How prediction market prices work, and how to read them as probabilities

Prediction markets let people trade contracts on the outcome of real-world events: an election, a Federal Reserve decision, a football game, tomorrow's high temperature. In the US, Kalshi is a CFTC-regulated exchange for these event contracts.

Their most useful property for research is simple: the price of a contract can be read as a probability. This guide explains why, and how to do it carefully.

#The $1 contract

A standard binary event contract pays $1 if the event happens and $0 if it does not. Prices are quoted in cents between 1 and 99.

Suppose a contract on "Team A wins" trades at 62¢. If you buy it:

• If Team A wins, you receive $1, a profit of 38¢.
• If Team A loses, you receive nothing, a loss of 62¢.

You would only pay 62¢ if you believed the chance of winning was at least 62%. A seller would only sell at 62¢ if they believed it was at most 62%. The price where they meet is the market's consensus: about a 62% implied probability.

Implied probability ≈ price in dollars. 62¢ → 0.62 → 62%.

#YES and NO are two sides of one contract

Every binary market has a YES side and a NO side. Buying NO at 38¢ is economically the same as selling YES at 62¢: you profit if the event does not happen. So the NO price is roughly 1 − YES price.

#Bid, ask and the midpoint

A live market has two prices, not one:

• Bid: the highest price someone is willing to pay for YES right now.
• Ask: the lowest price someone is willing to sell YES for right now.

If the YES bid is 60¢ and the ask is 64¢, which number is "the probability"? The best single estimate is usually the midpoint:

Mid = (Bid + Ask) / 2 = (0.60 + 0.64) / 2 = 0.62

The spread (ask − bid, here 4¢) is a measure of uncertainty and liquidity. A 1¢ spread on an active market tells you much more than a 20¢ spread on a quiet one. When a market is very thin, the last traded price can be stale, and even the mid can be misleading.

In a spreadsheet:

D2  Mid            =(B2 + C2) / 2
E2  Spread         =C2 - B2
F2  Implied prob.  =D2

#Multi-outcome markets and normalization

Many questions have more than two outcomes. A Fed decision market might list "cut 50+", "cut 25", "hold" and "hike" as separate contracts. Exactly one will settle YES, so their true probabilities must sum to 100%.

Market prices rarely sum to exactly 100%. Here is a hypothetical example using YES asks:

The asks sum to 105%. That extra 5% is the cost of buying every outcome at the ask, often called the overround. The bids sum to less than 100% for the same reason in reverse. Neither set is a clean probability distribution.

To compare outcomes fairly, normalize: divide each outcome's mid by the sum of all mids.

F2  Normalized prob.  =D2 / SUM($D$2:$D$5)

In this example the mids already sum to 1.000, so normalization changes nothing. In real markets they often sum to slightly more or less than one, and normalizing makes the probabilities add up to exactly 100%.

#Fees change the breakeven, not the probability

Exchanges charge trading fees. On Kalshi, the fee depends on the contract price and the number of contracts, and the venue publishes its current fee schedule. Fees do not change what the price implies about probability, but they change the probability you need to be right to break even:

Breakeven probability = (Price + Fee per contract) / $1

Buying at 62¢ with a 2¢ fee means you need the event to happen more than 64% of the time to profit. See Expected value and edge for the full calculation.

#Is the market's probability right?

An implied probability is what the market is willing to pay, not what will happen. Prices can be biased in consistent ways. A well-known hypothesis in betting markets is the favorite–longshot bias: longshots tend to be overpriced and favorites slightly underpriced. Whether that holds on a given venue, in a given category, in a given period, is an empirical question.

Because every contract eventually settles to 0 or 1, you can test it. Collect many settled markets, group them by price, and compare the implied probability with how often they actually resolved YES. That is a calibration study, explained step by step in Market calibration.

#Reading prices in practice

A short checklist before treating a price as a probability:

1. Use the mid, not the last trade, unless the market is very active.
2. Check the spread. Wide spreads mean low confidence in any single number.
3. Check the volume. A price that has not traded in hours is an old opinion.
4. Normalize multi-outcome markets before comparing outcomes.
5. Note the time to expiry. Prices a week out and an hour out answer different questions.

In Jordan, you can ask "What does the market think the Fed will do next?" and Jordan loads the open markets onto a sheet with bid, ask, mid and normalized probability columns as formulas, plus a chart, for you to review before it is applied.

#Frequently asked questions

#Why does a prediction market price equal a probability?

Because the contract pays exactly $1 if the event happens. A risk-neutral buyer would pay up to p dollars for a contract that pays $1 with probability p. In practice prices also reflect fees, spreads and trader biases, so the price is an approximate, market-implied probability rather than a perfect one.

#Should I use the last traded price or the bid-ask midpoint?

Use the midpoint for most purposes. The last traded price can be stale in quiet markets, and it bounces between bid and ask in active ones. For a backtest, use the price you could actually have traded at: the ask when buying and the bid when selling.

#Why do the prices in a multi-outcome market add up to more than 100%?

Each outcome's ask includes a small premium to the seller. Summing the asks adds those premiums together. Divide each outcome's midpoint by the total of all midpoints to get probabilities that sum to exactly 100%.

#Are prediction markets accurate?

Research has generally found prediction markets to be reasonably well calibrated on liquid questions, but accuracy varies by venue, category, liquidity and time to expiry. The way to know for a specific set of markets is to measure calibration on settled contracts.