Scientific background

The study LabOfMetrics is built on. It compares six forecasting models (including the in-house Veto and Balance heuristics) and five betting strategies over 539 real matches, with 300,000 model–strategy simulations in total. In the study the Veto + Kelly combination came out best; our later walk-forward research over 50,000+ matches did not confirm this as a durable edge.

One of the best-known football-statistics papers, which introduces the Poisson-based goal model (with the joint distribution of home/away goal counts) and demonstrates the efficiency gaps of the betting market.

An Elo-based model for predicting football matches. It examines whether Elo ratings can beat the bookmakers, and under what conditions.

A foundational paper on evaluating probabilistic forecasts: calibration (how accurate the probabilities are) vs. sharpness (how “bold” they are). The theoretical background of the Brier Score and Log Loss metrics.

A machine-learning framework for predicting sports results: it presents the steps of the learning process and the application of methods (e.g. neural networks) in sports forecasting.