Can simple models predict football — and beat the odds? Lessons from the German Bundesliga

This study examines whether a simple model based on expected goals (xG) can generate accurate forecasts and identify profitable signals in football betting markets. The model uses recent xG to estimate win–draw–loss probabilities via a Skellam distribution, with isotonic regression applied for calibration, and evaluates the performance over eleven Bundesliga seasons (2014/15 through 2024/25). While bookmaker odds tend to exhibit superior statistical calibration, the xG-based model captures certain signals not fully reflected in market prices. In simulated betting, the model yields a return on investment of approximately 10% using average market odds, increasing to nearly 15% under the best available prices. Profits stem predominantly from home win bets, while backing away wins is consistently loss-making. The results show some robustness to modelling choices, though profitability varies considerably by season and bet type. The study highlights the potential of simple models as practical tools for identifying predictive value in structured football data.