what's the probability
it rains tomorrow?
would you bet money on it?
that bet is a prediction market.
a pot of money. two outcomes.
the split is the probability.
bet on a single US election on Polymarket
three days before polls closed,
markets called it correctly.
this guide teaches you how they work —
from an empty market to resolution.
A prediction market is a pot of money split between two outcomes — YES and NO. The ratio of the split IS the probability. Not a formula. Not a calculation. Just how much money is on each side.
YES shares pay $1.00 if the event happens. NO shares pay $1.00 if it doesn't. Their prices always sum to $1.00. The price IS the implied probability. If YES costs $0.62, the market says there's a 62% chance it happens.
If you genuinely believe an event has an 80% chance and YES shares cost $0.62, your expected value is (0.80 × $0.38) − (0.20 × $0.62) = +$0.18. Rational traders buy until the price reaches their true belief — that's why price converges to probability.
You don't lose a little. Your shares expire at exactly $0.00. But if you're right, you win exactly your edge — the gap between what you paid and $1.00. No partial outcomes. No averaging. This sharpens thinking in a way soft opinions never do.
real money beats pundit predictions. when politicos felt confident, money said toss-up.
an lmsr market maker. the more shares already bought on a side, the more each next share costs. price = belief.
The AMM uses a constant product formula — x · y = k — the same that powers token exchanges. When you “buy YES” you're trading against the pool: you bring collateral, the pool gives you YES tokens, and the price moves against you. You're not adding to a pot — you're taking a position against a curve.
Parimutuel: shared pot, odds shift until close. AMM: liquidity pool, price locked at trade. LMSR: deterministic formula, guaranteed fill, bounded loss.
Clicking “Buy YES” sends collateral to the pool and receives YES tokens. The pool size stays constant. Only the YES/NO ratio shifts — and that ratio is the price.
LPs deposit equal YES and NO tokens and earn 1–2% on every trade. But at resolution, one side of their holdings goes to zero. Their accumulated fees have to offset that loss. They're the silent counterparty to every trade — taking real directional risk to earn those fees.
In a token AMM, IL can reverse if prices revert. In a prediction market AMM, it never reverts — it resolves. The losing side expires at $0.00 permanently. An LP must earn enough in fees before the outcome is known to justify the position.
Once you understand slippage, sizing matters as much as being right. Kelly: divide expected profit per share by profit if you win. If you believe 80% and the market says 62%, Kelly says size at 0.18/0.38 ≈ 47% of bankroll. In practice, use half-Kelly to account for model uncertainty. Being right is necessary. Sizing correctly is what makes it profitable.
Market says 62%. After fees + slippage, you need 64.5% to break even. Your edge just shrunk.
every signal nudges the price. the market is a real-time poll of belief. an oracle settles it.
News breaks. Traders who understand the significance faster buy at better prices. The market price at any moment is the crowd's best estimate — but it's not a poll. Every opinion is backed by money.
In a poll, lying is free. In a prediction market, your opinion has a price tag. If you think X wins but bet the other way, you lose money. This “skin in the game” extracts honest beliefs in a way polls fundamentally cannot.
When you buy YES because you believe the market underestimates, your purchase moves the price up — visible to everyone. Your private knowledge has become a public signal. This is how distributed knowledge aggregates into a single number.
Polymarket had Trump at ~65% three days before the 2024 US election closed, while polling averages showed a near-tie within margin of error. The Iowa Electronic Markets, since 1988, have consistently outperformed polls on presidentials. When being wrong costs money, you think harder about being right.
If you're trading on Polymarket or Limitless, you're competing against professional quants, political analysts with insider access, and algorithms that parse news in milliseconds. The retail prediction trader is in the same position as the retail perp speculator: providing the liquidity that informed participants extract from. Trade only when you genuinely know something the market doesn't — not when you have an opinion.
Three ways the chain
touches reality.
Every prediction market depends on one thing outside the mechanism: someone — or something — that reads the real world and reports the outcome. This is the oracle. The point where onchain touches reality.
A single entity reads the result and settles all bets — Kalshi, a sportsbook, PredictIt. Fast and simple. But you're trusting them not to act in their own interest, freeze withdrawals, or make mistakes. Counterparty risk concentrated.
Used by Polymarket via UMA. Anyone can propose a resolution. If no one disputes within 48 hours, it's accepted. If disputed, token holders vote. The system assumes honesty by default and uses economic incentives to punish bad actors.
Chainlink and similar networks aggregate data from many independent nodes. No single point of failure. Widely used for price feeds. For prediction markets, the challenge is subjective outcomes — a price feed is objective, a political outcome requires interpretation.
Prediction markets fail when questions are ambiguous, outcomes contested, or the oracle corrupt. Good market design anticipates this: specific resolution criteria, objective sources, dispute mechanisms. The mechanism is only as good as the question it's pricing.
A $10K pool can be moved by a single $1K trade. Small markets are easily manipulated. Volume and depth are prerequisites for reliable pricing.
Capital is locked until resolution. A 2-year market ties up funds that could be deployed elsewhere. Discourages participation, thins the market, degrades accuracy. Best for near-term, high-salience events.
“Will the economy improve?” isn't a prediction market question. No objective oracle. Good markets need unambiguous, verifiable resolution criteria — who won, what price, which date.
Same question on Polymarket at 62% and Limitless at 58%. Buy YES on Limitless, sell YES on Polymarket. Lock in 4% regardless of outcome. The prediction-market carry trade.
Deposit equal YES and NO into the AMM. Earn 1–2% on every trade. Risk: at resolution, one side goes to $0. You need to earn more in fees than you lose on the position.
A 6-month resolution ties up capital for 6 months. A 5% edge annualises to 10% — compare to lending yields. This opportunity cost is why long-dated markets are chronically mispriced. Time is the variable most traders ignore.
what if every company ran an internal prediction market on whether their product would ship on time and executives couldn't ignore the answer?
what if scientists could bet on whether a study would replicate and the market odds told you more than the p-value?
what if insurance was just a prediction market you buy no on ‘my house will be fine this year’ and get paid if disaster strikes?
what if the probability of a pandemic, a rate hike, or a diplomatic breakthrough was priced every second and you could trade on your conviction?
what if a factory worker could hedge their own job loss by holding no on ‘the plant stays open this year’ — and get paid if it closes?
A YES share that pays $1.00 if an event occurs is structurally identical to a binary cash-or-nothing option. Traditional finance has had these for decades — regulated, standardised, cleared. Prediction markets didn't invent the payoff. They made it permissionless, composable, and global.
In a centralised market, your winnings are an IOU. The platform can freeze withdrawals or block your account. Onchain, your YES tokens are yours. Once the oracle confirms, the contract executes automatically — no human decides whether to pay.
Kalshi requires US residency and KYC. Bookies require local regulation. An onchain prediction market requires a wallet. Anyone, anywhere, on any verifiable question.
The market is the crowd, priced.
The oracle is the bridge.
What you bet on
is up to you.
10 questions. No going back.
Your result is shareable.