example of a tennis match, we can describe a game, tiebreaker, set and match through
the graphs depicted in Figures 1,2,3,4. In our case, the Markov models operate on states
representing the current score in context of the game, set, tiebreaker or match. Given a
game, a transition between any two states is caused by one of the players scoring a point
in a rally during which she or her opponent were serving. Associated with the transitions
are appropriate probabilities of the players winning a point on serve and return. The
outcome of any of those models is the fact that the player has won or lost the given event.
This can be scaled up to other parts of the match as shown in the Figures mentioned.
Basing on the idea of hierarchical, stochastic Markov models, both O’Malley [2] as well
as Barnett [15] arrive with formulae for calculating the estimated probabilities of players
winning each stage of a tennis match. Those are presented in the subsequent subsections.
The idea of a tennis match as a Markov chain and modelling it as a Markov model gives
elegant, efficient and easy to automate solutions. Thanks to the work of Barnett [15] and
O’Malley [2] we can easily code up the calculations in the form of software and construct
prediction programmes requiring a minimal number of parameters to produce accurate
and reliable results