Prices Are Probabilities: How to Read Prediction Market Odds (2026)

The one-line rule: a 62¢ share means about a 62% chance

Many prediction-market contracts are yes/no questions. A Yes share pays $1 if the event happens and $0 if it doesn't; a No share pays the other way round. Prices run between $0 and $1, so a Yes price of 62¢ reads as about 62%, and the matching No price of about 38¢ reads as about a 38% chance that the event doesn't happen.

Why it works: because the payout is fixed at $1, a share is worth the chance of winning times $1. A trader who thinks the chance is 70% values a Yes share at 0.70 × $1 = 70¢ and is willing to pay up to about that; one who thinks it is 55% values it at 55¢ and is willing to sell above that. Buying by people who think the price is too low and selling by people who think it is too high pushes the price toward the traders' combined view.

The platforms and the US regulator describe prices the same way. Polymarket's docs say "the price directly represents the market's belief in the probability of that outcome". Kalshi's help center gives the example of a 70% consensus showing Yes at 70¢ and No at 30¢, and its blog says a Yes price of p cents implies "about p%". The CFTC's guide to prediction markets also reads a 70-cent price as about a 70% implied likelihood.

Yes and No mirror each other. On Polymarket, a Yes order at $0.60 and a No order at $0.40 add up to $1.00 and can be matched by creating one new Yes and one new No share. Kalshi's API docs state the same link from the other side: a Yes bid at X is equivalent to a No ask at $1.00 − X, so a best No bid of $0.56 means a best Yes ask of $1.00 − $0.56 = $0.44.

Which number are you looking at?

A market has several prices at once, and the big percentage on the page is only one of them. These are the ones that matter:

• Bid: the highest price someone is currently offering to pay. It is what you would receive if you sold immediately.
• Ask: the lowest price someone is currently willing to sell at. It is what you would pay if you bought immediately.
• Spread: the ask minus the bid. Polymarket's docs note that tighter spreads indicate more liquid markets.
• Midpoint: halfway between the bid and the ask, (bid + ask) ÷ 2. Nobody necessarily trades at this price.
• Last trade: the price of the most recent fill. It can be minutes, hours or days old.

What Polymarket displays: Polymarket's docs and help center say the displayed price is the midpoint of the bid-ask spread, and that if the spread is wider than $0.10 the last traded price is shown instead. In Polymarket's own example, a 34¢ bid and a 40¢ ask give (0.34 + 0.40) ÷ 2 = 0.37, shown as 37%. But an immediate buy pays the 40¢ ask and an immediate sell receives the 34¢ bid; nobody trades at 37¢ in that moment.

What Kalshi displays: Kalshi's help center and blog say its prices reflect the market's probability, but we found no Kalshi page that says whether the headline chance is the last trade, the midpoint or something else. Don't assume it follows Polymarket's rule; check the bid and ask shown for the market itself.

Step-by-step guides to the order book: how to buy on Polymarket and how to sell on Polymarket. How each platform works overall: Polymarket and Kalshi.

Spread and depth: when the displayed percentage is unreliable

A narrow spread means buyers and sellers roughly agree, so the midpoint is a fair summary. A wide spread means the market is really saying "somewhere between the bid and the ask", and a single percentage hides that.

Depth: the best bid and ask hold only for the shares resting at those prices. Polymarket sets no maximum order size, but it warns that there is no guarantee you can trade a given amount without moving the price significantly, or at all. For example, with 100 shares offered at 40¢, 100 at 42¢ and 500 at 45¢ (hypothetical), $100 buys 100 × $0.40 = $40.00, then 100 × $0.42 = $42.00, then $18.00 ÷ $0.45 = 40 shares: 240 shares at an average of $100 ÷ 240 = 41.67¢, before fees.

So before reading a percentage as the market's view, look at the bid, the ask, how many shares sit at each, and when the last trade happened. The checklist below gathers these.

Converting prices to decimal, fractional and American odds

The implied probability is simply the price in dollars × 100, so $0.37 is 37%. Sports bettors often think in odds instead. The standard conversions below use the raw price p in dollars (between 0 and 1) and ignore fees:

• Decimal odds = 1 ÷ p. The total you get back per $1 staked, including the stake.
• Fractional odds = (1 − p) ÷ p. Profit per unit staked, written as profit/stake.
• American odds: below 50¢, +100 × (1 − p) ÷ p, the profit on a $100 stake; above 50¢, −100 × p ÷ (1 − p), the stake needed to make $100 profit; at exactly 50¢, +100 (the same as −100).

Kalshi's own blog gives matching fractional examples: buying Yes at 25¢ risks 25¢ to win 75¢, or 3-to-1; 50¢ is 1-to-1; 75¢ is 1-to-3.

Going back from odds to a probability: decimal odds d → 1 ÷ d; American +A → 100 ÷ (A + 100); American −A → A ÷ (A + 100). So +150 is 100 ÷ 250 = 0.40, or 40%, and −200 is 200 ÷ 300 = 0.667, or about 66.7%.

With fees: replace p with what a share really costs you, the price plus the fee per share (next sections). At Polymarket's politics rate of 0.04, a 25¢ share carries a taker fee of 0.04 × 0.25 × 0.75 = $0.0075, so it costs $0.2575. Decimal odds fall from 4.00 to 1 ÷ 0.2575 = 3.88, and American odds from +300 to +100 × 0.7425 ÷ 0.2575 = +288.

Multi-outcome markets: do the prices add up to 100%?

Questions with several possible answers, such as who will win an election, are a set of yes/no markets, one per outcome, each with its own Yes and No prices. In Polymarket's "negative risk" groups exactly one market resolves Yes and every other resolves No, and a No share in one outcome can be converted into one Yes share in every other outcome.

Neither Polymarket nor Kalshi publishes a rule that the displayed percentages must add up to 100%, so what follows is arithmetic, not a platform rule. If exactly one outcome pays $1:

• The asks should add up to about $1 or more. If they added up to less, buying one Yes of every outcome would cost less than the $1 that one of them is certain to pay, before fees, and buying would push prices up.
• The bids should add up to about $1 or less, by the same argument in reverse.
• Displayed numbers can land on either side of 100%, because midpoints, stale last trades and rounding don't have to add up exactly.

The extra 3¢ on the asks is the cost of buying every outcome at once. It comes from the spreads, not from a fee the exchange charges. Dividing each ask by the total rescales them to 100%: 0.52 ÷ 1.03 = 50.5%, 0.31 ÷ 1.03 = 30.1% and 0.20 ÷ 1.03 = 19.4%.

Real markets show the same thing. On the evening of New York City's 2025 Democratic mayoral primary, Forbes reported Kalshi at 53% for Cuomo and 48% for Mamdani, 101% in total.

Break-even after fees: price plus fee per share

When you buy immediately (as a "taker"), you pay the ask plus a fee in most markets. If you hold to resolution, the break-even probability is (price paid + fee per share) ÷ $1: the outcome has to be more likely than that for the position to be worth what it cost. This ignores interest and holding rewards; selling early costs the spread plus, in most markets, another taker fee. The rates below are as of September 2026 and change often; check Polymarket fees explained, Kalshi fees explained or the fee calculator.

• Polymarket (polymarket.com): fee = shares × fee rate × price × (1 − price), charged to takers only; "makers are never charged fees". Rates by category: crypto 0.07; sports, economics, culture, weather and other 0.05; finance, politics, mentions and tech 0.04; geopolitics 0. Fees are rounded to 5 decimal places.
• Polymarket US (the separate US app): fee = 0.0695 × contracts × price × (1 − price) for takers, at most $1.74 per 100 contracts at 50¢, rounded to the cent (banker's rounding). Makers receive a rebate at 0.0125 on the same formula.
• Kalshi: taker fee = round up(0.07 × contracts × price × (1 − price)), for example $1.75 per 100 contracts at 50¢ in Kalshi's own table. The fee schedule also has a maker formula at 0.0175, but its default maker multiplier is 0, so makers generally pay no fee; some series charge one, and some have no fees at all. There is no settlement fee.

Fees matter less near the extremes in cents, but not always in proportion. At a 90¢ ask, Polymarket's politics fee is 0.04 × 0.90 × 0.10 = $0.0036, so break-even is 90.36%; Kalshi's table gives $0.63 per 100 contracts, so break-even is 90.63%. At a 5¢ ask, the politics fee is 0.04 × 0.05 × 0.95 = $0.0019, so break-even is 5.19%, which is 0.19 ÷ 5 = 3.8% above the price in relative terms.

Kalshi's rounding is not clear-cut. Its fee schedule (effective July 7, 2026) says fees are rounded up so that fee plus cost lands on a centicent ($0.0001), while its own table rounds to the cent, and its API docs describe yet another method. For large orders this barely matters, and the 100-contract examples above match Kalshi's table. For one contract at 5¢ it does: 0.07 × 0.05 × 0.95 = $0.003325 becomes $0.0034 (break-even 5.34%) if rounded up to the centicent, or $0.01 (break-even 6.00%) if rounded up to the cent.

The midpoint trap: Polymarket shows 37% (bid 34¢, ask 40¢) in a politics market. An immediate buy costs 0.40 + 0.0096 = $0.4096 per share, so the break-even is about 41%, not 37%.

Orders that wait: a limit order that rests on the book and is filled later pays no fee on Polymarket, and none on Kalshi by default, so if it fills at 38¢ the break-even is 38%. On Polymarket US the maker receives a rebate: −0.0125 × 100 × 0.38 × 0.62 = −$0.2945, about $0.29 back per 100 contracts, so the break-even is about 37.7%.

Selling before resolution: the formula applies to sells too. Buying at 40¢ and later selling at 40¢ in a politics market, both as a taker, costs 0.0096 + 0.0096 = $0.0192 per share, 1.92¢ lost although the price didn't move. The examples ignore Polymarket's taker-rebate programme and Polymarket US volume rebates, which return part of the fee to traders above set volume thresholds.

When the price isn't the probability

Reading a price as a probability is a good first approximation, not an identity. Economists have studied when it holds:

• Risk attitudes and disagreement: Wolfers and Zitzewitz show that for a broad class of models prices are usually close to traders' average belief, typically within a percentage point, and exactly equal to it in a benchmark case. The gap is largest for prices closest to $0 and $1. Manski shows that if traders are risk-neutral and disagree, the price says nothing about how widely their beliefs are spread, and only brackets their average belief within an interval centred on the price.
• Time value and holding rewards: money in a position is tied up until resolution. With an outside interest rate r and T years to go, a risk-neutral trader pays about q ÷ (1 + r)^T for a share worth q at the end, so a 90¢ price with r = 4% and one year to go is consistent with a belief of about 0.90 × 1.04 = 0.936, or 93.6% (our arithmetic). Both platforms now pay interest-like rewards that, by our reasoning, shrink this gap: Polymarket pays variable Holding Rewards, 4.00% a year at the time of writing, on eligible markets, and Kalshi pays a variable 3.25% on cash and open positions to eligible US members with balances of $250 or more.
• Fees: a buyer's break-even sits above the displayed price, and a seller's below it, as shown above. A price can sit a little off the true probability simply because moving it would cost more in fees than it earns.
• Thin books and large traders: in a shallow market one order can move the price a long way. In 2012 one Intrade trader accounted for more than a third of all bets on Romney to win, spent over $375,000 on Election Day holding prices in place and lost close to $7 million, according to Rothschild and Sethi. Research on the Iowa Electronic Markets, by contrast, found that manipulation attempts had only short-lived effects. In 2024 one trader held about $28.6 million on Donald Trump across four Polymarket accounts (NBC News); Polymarket said it had not identified any information suggesting that this trader manipulated the market.
• Rule wording: a price is the probability that the contract resolves Yes under its written rules, which is not always the headline question. A Kalshi market asked whether Ali Khamenei would be out as Iran's Supreme Leader before March 1, 2026. After his death on February 28, 2026, Kalshi applied a carve-out in its rules for death and settled at the last traded price before his death; the market had about $54 million in volume and traders filed a class action on March 5, 2026 (as reported by CasinoBeats on March 9, 2026). Polymarket paid out on its own Khamenei markets. Polymarket has had disputed resolutions of its own, such as the 2025 markets on a Ukraine minerals deal and on whether Zelenskyy wore a suit.
• Longshot bias: cheap contracts have often won less often than their price implies. The evidence is in the next section.

How accurate have prediction markets been?

A single result can't show that a price was right or wrong: a 53% favourite should lose about 47% of the time. Accuracy is judged over many markets. Calibration asks whether events priced at X% happened about X% of the time. The Brier score is the average of (forecast − outcome)², where the outcome is 1 or 0: 0 is perfect and always saying 50% scores 0.25. A 70% forecast scores (0.70 − 1)² = 0.09 if the event happens and (0.70 − 0)² = 0.49 if it doesn't.

What independent research finds

• Early evidence: in a 2004 review, Wolfers and Zitzewitz found market forecasts reasonably accurate, outperforming "most moderately sophisticated benchmarks". Comparing the Iowa Electronic Markets with 964 polls across five US presidential elections from 1988 to 2004, Berg, Nelson and Rietz found the market closer to the result 74% of the time. Arrow and co-authors report an average election-eve error of about 1.5 points for the market against 2.1 for the final Gallup poll.
• Long horizons: Page and Clemen, studying 1,787 Intrade markets, found prices reasonably well calibrated close to expiry but significantly biased for events further in the future. Overall, a price of 0.20 went with a 15.3% frequency and a price of 0.80 with 87.4%. A 2026 preprint by Le on 292 million Kalshi and Polymarket trades likewise finds long-dated prices compressed toward 50%.
• The 2024 US election: a Vanderbilt working paper by Clinton and Huang covered more than 2,500 markets on the Iowa Electronic Markets, Kalshi, PredictIt and Polymarket, with $2.4 billion traded in the final five weeks. It found that prices for identical contracts diverged across exchanges, with arbitrage opportunities peaking in the final two weeks. Media reports give accuracy shares of 93% for PredictIt, 78% for Kalshi and 67% for Polymarket, but the platforms listed different markets, so this is not a like-for-like ranking, and Kalshi disputes the method.

The favorite-longshot bias

In betting markets, longshots tend to be overbet and favourites underbet, so cheap contracts win less often than their price implies. Snowberg and Wolfers, studying horse racing, conclude that misperception of probabilities drives the pattern.

• Kalshi: a working paper by Bürgi, Deng and Whelan (University College Dublin) finds an average return of about −20% on Kalshi contracts: −9.64% for makers and −31.46% for takers. Buyers of contracts under 10¢ lost over 60% of their money, and a 5¢ contract won about 2% of the time, while contracts above 50¢ earned a small positive return. Prices became more accurate as expiry approached.
• Polymarket: the evidence is mixed. A working paper by Reichenbach and Walther, covering more than 478 million trades, finds prices closely track realised probabilities, slightly outperform bookmaker odds, and show no general longshot bias. A September 2026 preprint by Cardozo and Rivero-Wildemauwe, on 588 million trades, finds contracts under 10¢ returned −6.30% equal-weighted and −19.35% dollar-weighted, but +4.09% when grouped by parent event, with sports as an exception. The answer depends on horizon, category and how returns are counted.

What the platforms say about themselves (self-published)

• Polymarket's accuracy page (self-published): the leading outcome matched the result 90.2% of the time one month before resolution, 94.2% one week before, 96.4% one day before, 97% twelve hours before and 98.5% four hours before, with a Brier score of 0.0627. The page does not state its sample size, date range or which price it uses. It also shows that 77.8% of resolved markets resolved No, so always picking No would have scored about 78% with no skill; compare such figures with that base rate, not with 50%.
• Kalshi's calibration study (self-published, August 2026): Kalshi says its predicted probabilities match real-world outcomes closely and improve with time, participation and volume, and that even markets with about $10,000 in volume or as few as 20 traders give useful forecasts. The study's detailed figures appear only in chart images; as reported by Yellow, it covers 2,243,741 resolved markets from 2021 to mid-2026. We have not checked its Brier scores against Kalshi's charts, so we don't quote them. Kalshi says all its data except trader counts is public.

Self-published and independent figures measure different samples, times and definitions, so a 97% headline and a 67% study result can both be true. Neither is a guarantee about the next market.

Notable misses

• 2016 US presidential election: on election night PredictIt traders had Hillary Clinton at about 80% and the PredictWise aggregator had her at about 89%, according to PBS NewsHour.
• 2016 Brexit referendum: betting markets gave Leave about 25% in the final week. Researchers Gelman and Rothschild argued afterwards that belief in the markets' correctness made them too stable.
• 2025 New York City mayoral primary: at about 6 p.m. on primary day Polymarket had Cuomo at 53% and Mamdani at 47%; Mamdani won. A 53% favourite losing is what a 53% forecast allows about 47% of the time, so on its own this is not evidence of a bad forecast.
• 2024 US presidential election: the markets favoured Donald Trump, who won, but identical contracts traded at different prices on different exchanges. Good Authority's summary of the Clinton and Huang paper notes that, across exchanges, the Harris and Trump contracts, which should have summed to $1, did not on 62 of the 65 days before the election.

Before you trust a price: a checklist

• Volume and traders: in Kalshi's self-published data, markets with more volume and more traders were more accurate. Treat volume totals with care: check who measured them and when, and remember that a Columbia University study estimated about 25% of Polymarket's historical volume was likely wash trading (reported by Decrypt, November 7, 2025).
• Spread: look at the bid and the ask, not just the headline number. If they are more than 10¢ apart, Polymarket is showing you the last trade.
• Last trade time: a price from hours or days ago may be far from where anyone would trade now.
• Depth: how many shares sit at the best prices? A small book can be moved by one order.
• Resolution rules: read exactly what counts as Yes, the source used, the deadline and any exceptions. The price is for the contract as written, not the headline.
• Time to resolution: long-dated prices have tended to be compressed toward 50% and are affected by interest and rewards.
• Fees: add the fee per share to the price to see the real break-even.
• Other platforms: compare the same question elsewhere, checking that the rules really match. Our Polymarket vs Kalshi comparison explains how the two differ.

Frequently asked questions

What do Polymarket odds mean?

The price of a Yes share, in cents, is the market's rough percentage chance that the event happens, because the share pays $1 if it does and $0 if it doesn't. A Yes share at 62¢ means about a 62% chance. It is the traders' combined estimate, not a guarantee.

How do you read Polymarket odds?

Read the percentage as the chance of Yes, then check the bid and the ask. Polymarket shows the midpoint between them, or the last trade if they are more than 10¢ apart, so a market showing 37% with a 34¢ bid and a 40¢ ask costs 40¢ to buy immediately, plus any fee.

Is Polymarket accurate?

Polymarket's self-published figures show its leading outcome matching the result 90.2% of the time a month before resolution and 98.5% four hours before, but they don't state the sample size, and most of its markets resolve No. Independent studies are mixed: one working paper finds prices closely track outcomes, while a study of 2024 election markets found prices for identical contracts diverged across exchanges.

How do you convert a Polymarket price to odds?

With p as the price in dollars: decimal odds = 1 ÷ p, fractional odds = (1 − p) ÷ p, and American odds = +100 × (1 − p) ÷ p below 50¢ or −100 × p ÷ (1 − p) above 50¢. So 25¢ is 4.00, 3/1 or +300, and 75¢ is 1.333, 1/3 or −300, before fees.

Why do Polymarket and Kalshi show different odds?

They are separate markets with different traders, fees, interest rates, money (crypto-based pUSD or US dollars), rules and resolution sources, and the headline number may be calculated differently. polymarket.com is closed to US residents (Polymarket US is a separate app), Kalshi is a US-regulated exchange, and many countries limit access to one or both (see where Polymarket and Kalshi are available); Polymarket's terms prohibit using a VPN to get around its restrictions. A gap smaller than the combined fees and spreads can persist because closing it would cost more than it earns.

What does a price mean on Kalshi?

The same thing: Kalshi says a Yes price of p cents implies about a p% chance, and its help center shows a 70% consensus as Yes at 70¢ and No at 30¢. Kalshi doesn't document whether its headline number is the midpoint or the last trade, so check the bid and ask.

Do prediction market odds add up to 100%?

Not exactly. In a multi-outcome market the asks tend to add up to a little over $1 and the bids a little under, while displayed midpoints or last trades can land either side of 100%. Neither platform publishes a sum-to-100% rule.

Is the percentage on Polymarket the price I pay?

Not usually. An immediate buy pays the ask, which is above the displayed midpoint, and larger orders also fill at higher asks. In most categories a taker fee is added, so a politics share shown at 37% with a 40¢ ask costs $0.4096.

What is the favorite-longshot bias?

The tendency for cheap longshot contracts to win less often than their price implies. On Kalshi, a UCD study found buyers of contracts under 10¢ lost over 60% of their money; on Polymarket the evidence is mixed.

Are prediction markets more accurate than polls?

Often, in the research so far: the Iowa Electronic Markets were closer to the result than 964 polls 74% of the time across five US presidential elections. But markets have missed too, as in 2016, and their accuracy depends on volume, horizon and how the market is written.

Sources & verification

Every fact on this page comes from the sources below, checked on September 24, 2026; see our methodology. Figures are as of September 2026, and fees and platform rules change often. Statistics that Polymarket or Kalshi publish about themselves are labelled as self-published. Worked examples marked hypothetical use made-up order books, and all arithmetic is shown.

Polymarket: prices and order book, fees, markets and events, negative risk, positions and tokens, How Are Prices Calculated?, Does Polymarket Have Trading Limits?, Geographic Restrictions, accuracy page (self-published) and Terms of Use.

Polymarket US: fee schedule.

Kalshi: How are prices determined?, APY on Kalshi, How to read probabilities (self-published), calibration study (self-published), order book API docs and fee schedule (effective July 7, 2026).

Regulator: CFTC, Understanding Prediction Markets and Event Contracts.

Peer-reviewed and published research: Wolfers & Zitzewitz, Prediction Markets (2004) and Interpreting Prediction Market Prices as Probabilities (2006); Manski, Interpreting the Predictions of Prediction Markets; Page & Clemen, Do Prediction Markets Produce Well-Calibrated Probability Forecasts? (2013); Snowberg & Wolfers, Explaining the Favorite-Longshot Bias (2010); Berg, Nelson & Rietz, Prediction market accuracy in the long run (2008); Arrow et al., The Promise of Prediction Markets (2008); Berg & Rietz, manipulation and accuracy on the Iowa Electronic Markets (2014); Rothschild & Sethi, Trading Strategies and Market Microstructure.

Working papers and preprints (not peer-reviewed): Bürgi, Deng & Whelan, Makers and Takers: The Economics of the Kalshi Prediction Market and VoxEU column; Reichenbach & Walther, Accuracy, Skill, and Bias on Polymarket; Cardozo & Rivero-Wildemauwe, The Favorite–Longshot Bias in Prediction Markets; Le, Decomposing Crowd Wisdom; Clinton & Huang, The Accuracy and Efficiency of $2.4 Billion in the 2024 Presidential Election.

Reporting: Yellow on Kalshi's calibration study, Good Authority on Clinton & Huang, DL News on Clinton & Huang, PBS NewsHour (2016), Slate (2016), Forbes (2025), CasinoBeats (2026), NPR (2026), NBC News (2024), Decrypt on wash trading (2025), CoinDesk on the Ukraine minerals market (2025) and Decrypt on the Zelenskyy suit market (2025).

PolyKal is an independent research tool, not affiliated with Polymarket or Kalshi. This page explains how prediction-market prices work; it is not financial, legal or tax advice, and not an invitation to open an account or trade.