2026
prediction market pricing engine
quantitative research and paper trading
a research and paper trading setup for polymarket combination markets. i use it to price a combination, decide how to quote it, and then check that decision against what really happened in the market.
- python
- numpy
- scipy
- fastapi
- timescaledb
- redis
- next.js
illustrative model and market series
pricing outcomes together
pricing uses soccer score-grid models, normal approximations for other sports and gaussian copulas for dependent events. when outcomes are related, the combination price reflects that, and probability uncertainty is carried through the math.
from events to a usable market history
market and on-chain events are ingested asynchronously into postgres and timescaledb. redis streams link the services, fastapi exposes the backend and a next.js dashboard shows prices, positions and simulation results.
replaying history
to evaluate it i replay historical executions through the pricer and the quoting policy. paper execution tracks inventory and exposure, and i check calibration and uncertainty over several historical windows. prometheus, grafana and structured logs cover monitoring.
research scope
everything is historical replay or paper execution. the simulated results are research outputs and aren't real trading returns.
the thread through my work
the model turned out to be only part of the work; the pipeline, risk rules and evaluation took as much.
development history, from 2019