How Prediction Markets Work

What Are Prediction Markets?

A prediction market is a contract that pays out based on the outcome of a future event. Participants buy shares representing possible outcomes, and the price of each share reflects the market's collective estimate of the probability that outcome will occur.

The core mechanic is elegantly simple. Take a binary question: "Will Candidate X win the election?" The market issues two types of shares — YES and NO. YES shares pay $1 if Candidate X wins; NO shares pay $1 if they don't. If YES shares are currently trading at $0.62, the market is saying: there is approximately a 62% probability Candidate X wins.

This is not just convention — it is the economic logic of the system. If you believe the true probability is 75% but YES shares trade at $0.62, buying YES at $0.62 gives you positive expected value. Enough traders acting on this discrepancy will push the price toward the true probability. This is prediction markets' fundamental claim to accuracy: prices aggregate the private information of every participant, weighted by their willingness to stake money on their belief.

Prediction markets have existed for decades. The Iowa Electronic Markets (IEM), run by the University of Iowa, has been forecasting US elections since 1988 with remarkable accuracy. The modern era brought crypto-native platforms like Polymarket (2020), which combined permissionless market creation, stablecoin settlement, and decentralised resolution — achieving over $1 billion in monthly volume by 2024.

How They Work

Prediction markets use one of two mechanisms to set prices: automated market makers (AMMs) or order books. The choice affects liquidity, price discovery, and how easy it is to enter and exit positions.

Automated Market Makers (AMMs) are the dominant mechanism on crypto prediction markets. The most common variant is the Constant Product Market Maker (CPMM) — the same mechanism underlying Uniswap. The market holds a pool of YES and NO shares. When you buy YES shares, you add NO shares to the pool and remove YES shares, causing the YES price to rise and the NO price to fall by the same amount. The price at any moment is determined by the ratio of shares in the pool.

An earlier design is the Logarithmic Market Scoring Rule (LMSR), used by Augur and academic prediction markets. In LMSR, a market maker (who may be the platform itself) provides liquidity across all prices. The market maker subsidises liquidity but captures fees. LMSR guarantees that you can always buy or sell at a price, even for obscure events — but it requires upfront capital to bootstrap.

Order books work like traditional financial exchanges: buyers post bids and sellers post asks; trades execute when they match. This is how Kalshi operates. Order books provide precise price discovery in liquid markets but suffer from wide spreads and poor execution in illiquid ones.

Probability & Price

In a well-functioning prediction market, price and probability are the same thing. A YES share at $0.70 does not just imply 70% probability — for a risk-neutral, rational agent, buying YES is exactly equivalent to betting at 70% odds.

In practice, prices deviate slightly from true probabilities due to:

The key signal to watch is volume. A market with $50 at stake is nearly meaningless — one opinionated person can set any price. A market with $10 million at stake has forced many independent actors with skin in the game to reveal their beliefs. Polymarket's most liquid election markets in 2024 had hundreds of millions of dollars in volume — generating some of the most accurate forward-looking probability estimates available anywhere.

Market Resolution

Resolution is the hardest part of prediction markets. Someone or something must determine what actually happened — and must do so in a way that participants trust and cannot easily manipulate.

There are three resolution models in use today:

1. 1
   Centralised oracle (most common)

   A single trusted entity reads the real-world outcome and reports it on-chain. Polymarket uses UMA's Optimistic Oracle: a proposed outcome is posted, and anyone can dispute it within a challenge window. If undisputed, it resolves automatically. If disputed, UMA token holders vote. Fast, cheap, and usually accurate — but relies on the oracle's honesty and competence.

2. 2
   Decentralised token-holder vote (Augur model)

   REP (Reputation) token holders report and dispute outcomes. Reporters stake REP on their answer; the minority in a dispute loses their stake. This creates financial incentives to report truthfully without relying on any single trusted party. Slower (7+ day dispute windows) but genuinely censorship-resistant.

3. 3
   Automated on-chain resolution

   For events with verifiable on-chain data (ETH price above $X, BTC halving date, a specific transaction occurring), smart contracts can resolve markets automatically using Chainlink or similar oracle networks — no human judgment required. Most reliable, but only works for clearly quantifiable, on-chain-verifiable outcomes.

4. 4
   Centralised platform resolution (Kalshi, PredictIt)

   The platform itself determines outcomes, guided by its resolution rules. Simplest operationally, but introduces platform counterparty risk. Mitigated when the platform is a regulated entity subject to CFTC oversight.

When resolution is genuinely ambiguous — the event was cancelled, the question was poorly worded, or the outcome is contested — most markets resolve as "N/A" (invalid), returning all stakes to participants minus trading fees already paid. Clear resolution criteria in the market description are essential to avoid disputes.

Liquidity & Incentives

Who provides the other side of every trade, and why? Prediction markets only work if someone is willing to take opposing positions — which requires incentives beyond the desire to speculate.

Liquidity providers (LPs) deposit capital into YES/NO pools and earn a fee (typically 1–2%) on every trade that moves through the pool. For high-volume markets this can be substantial. The risk LPs face is similar to impermanent loss in DeFi: if the market resolves strongly in one direction, LPs who held the losing side lose more than they earned in fees. LPs must be compensated for this risk through fees.

Market creators seed initial liquidity and often earn a slice of trading fees as compensation. On Polymarket, anyone can create a market — the creator sets the initial pool and resolution source. On Kalshi, markets are curated by the platform.

Arbitrageurs are perhaps the most important participants: they buy underpriced shares (relative to their private information or cross-market prices) and sell overpriced ones. Their activity is what makes prices converge to true probabilities. An active arbitrage community is a sign of a healthy market.

Platforms & Protocols

The prediction market landscape splits into two camps: crypto-native platforms offering permissionless markets and stablecoin settlement, and regulated platforms that US residents can legally use.

The dominant platform by volume is Polymarket, which processed over $3.5 billion in trading volume during the 2024 US election cycle alone. Its combination of USDC settlement, low fees (~1% spread), and UMA oracle resolution created a genuinely liquid market — prices on Polymarket were consistently better calibrated than most mainstream polling aggregators.

Kalshi is the most significant development in regulated prediction markets. After winning a lawsuit against the CFTC in 2024 that affirmed its right to list election contracts, Kalshi opened the door for prediction markets to operate openly with US retail investors under CFTC oversight — a model that could eventually scale to rival sports betting in terms of legal clarity and consumer access.

Accuracy & Track Record

Do prediction markets actually work as forecasting tools? The evidence, accumulated over three decades, is strong — though not without caveats.

The research consensus is that prediction markets are well-calibrated — when markets say 70%, the event happens roughly 70% of the time, across many different events. They are not perfectly accurate on any single prediction, but they are less systematically biased than polls, expert panels, or quantitative models.

Limitations are real: thin markets are easily manipulated and generate noisy signals. Correlated events (e.g., election outcomes and economic outcomes) can cause cascading mispricing. And prediction markets cannot predict genuine black swan events that no participant has information about.

Legal & Regulatory Landscape

Prediction markets occupy a legally ambiguous zone between gambling, derivatives trading, and information markets — and regulators around the world have treated them very differently.

The Kalshi precedent is the most important recent development. By winning its legal argument that election contracts are federally regulated derivatives — not gambling — Kalshi established that prediction markets can operate openly in the US under CFTC jurisdiction. This regulatory clarity should, over time, attract more capital, better liquidity, and wider adoption than is possible under the current patchwork of offshore crypto platforms blocking US users.

The fundamental regulatory question remains: at what point does a prediction market become socially harmful? Critics argue that markets on political outcomes could incentivise participants to manipulate the outcome they bet on (e.g., betting on a candidate and then working to elect them). Proponents argue this risk is small relative to the forecasting value, and that existing markets (betting, derivatives) create similar incentives without generating systemic harm.

Why They Matter

Prediction markets are one of the most powerful information aggregation tools ever designed — and they are still in the early stages of their potential applications.

The deeper significance of prediction markets is philosophical: they are a mechanism for converting private information into public knowledge. Every participant in a prediction market has some private information — a conversation they had, a dataset they analysed, an intuition from experience. The financial incentive to bet on that information causes it to flow into the price, where everyone can observe it.

In a world awash in opinion, misinformation, and motivated reasoning, a well-functioning prediction market produces something rare: a probability estimate that has been tested against the financial interests of many independent actors who had every incentive to get it right.