Title: Forecasting tennis betting odds by artificial neural networks
Authors: Vincenzo Candila - University of Salerno (Italy) [presenting]
Abstract: In sports literature, published betting odds are considered the most accurate source of probability forecasts. However, due to the presence of the longshot bias and bookmaker's over-round, these betting odds do not represent true bookmaker expectations about the outcome of the event under consideration. Artificial neural networks (ANNs) are employed to forecast betting odds in tennis betting market, starting from variables observable at the beginning of the matches. The ANNs are capable of handling a variety of input variables, contrary to standard approaches in the context of sport outcome forecasting. Moreover, the forecasted betting odds by ANNs are odds directly related to the probability of winning, such that there is no bookmaker's over-round and no longshot bias. In terms of probabilities, the proposed ANN model provides forecasting performances generally superior to the benchmarks considered. Moreover, forecasting betting odds by ANNs generally allows us to achieve positive returns. For robustness purposes, the analysis is repeated considering different datasets consisting of matches randomly selected.