Approach
Price the outcome. Respect the path. Survive the jump.
Prediction markets look simple: a contract pays $1 if something happens. Trading them well is not. Three disciplines do most of the work.
01 · Market making
Two-sided, always, including at the moments that matter.
We quote bids and offers across the full life of a contract, from the first minute after listing until the final print. Spreads are set by a model of uncertainty rather than by habit: wide where a single headline can move the price twenty points, tight where the outcome is grinding slowly toward certainty.
Inventory is managed actively. As we accumulate a position, quotes skew to attract the other side; when exposure to a single theme (an election, a central-bank meeting, a hurricane season) approaches its limit, hedges are placed in correlated contracts rather than simply pulling quotes. The result is liquidity that stays when it is most needed.
The economics are the economics of every good market maker: many small, uncorrelated edges, compounded with discipline. We go into detail in Market making in prediction markets.
02 · Systematic forecasting
A fair value for every question we touch.
Before we quote, we need a number. Each category has its own modelling tradition: nowcasts and release-surprise models for macro data, poll aggregation with house-effect correction and structural turnout models for elections, ensemble numerical weather prediction for temperature and storms, rating systems and live-state models for sport.
Those category models are combined with the information in the market itself: order flow, cross-venue prices and the term structure of related contracts. The market is often right; our models exist to know when it is not, and by how much.
Every model is scored the same way: out-of-sample, on resolved contracts, with Brier score, log loss and calibration error. Models that stop earning their place are retired.
| Category | Primary inputs | Share of quotes |
|---|---|---|
| Economics & rates | Release calendars, nowcasts, rates curves | 31% |
| Politics & policy | Polls, fundamentals, legislative calendars | 24% |
| Sport | Ratings, live game state, official data | 19% |
| Weather & climate | Ensemble NWP, station data | 14% |
| Culture, tech & other | Tracking data, structured web signals | 12% |
Approximate split of quoting activity by category, trailing twelve months.
03 · Cross-venue relative value
The same question is rarely the same contract.
"Will the Fed cut in December?" can trade on a regulated exchange, on an on-chain venue and implicitly in interest-rate futures, each with different resolution wording, fees, settlement timing and collateral. Naive arbitrage ignores those differences. We price them explicitly.
Our resolution-criteria library maps tens of thousands of contracts to normalised outcome definitions, flags edge cases (revisions, postponements, ties, disputed sources) and attaches a basis model to each pair. Where the residual difference is real, we trade it; where it is a rules mismatch, we leave it alone. See Cross-venue arbitrage: what is real and what is not.
Risk
Binary outcomes demand a different kind of risk management.
Per-event caps
Hard limits on net exposure to any single outcome, sized to the worst case rather than the expected one.
Theme correlation
Contracts are grouped into themes (a meeting, a race, a storm) and limited in aggregate, because outcomes cluster.
Jump protocols
Before scheduled announcements, quotes widen or pause by rule rather than by discretion, then return within milliseconds of the print.
Resolution review
Every contract is reviewed for ambiguous wording before we take meaningful size in it.
Questions
On our approach.
What kinds of event contracts do you trade?
Economic releases (CPI, payrolls, central bank decisions), elections and legislation, weather and climate, sport, entertainment and awards, technology and corporate milestones. We avoid contracts with ambiguous resolution or where we cannot form an independent view.
Do you trade on inside information?
No. Our models use public data and the order book only. We maintain strict information barriers and decline to trade categories where we, our staff or our partners could plausibly hold material non-public information.
How do you manage risk on binary outcomes?
By diversifying across thousands of weakly correlated outcomes, capping exposure per event and per theme, hedging correlated contracts against each other, and widening or pausing quotes into binary jump risk such as a scheduled announcement.