Football Market Intelligence

Betting-market modelling & backtest

What this is

An honest negative result. The custom models do not beat the market, and it says so. Built with walk-forward CV, isotonic calibration and explicit feature-leakage and temporal-split tests.

Metrics

Best config

+6.72% ROI over 118 bets (Market LR, Pinnacle)

Reality

Most configs negative; zero bets in 2023/24 on a structural break

Testing

80 tests incl. leakage and temporal-split

Stack

DuckDBWalk-forward CVIsotonic calibrationLeakage tests

What I would not claim

xgboost
statsmodels
plotly
any positive market edge