What a prediction market is, and how one works

Every other page on this site answers "which tool". This one answers the question underneath that, because a reader who does not already know what is being bought cannot use any of the answers: they will pick a venue, deposit, and discover that the thing they thought they had bought is not what settled.

The generic explanation — markets aggregate information, the price is the crowd's probability — is true, available in a hundred places, and useless at the point of deciding. It describes a category as if every member of it worked the same way. They do not. The same question, on two venues, is a different instrument with a different counterparty, a different party deciding the outcome, a different fee shape and a different answer to whether you may trade it at all. That divergence is the whole subject of this page.

How it works

What you are actually buying

A share of a fixed payout, priced in cents.

The mechanism is written most plainly in the exchanges' own rulebooks. The Polymarket US rulebook dated 14 September 2026 defines a Contract Outcome this way: if the Expiration Value satisfies the Payout Condition, the Contract Outcome is $1.00 and the settlement amount goes to holders of long positions; if it does not, the Contract Outcome is $0.00 and the settlement amount goes to the holders of short positions. The ForecastEx rulebook dated 17 September 2026 says the same thing in its Rule 603 — a contract with a Settlement Value of $1.00 is extinguished and the collateral account is credited $1.00.

Three consequences follow immediately, and they are the ones worth holding on to:

1. The most you can make and the most you can lose are both known before you trade. A contract bought at 62 cents risks 62 cents and can return 100. Nothing about that changes if the event turns out to be far more dramatic than expected.
2. Yes and No are one dollar between them. ForecastEx makes this structural rather than incidental: under Rule 401(c) participants bid prices between $0.01 and $0.99, and a Yes bid and a No bid are paired when their combined price reaches $1.00. Rule 604 closes the loop — when opposite contracts are offset, the account is credited $1.00 for each pair.
3. You can leave before the event is decided. This is the property that makes the thing a market rather than a ticket. On a venue with a live two-sided market you sell your position to somebody else at the going price, take the difference, and never find out how the question ended.

That last one is also the first place a reader's mental model breaks, so it is worth saying in the negative: on a thin market, or a pool, the exit may not exist at any price you would accept.

Why the price reads as a probability

Because the payout is fixed at a dollar, the price is a number between zero and one that the market is willing to pay for that dollar. The CFTC has described the reading in its own documents twice, twenty years apart. Its 2008 concept release says binary event contracts typically pay out a fixed amount when an outcome occurs or does not, and that trading them "can facilitate the discovery of information by assigning probabilities, through market-derived prices, to discrete eventualities". Its March 2026 advance notice opens by saying prediction markets "function as information aggregation vehicles" because contract prices reflect participants' aggregate beliefs, and cites the 2008 release for it.

Note what those sentences are careful about. They describe what the price can do, under conditions neither document enumerates. The section below on where the reading breaks is that missing enumeration, and it is the part a reader arriving at a venue for the first time actually needs.

Who is on the other side — four different answers

The single most useful thing to know about a venue, and the one least visible from its screen, because all four mechanisms render as a percentage.

• An order book. Another user's resting order is your counterparty; the venue matches and charges a fee. If nothing is resting, there is no price, only a wide gap.
• An automated market maker. A curve funded by liquidity providers quotes you. There is always a price, and it always moves against you as your order grows.
• A pool. Nobody quotes anything. Everyone pays in, the pool is divided among the winners after the close, and the percentage shown while it was open was an estimate that kept moving.
• The exchange itself. ForecastEx runs a book with no seller. Rule 401(c) has the exchange execute the contract once a Yes bid and a No bid add to a dollar, and ForecastEx then carries the obligation to pay at settlement. Rule 608 requires every position to be fully collateralised at the clearing house, and Rule 401(i) adds that contracts are non-marginable and have no loan value — so the money behind both sides is already there.

Which one you are facing determines what your order does to the price, what a large order costs, and whether an exit exists at all. The full version, with the numbers to check on each, is where the liquidity comes from; the prediction market venues listing names the mechanism on every card.

Who decides the outcome — three more different answers

Resolution is not a formality at the end. It is the product, and the venues answer it in ways that have almost nothing in common.

• The exchange, under a filed rulebook. On the CFTC-regulated venues the exchange reads the source named in the contract terms, states the outcome, and its statement is the outcome. There is a named accountable entity and a published procedure. There is also no appeal.
• An optimistic oracle with a bond behind it. UMA's own documentation describes the design: a proposer posts an answer with a refundable bond, anyone may dispute it during a challenge period by posting a bond of their own, and a dispute escalates to the Data Verification Mechanism, where token holders vote. UMA puts the share resolved without a dispute at 99.8%. This is the only mechanism in this catalogue a third party can actually enter — and the only one where the deciding party has no support inbox and no obligation to you.
• A price feed. No committee has an opinion, which also means no committee notices when the feed prints something absurd for one second.

A fourth answer exists off to one side. On a play-money platform such as Manifold the person who wrote the question resolves it, which is a real mechanism rather than the absence of one, and it works exactly as well as the author's continued interest.

The clause-by-clause version, including what each venue does when a question turns out to have no clean answer, is who decides the outcome. Read it before you size a position, not after.

Who decides which questions exist

Not the trader, and on a US exchange not only the exchange either.

A designated contract market self-certifies a new contract with the Commission before listing it, which is why the contract terms and the named settlement source are public documents before trading opens. Against that sits CEA section 5c(c)(5)(C), the provision the Commission's June 2026 proposed rule is built around: the Commission may determine that an event contract is contrary to the public interest where it involves "(I) activity that is unlawful under any Federal or State law; (II) terrorism; (III) assassination; (IV) war; (V) gaming; or (VI) other similar activity determined by the Commission, by rule or regulation, to be contrary to the public interest." That proposal — published 12 June 2026 at 91 FR 35806, comment period closed 27 July 2026 — would define "gaming" and set out when a contract "involves" one of those activities. Nothing in it binds anyone until a final rule is published under the same RIN, 3038-AF65.

The constraint is real enough that exchanges write it into their own rules: ForecastEx Rule 401(b) says outright that it will not list markets deemed prohibited by the CFTC under that section.

Away from the regulated venues the answer is different again — whoever can deploy the contract, or any user the platform allows to create a question. Neither arrangement is a detail. It decides which questions you will ever see, and it is why the same event is listed on one venue and absent from the next.

Whether you can trade it at all

This is a fact about your jurisdiction, not about the product, and merging the two is the most expensive mistake a new reader makes. The same venue is a real-money venue in one country and unreachable in the next, and the difference is a regulator's order, a clause a court read a particular way, or a settlement signed last quarter — never a release note. It also fails at four different moments: a location check at signup, a representation you make in a member agreement, a payment rail that will carry money in but not out, and a single state removed from an app.

That is a page of its own: why a venue is unavailable where you are. The narrower legal question underneath it — whether these contracts are swaps, which is what decides who regulates them — is event contract or swap.

Where the cents stop reading as a probability

Five conditions, each of which puts a wedge between the number on the screen and the odds a careful reader would quote.

1. Nothing is resting on the other side. A price with no depth behind it is the last thing somebody was willing to do, not a consensus. The measurements that tell you which you are looking at — the gap between best bid and best ask in cents, and the money resting within two cents of the midpoint — are in where the liquidity comes from.

2. The quote comes from a curve, not from a crowd. On an automated market maker the displayed price is a function of the pool's inventory. It will quote you when nobody else would, which is the point, and your own order moves it, which is the cost.

3. The fee is taken twice, and it is largest exactly where you are least certain. The Polymarket US fee schedule in force from 17 September 2026 computes the charge as a coefficient times the number of contracts times price times one minus price. At a taker coefficient of 0.0695 that is $1.74 on 100 contracts at 50 cents and seven cents on 100 contracts at one cent. A 50-cent contract therefore costs about 51.7 cents to enter, and you pay again to leave. The shape is deliberate and it is worth absorbing: the fee peaks where the market is most uncertain, which is precisely where the probability reading was supposed to be most informative.

4. The dollar is locked until resolution, and it earns nothing while it waits. ForecastEx Rule 608 requires the position to be fully collateralised and Rule 401(i) states that contracts are non-marginable and have no loan value. A contract at 90 cents resolving in a month and one at 90 cents resolving in two years are not the same claim about the world, because the second ties up your money for twenty-four times as long at the same maximum return.

5. The rule is narrower than the headline. The price answers the payout condition written in the contract terms — a named source, a named release, a stated time — and not the sentence you read at the top of the page. Where those two come apart, the price is right and your reading of it was wrong. This is common enough that it has its own page.

None of this means the number is useless. It means the number is a price first and a probability second, and the gap between them is made of things you can actually measure before you trade.

What a prediction market is not

Draw this line by mechanism rather than by tone, because tone is what people argue about and mechanism is what determines whether you can get your money back.

An instrument on this side of the line has a two-sided price, lets you take either side of the same question, and lets you exit before the event is settled. Those three together are what make the price informative and what make the position tradeable. Remove the third and you have something that only pays at the end; remove the second and the price is a quote rather than a market.

A forecasting platform is a different product, and it is a separate category here for that reason. You are scored rather than paid, the settlement is in points, and there is usually no position to close. The overlap is visual — a question, a percentage, a crowd — and almost nothing else agrees. What that score actually measures, and why a good score on one platform means very little on another, is what a forecasting score measures.

A product that sells you an outcome, or tells you which side to take, is outside this catalogue altogether. This site lists software and venues; it does not list advice, and it never issues any. The distinction is not squeamishness. A venue publishes a rulebook, names a settlement source and can be checked against both; a recommendation publishes a claim about the future and cannot be checked against anything until it is too late to matter.

What it costs

Everything here is denominated against a fixed one-dollar payout, which makes small-sounding numbers large. Convert every figure to the same measure — cost per dollar of payout — before comparing two venues at all.

• Polymarket US charges, per its fee schedule effective 17 September 2026, a coefficient times contracts times price times one minus price. Taker coefficient 0.0695: $1.74 per 100 contracts at a price of $0.50, falling to $0.07 per 100 at $0.01. Makers receive a rebate at a coefficient of −0.0125, which is $0.31 per 100 contracts at $0.50, applied at the point of trade.
• ForecastEx charges a flat $0.01 per contract, and its fee schedule states that the charge falls on both the Yes and the No side of each executed transaction, independently of resolution, netting or settlement. On a contract entered at 50 cents that is 2% of what you paid; on one entered at 5 cents it is 20% of what you paid, for the same one cent.
• The spread is a cost on both legs and appears on no fee schedule. On a market quoted 40 bid, 44 offered, buying and selling immediately costs four cents on a dollar of payout before any fee at all.
• The capital is fully committed for the life of the contract, with no leverage and no loan value against it, so the real cost of a long-dated position includes whatever that money would otherwise have earned.

Venues publish these in units that do not compare — a coefficient, a percentage of what the fill cost, a flat cent, a share of profit — and the conversion arithmetic for each, with the current figures, is in what a trade actually costs.

What you can do about it

Read the rest in this order. Each page assumes the one before it, and together they are the whole decision.

1. Where the liquidity comes from — who is on the other side, and how to tell a real price from a lonely one. This is the page that changes what you think the percentage means.
2. Who decides the outcome — the resolution mechanisms clause by clause, including what each venue does with a question that cannot be answered as written.
3. What a trade actually costs — four incompatible fee units converted to one number, so that 3.00% on one venue can be compared with 1.75% on another.
4. Why a venue is unavailable where you are — the four separate restrictions that all get called "not available in your country".
5. Event contract or swap — only if you want the legal frame under all of it. It is the most contested question in the sector and the least urgent one for a first deposit.

Then the prediction market venues listing, where the counterparty mechanism, the settlement currency and the resolution source are on every card.

Check exactly three things before you fund an account. In this order, because each can end the exercise on its own.

1. Can you open, fund and withdraw where you physically are? Not one question, three. A location check fails before you sign up and costs you nothing; a withdrawal rail fails after your money is inside. Read the venue's own terms and its geographic-restrictions page, not a summary, and check whether the list that binds is the one you read or a separate document it incorporates.
2. Who resolves, against which named source, on what timetable, and what happens if the question is ambiguous? All four are published before you can trade on every venue in this catalogue. The fourth is the one nobody advertises, and the answer is usually that you get something near what you paid rather than the dollar you thought you had won.
3. What a fill costs on this venue's mechanism, in cents per dollar of payout — the fee at the price you actually want, plus the spread you cross to get in, plus the spread you will cross to get out, plus how much is resting at that price. A venue advertising no commission still charges you the spread twice.

Start smaller than the position you have in mind, and hold it through one resolution. The mechanisms above are cheap to read about and expensive to learn from a support ticket. One contract carried from entry to settlement tells you what the fee actually was, how long the money was locked after trading stopped, and whether the resolution matched the headline — the four facts that no page, including this one, can establish for your particular market.

And know what this catalogue is and is not standing behind. Every card here currently reports Tested hands-on: no. The fields on them are read from each venue's own terms, rulebook, fee schedule and documentation on a stated date, and the pages above cite the documents rather than paraphrasing them. Where a sentence here rests on a primary source, the source is listed with its date so you can check whether it has since been replaced.