A Markov process approach to untangling intention versus execution in tennis

Abstract Value functions are used in sports to determine the optimal action players should employ. However, most literature implicitly assumes that can perform prescribed with known and fixed probability of success. The effect varying this or, equivalently, “execution error” implementing an (e.g., hitting a tennis ball specific location on court) design strategies, has received limited attention. In paper, we develop novel modeling framework based Markov reward processes decision investigate how execution error impacts player’s value function strategy tennis. We power our models hundreds millions simulated shots 3D 2D player tracking data. find shot selection strategies become more conservative as grows, having perfect empirical is roughly equivalent choosing one or two average error. backhand costly than forehand shots, serve return valuable any other shot, over all values