Prediction Market Math
APPLIED PROBABILITY INSTITUTE // QUANTITATIVE GLOSSARY

Prediction Market Math Glossary

Authoritative mathematical definitions, closed-form equations, and worked case examples for 35 foundational concepts in binary event derivatives and market microstructures.

Contract Pricing & AMM Mechanics

(10)

Implied Probability

#
P_{implied} = P_m / \sum_{j=1}^n P_j

The probability of an event occurrence discounted by the market clearing price of an Arrow-Debreu contract share that pays $1.00 upon favorable resolution.

Worked Case Example: A Polymarket Yes share trading at $0.63 denotes a market-implied event probability of exactly 63.0%.

Arrow-Debreu Security

#
\Phi(S_i) = \mathbf{1}_{\{S = S_i\}} \cdot \$1.00

A state-contingent financial derivative that pays exactly one unit of currency if a specific state of the world occurs and zero otherwise.

Worked Case Example: Binary prediction market shares are pure Arrow-Debreu assets paying $1.00 on truth and $0.00 on falsehood.

Risk-Neutral Measure (Q-Measure)

#
P_t = e^{-r(T-t)} \mathbb{E}^{\mathbb{Q}}[\Phi(S) \mid \mathcal{F}_t]

A mathematical probability measure under which the discounted asset price process behaves as a martingale, equating price to expected payoff.

Worked Case Example: In binary prediction contracts with negligible time to expiry, the Q-expectation simplifies directly to P_t = Q(Event = True).

Constant Product Market Maker (CPMM)

#
x \cdot y = k, \quad p_{Yes} = \frac{y}{x+y}

An automated market maker algorithm maintaining invariant collateral reserves x and y such that their product remains constant, setting price via reserve ratios.

Worked Case Example: Polymarket binary liquidity pools utilize CPMM so that marginal Yes and No prices deterministically sum to 1.00 at all points.

Logarithmic Market Scoring Rule (LMSR)

#
C(\mathbf{q}) = b \ln\left(\sum_{i=1}^n e^{q_i / b}\right), \quad p_i = \frac{e^{q_i / b}}{\sum e^{q_j / b}}

Robin Hanson's bounded-loss scoring rule for combinatorial prediction markets where price is the softmax of outstanding contract shares scaled by liquidity b.

Worked Case Example: LMSR guarantees infinite liquidity across n combinatorial outcomes while bounding the market organizer's maximum loss to b * ln(n).

Overround (Vigorish / Vig)

#
\text{Overround} = \sum_{i=1}^n \frac{1}{d_i} - 1.000

The theoretical percentage margin by which the sum of bookmaker implied probabilities exceeds 100%, representing operator profit extraction.

Worked Case Example: A sportsbook offering Yes at 1.85 and No at 1.85 has raw sum 54.05% + 54.05% = 108.10%, extracting an 8.10% overround.

Shin's Information-Theoretic Model

#
\pi_i = \frac{\sqrt{z^2 + 4(1-z) \frac{P_i^2}{\sum P_k}} - z}{2(1-z)}

A mathematical normalization framework proving bookmakers skew margins toward longshots to protect against informed insider traders.

Worked Case Example: Shin normalization recovers true objective probabilities by solving for insider fraction parameter z from market price distributions.

Bonding Curve

#
P(x) = f(x), \quad \text{Cost} = \int_0^{\Delta x} f(x) \, dx

A deterministic mathematical curve establishing the continuous functional relationship between contract token supply and marginal price.

Worked Case Example: AMM bonding curves enforce deterministic pricing without requiring centralized order-matching servers.

Discrete Expected Value (EV)

#
\text{EV} = S \cdot \left[ \frac{p_{true} - P_m}{P_m} \right]

The probability-weighted mean payoff of an event contract minus initial capital commitment, measuring mathematical trading edge.

Worked Case Example: Buying $0.40 shares with a true 50% probability yields an expected return of (0.50 - 0.40) / 0.40 = +25.0% EV per dollar risked.

Price Impact & Execution Slippage

#
P_{avg} = P_{spot} \cdot \left[ 1 + \frac{\Delta c}{2L} + \mathcal{O}\left(\frac{\Delta c^2}{L^2}\right) \right]

The adverse shift in average execution price caused by consuming available liquidity depth along an AMM bonding curve or order book.

Worked Case Example: Deploying $50,000 into a $200,000 liquidity pool pushes the effective entry price from 50¢ to 56¢, degrading net trading edge.

Cross-Venue Arbitrage & Spreads

(7)

Synthetic Cross-Market Arbitrage

#
P_A(Yes) + P_B(No) < 1.000 - \gamma_{total}

A risk-free trading strategy executing simultaneous complementary positions across fragmented venues when the sum of purchase prices is strictly under $1.00.

Worked Case Example: Buying Yes at $0.46 on Venue A and No at $0.48 on Venue B guarantees a riskless gross profit of +7.53% upon event resolution.

Dual-Leg Capital Sizing

#
S_A = C \cdot \frac{P_A}{P_A + P_B}, \quad S_B = C \cdot \frac{P_B}{P_A + P_B}

Closed-form capital allocation formulas ensuring invariant dollar payout across both resolution states in a cross-market arbitrage trade.

Worked Case Example: For $10,000 capital across Yes at $0.40 and No at $0.55, allocate $4,210.53 to Leg A and $5,789.47 to Leg B to guarantee identical returns.

Negative Synthetic Spread

#
\text{Spread} = 1.000 - [ P_A(Yes) + P_B(No) ] > 0

A transient market inefficiency where buying the full set of exhaustive outcome shares costs less than the guaranteed $1.00 redemption settlement.

Worked Case Example: A spread of +3.5% indicates an investor can capture $1.035 in guaranteed cash for every $1.000 of combined capital deployed.

Legging-In Risk

#
\Delta P_{\text{leg}} = P_B(t + \Delta t) - P_B(t)

The operational vulnerability where price movements or latency delays occur between executing the first leg and confirming the second leg of an arbitrage trade.

Worked Case Example: A 10-second blockchain confirmation delay on Leg A can allow Leg B to move from 48¢ to 56¢, turning a +4% arbitrage into an unhedged loss.

Fee Hurdle Rate

#
\gamma_{total} = \phi_{taker, A} + \phi_{taker, B} + \frac{\text{Gas}}{C} + \phi_{cashout}

The minimum gross spread threshold required to cover all exchange commissions, network gas fees, and capital rebalancing costs.

Worked Case Example: If exchange fees are 2.0% and gas costs absorb 0.8%, an arbitrage spread must exceed 2.8% to deliver positive net alpha.

Fill-or-Kill (FOK) Order

#
\text{Order Condition} = \{ \text{Fill} \text{ if } V_{available} \ge V_{order} \text{ else } \text{Cancel} \}

An order execution parameter requiring the immediate total fulfillment of an order or its immediate total cancellation to prevent partial execution.

Worked Case Example: Arbitrage algorithms mandate FOK execution on off-chain order books to avoid partial fills that leave capital exposed.

Law of One Price

#
P_A(E) = P_B(E) \quad \forall E \in \Omega

An economic theorem stating that identical financial assets with identical future payoffs must clear at identical prices in efficient markets.

Worked Case Example: Violations of the Law of One Price between decentralized and centralized prediction venues create systematic arbitrage opportunities.

Kelly Sizing & Capital Preservation

(7)

Binary Kelly Criterion

#
f^* = \frac{p - P_m}{1 - P_m}

The information-theoretic formula determining the optimal fraction of bankroll to wager on binary prediction contracts to maximize geometric wealth growth.

Worked Case Example: When a $0.40 contract has a true 55% probability, optimal Kelly fraction is (0.55 - 0.40) / (1 - 0.40) = 0.15 / 0.60 = 25.0% of bankroll.

Fractional Kelly (Half / Quarter Sizing)

#
f_{\text{frac}} = c \cdot f^*, \quad c \in [0.25, 0.50]

A risk-mitigation practice scaling down full Kelly fractions by factor c to safeguard against model estimation errors and severe drawdowns.

Worked Case Example: Half Kelly (c = 0.50) achieves 75% of maximum growth rate while slashing portfolio return variance by 50%.

Logarithmic Utility

#
U(W) = \ln(W), \quad \mathbb{E}[U(W_T)] = \max_f \sum p_i \ln(1 + f \cdot R_i)

A concave von Neumann-Morgenstern utility function reflecting constant relative risk aversion (CRRA = 1), preventing overbetting ruin.

Worked Case Example: Log utility places an infinite penalty on bankruptcy (ln(0) = -inf), mathematically barring strategies with positive ruin probability.

Volatility Drag

#
R_{\text{geom}} \approx R_{\text{arith}} - \frac{\sigma^2}{2}

The mathematical reduction in compound geometric return caused by portfolio variance, proving that high arithmetic gains can yield negative compound growth.

Worked Case Example: A +50% gain followed by a -50% loss yields an arithmetic mean of 0%, but a geometric reality of -25% capital erosion.

The Longshot Fade Trap (Polymarket Fade Trap)

#
\text{Lock} = \$0.95, \quad \text{Yield} = +5.26\%, \quad \text{Tail Loss} = -100\%

The capital inefficiency trap of purchasing No shares on 3¢–5¢ contracts, locking up vast capital for meager yields while absorbing massive black swan risk.

Worked Case Example: Buying No at $0.95 risks $10,000 to win $526; a single upset erases the cumulative profits of 19 consecutive wins.

Drawdown Probability Formulation

#
P(\text{Drawdown} \ge D) = (1 - D)^{2c - 1}

The probability that an investment account will experience a peak-to-trough decline of at least fraction D before doubling capital under fractional Kelly c.

Worked Case Example: Under full Kelly (c = 1.0), the probability of a 50% bankroll drawdown is (1 - 0.50)^1 = 50.0%, whereas under Half Kelly (c = 0.5) it is heavily suppressed.

Portfolio Covariance Sizing

#
\max_{\mathbf{f}} \left[ \mathbf{f}^T \boldsymbol{\mu} - \frac{1}{2} \mathbf{f}^T \boldsymbol{\Sigma} \mathbf{f} \right] \quad \text{s.t.} \quad \sum f_i \le 1.0

Multivariate capital allocation optimization across correlated binary event markets incorporating covariance matrix Sigma to prevent simultaneous drawdowns.

Worked Case Example: Electoral outcomes across swing states exhibit high positive correlation (rho > 0.70), requiring downscaled simultaneous Kelly allocations.

Brier Scoring & Empirical Calibration

(8)

Brier Score

#
\text{BS} = \frac{1}{N} \sum_{t=1}^N (f_t - o_t)^2

A strictly proper quadratic scoring rule measuring the mean squared error between probabilistic forecasts f_t in [0, 1] and binary realizations o_t in {0, 1}.

Worked Case Example: A perfect forecaster scores 0.000, while a totally uninformative 50/50 baseline coin toss scores 0.250.

Murphy Decomposition

#
\text{BS} = \text{Reliability} - \text{Resolution} + \text{Uncertainty}

An algebraic identity partitioning the aggregate Brier Score into three orthogonal components: Reliability (calibration error), Resolution, and Uncertainty.

Worked Case Example: Decomposing prediction market scores proves whether misforecasting stems from market overconfidence or inherent domain entropy.

Reliability (Calibration Error)

#
\text{Reliability} = \frac{1}{N} \sum_{k=1}^K n_k (f_k - o_k)^2

The component of the Murphy decomposition measuring how closely observed frequencies match forecast probabilities across discrete probability bins.

Worked Case Example: In a well-calibrated market, events priced at 70% resolve positively 70% of the time, resulting in near-zero Reliability penalty.

Resolution (Discriminatory Power)

#
\text{Resolution} = \frac{1}{N} \sum_{k=1}^K n_k (o_k - \bar{o})^2

The component measuring a forecasting mechanism's ability to sort outcomes into distinct subsets whose frequencies differ from the base rate.

Worked Case Example: Higher Resolution reduces the Brier Score, indicating the market makes decisive, informative probability assignments.

Uncertainty (Base-Rate Entropy)

#
\text{Uncertainty} = \bar{o}(1 - \bar{o})

The irreducible environmental variance of the event sample space, peaking at 0.250 when historical base occurrence rate is exactly 50%.

Worked Case Example: Uncertainty depends exclusively on natural event frequencies and cannot be mitigated by forecaster skill.

Favorite-Longshot Bias (FLB)

#
P_m(E_{\text{rare}}) > p_{\text{true}}(E_{\text{rare}}), \quad P_m(E_{\text{fav}}) < p_{\text{true}}(E_{\text{fav}})

An empirical pricing distortion where low-probability outcomes trade at substantial premiums while high-probability favorites are underpriced.

Worked Case Example: Contracts trading at 3¢ resolve positively only 1.2% of the time, while 92¢ contracts resolve positively 95.8% of the time.

Log-Loss (Cross-Entropy)

#
\text{LogLoss} = -\frac{1}{N} \sum_{t=1}^N [ o_t \ln(f_t) + (1 - o_t) \ln(1 - f_t) ]

An information-theoretic proper scoring rule that heavily penalizes overconfident incorrect predictions with asymptotic infinity.

Worked Case Example: Pricing a contract at 99.9% that ultimately fails incurs a severe log-loss penalty of -ln(0.001) = 6.908.

Kullback-Leibler (KL) Divergence

#
D_{KL}(P \parallel Q) = \sum_{x} P(x) \ln\left( \frac{P(x)}{Q(x)} \right)

The measure of information entropy lost when approximating the true probability distribution P with market belief distribution Q.

Worked Case Example: Minimizing KL divergence between proprietary forecasts and market prices directly maximizes asymptotic Kelly capital compounding.

Oracles, Settlement & Risk Mechanics

(3)

Optimistic Oracle Architecture

#
\text{State} = \{ \text{Proposed} \to \text{Dispute Window } \Delta t \to \text{Finalized} \}

A game-theoretic mechanism where proposed event outcomes are assumed truthful unless challenged with a collateral bond during a challenge period.

Worked Case Example: UMA protocol allows anyone to propose a market resolution bond, which finalizes in 2 hours if uncontested.

UMA Tokenholder Dispute Voting

#
\text{Voting Result} = \text{Mode}\left( \{ v_i \cdot w_i \} \right), \quad w_i = \text{Token Weight}

The Schelling-point consensus protocol where tokenholders vote on disputed resolutions, rewarding voters aligned with the majority consensus.

Worked Case Example: When a Polymarket resolution is disputed, UMA tokenholders cast commit-reveal votes to establish final canonical ground truth.

Settlement Semantic Ambiguity

#
\mathcal{S}(E) = \{ s \in \Omega \mid \text{Text}(E) \text{ is ungrounded} \}

The systemic risk occurring when event resolution rules contain linguistic or boundary edge cases not anticipated in the market specification.

Worked Case Example: An election market asking 'Will Candidate X resign before Date Y?' creates ambiguity if the candidate announces resignation effective Date Y+1.