Can Machine Learning Beat the Swedish Betting Market? A Statistical Analysis of Market Efficiency in Allsvenskan

Sammanfattning

In this thesis we examine whether inefficiencies exist in the Swedish football league Allsvenskan using a machine learning approach. While prior research on betting market efficiency has primarily been concentrated towards major European football leagues, smaller and less liquid markets such as Allsvenskan have received considerably less attention. Drawing from the size effect documented in financial markets, the thesis investigates whether lower analyst coverage and thinner betting volumes constitutes a foundation for identifying mispricings. We employ a Random Forest classifier which trains on historical data from Allsvenskan in the time frame 2018-2023 and predicts the outcome of matches in 2024-2025 using a rolling window approach. The models estimated probabilities are then compared with the bookmakers' implied probabilities to identify outcomes with positive expected value. The resulting betting strategy yields a 6.06% return over the test period, where the vast majority of the return stems from away-team wagers and underdogs. The findings constitute evidence against the semi-strong form of the efficient market hypothesis and are consistent with well documented behavioral patterns in football betting markets, such as the home bias.