When Data Lies: Finding Optimal Strategies for Penalty Kicks with Game Theory

Introduction

Penalties are amongst probably the most decisive and high-pressure moments in football. A single kick, with only the goalkeeper to beat, can determine the consequence of a whole match or perhaps a championship. From a knowledge science perspective, they provide something much more interesting: a uniquely controlled environment for studying decision-making under strategic uncertainty.

Unlike open play, penalty kicks feature a set distance, a single kicker, one goalkeeper, and a limited set of clearly defined actions. This simplicity makes them an excellent setting for understanding how data and strategy interact.

Suppose we wish to reply a seemingly easy query:

At first glance, taking a look at historical data appears to be sufficient to reply this query. As we are going to see, nevertheless, relying solely on raw statistics can result in misleading conclusions. When outcomes rely upon strategic interactions, optimal decisions can’t be inferred from averages alone.

By the tip of this text, we are going to see why probably the most successful technique to kick a penalty will not be the one suggested by raw data, how game theory explains this apparent paradox, and the way similar reasoning applies to many real-world problems involving competition and strategic behavior.

The Pitfall of Raw Conversion Rates

Imagine getting access to a dataset containing many historical observations of penalty kicks. A natural first quantity we’d consider measuring is the scoring rate related to each shooting direction.

Suppose we discover that penalties aimed toward the middle are converted more often than those aimed toward the edges. The conclusion might sound obvious: kickers should all the time aim at the middle.

The hidden assumption behind this reasoning is that the goalkeeper’s behavior stays unchanged. In point of fact, nevertheless, . They’re strategic interactions through which each players constantly adapt to one another.

If kickers suddenly began aiming centrally each time, goalkeepers would quickly respond by staying in the center more often. The historical success rate of center shots subsequently reflects past strategic behavior fairly than the intrinsic superiority of that alternative.

Hence, the issue will not be about identifying the very best motion in isolation, but about In game theory, this balance is often known as a Nash equilibrium.

Formalizing Penalties as a Zero-Sum Game

Penalty kicks can naturally be modeled as a two-player zero-sum game. Each the kicker and the goalkeeper have to concurrently select a direction. To maintain things easy, allow us to assume they simply have three alternatives:

• Left (L)
• Center (C)
• Right (R)

In making their alternative, .

If $P$ denotes the probability of scoring, then the kicker’s payoff is $P$, while the goalkeeper’s payoff is $-P$. The payoff, nevertheless, will not be a set constant, because it is dependent upon the combined alternative of each players. We will represent the payoff as a matrix:

$P= begin{bmatrix} P_{LL} & P_{LC} & P_{LR} P_{CL} & P_{CC} & P_{CR} P_{RL} & P_{RC} & P_{RR} end{bmatrix}$,

where each elements $P_{ij}$ represents the probability of scoring if the kicker chooses direction $i$ and the goalkeeper chooses direction $j$.

Later we are going to estimate these probabilities from past data, but first allow us to construct some intuition on the issue using a simplified model.

A Toy Model

To define an easy yet reasonable model for the payoff matrix, we assume that:

• If the kicker and the goalkeeper select different directions, the result’s all the time a goal ($P_{ij}=1$ for $ine j$).
• If each select center, the shot is all the time saved by the goalkeeper ($P_{CC}=0$).
• If each selected the identical side, a goal is scored $60$ of the times ($P_{LL}=P_{RR}=0.6$).

This yields the next payoff matrix:

$P= begin{bmatrix} 0.6 & 1 & 1 1 & 0 & 1 1 & 1 & 0.6 end{bmatrix}$.

Equilibrium Strategies

It is straightforward to grasp that having a set strategy, i.e. all the time making the identical alternative, can’t be optimal. If a kicker all the time aimed in the identical direction, the goalkeeper could exploit this predictability immediately. Likewise, a goalkeeper who all the time dives the identical way could be easy to defeat.

So as , which is what in game theory known as having a mixed strategy.

A mixed strategy is described by a vector, whose elements are the chances of constructing a specific alternative. Let’s denote the kicker’s mixed strategy as

$p = (p_L, p_C, p_R)$,

and the goalkeeper’s mixed strategy as

$q = (q_L, q_C, q_R)$.

Equilibrium is reached when neither player can improve their consequence by unilaterally changing their strategy. On this context, it implies that kickers must randomize their shots in a way that makes goalkeepers indifferent to diving left, right, or staying center. If one direction offered the next expected save rate, goalkeepers would exploit it, forcing kickers to regulate.

Using the payoff matrix defined earlier, we are able to compute the expected scoring probability for each possible alternative of the goalkeeper:

• if the goalkeeper dives left, the expected scoring probability is:

$V_L = 0.6 p_L + p_C +p_R$

• if the goalkeeper stays in the middle:

$V_C = p_L +p_R$

• if the goalkeeper dives right:

$V_R = p_L + p_C + 0.6 p_R$

For the strategy of the kicker to be an equilibrium strategy, we’d like to search out $p_L$, $p_C$, $p_R$ such that for goalkeepers the probability of conceding a goal doesn’t change with their alternative, i.e. we’d like that

$V_L = V_C = V_R$,

which, along with the normalization condition of the strategy

$p_L+p_C+p_R=1$,

gives a linear system of three equations. By solving this method, we discover that the equilibrium strategy for the kicker is

$p^* simeq (0.417, 0.166, 0.417)$.

Interestingly, though central shots are the simplest to save lots of when anticipated, shooting centrally about $16.6$ of the times makes all options equally effective.

Now that we’re armed with the knowledge of game theory and Nash equilibrium, we are able to finally turn to real-world data and test whether skilled players behave optimally.

Learning from Real-World Data

We analyze an open dataset (CC0 license) containing 103 penalty kicks from the 2016-2017 English Premier League season. For every penalty, the dataset records the direction of the shot, the direction chosen by the goalkeeper, and the ultimate consequence.

By exploring the information, we discover that the general scoring rate of a penalty is roughly $77.7$, and that center shots seem like probably the most effective. Specifically, we discover the next scoring rates for various shot directions:

• Left: $78.7$;
• Center: $88.2$;
• Right: $71.2$.

So as to derive the optimal strategies, nevertheless, we’d like to reconstruct the payoff matrix, which requires estimating nine conversion rates — one for every possible combination of the kicker’s and goalkeeper’s decisions.

Nonetheless, with only 103 observations in our dataset, certain mixtures are encountered quite rarely. As a consequence, estimating these probabilities directly from raw counts would introduce significant noise.

Since there isn’t a strong reason to consider that the left and right sides of the goal are fundamentally different, we are able to improve the robustness of our model by imposing and aggregating equivalent situations.

This effectively reduces the variety of parameters to estimate, thus lowering the variance of our probability estimates and increasing the robustness of the resulting payoff matrix.

Under these assumptions, the empirical payoff matrix becomes:

$Psimeq begin{bmatrix} 0.6 & 1 & 0.86 0.94 & 0 & 0.94 0.86 & 1 & 0.6 end{bmatrix}$.

We will see that the measured payoff matrix is kind of just like the toy model we defined earlier, with the principal difference being that in point of fact kickers can miss the goal even when the goalkeeper picks the incorrect direction.

Solving for equilibrium strategies, we discover:

$begin{aligned} p^* &simeq (0.39, 0.22, 0.39) q^* &simeq (0.415, 0.17, 0.415) end{aligned}$.

Are Players Actually Optimal?

Comparing equilibrium strategies with observed behavior reveals an interesting pattern.

Kickers behave near optimally, although they aim at the middle barely less often than they need to ($16.5$ of the times as an alternative of twenty-two%).

However, goalkeepers deviate significantly from their optimal strategy, remaining in the middle only $6$ of the times as an alternative of the optimal $17$.

This explains why center shots appear unusually successful in historical data. Their high conversion rate doesn’t indicate an intrinsic superiority, but fairly a scientific inefficiency within the goalkeepers behavior.

If each keepers and goalkeepers followed their equilibrium strategies perfectly, center shots could be scored roughly $77.8$ of the time, which is near the worldwide average.

Beyond Football: A Data Science Perspective

Although penalty kicks provide an intuitive example, the identical phenomenon appears in lots of real-world data science applications.

Online pricing systems, financial markets, advice algorithms, and cybersecurity defenses all involve agents adapting to one another’s behavior. In such environments, A pricing strategy that appears optimal in past data may stop working once competitors react. Likewise, fraud detection systems change user behavior as soon as they’re deployed.

In competitive environments, learning from data requires modeling interaction, not only correlation.

Conclusions

Penalty kicks illustrate a broader lesson for data-driven decision-making optimization.

. When outcomes emerge from strategic interactions, observed data reflects an equilibrium between competing agents fairly than the intrinsic quality of individual actions.

Understanding the mechanism that generates the information is subsequently essential. Without modeling strategic behavior, descriptive statistics can easily be mistaken for prescriptive guidance.

The true challenge for data scientists is subsequently not only analyzing what happened, but understanding why rational agents made it occur in the primary place.