Predicting Asian handicap outcomes of the German Bundesliga with artificial and recurrent neural networks

Abstract

Artificial and recurrent neural networks are increasingly being applied to solve problems in all sorts of industrial domains. This study aims to take a look at the theoretical foundations of artificial and recurrent neural networks as well as applying them to a reallife problem. In particular, different neural networks are applied to predict the outcome of football matches. For this purpose, we look at Germany’s Bundesliga for seasons 2013/14 to 2016/17. We used the first three seasons starting from August 2013 as our training set, and the last full season starting from August 2016 as our test set. The predicted number of goals for every team in every match of season 2016/17 is obtained by the neural networks under test. As a result, the predicted goal difference of a football match is compared with the handicap line given by five different bookmakers. A betting strategy applied to the Asian handicap market, enlightened by goal difference predictions is used to investigate the predictability and profitability of the models under test for the mentioned betting market. The performance of all models was analysed by means of three important results; rate of interest, percentage of bets won and percentage of bets lost. The conclusion reached was that an artificial neural network, and even more so a recurrent neural network approach, provided a profitable betting strategy and a great level of predictability accuracy.