Why don't Polymarket prices add up to 100%?
Last updated August 2026 · The Cent Signals desk
TL;DR
In a binary Polymarket market they almost always do. The No price shown is one hundred minus the Yes price, so the displayed pair sums to 100 by construction, and one Yes share plus one No share redeems for exactly one dollar by contract. The sums that genuinely drift sit in the order books underneath, and in multi-outcome fields. Cent Signals is a free, independent desk that tracks Polymarket activity and explains how prediction markets price probability, not trading advice.
The number you see is one price written twice
The question usually arrives after someone adds two numbers on a market page and expects the arithmetic to misbehave. On a binary Polymarket contract it will not, and the reason is worth knowing before anything else. The interface and the public Gamma API both carry an outcome price array in which the No value is the complement of the Yes value. A market quoting Yes at 44.5 cents reports No at 55.5 cents because that is what one minus 0.445 equals, not because a separate book settled there.
We can put a number on how absolute that is. Every Cent Signals snapshot records the Yes and No price for the 400 highest-volume markets it tracks. In the snapshot generated on July 31, 2026, all 400 pairs summed to 1.0000 exactly. Not close to a dollar, not within a cent, but 400 out of 400 with zero deviation. A dataset that clean is not a finding about market efficiency. It is a finding about what the field contains, and it is the single most common source of confusion on this topic.
Why a dollar is the anchor at all
Underneath the display sits a mechanism that makes a dollar the natural reference point. Polymarket outcome shares are issued through the Gnosis conditional token framework, a set of public contracts on Polygon. One USDC of collateral can be split into a full set of one Yes share and one No share, and a full set can be merged back into one USDC at any point before the market resolves. Both operations are functions on the contract, available to anyone, and they carry no trading fee of their own.
That is a stronger statement than the usual claim that probabilities must sum to one. It means the pair has a redemption value fixed at a dollar regardless of what anyone believes about the underlying event. Exactly one of the two shares pays out a dollar at resolution and the other pays nothing, which is also why nobody has to fund the winning side separately, a point covered in who pays you when you win on Polymarket. The dollar is structural. The prices quoted against it are not.
Where the sum actually moves
One layer below the displayed pair, Yes and No are two separate ERC-1155 tokens trading on two separate order books. Each book has a best bid and a best ask, so there are four live numbers rather than two, and they do not have to reconcile. Add the two asks and the total will normally exceed a dollar by the width of the spreads. Add the two bids and it will fall short by the same mechanism. Neither is a pricing error. It is what a spread looks like when the same instrument is quoted from both ends, and it is the same friction described in what is a Polymarket order book.
Three further frictions widen the gap between an advertised number and a realised one. Depth is uneven, so the size available at the top of the book on a quiet contract can be small. Update timing differs between the two books, so one side can carry a stale quote for minutes while the other has already moved. And Polymarket applies taker fees on many markets as of 2026, set out in Polymarket fees explained, which sit on top of whatever the book showed.
Multi-outcome fields, where the arithmetic gets harder
The version of this question that has a real answer concerns events with more than two outcomes. A field such as the September 2026 Fed rate decision is not one contract with several halves. It is a set of separate contracts, each with its own book, its own spread, and its own last trade, that happen to describe mutually exclusive outcomes of the same event. Nothing in the contract forces their quoted prices to sum to a dollar the way a full set does.
Polymarket's negative risk structure exists partly to bind those legs together, and the arithmetic only carries a guarantee when the set of outcomes is exhaustive, mutually exclusive, and fully visible to whoever is adding it up. The failure mode is subtle: a field read partially will sum to less than a dollar simply because legs are missing, which looks identical to a mispricing and is not one. Our guide to negative risk on Polymarket works through that trap with live examples. The academic literature notes a related structural fact: more than 90 percent of the liquidity in a multi-outcome market typically sits in its top four legs, so the long tail contributes to the sum while barely trading.
What a year of executed trades showed
The best public measurement of how far the sums really drift comes from Unravelling the Probabilistic Forest, by Oriol Saguillo, Vahid Ghafouri, Lucianna Kiffer, and Guillermo Suarez-Tangil, published at the 2025 Conference on Advances in Financial Technologies. The authors read 86 million executed bids on markets that resolved between April 1, 2024 and April 1, 2025, computed a volume-weighted average price for each position, and counted the windows where the sum sat more than two cents from a dollar.
Of 17,200 conditions in that dataset, 7,051 showed at least one such window. Every instance they found within a single condition was on the low side, meaning the two halves together changed hands for less than the dollar they redeem for. Within multi-outcome markets the pattern was denser: 662 of 1,578 such markets carried at least one window, averaging roughly a hundred per market. Estimating from executed bids rather than quotes, the authors put the total actually captured across every strategy they identified at 39,587,585 dollars for the year, with the single most active address accounting for 2,009,631 dollars across 4,049 executed bids.
Two caveats belong with those figures. The measurement window predates Polymarket's per-trade taker fees, which the authors note were not charged during the period they studied, so the same gaps would net differently today. And cross-market inconsistencies, where two logically linked events priced incompatibly, turned out to be far rarer than the single-market kind: of thirteen dependent market pairs identified around the 2024 US election, only five showed any executed activity at all, the largest at 60,236 dollars.
Four numbers, side by side
Most of the confusion here dissolves once the four different things called the Polymarket price are separated. The table sets them against each other. Rows describing costs are marked as of 2026, because the fee structure has changed inside the last year.
| Attribute | Displayed pair | Order book quotes | Multi-outcome field | Executed prices |
|---|---|---|---|---|
| What the number is | The Yes price and its complement, as shown in the interface and returned by the public Gamma API | The best resting bid and best resting ask on each of the two token books | The quoted prices of every leg in a multi-outcome event, added together | Volume-weighted average of trades actually filled over a time window |
| Does it sum to exactly 100 | Yes, by construction. 400 of 400 markets in our July 31, 2026 snapshot summed to 1.0000 | No. The two asks together typically exceed a dollar and the two bids fall short of it | Only when the field is exhaustive, mutually exclusive, and fully visible | Frequently not. The AFT 2025 study found 7,051 of 17,200 conditions off by more than two cents at some point |
| Why it can drift | It cannot. It is one number written twice | Two independent books, each with its own spread, depth, and update timing | Each leg has its own book, and long-tail legs go stale between trades | Trades land at different moments, and one side can fill before the other moves |
| What it is useful for | Reading implied probability at a glance and comparing markets to each other | Judging what size could actually change hands near the quoted level | Seeing how an event distributes probability across its outcomes | Measuring what participants paid rather than what was advertised |
| Costs sitting on top (as of 2026) | None. It is a display value, not a fill | The spread on each side, plus taker fees on many markets | One spread per leg, so a sixteen-leg field means crossing sixteen spreads | Already reflects what was paid, but not fees charged outside the fill |
| Where Cent Signals uses it | Market pages, the markets index, and the signals list | Not collected. We record liquidity as reported rather than book depth | Read at the event level, never leg by leg in isolation | The whale trades feed, which records executed size on chain |
Displayed-pair figures computed from the Cent Signals snapshot generated July 31, 2026. Executed-price figures from Saguillo and others, AFT 2025, covering markets resolved April 1, 2024 to April 1, 2025.
What a gap does and does not represent
A sum that reads 96 cents does not describe four cents sitting on a table. Between the arithmetic and anything realisable stand the spread on each leg, the depth actually available at the quoted level, taker fees on many markets, and the fact that filling two legs on an order book is not a single atomic action, so one side can complete while the other moves away. The AFT authors classify this as non-atomic arbitrage and treat the execution risk as a real cost rather than a rounding detail.
The strongest evidence that these gaps are harder to act on than they look is in the same paper: around the 2024 US election, roughly 1 percent of the opportunities the authors detected were acted on by anyone. The rest expired unclaimed in a market watched continuously by automated participants, which is not what a genuinely free four cents would do. We describe the arithmetic because it explains how the venue works. We do not suggest anyone act on it, and the markets flagged on this site are flagged as worth a second look, never as instructions. How that flagging works is set out in how to tell if a Polymarket market looks mispriced and on the methodology page. The prices themselves, with their snapshot dates attached, are on the markets index.
Frequently asked questions
Why don't Polymarket prices add up to 100%?
In a binary market the displayed pair almost always does add up to 100, because the No price shown in the interface and returned by the public Gamma API is derived as one hundred minus the Yes price rather than quoted independently. Across all 400 markets in the Cent Signals snapshot generated on July 31, 2026, the Yes and No prices summed to exactly 1.0000 with no exceptions. The sums that genuinely drift away from a dollar live one layer down, in the separate order books for the two tokens, and in multi-outcome fields where several contracts share the same dollar.
Do Yes and No always sum to exactly one dollar on Polymarket?
The redemption value does, by contract. Polymarket outcome shares are issued through the Gnosis conditional token framework on Polygon, where a full set of one Yes share and one No share can be created from one USDC of collateral and merged back into one USDC at any time before resolution. That relationship is enforced by the smart contract, not by anyone's forecast, which is why the pair has a fixed anchor. Traded prices are a different quantity and can sit either side of it while a market is open.
Why does a Polymarket multi-outcome field sum to more or less than 100?
Because the field is many separate contracts on separate books rather than one contract with two halves. Each leg is quoted on its own, each has its own spread and its own depth, and low-probability legs are often stale. A field only carries the same arithmetic guarantee as a binary pair when the set of outcomes is exhaustive and mutually exclusive and every leg is visible, which is the structure Polymarket's negative risk markets are built to express. Our guide to negative risk on Polymarket covers the partial-field trap in detail.
How often do Polymarket prices actually deviate from a dollar?
More often than the interface suggests. A peer-reviewed study published at AFT 2025 by researchers at IMDEA Networks and the Oxford Internet Institute examined 86 million executed bids on markets that resolved between April 1, 2024 and April 1, 2025. Of the 17,200 conditions in that dataset, 7,051 showed at least one window where the volume-weighted sum of Yes and No prices sat more than two cents away from a dollar. Every single-condition instance they observed was on the low side, meaning the two halves together traded for less than the dollar they redeem for.
Does a gap in the sum mean a guaranteed profit?
No, and the same study is the clearest evidence of that. It estimated that only about 1 percent of the opportunities it detected around the 2024 US election were acted on at all. A quoted gap sits behind the bid-ask spread on each side, thin depth on long-tail legs, taker fees on many markets as of 2026, and the plain fact that filling two legs on an order book is not atomic, so one side can execute while the other moves. The authors describe this as non-atomic arbitrage carrying real execution risk.
Where can I see the Yes and No prices Cent Signals records?
Every market page on this site carries the Yes and No price captured in our most recent snapshot, together with volume, liquidity, and the one-day and one-week price change, and the markets index lists all 400 markets in the current snapshot. Each figure is stamped with the snapshot date, because these are point-in-time captures from Polymarket's public API rather than a live feed. The methodology page sets out exactly what is collected, how often, and what is deliberately left out.
Related reading
This explainer is editorial reference about public market data and published research. It is not financial advice, a tip, or a recommendation to take any position. Cent Signals does not facilitate trades, custody funds, or process payments. For how the Polymarket figures on this site are collected, see the methodology page.