Economia

Predicting sports events from past results
Towards effective betting on football matches
Douwe Buursma
University of Twente
P.O. Box 217, 7500AE Enschede
The Netherlands

[e-mail]
ABSTRACT
A s ystem for predicting the results of football matches that beats the bookmakers’ odds is presented. The predictions for the matches are based on previous results of the teams involved.

Keywords
Prediction, odds, bookmakers, football, probability, profit,
Weka, Bayesian networks, regression analysis.

1. INTRODUCTION
Betting on sports events has gained a lot of popularity recently and bookmakers and betting offices are now omnipresent at the internet. One of the most popular sports to bet on is football, because hundreds of football matches are played every week in professional competitions around the globe.

be possible to predict football matches even more accurately and therefore gain a higher profit by betting. Goddard [5], in
2004, did use a much bigger feature set for his football match predictions. This feature set includes the geographical distance between the teams involved and their average spectator attendances. However, the profit of their system was a lot lower than the one by Buchdahl. We believe the main reason for this is that they simply bet an amount of money on the most probable outcome for each match. Reasonable though this may seem, it is not too hard to understand that, because of bookmakers’ unfairness, it may sometimes be better not to bet on a match at all. In our investigation an attempt will be made to accurately choose which bets to take on in order to maximize the profit.

1.2 Research Questions

The method is always the same, the bettor wagers an amount of money on a bet of their choice. If their prediction turns out to be correct they win an amount of money, equal to the odds of the bet, per unit of money invested. For instance, if a bettor wagers
3 units of money on a bet with