Leverage & Collateral Efficiency in Prediction Markets

Problem statement

Polymarket's NegRiskAdapter shows the payoff of netting collateral across a declared mutually-exclusive claim set: roughly 9.5x capital efficiency versus fully separate positions. That's the easy case. The open problem is the general one — given a portfolio of claims with an arbitrary correlation or subsumption structure, not just clean partitions, how should a protocol size collateral? Open questions for the group: what's the right correlation primitive to condition margin on (pairwise correlation, a subsumption DAG, factor-model concentration); worst-case vs. risk-based/VaR-style margin, and where the right point on that trade-off sits for a permissionless market; whether the mechanism generalizes past declared-exclusive partitions to statistically-inferred correlation, or hits a hard identifiability wall for claims about a specific, never-repeated cluster; how short positions and price asymmetry change the margin calculation; and whether regulatory leverage-cap constraints should shape the design space from the start. The goal isn't one correct answer — it's stress-testing framings against people who've hit adjacent versions of this problem.

Submitted by Krzysztof Paruch (Token Engineering Labs)

Resources

• Polymarket NegRiskAdapter docs (the declared-exclusive-set reference case)
• Chen, Fortnow, Lambert, Pennock, Wortman — Complexity of Combinatorial Market Makers (EC 2008)

Sign up for this topic

Low-barrier — this just tells us you're interested so we know how many people to expect in this group at the kickoff. Team formation itself happens live, plus async in the coordination channel beforehand.

You're on the list for this topic. We'll see you in the coordination channel before the kickoff.

RSVP on Luma — link TBD Join the coordination channel