Headlines scream “prediction market manipulated!” when odds lurch on thin volume. Game theory asks a colder question: Was manipulation a winning strategy ex ante—and for whom?
This chapter models manipulation as a costly move in a game, explains why pump-and-dump often fails on real contracts, and when attacks still work.
Manipulation as a strategic move
Define manipulation (practical, not legal): trading or quoting intended to move displayed price away from your true belief to influence others, settlement-adjacent bets, or media narrative—not because you want exposure at that price.
Manipulator’s possible payoffs:
- Profit on pre-positioned shares when price reverts or when others buy your exit liquidity
- Political signaling — 80% YES headline shapes coverage (may accept trading loss)
- Hedging elsewhere — move public odds while real hedge is off-platform
- Attention / fraud — pump related tokens, referral traffic
Legal manipulation (spoofing, insider trading) carries enforcement payoffs—negative. This chapter is economics; consult lawyers for compliance.
The cost function of moving price
Moving implied probability from 50% → 80% requires buying YES (or removing NO liquidity). Cost includes:
Manipulation budget = sum of layers. On a $50k-depth election market, budget might be six figures for a durable 20-point move. On a $800 pool, budget might be $200—which is why small markets are fragile.
Pump-and-dump structure (extensive form)
Typical story:
- Accumulate YES cheaply (stealth).
- Pump — aggressive buys move price up.
- Outsiders see “market says 75%” and buy YES (narrative cascade).
- Dump — sell into new demand at higher prices.
For this to work:
- Step 3 must happen (others must trade your way)
- Step 4 must exceed step 2 costs + inventory left behind
- No arb bot sells YES on another venue at old price
On liquid political markets, step 3 is weak—sophisticated watchers know the move is size-driven; step 4 faces arbitrage .
Why arbs break pumps
Suppose manipulator pushes Polymarket-style YES to 75% while Kalshi-style contract still 58%.
Arbitrageur game:
- Buy YES where 58%, sell or buy NO where 75% (venue-dependent)
- Lock risk-free edge if resolution rules match
Manipulator becomes donor to arb P&L. Pump stops when marginal cost > arb gap.
Key inequality for sustainable pump:
Manipulation cost ≥ Cross-venue arb depth × gap
Election week gaps close in seconds when rules align.
Thin pool vs thick book
Spoofing and layering (order book games)
Manipulator posts large fake bids to scare sellers, cancels before fill—classic spoofing (illegal on regulated markets). Game:
- If surveillance low → short-lived illusion
- If surveillance high → criminal payoff dominates
Prediction markets copying exchange surveillance reduce this equilibrium.
Wash trading
Self-trades fake volume and last price. Payoff: attention, referral bonuses. Cost: fees + gas. Platforms filter graph clusters; still happens on weak venues.
Trader lesson: volume without depth is cheap talk.
Signaling manipulation (non-profit motive)
Campaigns may want headline odds more than trading profit. Payoff is media, not settlement. Budget can be advertising spend in disguise.
Detect:
- Price spike without cross-venue alignment
- Spike on non-betting audience platforms citing “markets”
- Revert after news cycle while fundamentals unchanged
Defense mechanisms (mechanism design)
Equilibrium: when manipulation persists
Manipulation equilibria survive when:
- Arb blocked — different resolution, capital controls, no second venue
- Audience is captive — readers never see arb-free reference
- Profit is off-market — crypto token tied to “sentiment index”
- One-shot — attacker exits platform forever after scam
Otherwise honest trading + arb is cheaper than sustaining fake odds.
Case patterns (hypothetical composites)
Failed pump — Wallet buys $12k YES on obscure prop; price 40% → 68%; within 20 minutes arb + makers push composite back to 44%; wallet underwater.
Successful narrative spike — $3k on $4k pool moves UI to 91%; cable cites “prediction market”; no arb venue exists; price mean-reverts slowly after attention fades; manipulator sold $2k into spike—small profit, large publicity.
Insider trade — Private info, not pump; price moves and stays because information was real; prosecutors later allege crime—different game.
What comes next
Next: Arbitrage as the market’s self-correction mechanism.