The Market Making Book

10. Polymarket: On-Chain Prediction Markets

A hybrid CLOB settling in USDC on Polygon — where the house pays you a daily salary just for resting good quotes.

Part III · Chapter 10
Instrument
Outcome tokens (ERC-1155, CTF)
Collateral
pUSD on Polygon (1:1 USDC)
Book
Off-chain CLOB, on-chain settle
Maker fee
Never charged
Taker fee
rate·p·(1−p), by category
Resolution
UMA Optimistic Oracle

The plumbing

Every YES/NO pair is fully collateralized: exactly $1.00 of collateral sits locked in Gnosis's Conditional Tokens Framework per share pair, so YES + NO = $1 is enforced by a smart contract, not by trust. (Since the April 28, 2026 exchange upgrade — CTF Exchange V2 — that collateral is pUSD, Polymarket's own dollar token, itself backed 1:1 by USDC with the backing enforced on-chain.) Taker fees follow the same parabola shape as Kalshi's, with the rate set per category — roughly: crypto 0.07, sports 0.03, politics/finance 0.04, most others 0.05, geopolitics fee-free — and makers are never charged. Resolution flows through UMA's optimistic oracle: a bonded proposer asserts the outcome, a ~2-hour liveness window allows bonded disputes, and contested cases escalate to a multi-day UMA token-holder vote. Since November 2025, proposing is no longer permissionless — a whitelist of vetted proposers (Managed Optimistic Oracle V2) asserts outcomes, though anyone can still dispute. This is the venue's distinctive tail risk: resolution risk — ambiguous wording or a contentious dispute can settle a market against the "obvious" answer (over a thousand markets were disputed in the first half of 2026 alone). Price it into every position you carry to resolution.

And the US question: since December 2, 2025 there are effectively two Polymarkets — the global crypto-native venue this chapter describes, and Polymarket US, a CFTC-regulated exchange (built on the QCEX license acquired for $112M) with intermediated access for US users. The Kalshi chapter's "US legal" edge is no longer unique; check which venue, and which rulebook, you're actually quoting on.

The liquidity rewards program: getting paid to exist

Polymarket's signature mechanism: a daily per-market budget, divided among makers by a per-minute sampling score. The score rewards orders by a quadratic function of closeness to the midpoint — an order 1¢ from mid scores dramatically more than one 3¢ away — and rewards two-sidedness: with the midpoint inside the 10–90¢ band, quoting only one side is worth exactly a third (the score divides the single side by 3); outside that band, single-sided quoting scores zero and only genuinely two-sided depth still earns. Crucially, you earn whether or not you're filled. April 2026 alone distributed over $5M in incentives, plus a maker-rebate program recycling 20–25% of taker fees back to makers (25% standard, 20% on crypto markets), paid daily.

interactive — reward score vs. distance from mid
Relative score
Reading
The score S ∝ ((max−d)/max)² collapses quadratically with distance d from the mid and hits zero outside the max scoring spread. The economic tension of the whole venue in one curve: the closer you quote, the more salary you collect — and the more adverse selection you absorb. Proposal C (Chapter 18) optimizes exactly this trade-off.
Polymarket MM doctrineTreat rewards as a yield on parked two-sided depth. Farm calm, long-dated markets (low jump risk, full salary); in news-driven markets, quote at the distance where reward ≥ expected adverse selection, not a tick closer. Respect resolution risk as a position limit, not a footnote.

On this page

GitHubGitHub repository