A brass balance scale weighing green and red tokens in front of a blurred chart

How Do Prediction Markets Work? Order Books, Prices and Settlement

Prediction markets work by matching people who disagree about an event. Traders buy “Yes” or “No” contracts that pay $1 if they’re right, either from each other through an order book or from a pricing formula called an automated market maker. The going price reads as an implied probability, and when the event ends, a named source decides the answer and the exchange pays the winners.

If you’re new to the idea itself, start with what prediction markets are. This page is the plumbing: how trades get matched, how a price turns into a probability, why the price you see isn’t always the price you get, and who decides the result.

Two ways to match trades: order books and automated market makers

Every prediction market needs a way to put a buyer and a seller together. There are two main designs: a central limit order book, where traders quote prices to each other, and an automated market maker, where a formula always quotes a price.

Order books

An order book is a live list of offers. Bids are prices people will pay; asks are prices people will accept to sell. When a bid meets an ask, a trade happens. This is how stock exchanges work, and it’s what Polymarket uses: its docs say prices come from buyers and sellers on a central limit order book, not from Polymarket itself.

On an order book you choose between two kinds of orders:

  • Limit order: you name your price and wait. It fills only at that price or better, and it can fill in pieces.
  • Market order: you take whatever the best available price is right now. Polymarket notes that a market order is really a limit order priced to match resting orders immediately.

The people posting resting orders are called makers, because they make liquidity. The people who trade against those orders are takers. That split matters for fees, as you’ll see below.

Automated market makers

An automated market maker (AMM) replaces the other trader with a formula. It always has a price for you, and every purchase nudges that price up. The best-known design in prediction markets is the market scoring rule developed by economist Robin Hanson, described in his 2002 paper on logarithmic market scoring rules, which frames it as a market maker that always has public offers to buy and sell and updates them after each trade.

The trade-off is cost. Stanford’s Tim Roughgarden explains that an automated market maker can lose money, unlike a plain order book, but its loss is capped at a fixed amount set by the design. Whoever runs it is paying for liquidity, so that traders in a quiet market always find a counterparty.

Order book Automated market maker
Who quotes the price Other traders A formula
Always a price available? Only if someone has posted an order Yes
Who takes the other side Another trader The market maker’s pool
Main weakness Thin markets can have wide gaps or no offers The operator funds possible losses
Example Polymarket’s central limit order book Hanson’s logarithmic market scoring rule

How a price becomes an implied probability

Because a winning contract pays exactly $1, a price in dollars converts straight to a probability: 25 cents is 25%, 75 cents is 75%. Polymarket’s docs use those same numbers.

The math behind it is a “complete set.” On Polymarket, a Yes buyer at 60 cents and a No buyer at 40 cents can be matched because their prices sum to $1. That match creates one Yes and one No share from $1 of collateral, and the pair can be merged back into $1 later. One of the two will be worth $1 at the end, so a Yes plus a No always pays out exactly $1.

That’s why the two sides of a market move together. If Yes climbs to 65 cents, No has to slide toward 35 cents. Roughgarden’s notes make the same point: if the two prices summed to much less than $1, anyone could buy both and lock in a profit, and traders would close that gap fast.

Read the number as an estimate, not a fact. The CFTC’s explainer says the price “reflects the market’s expectation,” and fees, spreads and the mix of people trading all bend it a little.

Liquidity and spreads: the price you see vs. the price you get

The quoted price is usually a midpoint, and you’ll pay a little more to buy or get a little less to sell. The gap between the best bid and the best ask is the spread, and a narrow spread is the sign of a liquid market.

Polymarket shows the midpoint of the best bid and ask, unless the spread is wider than 10 cents, in which case it shows the last traded price. Its own example: with a bid of 34 cents and an ask of 40 cents, the market displays 37%, but you’d pay 40 cents to buy and get 34 cents to sell.

Depth matters too. Polymarket says it has no size limits, but large orders can move the price a lot because they eat through the resting offers one level at a time. Exchanges pay to keep books full. Polymarket runs a liquidity rewards program that pays makers daily for resting orders close to the midpoint, and it hands part of its taker fees back to makers as rebates.

Fees are the other cost between you and the quoted price. On Polymarket’s international site, a taker fee applies on most categories and follows a formula that peaks at a 50-cent price and shrinks toward 0 and $1, per its fee page. We break those numbers down on how Polymarket makes money, and Kalshi’s on how Kalshi makes money.

Settlement: who decides the answer

Every market names, in its rules, where the final answer comes from. When the event ends, that source decides which side wins, winning contracts pay $1 and losing ones pay $0. The two Polymarket products show the two main approaches.

An oracle and a dispute window (Polymarket international). Its resolution docs, checked November 8, 2026, describe the UMA Optimistic Oracle. Anyone can propose an outcome by posting a bond, typically $750. Anyone else can dispute it within a 2-hour challenge period. An undisputed market resolves about 2 hours after the proposal. If it’s disputed twice, UMA token holders vote, and the whole process takes about 4 to 6 days. In rare cases a market resolves 50/50 and each share pays 50 cents.

The exchange decides (Polymarket US). The US exchange’s resolution page says the exchange determines the result from the sources listed in each market’s rules, such as a league’s official result for sports, government data like Bureau of Labor Statistics figures, or a named price index for crypto. Its rulebook treats all resolutions as final. Kalshi also settles under its own exchange rules; see how Kalshi works.

Timing follows the event, not the clock. Sports markets often settle soon after the final whistle, while election or news markets can wait on certification or official publication. Before you trade any market, read its rules and check which source decides it.

Frequently asked questions

Who sets the price in a prediction market?

On an order-book exchange, traders do. Buyers post bids, sellers post asks, and a trade happens when they meet. On a market run by an automated market maker, a formula quotes the price and moves it as people buy and sell. Polymarket's docs say it does not set prices; they come from orders on its book.

Why don't Yes and No prices always add up to exactly $1?

Because there is a gap between the best price someone will pay and the best price someone will accept, called the spread. The midpoint of Yes and No should sit near $1, but the price you can actually buy at is usually a little higher and the price you can sell at a little lower. Fees widen the gap further.

How long does it take a prediction market to settle?

It depends on the platform and the event. On Polymarket's international site, an undisputed result typically resolves about 2 hours after someone proposes it, and a disputed one takes about 4 to 6 days, per its docs. Polymarket US says markets resolve once the outcome is publicly confirmed, which can take longer for election or news results.

What happens if a prediction market question is unclear?

Exchanges publish rules for each market, including the source that decides the answer, and some issue clarifications after trading starts. Polymarket's docs say clarifications cannot change the basic intent of a question. In rare cases its markets resolve 50/50, with each share paying $0.50.

Sources

  1. Polymarket Docs: Prices and order book (checked November 8, 2026)
  2. Polymarket Docs: Resolution (checked November 8, 2026)
  3. Polymarket Docs: Liquidity rewards (checked November 8, 2026)
  4. Polymarket Docs: Trading fees (checked November 8, 2026)
  5. Polymarket US Docs: Market resolution (checked November 8, 2026)
  6. CFTC: Understanding Prediction Markets and Event Contracts (checked November 8, 2026)
  7. Robin Hanson, Logarithmic Market Scoring Rules for Modular Combinatorial Information Aggregation (2002 draft) (checked November 8, 2026)
  8. Tim Roughgarden, CS269I Lecture 18: Prediction Markets (Stanford, 2016) (checked November 8, 2026)

Jordan Whitfield

Editor, Trading the Odds

Jordan Whitfield edits Trading the Odds, which covers stock market statistics (seasonality, volatility and how markets behave around recurring events) and how prediction markets work and where they are legal. Market statistics on the site are calculated from public index data and say when the data runs through; every legal claim carries the date it was checked and a link to its source. How we check facts.