How Prediction Market Prices Are Set

A prediction market price is set the same way any price is: by what a buyer and a seller agree to trade at. The clever part is the machinery that finds that agreement, because the price also has to read as a probability.

On a market that pays 1 when an event happens and 0 when it does not, a share trading at 63 cents is the crowd saying the event is about 63% likely. So the question "how is the price set" is really two questions: who is on the other side of your trade, and what keeps the number behaving like a well-formed probability? There are two main answers, and most platforms use one or the other.

Two ways a market turns trades into a price

Every prediction market needs a way to answer one question millions of times a day: what is this contract worth right now? There are two families of answer. The first, an order book, waits for buyers and sellers to meet. The second, an automated market maker, is a piece of math that always stands ready to quote you a price. Understanding the difference is the whole game.

Order books: matching buyers and sellers

An order book is the same mechanism a stock exchange uses. Everyone who wants to buy posts a bid ("I will pay up to 60 cents for a YES share"). Everyone who wants to sell posts an ask ("I will sell YES at 62 cents"). The book stacks those orders by price, and whenever the highest bid meets the lowest ask, a trade happens and that becomes the last traded price.

The price you see quoted is really two numbers. The gap between the best bid and the best ask is the spread, and the pile of orders waiting at each level is the depth. Both tell you how real the price is.

• A tight spread (bid and ask almost touching) means buyers and sellers roughly agree on the value.
• Deep books, with lots of orders stacked near the current price, mean you can trade a big size without moving the number much.
• The catch: an order book only works when there are traders on both sides at the same time. Early in a market, or on an obscure question, the book can be empty and there is simply no price.

The liquidity problem with thin order books

This empty-book problem is the reason order books alone are a poor fit for many prediction markets. Real news questions are numerous, niche, and short-lived. You cannot count on a crowd of traders showing up for "will this specific bill pass committee by Friday" the moment it opens. If nobody has posted an order, a newcomer who wants to trade has nobody to trade against, and the market never gets off the ground. That cold-start problem is exactly what the second approach was built to solve.

Automated market makers: a bot that always quotes

An automated market maker (AMM) replaces the crowd on the other side of your trade with a formula. Instead of waiting for a human seller, you trade against the market maker itself, and a rule decides the price. The rule reads the current state of the market (how many YES and NO shares are outstanding) and outputs a price for the next share. Buy YES and the price of YES rises; buy NO and it falls. The market maker is always open, always quoting, and never needs a matching human on the other side.

That solves the cold-start problem instantly. A brand-new market has a price from the first second, so the first trader has something to push against. The design question becomes: what formula? A naive one could hand out unlimited free money to a clever trader, or produce prices that do not add up to a coherent probability. The most influential answer is a scoring rule.

LMSR: Robin Hanson’s market maker

The best known AMM for prediction markets is the Logarithmic Market Scoring Rule, or LMSR, introduced by economist Robin Hanson around 2002 in his paper on logarithmic market scoring rules. It is the pricing engine behind many subsidized prediction markets, and it is worth knowing conceptually even if the algebra stays under the hood.

The intuition is simple. LMSR keeps a running cost that depends on how many shares of each outcome have been bought. To buy shares, you pay the increase in that cost; to sell, you receive the decrease. Because the cost function is smooth, the price moves a little with every share rather than jumping, and it always slides in the right direction: buying an outcome makes it more expensive, which is exactly how a probability should behave when new money backs it.

Three properties are what made LMSR the default for so long:

• Continuous liquidity: there is always a price and you can always trade any size, because the formula never runs out of the other side. No matching buyer is needed.
• Bounded loss: the market operator subsidizes the market, but the formula caps the worst case in advance. A single parameter (often called b) sets both how much liquidity there is and the maximum the operator can lose, so the subsidy is known before the market opens.
• Prices that behave like probabilities: the quoted prices for all outcomes always sum to 1 (100%), so you can read them directly as the crowd’s probability without any extra normalization.

The b parameter is the dial worth understanding. A large b makes the market deep and hard to move (each trade barely nudges the price) but costs the operator more subsidy. A small b makes the market cheap to run but jumpy, so a single trade can swing the odds. Choosing b is the operator choosing how much liquidity to pay for.

Thin versus deep markets, and how prices move

Whichever mechanism a platform uses, the same truth holds: depth is what makes a price trustworthy. In a thin market (a shallow order book, or an AMM with a small b), a single modest trade can move the price several points. That number is fragile, easily distorted by one motivated or careless participant. In a deep market, the same trade barely registers, so the price reflects a genuine consensus rather than the last person to click.

This is why you should read a price together with its liquidity. A YES at 70 in a deep market is a strong signal. A YES at 70 in a market with almost no volume is barely more than one person’s guess dressed up as a probability. The mechanism sets the price, but the depth sets how much you should trust it.

That link between depth, incentives, and accuracy is the subject of why prediction markets are accurate, which picks up where this guide leaves off.

A note on crypto AMMs

If you have met automated market makers before, it was probably in crypto, where AMMs price token swaps on decentralized exchanges. The family resemblance is real: both replace a human counterparty with a formula over the pool’s current balances. But the goals differ. A crypto swap AMM is built to price the exchange rate between two assets and is happy for the pool to drift with supply and demand. A prediction-market AMM like LMSR is tuned so the outputs stay pinned to a valid probability and so the operator’s subsidy is bounded. Same idea, different constraints.

Plenty of platforms make these mechanics visible while you trade. Manifold Markets runs on an automated market maker, so you can watch the price shift as you buy.

And not every forecasting platform uses prices at all: Metaculus aggregates explicit probability estimates from forecasters instead of running a market, a useful contrast to keep the two models straight.

If any of the underlying vocabulary is new, prediction markets explained covers what a share, a resolution, and a payout actually are.

See it happen for yourself

The fastest way to understand price formation is to move a price yourself and watch it settle back as others trade against you. You do not need real money to feel the mechanics.

Clutch lets you trade real news and sports markets with in-app credits and shows the price update on every trade. Get the app and watch a price form in real time.