Prediction market arbitrage: how it actually works
Arbitrage is the most misused word in prediction markets. People use it to mean any position that feels safe, and screenshots of "guaranteed" edges circulate constantly. Real arbitrage does exist here, and the public, settle-to-truth nature of these markets makes it cleaner to reason about than in most places. But the edge is thinner than it looks, the costs are larger than beginners expect, and most of what gets called arbitrage is a data artifact that disappears the moment you try to trade it. This is the honest version: where the edge comes from, what eats it, and how to tell the real thing from the mirage.
What arbitrage means in a prediction market
A prediction market share pays exactly $1 if its outcome happens and $0 if it does not. That single fact is the foundation of every arbitrage here. If you can assemble a bundle of shares that is guaranteed to pay $1 at resolution no matter what happens, and buy that bundle for less than $1 all-in, you have locked a profit that does not depend on being right about the event. You are not predicting anything. You are buying a dollar for ninety-eight cents and waiting for it to settle.
The word for that bundle is delta-neutral: your profit and loss do not move with the outcome, because whichever side wins, one leg of the bundle pays and the other expires worthless, and the two were sized to cover each other. The whole game is getting the all-in cost of that dollar below a dollar after every fee and every spread, and keeping it there until the market resolves.
The two edges that are actually real
Cross-venue: the same market, two prices
The same event often trades on more than one platform. A Polymarket market and its mirror on a competing venue, or a Polymarket price and the same market surfaced through a terminal, can drift apart because each venue has its own order book and its own flow. When they disagree, you can buy the cheap side on one venue and buy the opposite outcome on the other, so the two positions together are guaranteed to pay $1. If the combined cost is under a dollar, the gap is your edge. This is the cleanest and most common form, and it is exactly the structure our own team studied when building tooling around venues like Polymarket and its BNB-chain fork.
Complementary outcomes: buying both sides under a dollar
Within a single binary market, YES and NO are complementary: one of them always pays $1. If a thin or dislocated book ever lets you buy YES and NO for a combined price below one dollar including fees, that is a locked profit on the same venue. It sounds too easy, and usually it is, because a well-run exchange prevents it and a negative-risk framework on multi-outcome markets closes the equivalent gap across a whole event. When it appears, it is almost always on a stale or barely-traded book, which is the exact place the number is least trustworthy.
The costs that quietly eat the edge
The reason arbitrage screenshots lie is that they price the edge before costs. Here is what stands between a two-cent gap and two cents of profit.
- Fees. Every venue takes a cut, and it applies on both legs. A platform charging a flat 0.5 percent per side turns a "two percent" gap into something closer to one. Cross two venues and you pay two fee schedules, not one.
- The spread. The price you see is usually the midpoint or the best quote for a tiny size. To actually fill, you cross the spread, and on a thin market the spread can be wider than the edge you were chasing. Taking liquidity is expensive precisely where arbitrage looks most tempting.
- Slippage and depth. A quote is good for the size resting at it. Try to put real money to work and you walk up the book, paying worse prices as you go. A gap that exists for ten dollars often does not exist for a thousand.
- Capital lock and time. A delta-neutral pair is only guaranteed at resolution. If the market settles in six months, your money is tied up for six months to earn that small edge once. Annualize it and a headline two percent can be a mediocre return, which is why the timing of resolution matters as much as the size of the gap.
- Gas and execution friction. On-chain venues add transaction costs and the risk that one leg fills while the other moves, leaving you briefly exposed on the exact position you were trying to neutralize.
Why most "arbitrage" is fake
When you scan for the biggest gaps, you will find eye-watering numbers: fifteen, twenty, forty percent. They are almost never real, and the tell is always the same. The book on one side is empty, with a stray bid parked at a penny so the math computes a huge discount that nobody can actually sell into. Volume is zero, the spread is enormous, and the market resolves far in the future. Every one of those is a symptom of a dead or stale book, not an opportunity. The genuine, tradeable edge lives in the boring middle: liquid markets with tight spreads where the two venues disagree by a fraction of a percent, deep enough to fill and close enough to resolution to recycle your capital. If a gap looks too good, the correct assumption is that you are misreading a book nobody is trading, not that you found free money the whole market missed.
SmartX is an independent AI trading terminal for prediction markets. It surfaces live Polymarket odds, ranks smart-money wallets, and lets you trade the same markets from one place, which is where you would spot a genuine cross-venue dislocation instead of a dead-book mirage. Fees are a flat 0.5 percent, and you fund your account in USDC.
Open SmartX →How real arbitrage is actually captured
The traders who make this work do a few unglamorous things consistently. They favor maker orders where they can, resting a limit at the best price to avoid paying the spread, because on a thin edge the difference between paying and earning the spread is the difference between profit and loss. They keep the position delta-neutral end to end, hedging the second leg immediately so they are never left holding a naked directional bet on the market they meant to arb. They prefer markets that resolve soon, so capital comes back quickly and the edge compounds instead of sitting locked for a year. And they size to the depth that actually exists, not the depth the top-of-book quote pretends to offer. None of it is exciting. All of it is the reason the edge survives contact with real costs. If you want the conceptual toolkit behind reading a price as a probability first, start with how to read prediction market prices.
Arbitrage versus hedging
These get confused constantly, so it is worth separating them. Arbitrage is opening a delta-neutral pair for a locked profit because two prices disagree. Hedging is neutralizing risk on a position you already hold, usually to protect a gain or cut exposure rather than to bank a guaranteed edge. They share the same delta-neutral machinery, but the motive is different: arbitrage is offense, hedging is defense. If the hedging side is what you actually need, the mechanics carry over directly to hedging a prediction market position.
Common mistakes
- Pricing the edge before costs. A gap is not a profit. Subtract both fees, the spread you will cross, and slippage at your intended size before you believe any number.
- Trusting a dead book. The largest gaps sit on markets with no volume and a penny bid. That is a broken quote, not an opportunity, and trying to fill it is how you learn the difference.
- Ignoring time to resolution. A small edge locked for a year is a weak return once annualized. Favor markets that settle soon so capital recycles.
- Leaving a leg naked. If one side fills and you delay the other, you are holding the exact directional risk you were trying to remove. Hedge the second leg immediately or do not open the first.
- Underestimating min-order and rounding frictions. Small residual amounts that fall below a venue's minimum can leave you with dust you cannot cleanly close, quietly turning a tidy pair into a lopsided one.
FAQ
Is prediction market arbitrage risk-free?
In theory a fully delta-neutral pair bought below a dollar all-in is risk-free at resolution, because one leg always pays $1. In practice the risks are execution and cost: fees and spread can erase the gap, one leg can fill while the other moves, a venue can have counterparty or resolution risk, and capital is locked until settlement. The event outcome is neutralized, but the trade around it is not free of risk.
Why do the biggest arbitrage gaps never seem tradeable?
Because they usually sit on dead books. A market with no volume, a huge spread, and a stray one-cent bid will compute an enormous discount that nobody can actually sell into. The real, tradeable edge is small and lives on liquid markets with tight spreads, deep enough to fill. A gap that looks too good is almost always a misread quote, not free money.
How much can you actually make on prediction market arbitrage?
After fees, spread, and slippage, genuine cross-venue edges are typically a fraction of a percent to a couple of percent per pair, and only on markets deep enough to size into. Because capital is locked until resolution, the return that matters is the edge divided by how long your money is tied up, which is why traders favor markets that settle soon.
What is the difference between arbitrage and hedging?
Arbitrage opens a delta-neutral pair to lock a profit because two prices disagree. Hedging neutralizes risk on a position you already hold, to protect a gain or reduce exposure rather than to bank a guaranteed edge. Same delta-neutral mechanics, different motive: arbitrage is offense, hedging is defense.
Arbitrage in prediction markets is real, but it is a discipline, not a shortcut. The edge is small, the costs are the whole story, and the flashiest opportunities are the least real. Read the book before you trust the gap, subtract every fee and spread, size to the depth that exists, and neutralize both legs the moment you open. Nothing here is financial advice, and prediction markets carry risk. Treated as a craft rather than a hack, arbitrage is one of the few honest edges these markets offer.