Prediction markets are not physics experiments where passive particles bounce off each other. They are games: players choose actions under uncertainty, knowing others are doing the same, with payoffs tied to outcomes and prices.
Game theory is the toolkit for asking: Given what everyone wants, what behavior should we expect—and when does the system still produce useful odds?
This chapter installs the vocabulary used throughout Module 03. Later lessons apply it to manipulation, arbitrage, herds, news jumps, and honest reporting.
Players, strategies, and payoffs
Every market session involves:
- Players — retail takers, market makers, arbitrageurs, platforms, oracles, regulators, sometimes manipulators
- Strategies — buy YES, sell YES, quote liquidity, arb across venues, delay trading until news, spoof limits (illegal), report truthfully to an oracle
- Payoffs — P&L on contracts, fees earned, reputation, political influence, data sales, enforcement penalties
A strategy is a plan contingent on information and others’ actions. “Always buy my candidate” is a strategy; so is “quote 2¢ spread unless volatility doubles.”
Payoffs make prediction markets different from Twitter polls: beliefs have a price tag.
Information sets
What you know when you act defines your game:
Game theory calls this your information set. Markets price expected value given public information; edges live in better processing or private info (with legal boundaries).
Common knowledge — “everyone knows X, and everyone knows everyone knows X”—matters for information cascades later in this module. A headline on CNN becomes common knowledge in minutes; a Discord leak may not.
Simultaneous vs sequential games
Simultaneous — players move without seeing others’ current action. Sealed bids resemble this; so does a race to arb a stale quote before others.
Sequential — players observe prior moves then act. Stack:
- Market maker posts quotes
- Informed trader lifts ask
- Maker widens spread
- Arb bot sells on another venue
Extensive-form trees model sequential games; normal-form matrices model simultaneous choices. Both appear in market microstructure stories.
Dominant strategies and Nash equilibrium
A strategy is dominant if it is best regardless of what others do. Rare in forecasting markets except trivial cases (“never pay > $1 for a $1 binary”).
A Nash equilibrium is a profile where no player gains by unilaterally deviating. Example intuition:
- If everyone believes the crowd is accurate, quoting tight spreads as a market maker can be equilibrium—until informed flow arrives and makers widen spreads (new equilibrium).
Prediction markets often have many equilibria: same contract trades 55% or 62% depending on liquidity and narrative. Equilibrium is not “truth”; it is stability given incentives.
Zero-sum vs positive-sum framing
Pure zero-sum — one dollar lost by A is gained by B (before fees). Binary settlement is zero-sum among final holders of YES vs NO.
Positive-sum (society-level) — better forecasts improve decisions (policy, risk management). Negative-sum with fees, spreads, and failed manipulation—most active trading is negative-sum net of fees; participants buy information and entertainment.
Platforms are positive-sum on fees; LPs and takers fight over the surplus.
Incentives vs outcomes (mechanism design)
Mechanism design asks: What rules produce the behaviors we want?
Honest probability reports use proper scoring rules and LMSR-style subsidies. Deep liquidity uses maker rebates and hybrid automated market maker and central limit order book designs. Integrity uses surveillance, position limits, and KYC. Fast discovery uses continuous trading and news integrations. Manipulation resistance raises the cost of moving price and relies on arbitrage depth.
Module 03 is applied mechanism design for traders—reading which equilibria a rule set favors.
Prediction markets as repeated games
One-shot election contract vs career of a market maker across thousands of events changes behavior:
- Reputation — makers return daily; manipulators may burn accounts
- Learning — traders calibrate after Brier scoring the probability toolkit module
- Collusion harder when identities persist (KYC venues)
Folk theorem intuition: in repeated games, cooperation (honest quoting, not spoofing) can sustain if future payoffs matter. Anonymous wallets shorten the shadow of the future—more scam, more arb opportunity.
Risk preferences and utility
Game theory often assumes expected utility: players maximize probability-weighted payoffs. Real humans use:
- Loss aversion — hate losing $100 more than liking winning $100
- Probability weighting — overweight tails and near-certainty
- Kelly vs fractional Kelly — position sizing the probability toolkit module
Two traders with identical forecasts trade different sizes because utility differs, not because either is “irrational.”
Adverse selection (preview)
When your counterparty might know more, you protect yourself:
- Market makers widen spread after news
- LPs withdraw from AMM pools
- Retail stops trading the open
This is adverse selection—a game where accepting trades is dangerous. later chapters on manipulation and arbitrage show how arbs and costs push back.
What game theory does not do
It does not replace statistics, guarantee moral truth, or ignore law and compliance. It does not model every cognitive bias—behavioral finance fills gaps. It does clarify why pump-and-dump often fails, why arb bots exist, and why herds form after headlines.
Roadmap for the rest of this module
Later chapters cover manipulation as costly deviation, arbitrage as equilibrium pressure, cross-market and cross-contract consistency, structural coherence, cascades, news shocks, and reporting incentives at settlement.
The “if I were them” drill
Before trading a headline market, ask who profits if YES rips, who loses, what their next move is, and whether deviation stays profitable after fees. If not, a price spike may mean revert. That drill is game theory without Greek letters.
Markets are strategic interactions with information, timing, and payoffs—not lone numbers. Nash equilibrium explains stability, not moral truth. Repeated play and identity change manipulation and liquidity. Mechanism design links rules to forecast quality.
What comes next
Next: The cost of manipulation—why moving price is often a losing strategy.