Prediction Markets

01 · Concept — what problem does it solve?

How do you turn scattered opinions about an uncertain future event — who wins an election, whether a protocol gets exploited, whether a macro print beats consensus — into a single, tradeable number? A prediction market lets anyone buy and sell shares tied to a yes/no outcome; the share's price is the market's collective probability estimate. Unlike a poll, participants back their opinion with capital, which tends to make the price more accurate than punditry — the same "wisdom of crowds" logic that makes order books good at pricing stocks now applies to discrete real-world events.

02 · Mechanics

• Outcome shares: a binary market has two complementary tokens, YES and NO. Depositing $1 of collateral mints one YES + one NO (a "complete set"). At resolution, the winning share redeems for $1 and the losing share for $0.
• Split / merge: anyone can split $1 into 1 YES + 1 NO, or merge 1 YES + 1 NO back into $1, at any time before resolution. This is a hard arbitrage constraint: price(YES) + price(NO) can never drift far from $1, or arbitrageurs mint/merge to capture the gap.
• Resolution: someone has to report what actually happened. On-chain venues typically use an optimistic oracle (e.g. UMA) — a proposer posts the outcome, a dispute window opens, and unresolved disputes escalate to a token-holder vote — or a simpler admin/multisig resolver for lower-stakes markets. This is the step with no on-chain source of truth to fall back on, unlike an oracle price feed.
• Order types:
  • Market order — fill immediately at the best resting price; guarantees a fill, not a probability. Takes liquidity, same as any order book.
  • Limit order — post a resting order at a chosen probability, e.g. "buy YES at 34¢" (34% implied odds); it only fills if the market trades down to meet it. Provides liquidity while you wait.
  • Range order — newer designs are experimenting with committing collateral across a probability band (e.g. 30¢–45¢) instead of one price, auto-quoting resting bids and asks throughout that range. This is the prediction-market analogue of Uniswap v3 concentrated liquidity: you earn the spread on trades that occur inside your range, but if the price exits the range your position stops earning and needs manual rebalancing.
  • AMM / LMSR alternative: some designs (early Gnosis Predict) skip the order book entirely and price against a market-scoring-rule curve, which always quotes both sides but charges more price impact on size than a deep order book would. Augur, by contrast, is order-book based (relayer-matched limit orders), not LMSR.

03 · Formulas

// no-arbitrage constraint between complementary shares
price(YES) + price(NO) ≈ $1

// price *is* the implied probability
price(YES) = P(YES resolves true)

// payout at resolution
winning share → $1        losing share → $0

// LMSR cost function (Hanson's market scoring rule) for outcome shares q_i
C(q) = b · ln( Σ e^(q_i / b) )
price_i = e^(q_i / b) / Σ e^(q_j / b)
   // b = liquidity parameter — larger b means deeper book, smaller price impact per trade

04 · Edge cases & risks

• Resolution disputes: ambiguously worded questions or contested real-world outcomes can leave a market frozen for days while an optimistic-oracle dispute plays out — capital is locked with no price discovery in the meantime.
• Thin long-tail markets: niche questions often have wide spreads and shallow books; limit and range orders can sit unfilled indefinitely, and a market order can move price sharply on modest size.
• Manipulation near resolution: because the price is read as "the market's odds," a large trade just before close can distort the perceived probability even if it doesn't change the underlying event — a risk for anyone citing the price as a signal rather than just trading it.
• Regulatory status varies by jurisdiction: event-contract venues have faced different regulatory treatment (e.g. CFTC actions, geofencing, or a path through registered exchanges) — availability and legal status differ by country and by venue.
• No arbitrage against reality: a mispriced perp can be arbitraged against spot instantly; a mispriced prediction market can only be corrected once the real-world event resolves, so the price can stay "wrong" for the market's entire duration.