Cross-venue pricing programme

Our own programme — pricing one venue’s contract from another’s, and the overlay that quietly became the position

Sector: Quantitative research · Year: 2026

Summary

A prediction market lists a yes/no contract on where an asset will be at a date. A futures venue prices the asset itself, continuously. The two describe the same world and can be made to price each other. This programme built that bridge, hedged the resulting exposure on the futures side, and ran it on paper — until its first settlements produced a result that was more useful as a methodology finding than as a return.

Challenge

Two venues, two clocks, two liquidity regimes and two settlement conventions. Pricing a binary from a futures-derived barrier model is the easy half; the hard half is that the hedge which makes the position market-neutral carries its own modelling assumption, and that assumption is easy to inherit from a textbook without ever testing it against the instrument in front of you.

Approach

Priced each binary from a futures-derived barrier model rather than from the prediction market’s own history, so the two venues could be compared on a common basis instead of a contract being judged against itself.
Recorded both venues continuously to a common clock, and held the entire realised history in a form that could be re-scored afterwards under different assumptions.
Re-scored the settled book with the hedge removed — admissible only because the hedge routine has no path to influence which positions were opened, which was verified in the code before the counterfactual was quoted.
Refused the flattering reading of the result: eight positions, seven settling in the same second, with the gain and the hedge loss driven by a single directional move, do not evidence an edge in either direction.
Published the post-mortem with its own caveats attached, including three operational defects unrelated to the pricing thesis.

Outcome

The first settlements returned a methodology result rather than a performance one. The hedge overlay had been computed once at fill time and never rebalanced, on an instrument whose hedge ratio is violently non-stationary — so it stopped hedging within days and became a stale directional position, short into a 15–25% rally. No test or backtest had ever validated it; it was wired in because it looked obviously correct. The programme now treats an overlay as a strategy component that must pass the same validation as the strategy. Everything ran on paper, the live execution path was never configured, and no capital was committed at any stage.

Stack

Python, asyncio, WebSockets, NumPy, pandas

Published papers

The Overlay Nobody Tested — A hedge was added because it seemed obviously correct, never validated, and cost more than the book it was protecting made. 8 pages, PDF.