A Markov process approach to untangling intention versus execution in tennis

Timothy C. Y. Chan, Douglas Fearing, Craig Fernandes, Stephanie Kovalchik

Abstract

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

Introduction

In sports analytics, determining the "optimal" move often relies on value functions that assume a deterministic relationship between a player's decision and the resulting outcome. However, tennis is characterized by significant physical variability. A player may intend to hit a winner down the line but, due to technical inconsistency, land the ball wide. This "execution error" means that a strategy optimal for a player with perfect precision may be suboptimal for one with average consistency. This study introduces a Markovian framework to decouple a player's intention (where they aim) from their execution (where the ball lands), allowing for the analysis of how technical skill reshapes strategic decision-making.

Methodology

The researchers developed a dual-modeling approach using Markov Reward Processes (MRP) for policy evaluation and Markov Decision Processes (MDP) for policy optimization. To generate the necessary transition probabilities, the study utilized the VON CRAMM framework, a generative model trained on 125,000 men’s and 80,000 women’s shots from the Australian Open. This framework "stitches" realistic 3D ball and 2D player trajectories to simulate over 75 million shots.

The tennis court was discretized into 84 cells to define the state space. The action space was defined by 39 aiming regions for rally shots and 6 for serves, as shown in the figure below.

Action set discretization defining the possible aiming regions (intentions) for the MDP model.

Execution error was modeled by fitting bivariate Gaussian distributions to landing locations, controlled by a scaling parameter $\epsilon$. Through calibration with real-world win percentages, $\epsilon=13$ was identified as the representative value for an average professional player.

Results and Discussion

The primary finding is the "conservative shift": as execution error ($\epsilon$) increases, the optimal distribution of shot intentions migrates toward the center of the court. This shift provides a safety buffer, reducing the risk of unforced errors at the cost of less aggressive placement.

Heatmap illustrating the distribution of optimal shot intentions across the court.

Key insights from the model include:

  • Wing Volatility: Execution error on backhand shots is more detrimental to win probability than equivalent error on forehands, confirming the backhand as a more volatile wing.
  • Strategic Leverage: The serve return was identified as the most critical shot to optimize. For an average player, choosing the optimal return target can increase point win probability from 0.575 to 0.704.
  • Skill vs. Strategy Trade-off: Interestingly, the model suggests that a player with average execution error can achieve win probabilities comparable to a "perfect" player simply by optimizing the targets of one or two key shots in a rally.

Conclusion

By explicitly modeling execution error, this research provides a more realistic lens for tennis strategy. The results suggest that "playing within oneself" by adjusting targets toward the center is a mathematically sound response to technical inconsistency. Future research could enhance the model by incorporating ball speed and spin, distinguishing between first and second serves, and modeling the interaction between two dynamic, optimizing opponents.

Summary updated

Abstract

Abstract Value functions are used in sports to determine the optimal action players should employ. However, most literature implicitly assumes that players can perform the prescribed action with known and fixed probability of success. The effect of varying this probability or, equivalently, “execution error” in implementing an action (e.g., hitting a tennis ball to a specific location on the court) on the design of optimal strategies, has received limited attention. In this paper, we develop a novel modeling framework based on Markov reward processes and Markov decision processes to investigate how execution error impacts a player’s value function and strategy in tennis. We power our models with hundreds of millions of simulated tennis shots with 3D ball and 2D player tracking data. We find that optimal shot selection strategies in tennis become more conservative as execution error grows, and that having perfect execution with the empirical shot selection strategy is roughly equivalent to choosing one or two optimal shots with average execution error. We find that execution error on backhand shots is more costly than on forehand shots, and that optimal shot selection on a serve return is more valuable than on any other shot, over all values of execution error.
BibTeX
@article{chan2022markov,
  author = {Timothy C. Y. Chan and Douglas Fearing and Craig Fernandes and Stephanie Kovalchik},
  title = {{A Markov process approach to untangling intention versus execution in tennis}},
  journal = {Journal of Quantitative Analysis in Sports},
  year = {2022},
  doi = {10.1515/jqas-2021-0077},
}
Figure