An Application of Bayesian Inference and Simulation in Golf

Professional golf has long been regarded as a difficult sport to model. The volatility of ever changing conditions of play, such as weather, makes predicting a player's round score challenging. The relatively large playing field in any given tournament also makes the task of predicting each player's final position an ambitious one. A wealth of performance statistics correlated with success are available for elite tournaments in the United States. In this work, we propose a method for modelling professional golf tournaments using only historic round score data and round by round placings. Using the 2012 US Masters as a case study, we developed a methodology for predicting round scores for each player through possible scoring distributions. These scoring distributions were estimated using historic data from the competing players. A player's score was generated randomly from a distribution with the likelihood estimated from the player's observed historic score data. We simulated the tournament round by round, updating each likelihood using a Bayesian analysis of current score. Each simulation provides a measure of the probability of success for each player. These probabilities were seen to converge as the tournament was played out and each player's actual score became known. We validated the model's effectiveness by comparing the predicted outcomes with actual outcomes and those predicted from publicly released market prices.