How Do Prediction Markets Work?

A prediction market is a marketplace where people buy and sell contracts tied to the outcome of a future event, such as an election, an interest rate decision, or an award show. Each contract pays a fixed amount (usually $1) if the event happens and nothing if it does not, so the price of a contract, between 0 and 100 cents, reads as the market's estimate of the probability that the event occurs. When traders think an outcome is 60 percent likely, the contract tends to trade near 60 cents.

This piece is about the machinery: how a market gets built, how a price becomes a probability, who sits on the other side of your trade, how trades actually match, and how the whole thing settles. For a primer on what a prediction market is and why the category exists, see What Is a Prediction Market?.

Key takeaways

• Everything starts with the wording. A market is only as reliable as the exact question it asks and the source of truth it names.
• Because a contract pays $1 if correct and $0 if wrong, its price reads directly as a probability, and Yes plus No adds up to roughly $1.
• Your order is matched either against another trader in an order book or against a pool of pre-committed capital in an automated market maker (AMM).
• You can usually exit before the event resolves by selling your position at the current price.
• Resolution, not trading, is where the sharpest problems live: someone or something has to decide what actually happened, and ambiguous outcomes can be disputed.

Step 1: how a market gets created and worded

A prediction market begins as a single question with a strictly binary answer: yes or no. "Will the Federal Reserve cut rates at its next meeting?" is a clean example. The question has to resolve to exactly one side by a stated deadline, with no room for "sort of."

The wording is the whole ballgame. Two things get pinned down when a market is created, and both decide whether the market pays out cleanly or descends into a dispute.

• The resolution criteria. These spell out precisely what counts as "yes." A market on "Will Candidate X win the election?" has to define whether it settles on a called race, a certified result, or an inauguration, and what happens if the outcome is contested. Vague criteria are how honest disagreements turn into fights over money.
• The source of truth. The market names, in advance, where the answer will come from: a named government agency, a specific data feed, a court ruling, or a designated official announcement. If the source is ambiguous or can be interpreted more than one way, the resolution step inherits that ambiguity.

A well-worded market reads like a contract because it is one. A poorly worded market can trade actively for weeks and then collapse into argument the moment reality turns out to be messier than the question assumed.

Step 2: how shares are priced, and why a price is a probability

Each market offers two contracts, a Yes and a No. Each pays $1 if its side is correct and nothing if it is wrong. Prices run from 0 to 100 cents.

The reason a price behaves like a probability comes straight from the payout. If you believe an event is 70 percent likely, a Yes contract is worth about 70 cents to you: 70 percent of the time you collect $1, the rest of the time you collect nothing, which averages out to 70 cents. Pay much more than that and you expect to lose money over many repeats. Pay much less and someone is handing you an edge. Buyers and sellers converging on that expected value is what pins the price near the crowd's probability estimate.

Because Yes and No cover every possible outcome and are mutually exclusive, their prices add up to roughly $1. If Yes trades at 65 cents, No trades near 35 cents. When that sum drifts away from a dollar, a trader can buy the cheap side of both and lock in a small guaranteed profit, and that activity pulls the two prices back into line.

Step 3: what happens when you place an order

When you buy a Yes contract, someone has to sell you one, and understanding who that counterparty is explains most of what follows.

You can enter an order two ways. A market order says "fill me now at the best available price," and it executes immediately against whatever is currently on offer. A limit order says "fill me only at this price or better," and it waits until someone is willing to trade at your number.

The counterparty is almost never the platform betting against you. It is another participant who holds the opposite view or who is providing liquidity for a fee. On a two-sided market, your purchase of Yes is somebody else's sale of Yes (or, equivalently, their purchase of No). The platform's job is to find that other side and execute the match, not to take the risk itself.

Step 4: how the order book or AMM matches your trade

There are two dominant ways a trade actually gets matched, and this is the main mechanical difference between platforms.

An order book works like a stock exchange. It is a live list of buy orders (bids) and sell orders (asks) at various prices. Your buy order is matched against the lowest ask that meets your terms. If you place a market order for Yes, it walks up the book, filling against the cheapest sellers first until your order is complete. If you place a limit order that no one currently wants to hit, it sits in the book as a resting bid until a seller crosses it. This is the model used by regulated US exchanges such as Kalshi, which run a central order book under the oversight of the Commodity Futures Trading Commission (CFTC), the federal agency that supervises derivatives markets.

An automated market maker (AMM) replaces the queue of human orders with a pooled reserve of capital and a formula. Instead of matching you against another trader, the AMM quotes a price from a mathematical curve based on how many Yes and No shares are currently in the pool. Buying Yes removes Yes shares from the pool, and the formula automatically raises the price of the next Yes share, which is how heavy buying pushes a probability up without any explicit counterparty on the other side of the desk. Some blockchain-based platforms have used this model so that a trade can execute even when no human is sitting on the opposite side at that instant.

Blockchain-based platforms (ones that settle on a public shared ledger rather than through a company's internal books) record each match on that ledger, which is why an outside analyst can reconstruct the trade flow rather than relying on a summary. For a platform-specific walkthrough of matching and settlement, see How Does Polymarket Work?.

Step 5: what happens between your trade and resolution

Once you hold a position, you are not locked in until the event decides itself. The price keeps moving as news arrives and as other traders buy and sell, and your position is worth whatever the current price is.

This is the part that makes a prediction market feel more like a live index than a one-off wager. Suppose you bought Yes at 40 cents and fresh news pushes the price to 65 cents. You can sell your Yes contracts at 65 cents and bank the 25-cent gain immediately, without waiting to find out whether the event actually happens. The mechanics of selling are the mirror image of buying: your sell order matches against a resting bid in the order book, or against the AMM pool, at the going price.

The same works in reverse. If the price falls and you want to cap the damage, you sell at a loss rather than ride the position to zero. Between placing a trade and final resolution, your money is not frozen; it is tied up in a position you can usually unwind at the prevailing market price.

Step 6: how resolution and settlement work

When the deadline arrives, the market has to be resolved: someone or something declares the outcome, winning contracts pay $1 each, and losing contracts pay nothing. This is the single most contested mechanic in the entire system, and it works differently depending on the platform.

On a regulated exchange, the exchange itself determines the result under its published rules, using the source of truth named when the market was created. If a contract was written to settle on an official government figure, the exchange reads that figure and settles accordingly. Disputes are handled through the exchange's own process and, ultimately, its regulator.

On a blockchain platform, resolution usually runs through a decentralized process rather than a single company. A designated mechanism proposes the outcome, and other participants can challenge it within a window by putting up their own capital, with the dispute escalated to a broader vote if it is contested. The intent is that no single party can unilaterally decide the answer.

That design has a known weak point. Because a disputed outcome is settled by whoever commits the most capital to their side of the argument, a well-funded actor can, in principle, push a resolution toward a result that the plain reading of events would not support, especially when the wording left room for interpretation. Bloomberg cited Allium's data on possible settlement manipulation on Polymarket, a concrete illustration of why the resolution layer, not the trading, is where the sharpest questions sit. When the criteria are airtight, resolution is boring and mechanical. When they are loose, this is where the money gets fought over.

Where the money sits the whole time

Follow the cash and the two models diverge again.

On a regulated US exchange, you fund an account with US dollars held under the exchange's regulated custody, subject to identity checks (KYC). Your funds sit in that account, your open positions are tracked against it, and payouts land back in it after settlement.

On a blockchain platform, trades are denominated in stablecoins, which are crypto tokens designed to hold a steady value of about one dollar, held in a wallet you control. When you buy a contract, the stablecoins backing both sides of the trade are locked in a smart contract (self-executing code on the ledger) for the life of the market. That locked collateral is what guarantees the winner can be paid: the $1 that settles each winning contract was already posted by the two sides when the position was opened. At resolution, the smart contract releases the collateral to the winning side. Because every step of this sits on a public ledger, an outside analyst can measure how much money is actually locked in a market rather than trusting an operator's summary. Independent onchain analysis of this kind is how US-linked activity persisting despite an access ban came to light.

A note on reading the numbers

Raw ledger data is available to anyone, but it is messy and has to be cleaned, labeled, and standardized before it reads as "volume" or "open interest." Allium is a data infrastructure company that ingests raw data from many blockchains and standardizes it into usable datasets, including prediction-market activity used by researchers and newsrooms. It is a read layer on this activity, not a venue, exchange, broker, or market maker, and it does not offer investment advice. For how these measurements are actually built, Allium has published practical breakdowns of Polymarket volume data and Kalshi historical data.

Frequently asked questions

What decides who is on the other side of my trade?

On an order-book platform, the other side is another trader whose resting order matches yours, or a liquidity provider quoting prices for a fee. On an automated market maker (AMM), there is no single human counterparty; your trade executes against a pooled reserve of capital, and a formula sets the price based on how many shares are in the pool. In neither case is the platform itself betting against you.

How does an order book differ from an AMM in matching a prediction-market trade?

An order book matches your order against the best available opposing order, filling a market order against the cheapest sellers or letting a limit order rest until someone crosses it. An AMM quotes a price from a mathematical curve tied to the shares in a pool, so buying a side removes shares and automatically raises the next price. Order books need active traders on both sides; AMMs can fill a trade even when no human is on the opposite side at that moment.

Why does a contract's price equal a probability?

Each contract pays exactly $1 if correct and $0 if wrong. A rational buyer pays up to the expected value of that payout, so a 70 percent chance is worth about 70 cents. Buyers and sellers converging on that expected value push the price toward the crowd's probability estimate, and because Yes and No cover all outcomes, their prices add up to roughly $1.

Can I sell a prediction-market position before the event resolves?

Usually yes. Your position is worth the current market price at any time, so you can sell to lock in a gain or cap a loss without waiting for the outcome. The sell matches against a resting bid in the order book or against the AMM pool at the prevailing price, the mirror image of buying.

How is a prediction-market outcome actually decided?

Every market names a source of truth in advance. On a regulated exchange, the exchange reads that source and settles under its published rules. On a blockchain platform, a decentralized process proposes the outcome and lets participants challenge it by committing capital, escalating to a broader vote if disputed. This settlement step is the most contested part, because ambiguous wording can be fought over and, in principle, swayed by whoever commits the most capital.

Where does the money sit between placing a trade and settlement?

On a regulated exchange, your funds sit in a US-dollar account under the exchange's regulated custody. On a blockchain platform, the stablecoins backing both sides of a trade are locked in a smart contract for the life of the market, which is what guarantees the winner can be paid. At resolution, the funds are released to the winning side. On a public ledger, an outside analyst can measure exactly how much is locked.