money-weighted market estimate rather than a measured fact.
• Profit depends on your entry price, fees and exit. Being right about the event and losing money is entirely possible.
• Traders can usually sell before resolution, though thin books and wide spreads make exiting expensive.
• Contract wording and resolution rules can matter more than forecasting skill, because you are trading the exact terms rather than the headline.
• Research finds that accuracy varies sharply by category and horizon. Federal Reserve research found Kalshi's policy-rate forecasts competitive with professional and market benchmarks, with particularly strong performance immediately before FOMC decisions, while other studies identify systematic favorite-longshot bias.
• Reported volume is a weak proxy for liquidity. A 2025 Columbia University paper estimated that transaction patterns consistent with wash trading accounted for roughly a quarter of Polymarket's historical volume, peaking near 60% of weekly volume in December 2024.
• Onchain markets add wallet, smart contract, stablecoin and oracle governance risks, and disputed resolutions decided by token voting have overturned outcomes that mainstream reporting treated as settled.
• Legal status is unsettled and moving. Federal and state courts issued conflicting preliminary rulings during 2026 on whether sports event contracts fall under federal commodities law or state gambling law.

What Is a Prediction Market?

A prediction market is a venue where participants trade contracts whose value depends on whether a defined future event occurs. Three features distinguish it from other markets:

• It settles on an event or measured outcome rather than delivering ownership of an underlying asset. Some contracts reference financial or economic measures (CPI, interest rates, index levels) but what determines the payout is whether a stated condition is satisfied, not the delivery of anything.
• Binary contracts have a fixed terminal payout: The common structure settles at $1 or $0, which caps both profit and loss per contract, although prediction markets can also use ranges, indexes and other payout functions.
• The price carries information: Because a $1 payout makes the price a share of the maximum value, it maps directly onto a probability estimate.

People use these markets for three different reasons, and the distinction matters when reading prices. Some are forecasting, trading a view they researched. Some are speculating on short-term price moves. Some are hedging, using a contract to offset a real-world exposure, such as a farmer buying a weather contract so that a bad season pays out even as the crop fails. The mix of motives in any given market shapes how informative its price is.

Not every prediction market is decentralized. Regulated exchanges with formal oversight and blockchain protocols using wallets and stablecoins both exist, and they carry very different protections. The category as a whole grew sharply through 2025 and 2026, with locked capital and trading volume rising on very different curves.

How Do Prediction Markets Work?

Despite these differences, prediction markets generally rely on the same basic mechanics for turning expectations about future events into tradable contracts.

Yes and No Contracts

A market poses a question with a defined resolution rule, then offers two complementary positions.

Those two asks add to $1.02 rather than $1.00. Buying both immediately would cost $1.02 because you cross the spread on both sides. Economically, a No contract provides the opposite terminal exposure to Yes. That does not necessarily make buying No operationally identical to short-selling Yes, because venues differ in collateral treatment, fees, order-book design and available liquidity on each side.

Where do the contracts come from? In some binary markets, complementary Yes and No positions can be created when opposing orders together provide the full collateral required for the $1 payout. Existing contracts can then trade between users in a secondary book. Other venues use different clearing, market-making or token mechanics, so treat the matched-pair picture as one common architecture rather than a universal one.

Beyond Binary: Other Contract Types

Range and index contracts reward closer reading than simple Yes/No shares. In a mutually exclusive bracket market, each contract price can be read as the implied probability of that interval, while the complete set of brackets expresses the market's view of the numerical outcome as a distribution rather than a single headline number. That is a feature worth using: a bracket market tells you about the market's uncertainty, not just its central estimate.

Conditional contracts need care of a different kind. A market on "if A wins, will B pass?" prices the probability of B given A, so it should not be compared directly with an unconditional market on B, and it may be void if the condition never occurs.

Payout, Profit and Maximum Loss

These three numbers are routinely conflated. Buying Yes at $0.64:

Two things follow. The settlement payout is $1 and your profit is 36 cents, which is a different figure from both the payout and the price. And a 56% potential return says nothing about whether the trade is worth taking. At a fair price of 64 cents you would win 64% of the time and lose 36% of the time, producing an expected value of zero before costs and slightly negative after them. Positive expected value requires the market price to be wrong in your favour by more than the cost of trading.

How Do Prediction-Market Prices Become Probabilities?

The conversion is simple arithmetic under a $1 settlement model:

Implied probability (%) ≈ contract price in dollars × 100

Everything interesting is in the qualifications.

Bid, Ask and Last Price

A market rarely shows one number. It shows several, and they mean different things.

So which is "the market probability"? Most interfaces and data feeds pick one convention, often the last trade or the midpoint. A five-cent spread means the current executable market is offering a five-point probability band rather than one clean number. A wide spread is itself information, commonly indicating low liquidity, uncertainty, adverse-selection risk or limited competition among market makers.

Quoting a market-implied probability to one decimal place from a book with a five-cent spread is precision theatre. It is the equivalent of giving your height to the nearest micron while standing on a trampoline.

Why Yes and No May Not Add to $1

In theory, complementary outcomes must sum to $1, since exactly one of them will pay. In practice, displayed figures often sum to slightly more or less, for mundane reasons: the displayed numbers may both be asks and therefore include the spread on each side, fee economics may affect where traders are willing to quote even when the venue charges the fee separately, liquidity can differ between the two sides, and on some venues each side has its own book that arbitrage aligns imperfectly. Small deviations persist because closing them is not free.

A persistent gap larger than a couple of cents in a supposedly liquid market is worth investigating rather than ignoring.

Fees, Makers and Takers

Fees do not just reduce your profit; they move the probability you need to be right. Venues differ in how they charge: some use a flat percentage, some a price-dependent formula, some a maker-taker schedule that rebates or discounts resting orders and charges the trader who crosses the spread. Posting a limit order and waiting can cost meaningfully less than taking the best available price, at the cost of an uncertain fill.

Before sizing a trade, work out the all-in cost of entry, exit and settlement, and add it to the price you are paying. That total, not the sticker price, is your break-even probability.

Implied Probability vs True Probability

The true probability of a specific one-off event cannot be observed before the fact, so no market price can be checked against it in real time. What a price reflects is the point at which the marginal buyer and marginal seller currently agree to transact. That balance responds to information, and also to spreads and fees, how much capital traders can commit, market-maker inventory and risk limits, liquidity and time remaining, who is legally allowed to participate, hedging demand unrelated to any forecast, risk preferences, attempted manipulation, and the risk that the collateral or the platform itself fails.

Key line: the price is the market's answer. Whether it is the correct answer is a separate question.

A Prediction-Market Trade Step by Step

1. Read the exact question, including the deadline and time zone.
2. Read the resolution rules, including the named source and what counts as a qualifying event.
3. Choose a side: Yes or No, or a specific bracket.
4. Check the bid, ask and depth, then decide whether to accept the best available price immediately or place a limit order at a price you choose, depending on the order types the venue offers.
5. Buy. Your fill price, plus costs, sets your break-even probability.
6. Watch the price move as information arrives, then either sell early or hold to settlement.
7. Trading closes at the stated time, and the outcome is determined under the resolution rules, which may take longer than the event itself.
8. Settlement pays $1 per winning contract, and losing contracts expire worthless.

On some crypto-native venues, the process additionally involves wallet connection, collateral-token permissions, transaction signatures and network fees. The exact authorization and payout mechanics depend on the blockchain and platform: some abstract or sponsor fees, some redeem automatically, some require no separate token approval. Where you are granting permissions, reviewing what each one authorizes matters as much as reviewing the trade.

Can You Sell Before the Event Ends?

Usually yes, if someone is willing to buy. This creates two independent paths to profit and two ways to lose. Say you buy Yes at the $0.62 ask. Good news arrives and the market rises to a $0.75 bid.

Two consequences that most explainers skip. You can forecast the outcome correctly and still lose money by overpaying at entry or exiting into a thin bid. And you can be entirely wrong about the eventual outcome while profiting from a temporary price move. Prediction-market trading and event forecasting are related skills rather than the same one.

The decision is partly forecasting and partly plumbing. As one Polymarket user explained to someone sitting on a position in the nineties, "" that leaving does not mean accepting a poor bid.

The arithmetic behind that is worth spelling out: a contract at 92 cents has eight cents of theoretical upside, a three-cent spread eats most of it, and the capital stays locked until resolution while it does so. Remaining upside, spread, depth and opportunity cost are four separate inputs to the hold-or-sell decision.

Order Books, Market Makers and Liquidity

The way orders are matched and liquidity is provided plays an important role in how easily traders can enter or exit a prediction market.

Order-Book Markets

Buyers and sellers post prices and quantities, and the exchange matches them. Advantages: visible depth, familiar mechanics, and prices set entirely by participants. Risks: books can be nearly empty in niche markets, spreads can be wide, and large orders move the price against themselves.

Automated Market Makers

A formula or a designated market maker quotes prices continuously from a liquidity pool or a scoring rule. Advantages: a price always exists, and new markets can launch without waiting for participants. Risks: prices depend on the initial parameters and pool size, price impact is mechanical and predictable, and the market maker or liquidity subsidy absorbs losses to informed traders.

Depth and Slippage

The quoted price applies only to the size available at it. A market showing a 62-cent ask might have this book:

Buying 300 contracts fills entirely at $0.62. Buying 3,000 contracts costs $2,010, an average of $0.67. The same market quotes a 62% probability for a small trade and delivers a 67% average for a large one. Anyone reading the headline number as the price at which real size can trade is reading a retail quote. A live book makes the point concrete.

One trader on a prediction-market forum compressed the practical lesson into eight words that belong above every order book: "." Entering meets resting orders that were already there. Exiting requires someone to want your position later, potentially after the news that moved you has reached everyone else, and after the market makers who were quoting have widened or withdrawn.

Volume, Liquidity and Open Interest

Volume is footprints. Open interest is how many people are currently in the room. A venue's locked value gives a rough read on outstanding exposure, and it moves on a very different rhythm from volume.

High volume does not establish deep liquidity, broad participation or reliable pricing, and on venues without trading fees it can be manufactured cheaply. A November 2025 Columbia University paper, "Network-Based Detection of Wash Trading," estimated that transaction patterns consistent with wash trading accounted for roughly 25% of Polymarket's volume between 2022 and 2025, coordinated or self-dealing accounts trading back and forth without changing net exposure.

The estimated share peaked near 60% of weekly volume in December 2024 and exceeded 90% in some weeks in sports and election markets. Roughly 14% of the 1.26 million wallets studied were flagged, including one cluster of more than 43,000 wallets trading mostly at sub-cent prices.

The researchers presented these as estimates, did not allege platform complicity, and suggested the likely motive was farming future incentives rather than moving prices. Keep the word "estimated" attached to the finding. The lesson for readers is narrower and still important: a large volume figure is not evidence that a market's price is well supported. For judging executability, visible depth is more informative than headline volume. Open interest adds information about outstanding exposure, but neither metric by itself establishes broad or independent participation.

How Are Prediction Markets Resolved?

Resolution is where prediction markets differ most from ordinary trading, and where the largest surprises happen.

Market Close vs Resolution

Four moments can fall on four different dates: trading closes to new orders, the real-world event occurs, the result becomes verifiable under the contract's rules, and the market resolves and pays.

An election happens on a Tuesday while certification takes weeks, recounts may run, and litigation may follow. A contract referencing an official certification, an inauguration or a specific data release resolves on that trigger rather than on the night the result became obvious. Your collateral stays locked throughout, earning nothing, which is a real cost on a long-dated contract even when you are right, because the same stablecoin could be earning a yield elsewhere.

Official Sources and Contract Wording

Before trading, the wording deserves more attention than the forecast:

• the precise event definition and what qualifies;
• the deadline and its time zone;
• the named source of truth;
• whether preliminary or revised data controls;
• tie, cancellation and postponement rules;
• what happens if the source is unavailable or changes methodology;
• conditions that make the market invalid;
• the dispute process and who has final say.

Consider a market asking whether two countries will sign a peace agreement by 31 December. Does an initialled memorandum count? Must both legislatures ratify it? Which time zone applies? Does a temporary ceasefire qualify? What if an announced deal is withdrawn a week later? None of those questions is about geopolitics, and all of them determine who gets paid. A real rulebook shows how specific this gets.

This gap between the real-world event and the qualifying event produces recurring disputes. In one market discussion, a trader warned that participants appeared to be buying Yes because fighting had stopped, even though the written condition required something more specific: the crowd was trading the news, "."

Key line: you are trading the contract wording, not the headline.

Oracles and Disputes

On a blockchain venue, settlement is automated once an outcome is determined. Determining the outcome ultimately depends on an external source or decision process. In straightforward markets that may be nearly mechanical; ambiguous or disputed markets can require human interpretation, and treating oracle settlement as automatic verification is the most common error in crypto-native explainers. When an outcome is challenged, some venues hand the final call to token holders.

Two documented cases show what that means in practice.

In March 2025, a roughly $7 million Polymarket contract asking whether Ukraine would agree to a Trump-backed minerals deal before April resolved Yes even though no qualifying agreement had occurred. After the outcome entered UMA's dispute process, token-weighted voting produced the final Yes resolution despite the apparent absence of the qualifying event. Polymarket described the episode as an unprecedented alleged governance attack, and holders of the losing side received no compensation.

In June and July 2025, a market on whether Ukraine's president would wear a suit before July resolved to No after UMA token holders voted that way, despite the BBC, the New York Post and other outlets describing his NATO summit outfit as a suit. WIRED reported roughly $210 million at stake. A menswear expert told WIRED that the outfit met the technical definition while most people would not call it a suit, so the question was genuinely ambiguous. Reporting at the time noted that four large holders controlled a substantial share of the UMA supply and that roughly 23 million tokens were staked in the vote. UMA's co-founder said there was no evidence of manipulation, and none has been proven.

Two structural points survive the specifics. UMA's voting incentives reward participation aligned with the eventual consensus outcome and can penalize votes that do not align with it; critics argue that such incentives may encourage voters to anticipate the majority rather than independently assess ambiguous evidence. And critics have also noted that nothing in the mechanism inherently prevents a voter from holding a position in the market being decided, which is a question about arbitrator independence rather than about any individual vote.

Comparing Resolution Models

Invalid and Ambiguous Markets

Some markets cannot resolve cleanly: the event is postponed indefinitely, the data source disappears, the wording admits two readings, or the outcome is partially satisfied. Treatment varies substantially by venue. Depending on the rules, an invalid or unresolvable market may be voided with collateral returned, settled at a predefined neutral value such as 50 cents, or handled under a venue-specific fallback provision. Whichever rule applies is set in advance, and it defines your downside in exactly the scenarios you did not anticipate — so read it before, not after.

Fixing ambiguous drafting creates its own problem, because correcting the wording after money has entered changes the trade people made. Reacting to a market whose title and detailed rules appeared to carry conflicting dates, one user described the perceived effect as "."

Whether the venue was contractually entitled to clarify the terms is a separate question from how the change felt to positioned traders. Platforms face a genuine dilemma between honouring flawed wording and altering the terms under which positions were opened, and readers should save a copy of the rules as they read at entry.

Are Prediction Markets Accurate?

These resolution challenges also affect how much confidence traders can place in prediction market prices as a measure of real-world probabilities.

Why Financial Incentives Can Help

The mechanism is straightforward. Anyone who believes a price is wrong can profit by trading against it, so mispricings attract capital. Traders have an incentive to research rather than to posture, prices update continuously as information arrives, and the resulting number aggregates many private views into one figure. Nobody has to be individually right for the aggregate to be good.

What the Evidence Shows

The research is more encouraging than critics allow and more conditional than platforms suggest. The single most useful finding is that accuracy varies sharply by category and by how far a contract sits from resolution.

Federal Reserve researchers examining Kalshi found its federal-funds-rate forecasts broadly competitive with professional forecasts at longer horizons and exceptionally accurate close to FOMC decisions, achieving a perfect forecast record one day before meetings within the sample studied. Performance varied across other macroeconomic categories, including CPI and unemployment. A separate study of 287 Kalshi unemployment contracts reported a Brier score of 0.1106, substantially below the 0.25 score produced by an always-50% forecast, with favourites well calibrated.

Recent Kalshi research also finds a favorite-longshot bias: low-priced contracts tend to win less often than their prices imply, while higher-priced contracts perform better. The size, and even the direction, of calibration error can vary by category, platform and time to expiry, which is why calibration is worth evaluating by domain, venue and horizon rather than assumed from price alone. Longer horizons tend to worsen the pattern, because traders dislike locking capital into expensive contracts for months.

On the comparison with polls and expert panels, forecasting tournaments have produced mixed results. Markets sometimes outperform individual or team forecasts, while well-designed probability aggregations can outperform market prices in other settings, and polls tend to do relatively better when resolution is far in the future.

When Prediction Markets Work Best

A useful filter: markets with very low volume are closer to one person's opinion than to a crowd forecast, and their prices should carry correspondingly little weight. More participation helps only when it adds information or genuine liquidity, since uninformed trading can also push prices away from fundamentals.

Prediction Markets vs Polls

A poll measures what a sample says. A prediction market measures the price at which traders will commit capital. Markets frequently use polls as an input, which makes them a complement to polling rather than a replacement, and it means a market can inherit a polling error rather than correct it.

Prediction Markets vs Betting and Sportsbooks

The mechanical differences are real. On exchange-style prediction markets, users generally trade against other participants or market makers rather than accepting a bookmaker's fixed odds; positions can be resold before the event; prices move continuously with order flow; and exchange-style venues typically earn fees rather than profiting from a built-in margin. Not every prediction market works this way, some use automated market makers, subsidized liquidity or designated market makers that do take the other side.

One Reddit user put the structural point less delicately: on a prediction market "," instead of against the house. Crude, and closer to the mechanics than most marketing copy, though it oversimplifies for the reasons just given.

The counterargument is equally real. Money is risked on an uncertain outcome, sports contracts in particular are functionally close to wagers, and state gaming regulators have argued that an exchange wrapper does not change the substance. This distinction is contested rather than settled, and courts reached opposing preliminary conclusions on it during 2026. A page that tells you prediction markets are definitively not gambling, or definitively are, is overstating what has been decided.

Prediction Markets vs Futures and Binary Options

Binary event contracts on a regulated exchange share arithmetic with retail binary options and differ in venue, oversight and counterparty structure. Using the phrase "binary options" for exchange-traded event contracts imports the wrong associations.

What Are Prediction Markets Used For?

Hedging is the use case that gives the category its clearest economic justification, because it transfers genuine risk rather than creating new exposure.

Centralized vs Crypto-Native Prediction Markets

These are two common architectures rather than mutually exclusive legal categories. A venue can be regulated and use blockchain infrastructure, centralized but crypto-funded, or decentralized at settlement while permissioned at the interface. The live landscape mixes casino-style operators with crypto-native exchanges.

Fewer restrictions do not automatically produce better forecasts. Open access can widen the pool of informed traders, and it can equally widen the pool of participants trading for entertainment, farming incentives or moving a headline number. Contract-level access and front-end access are also different things: an interface can block your jurisdiction while the underlying contracts remain reachable, which is a compliance question rather than a technical one.

Risks of Prediction Markets

Four of these deserve elaboration.

• Manipulation is broader than one large trade: It includes wash trading and fake volume, spoofing, coordinated buying, spreading false information, insider trading, bribing participants in the underlying event, oracle bribery and governance capture. Experimental and theoretical research suggests that purely financial attempts to push prices away from fundamentals can attract corrective trading in sufficiently liquid markets, which makes them expensive and often unprofitable. That protection weakens sharply in thin markets, and it does not address manipulation of the underlying event or of the resolution mechanism.
• Moving the price and controlling the payout are different powers: A large trader can push a displayed probability, particularly where books are thin. Buying contracts does not determine who gets paid, because settlement follows resolution. Where the same actor can influence both the price and a token-weighted resolution vote, those two powers converge, which is precisely what the disputed cases above illustrate.
• Prices can become part of the story: Media outlets, donors, campaigns and companies now cite market-implied probabilities. When a price influences coverage, and coverage influences behaviour, the market is partly shaping the event it is forecasting. That reflexivity also creates an incentive to move a price for reasons unrelated to expected profit on the position.
• Behavioural risk is not a footnote: These venues combine continuous price movement, short-duration outcomes and emotionally engaging topics, features that can encourage frequent trading and loss chasing in vulnerable users. The failure mode looks the same as it does anywhere else. In one first-person forum post, a user described a sports position as their "" after liquidating a conventional brokerage account to fund it: a textbook description of loss chasing. A commenter replied with the sentence that should be printed on every interface: . If that pattern sounds familiar, support is worth seeking before the next trade rather than after it.

Are Prediction Markets Legal?

Legality depends on jurisdiction, platform and the category of contract, and the position is unsettled. In the United States, the central question through 2026 has been whether event contracts are swaps under the Commodity Exchange Act, which would place them under exclusive CFTC jurisdiction, or wagers subject to state gambling law.

As of mid-2026, courts in different jurisdictions were applying sharply different preliminary views to that same federal-versus-state question. The federal regulator, meanwhile, has been actively reshaping its own event-contract framework.

Note that almost every decision below is preliminary. Courts assessing requests for injunctions ask whether a party is likely to succeed, which is not the same as finally deciding what the law is.

A snapshot as of early August 2026:

• In April 2026, the Third Circuit affirmed a preliminary injunction protecting Kalshi from New Jersey enforcement, concluding that Kalshi had shown a reasonable likelihood of success on its argument that federal commodities law preempts state regulation of its sports contracts. The decision was preliminary rather than a final merits ruling, and it binds only its circuit.
• The Ninth Circuit heard Nevada's appeal in April 2026, with a ruling pending that could create a circuit split and push the question toward the Supreme Court.
• Massachusetts obtained a January 2026 preliminary injunction barring Kalshi's sports contracts unless it complied with state sports-gaming law. A federal court in Tennessee went the other way in February 2026, granting Kalshi preliminary relief after finding it likely to succeed on its federal-law theory.
• The CFTC sued Arizona, Connecticut and Illinois in April 2026 seeking to block state enforcement, and a federal judge blocked Arizona from pursuing criminal charges against Kalshi.
• On 29 July 2026, a federal judge in Wisconsin denied the CFTC's request for preliminary relief against state enforcement, concluding that the agency had not shown a sufficient likelihood of success on its preemption theory. An appeal to the Seventh Circuit is expected.
• Litigation continued into late July and August 2026, including a Utah ruling permitting state anti-gambling enforcement to proceed and New York proceedings announced at the end of July.
• Three California tribes have argued that sports event contracts conflict with the Indian Gaming Regulatory Act.

Practical consequences for readers: Kalshi operates as a CFTC-designated contract market. Polymarket's U.S. operation runs through a separate registered venue and is distinct from its global crypto-native platform, which is not offered to U.S. users; do not treat the two as one regulated product. Both major U.S.-accessible platforms restrict access in certain states while litigation proceeds. Non-sports categories have generally faced less resistance than sports. Technical ability to reach a contract does not establish lawful access.

On tax: treatment can depend on the product, the venue and the taxpayer's own activity, and prediction-market taxation remains an area requiring product-specific professional advice. Rules outside the United States differ again, and some jurisdictions treat these products under gambling rather than investment rules. This section describes a moving position and should be checked against current sources before acting on it.

How to Evaluate a Prediction Market

Closing Thoughts

Prediction markets turn uncertainty into a price, which is a genuinely useful thing to do. A number that updates continuously, that costs money to be wrong about, and that anyone can check provides a continuously updating signal that can add information beyond punditry or polling, particularly in liquid, well-specified markets. The calibration research on liquid, well-specified contracts supports taking these prices seriously.

Taking them seriously means reading what sits behind them. The displayed figure may be a bid, an ask or a stale last trade. The depth behind it may be a few hundred dollars. The volume figure may be substantially recycled. The participants may all be reading the same three sources. The question may hinge on a definition nobody agreed on in advance, and the final call may rest with token holders whose independence is not guaranteed by the mechanism.