A common-opponent stochastic model for predicting the outcome of professional tennis matches

Tennis features among the most popular sports internationally, with professional matches played for 11 months of the year around the globe. The rise of the internet has stimulated a dramatic increase in tennis-related financial activity, much of which depends on quantitative models. This paper presents a hierarchical Markov model which yields a pre-play estimate of the probability of each player winning a professional singles tennis match. Crucially, the model provides a fair basis of comparison between players by analysing match statistics for opponents that both players have encountered in the past. Subsequently the model exploits elements of transitivity to compute the probability of each player winning a point on their serve, and hence the match. When evaluated using a data set of historical match statistics and bookmakers odds, the model yields a 3.8% return on investment over 2173 ATP matches played on a variety of surfaces during 2011.