Prediction markets explained: how Polymarket and Kalshi work and where arbitrage comes from
In short
- A prediction market contract pays $1 if the outcome happens and nothing if it does not, so its price reads as the market's estimate of the event's probability.
- Each platform settles markets by its own rules. Kalshi decides itself based on a named source; Polymarket uses the UMA oracle, where a proposed outcome can be disputed.
- The same question trades at different prices on different platforms, and prediction arbitrage is built on that gap. But a pair only works if the settlement rules match.
Prediction markets are platforms where people trade contracts on the outcome of a real event: an election, a match, a central bank decision, the price of bitcoin by a certain date. A contract costs between $0 and $1 and pays $1 at settlement if the outcome happens. That is why its price reads as the event's probability as judged by traders. The largest platforms are Polymarket and Kalshi.
Below we look at where the price comes from, who decides which outcome happened, how Polymarket differs from Kalshi and why the same question trades at different prices on two platforms. If the topic is new to you, start with the lesson What are prediction markets: it covers the basics, while this article covers how the platforms work.
How a market works: contract, order book and price
Every market is a yes-or-no question. Two contracts trade on it: YES pays $1 if the event happens, NO pays $1 if it does not. One of them always pays out and the other expires worthless.
That gives a simple rule: a YES + NO pair always pays exactly $1 at settlement. On Polymarket this is built into the design: $1 of collateral can be split into one YES and one NO, and a YES and NO pair can be merged back into $1. That is why the prices of the two sides stay close to a combined $1.
Who sets the price
The platform does not set the price; traders do. Buyers and sellers place orders, and these orders make up the order book. A trade happens when a buyer's and a seller's prices meet. If you buy at the best price already in the book, you are a taker; if you place your own order and wait, you are a maker. On many platforms this affects the fee, as explained in the glossary entry Maker and taker.
Buying NO at 40¢ is effectively the same as selling YES at 60¢: either way you win if the event does not happen. So the YES and NO prices on the same question move as mirror images: when one side gets more expensive, the other gets cheaper.
Price as probability, and where it misleads
Say YES trades at 62¢. That means traders on average put the event's chances at about 62%. It is a useful guide, not an exact probability. The price is affected by:
- the gap between buying and selling: the best price to buy YES might be 63¢ and to sell 61¢, and the "probability" lies somewhere in between;
- a thin order book: in an illiquid market a single large order moves the price by several cents;
- a long term: money in a contract is locked until settlement, so a market that settles in a year prices that in too;
- fees: they make small price corrections unprofitable, so a price can sit "in the wrong place" until the gap outweighs the costs.
Settlement: who decides which outcome happened
Settlement is the moment a platform determines the outcome and pays $1 on the winning contracts. Every market has rules: the exact wording of the question, the data source, the date and provisions for edge cases. The rules, not the headline, decide who gets paid.
The platforms settle markets in different ways.
- Kalshi determines the outcome itself, based on the contract rules and the source named in them: official statistics, government data, league results. The platform explains the process in its help article on market outcomes.
- Polymarket uses the UMA oracle. Anyone can propose an outcome by posting a bond, and someone else can dispute it with a bond of their own. If the dispute stands, UMA token holders decide by vote. The process is described in the Polymarket documentation.
Disputes happen under both models. When an event happened only partly or the wording can be read two ways, the decision rests on the text of the rules, and traders do not always agree with it.
Polymarket vs Kalshi: how the platforms differ
Both platforms trade YES and NO contracts on the same logic, but they are built differently.
| Polymarket | Kalshi | |
|---|---|---|
| What it is | A crypto platform on the Polygon blockchain | An event exchange in the United States |
| Regulation | The main platform operates outside the US. For US users there is a separate app run through an exchange licensed by the CFTC | A CFTC-designated exchange since November 2020 |
| Settlement currency | A dollar-pegged stablecoin | US dollars |
| Who settles markets | The UMA oracle: an outcome proposed with a bond, disputes, a vote | The exchange itself, by the contract rules and the named source |
| Fees | Taker trading fees apply in many market categories and depend on the price | A trading fee by a published formula that depends on the contract price |
| Access | Depends on the user's country and the platform's rules | Depends on the user's country and the platform's rules |
Regulation
In November 2020 Kalshi received designated contract market status from the US regulator, the CFTC, becoming the first regulated exchange built specifically for event contracts.
Polymarket's path was more complicated. In January 2022 it settled with the CFTC, paying a $1.4 million penalty for operating without registration and winding down some markets. In July 2025 Polymarket acquired QCEX, an exchange licensed by the CFTC, and since late 2025 it has been rolling out a separate app for US users in stages.
Fees
On both platforms the fee depends on the contract price: it is higher for contracts near 50¢ and lower for contracts close to $0 or $1. The platforms change rates and the categories that carry fees from time to time, so check their official fee pages before working out a trade. Account separately for the cost of deposits and withdrawals.
Besides Polymarket and Kalshi there are other platforms, such as Limitless, Predict.fun and Opinion. Each has its own set of markets, its own fees and its own settlement rules.
Why the same question trades at different prices
The platforms are not connected. Each has its own traders and its own liquidity, and money does not move between them instantly. So the same question can trade at 55¢ for YES on one platform and 60¢ on another, and the gap stays until someone closes it.
The main reasons:
- different audiences: one platform has more crypto traders, another more users with dollar accounts;
- different liquidity: in a thin order book the price reacts to every order;
- moving money is hard: closing the gap requires funds on two platforms at once;
- different rules: sometimes prices differ because the questions are not actually the same.
The last reason is the most dangerous. A large price gap more often signals different questions under similar titles than a gift from the market.
Prediction arbitrage: Direct and Cross
Prediction arbitrage is a trade on the price gap of one question across two platforms. There are two ways to build a pair.
| Method | What you do | When it fits |
|---|---|---|
| Cross | Buy YES on one platform and NO on the other | Always: one of the two contracts pays $1 whatever the outcome |
| Direct | Buy YES where it is cheaper and sell YES where it is more expensive | If you already hold YES contracts on the second platform |
An example with hypothetical numbers. YES on platform A costs 40¢ and NO on platform B costs 55¢. Together that is 95¢, and the winning contract pays 100¢. The spread is 5¢, a little over 5% on the 95¢ invested, before fees.
In Direct the spread in cents is the same, but it takes less capital, so the return on capital is higher. Both methods are covered in detail in the lesson Direct and Cross: two types of prediction arbitrage.
Risks of prediction arbitrage
The "payout whatever the outcome" math only holds on paper. Several risks stand between it and money in your account.
- Liquidity. The best price covers only a small number of contracts. Beyond that, buying gets more expensive: this is slippage, and in thin markets it eats the spread quickly.
- Fees. They are charged on both platforms, and some platforms' fees have to be estimated.
- Filling the second side. While you buy the second leg, the price can move. Then you are left with an ordinary bet on one outcome.
- Disputes and delays. Settlement can be disputed and postponed, and your money is locked all that time.
- Platform and access. Rules, fees and access from your country can change. You need funds on both platforms in advance.
How to check a pair before trading
Compare the settlement rules
Open the market rules on both platforms. The source, date, time zone and provisions must match. If they do not, these are different questions.
Work out the spread after fees
Add up the prices of both sides and the fees on both platforms. If the total for Cross is not below $1, there is no trade.
Check the order book for your size
Look at the average and worst price for the number of contracts you need, not just the best price. Check the minimum order.
Account for the term
Work out how many days your money will be locked and compare that with other options.
Buy the less liquid side first
Start with the side that has the thinner order book: if it does not fill, you will not need to buy the other one.
The guide Prediction arbitrage explains how to read the columns and risk tags, and the whole section is described in the Prediction Markets overview. How to select pairs and allocate money is covered in the lesson Strategies and risk management.
Questions and answers
Are prediction markets legal?
It depends on the country and the platform. Kalshi operates in the US as an exchange designated by the CFTC. Polymarket is rolling out a separate app for US users through a licensed exchange, while its main platform operates outside the US. Before opening an account, check the platform's rules and the laws of your country.
Why don't YES and NO add up to exactly $1?
At settlement a YES and NO pair always pays $1. Before that there is a gap between the best buy and sell prices, and in a thin order book the two sides can drift further apart. So at any given moment the sum of the prices can be slightly above or below $1.
Who settles markets on Polymarket?
A market participant proposes the outcome by posting a bond through the UMA oracle. If no one disputes it, the outcome is accepted. If it is disputed, UMA token holders decide by vote.
Can I exit an arbitrage pair before settlement?
Yes, by selling both sides on the platforms. But prices may have moved by then, and the sale goes through the order book with fees, so the result can be worse than planned. Pairs are usually held until settlement.