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.
Add money to either side. Watch the probability shift. At resolution, the winning side takes the entire pot — the losing side gets nothing.
YES shares pay $1.00 if the event happens. NO shares pay $1.00 if it doesn't. Their prices always sum to exactly $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 of happening and YES shares cost $0.62, your expected value per share is (0.80 × $0.38) − (0.20 × $0.62) = +$0.18. A rational trader will 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.
Everyone bets into a shared pot. Pot grows with each entry. Winners split at resolution. Odds keep shifting until close — your payout depends on what others bet after you.
A liquidity pool holds YES and NO tokens. You trade against the pool. Price is locked at the moment you trade. The pool stays the same size — only its composition shifts. Larger trades cost more.
The Logarithmic Market Scoring Rule uses a mathematical formula where a market maker takes every trade at a deterministic price. Guaranteed liquidity, bounded loss. Used in academic and corporate forecasting.
The AMM uses a constant product formula — x · y = k — the same formula that powers decentralised token exchanges. When you “buy YES” you are trading against the pool: you bring collateral, the pool gives you YES tokens in return, and the price moves against you. You are not adding money to a pot — you are taking a position against a curve.
Select a trade size. Drag the liquidity slider. A $1K trade in a $2K pool eats massive slippage — the same trade in a $100K pool barely moves the needle. That gap is the entire lesson.
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. The “Pool depth” slider is what adds liquidity.
Liquidity providers (LPs) deposit equal YES and NO tokens and earn a fee on every trade — typically 1–2% of volume. But at resolution, one side of their holdings goes to zero. Their accumulated fees have to offset that loss. LPs are the silent counterparty to every trade — and they're taking real directional risk to earn those fees.
In a standard token AMM, liquidity providers face impermanent loss that 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. This is the LP's bet.
Once you understand slippage, sizing matters as much as being right. The Kelly Criterion gives the optimal fraction of bankroll to risk: divide your expected profit per share by your profit if you win. If you believe 80% and the market says 62%, expected profit is +$0.18 and profit if correct is $0.38 — Kelly says size at 0.18/0.38 ≈ 47% of bankroll. In practice, use half-Kelly to account for model uncertainty and the real cost of slippage on large positions. Being right is necessary. Sizing correctly is what makes it profitable.
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 in it is backed by money.
Click Next to advance through events. Watch how each headline moves the market. Notice that the magnitude of movement reflects how much money flowed in response.
In a poll, lying is free. You can say anything — no consequences. In a prediction market, your opinion has a price tag attached. If you think X wins but bet the other way, you lose money. This “skin in the game” is the mechanism that extracts honest beliefs from people in a way polls fundamentally cannot.
When you buy YES because you believe the market underestimates the probability, your purchase moves the price up — visible to everyone. Your private knowledge has become a public signal. If you're right, you profit at resolution; those on the wrong side lose their stake. This is how distributed, decentralised knowledge aggregates into a single number.
Polymarket had Trump winning at ~65% three days before the 2024 US election closed, while most major polling averages showed a near-tie within margin of error. The Iowa Electronic Markets, running since 1988, have consistently outperformed polls on presidential elections. When being wrong costs you money, you think harder about being right.
Every prediction market depends on one thing outside the mechanism itself: someone — or something — that reads the real world and reports the outcome. This is the oracle. It's the point where the onchain world touches reality. How it works determines whether the market is trustworthy.
A single entity reads the result and settles all bets — a regulated exchange like Kalshi, a sports book, or a platform like PredictIt. Fast and simple. But you're trusting them not to act in their own interest, freeze withdrawals, or make mistakes. The counterparty risk never goes away — it's just concentrated in one place.
Used by Polymarket (via UMA). Anyone can propose a resolution. If no one disputes it 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 in DeFi. 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 are contested, or the oracle is corrupt. A question like “Will X happen by the end of the year?” sounds clear — until X happens on December 31st and the settlement date is disputed. Good market design anticipates this: specific resolution criteria, objective sources, and dispute mechanisms. The mechanism is only as good as the question it's pricing.
A market with $10K in the pool can be moved significantly by a single $1K trade. Small markets are easily manipulated — a well-capitalised actor can push the price to a misleading level. Volume and depth are prerequisites for reliable pricing.
Capital is locked until resolution. A market on a 2-year question ties up funds that could be deployed elsewhere. This discourages participation, thins the market, and degrades price accuracy. Prediction markets work best for near-term, high-salience events.
"Will the economy improve by year-end?" is not a prediction market question. There is no objective oracle. Good markets require unambiguous, verifiable resolution criteria — who won, what price, which date — leaving no room for interpretation.
Prediction markets have existed since the 1990s. The Iowa Electronic Markets have run since 1988. What does putting them onchain actually add?
In a centralised market, your winnings are an IOU. The platform can freeze withdrawals, go bankrupt, or block your account. In an onchain market, your YES tokens are yours. Once the oracle confirms the outcome, the contract executes automatically — no human decides whether to pay. The oracle can still be disputed (see above), but the settlement mechanism itself cannot be blocked or altered.
Kalshi requires US residency and KYC. Bookies require local regulation. An onchain prediction market requires a wallet. Anyone anywhere can create a market on any question with a verifiable answer — no licence, no approval, no geographic restriction. This is a meaningfully different world than what came before.
A YES share that pays $1.00 if an event occurs is structurally identical to a binary cash-or-nothing option. This payoff structure has existed in traditional finance for decades — exchange-traded binary options are standardised instruments with regulated brokers and clearing houses. Prediction markets didn't invent this. They made it permissionless, composable, and global.
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 exactly a binary call option. Finance has known this instrument for 50 years.
Pay a premium. Receive $1 or $0 based on whether a condition is met — asset above a price, rate above a threshold, default occurring. Exchange-traded binary options are standardised instruments available through regulated brokers with central clearing.
Same payoff structure. Two-sided market where buyers and sellers determine the price. The market price IS the probability — no broker sets the premium. Anyone can create a market on any verifiable question.
Full collateralisation (no leverage, no counterparty), automatic settlement via smart contract, composability (YES tokens can be used in other DeFi protocols), and censorship resistance.
A prediction market is a truth machine.
Not because it's always right —
but because being wrong is expensive.
Markets don't have opinions. They have prices. Prices aggregate the knowledge of every participant who has skin in the game. They update in real time as the world changes. And unlike polls, forecasts, or expert panels — they cost you something to be wrong.