Get Tokenomics

Prediction Markets: AMM, CLOB, and Betting Tokenomics

How prediction markets work: AMM vs CLOB comparison, LMSR pricing formula, betting fairness problem, and Sybil arbitrage.

Prediction markets are mechanisms that turn participant opinions into probabilities. Participants buy contracts on event outcomes, and the market price of a contract reflects the collective probability estimate. In 2024–2025, prediction markets experienced explosive growth: Polymarket processed roughly $3.7B on the single “US presidential winner” market and about $9B cumulative platform volume in 2024, with ~314,000 active traders (The Block; ChainCatcher). The platform runs hundreds of concurrent event markets at any given time. Growth was not monotonic — DLNews reported an ~84% drop in volume immediately after the November 2024 election, followed by a gradual recovery around sports, crypto, and macro events. This drawdown pattern is critical context for anyone designing tokenomics around event-driven platforms.

But behind the elegant idea lie serious tokenomics questions: how to set prices when liquidity is thin? Which mechanism is fairer — AMM or order book? How to defend against Sybil attacks?

How a Prediction Market Works

Core Mechanics

Each market is a binary (or multi-outcome) contract:

  • “Yes” is priced from $0.01 to $0.99
  • Price = participants’ probability estimate
  • When the event occurs, the “Yes” contract settles at $1.00, “No” at $0.00
P_outcome = Contract_price / Face_value
  • Contract_price — current trading price
  • Face_value — settlement value ($1.00)
  • P_outcome — implied probability (computed)
  • A contract price of $0.65 → the market estimates a 65% probability

Three Pricing Approaches

ApproachMechanismExamples
AMM (LMSR)Logarithmic cost function sets priceGnosis, Zeitgeist, XO.Market (LS-LMSR)
Odds-driven vAMMVirtual AMM with oracle-sourced odds and a singleton liquidity poolAzuro (LiquidityTree + vAMM)
AMM (Uniswap-style)Constant-product pool per outcomePolkamarkets
CLOBCentral Limit Order Book — participants set prices themselvesPolymarket, Kalshi, Limitless Exchange
HybridOff-chain order book + on-chain settlementSX Bet
AMM vs CLOB for prediction marketsAMM (LMSR) — formula, fixed liquidity. CLOB — participants, adaptive liquidityAMM (LMSR)formula determines pricevsCLOBparticipants determine pricefixed liquidityliquidityadaptive liquidityexponential scalingscalinglinear scalingPolymarket, Kalshi, SX Bet

AMM in Prediction Markets

LMSR — Logarithmic Market Scoring Rule

The classic AMM formula for prediction markets, proposed by Robin Hanson:

C(q) = b · ln(Σᵢ exp(qᵢ / b))
  • qᵢ — number of contracts sold for outcome i
  • n — number of outcomes; the sum runs over i = 1..n
  • b — liquidity parameter (higher b = lower slippage)
  • C(q) — cost function determining the current price (computed)

Price of a contract for a specific outcome:

pᵢ = exp(qᵢ / b) / Σⱼ exp(qⱼ / b)
  • pᵢ — current contract price (= probability estimate, computed)
  • The sum in the denominator runs over j = 1..n outcomes
  • Σᵢ pᵢ = 1 (probabilities are normalized)

LMSR Problems

Fixed liquidity. The parameter b determines the market maker’s maximum loss. If b is too small — slippage is enormous on small volumes. If b is large — the market maker incurs heavy losses. There’s no dynamic adjustment to changing demand.

Inefficiency on volatile markets. LMSR was designed for prediction markets with gradual probability refinement. When markets move sharply (election results, sports events), the formula can’t adapt fast enough, and arbitrageurs extract profit from the market maker.

Scaling. For multi-outcome markets (10+ options), LMSR requires exponentially more liquidity. For a market with 20 outcomes, the b parameter needs to increase proportionally.

AMM is losing to CLOB
Among the largest prediction market platforms, only a few use an AMM-family design: Polkamarkets (Uniswap-style AMM), XO.Market (LS-LMSR), and Azuro (odds-driven vAMM with LiquidityTree — not a classical LMSR). The majority — Polymarket, Kalshi, SX Bet, Limitless Exchange — run on CLOB or hybrid models. This signals systemic limitations of the AMM approach: insufficient pricing flexibility and market maker losses on volatile markets.

CLOB — Central Limit Order Book

Mechanics

CLOB works like traditional exchanges:

  1. A participant places a limit order: “Buy ‘Yes’ at $0.60”
  2. Another participant places a matching order: “Sell ‘Yes’ at $0.60”
  3. Orders match, the trade executes

Advantages Over AMM

CriterionAMM (LMSR)CLOB
SlippageDetermined by parameter b, fixedDepends on order book depth, adaptive
PricingFormula, lags the marketParticipants set price in real time
LiquidityLimited by initial market maker capitalSourced from all participants
Multi-outcomeExponential scaling requirementsLinear scaling
ArbitrageMarket maker bears lossesLosses distributed among counterparties

Why CLOB Won

Polymarket — the largest prediction market by volume — uses CLOB based on the CTF Exchange protocol. The reasons:

  1. Capital efficiency. Market makers place limit orders with managed risk, rather than pouring liquidity into a pool with unpredictable losses.

  2. Price accuracy. On election markets where probabilities shift 1–2% per day, CLOB participants reflect information more precisely than the LMSR formula with its lag.

  3. Composability. CLOB is compatible with professional trading strategies: market-making, cross-platform arbitrage, delta hedging.

The Fairness Problem: Normalized Volatility

On prediction markets, a fairness issue arises: outcomes with different volatility have different attractiveness for speculators.

Example

Two markets:

  • “Candidate A wins the election” — price fluctuates from $0.45 to $0.55 (low volatility)
  • “Bitcoin above $100K by June” — price fluctuates from $0.20 to $0.80 (high volatility)

A trader earns more on the second market with the same capital — not because they forecast better, but because the market is more volatile.

Normalized Volatility Formula

V_norm = Sigma_p / (P_avg × (1 − P_avg))
  • Sigma_p — standard deviation of contract price
  • P_avg — average contract price over the period
  • P_avg × (1 − P_avg) — maximum possible variance for a Bernoulli distribution
  • V_norm — normalized volatility, enables cross-market comparison (computed)

Normalized volatility enables comparing markets with different base probabilities and adjusting fees or margin requirements accordingly.

Sybil Arbitrage

Attack Mechanics

Sybil arbitrage — creating multiple accounts to exploit bonuses or subsidies on a prediction market:

  1. The platform subsidizes liquidity or offers bonuses for first bets
  2. The attacker creates N accounts
  3. Half the accounts bet “Yes,” the other half bet “No”
  4. One outcome wins, and the attacker collects bonuses from the winning accounts
Profit = N × Bonus − Fee × N × Bet_size
  • N — number of Sybil accounts (each makes one bet and receives one bonus)
  • Bonus — bonus per account
  • Fee — platform fee rate
  • Bet_size — bet amount per account
  • Profit — attack profit (computed); attack is profitable if Bonus > Fee × Bet_size
Ungated theoretical maximum
The formula above assumes zero capital cost, instant withdrawal, and no identity gate. Real deployments layer on minimum-volume requirements, cooldown periods, withdrawal fees, and KYC / proof-of-humanity checks — each of which reduces realized attacker profit and shifts the break-even. Treat the formula as an upper bound on attack economics, not a forecast.

Defenses

MethodMechanismWeakness
KYCDocument verificationReduces anonymity, loses crypto-native audience
Proof of HumanityBiometrics or social graphCostly, false positives
Deposit collateralMinimum stake to participateIncreases barrier to entry
Behavioral analysisAlgorithms detecting coordinated actionsArms race

Platform Tokenomics

Prediction market platforms use different tokenomic models:

Polymarket: No Governance Token, But a Platform Stablecoin

Polymarket operates without a governance token. Historically, bets were placed in USDC with minimal fees, and monetization relied on trading volume and data.

In April 2026, Polymarket announced CTF Exchange V2, introducing Polymarket USD — a platform-native, 1:1 USDC-backed stablecoin — together with Managed Optimistic Oracle V2 and a 37-address resolver whitelist (Crypto Times; Polymarket docs). The platform still has no freely traded governance token, but it is no longer “USDC-only”: users interact through a wrapped platform stablecoin that sits on top of USDC.

Advantage: no governance-token friction; users stay in stablecoins. Disadvantage: no on-chain governance instrument; value capture is indirect (volume and data).

Azuro: Token for Liquidity

Azuro uses its AZUR token for:

  • Staking in liquidity pools
  • Governance voting on market parameters
  • Incentivizing data providers (oracles)

On the mechanism side, Azuro is an odds-driven virtual AMM: data providers push odds to the protocol, a singleton LiquidityTree pool absorbs bets across many markets, and LP profit comes from the spread around the oracle odds — not from an LMSR cost function (Azuro Medium, “Introducing Azuro’s Liquidity Tree”; Messari, “Understanding Azuro”). Classifying Azuro as “AMM (LMSR)” is a common but incorrect shorthand.

Advantage: liquidity providers are motivated by the token, not just fees. Disadvantage: an additional asset increases complexity and risk for users; pricing quality depends on data-provider honesty.

When AMM, When CLOB

Choosing the pricing mechanism

  • AMM (LMSR): low-liquidity markets, launch phase, guaranteed liquidity needed without professional market makers
  • AMM (LMSR): multiple markets with predictable volatility (sports, fixed outcomes)
  • CLOB: high-liquidity markets with professional participants
  • CLOB: markets with variable volatility (politics, macroeconomics)
  • Hybrid: cold start on AMM with transition to CLOB as volume grows
  • Prediction markets and mechanism design

    Prediction markets are a special case of mechanism design. Pricing formulas, incentive systems, and attack defenses are built on the same principles as any tokenomic system.

    Mechanism design