Developing a tennis model that reflects outcomes of tennis matches

Many tennis models that occur in the literature assume the probability of winning a point on service is constant. We show this assumption is invalid by forecasting outcomes of tennis matches played at the 2003 Australian Open. Revised models are formulated to better reflect the data. The revised models improve predictions overall, particularly for the length of matches and can be used for index betting. Suggestions on further improvements to the predictions are discussed.