Is Blood Thicker than Water? The Impact of Player Agencies on Player Salaries: Empirical Evidence from Five European Football Leagues

3.1 Principal Agent Theory

In 1932, Berle and Means highlighted the classic problem of corporate governance, the separation of ownership and control. Many years later, Jensen and Meckling (1976) displayed the welfare-loss that emerges from transforming a hundred percent solely owned business to a widely shared stock market company, resulting in typical agency costs such as consumption on the job and/or empire building. While these costs are coming from the fact that external agents manage the owner’s money and investment, agency theory proposes incentives in order to lessen information and iatrogenic problems. In analogy to this situation of whether a single business is more successful than the alternative of going public, less is known about the performance and costs of football player agents managing a player’s contract and salary. A classic principal-agent problem arises (Mason 1999, 163–197). From this point of view being a professional football player can be interpreted as an agglomeration of human capital that might be managed and advised.

3.2 Asymmetric Information

The relationship between player and player agent is analogous to the lemons problem (Akerlof 1970, 488–500). Although the player agent can accurately evaluate the quality of a player in a market in which monitoring the workers’ performance is easy, it can be challenging for players to assess the competence of a player agency. This situation was exacerbated by the abolition of the licensing system in 2015, as this resulted in the absence of an opportunity for agencies to communicate their own expertise. The license serves as a proof of quality and associates a certain degree of competence in the areas of sports law, corporate law and economics. Moreover, there were exceptionally low barriers to market entry between 2015 and 2023, which attracted new player agencies. In turn, it leads to adverse selection. Within the market, there are qualitative imbalances, a lack of transparency and a high degree of complexity (Kelly and Chatziefstathiou 2017). This asymmetric information can be mitigated through targeted monitoring (Gohritz, Hovemann, and Ehnold 2023). If at all, it can only be determined over time, i.e. after the contract has been signed, whether the choice of player agency was worthwhile. It is therefore an economic relationship under uncertainty, whereby a pareto-optimal solution for both sides cannot be achieved easily (Shavell 1979, 55–73). Given this and based on the assumption that a player accumulates the greatest possible amount of information about his abilities and thus ultimately his value, the option of self-representation appears to be the most ideal. The further we move away from the player, the greater is the loss of information about the true human capital. This means that the player’s agency is the most distant, has the least information and therefore performs the lowest in salary negotiations. Family members, on the other hand, have an information advantage, but are still lagging behind the player himself. Consequently, we hypothesize:

3.3 Applied Theory

In general, the principal (player) is dependent on the performance of the agent (player agency). The transfer of responsibility takes place because the principal does not have the resources, such as time or the necessary expertise, to take on the task itself at a reasonable cost (Eisenhardt 1989, 532–550; Jensen and Meckling 1976; Sappington 1991, 45–66). In return, the agent receives remuneration from the principal. In this context, there is a risk that the agent may be tempted to deviate from previous agreements through opportunistic behavior (Holt, Michie, and Oughton 2006; Mason 1999, 163–197). Due to individual considerations of maximizing individual benefits, conflicting objectives can arise, which can lead to moral hazard (Sappington 1991, 45–66). The principal, respectively the player, could, for example, be interested in the long-term development of his sporting career, with match experience in the form of playing minutes being of particular importance. On the other hand, the agent, respectively the player agency, may consider maximizing the short-term benefit in terms of remuneration, which can lead to disadvantages for the principal (Gohritz, Hovemann, and Ehnold 2023). Although a player agency should pursue similar interests as the player due to the commission depending on the annual salary, we argue that the plurality of interests leads to an inefficient outcome. It should also be noted that many agencies manage a large number of players with different quality profiles. Player agencies pursue the goal of maximizing profits and accordingly try to maximize the return on the overall portfolio, hence all players. This can lead to a situation where more attention is paid to single promising investments or where a poor performance of other investments is tolerated in favor of a major transfer. In the case of agencies with a multi-client structure, this means that there might be a risk of a principal-agent conflict, which has already been proven in Major League Baseball (Krautmann, von Allmen, and Walters 2018). Based on the premise that the quality of care of the individual player decreases with an increasing number of players managed, we develop the following hypothesis:

4.1 Data

One of the appealing things about sports economics as a research discipline is the public availability of data. Professional football players are publicly exposed employees whose performance we can observe one weekend after another and therefore measure and analyze. In addition, certain individual features and professional backgrounds, as well as information on physiological or individual characteristics, are usually publicly accessible. We used various sources such as FBRef.com, Bundesliga.de and Kicker.de to collect the data. The main source of our data and in particular our data relating to the player’s features and agencies was collected via Transfermarkt.de. All public websites as well as Kicker as a sports magazine are well respected and have been frequently used by various sports economists in the past (Frick and Prinz 2006; Göke, Prinz, and Weimar 2014; Peeters 2018, 17–29). Our data was collected both manually and using a Python-based web-scraping tool.

The result is a unique data set with 3,778 different players from the top five leagues in Europe. We observe these players in an unbalanced panel over five survey dates. The first survey was conducted on October 31, 2022 and subsequently at each quarter. Based on these panel data we are able to manage possible player selection biases for one of the above-mentioned representation models. Table 2 shows the 237 changes of the representation models that appeared during our five survey dates.

In addition to the individual features of the players, the corresponding information on the agency model and salaries were collected each time. We also determined the team performance and individual performance indicators for the last season (t-1). We measure individual performance in 10 different categories: defence, goal and shot creation, general statistics, passing, passing types, playing time, possession, shooting, goalkeeping and advanced goalkeeping statistics. This gives us a total of 147 different individual performance indicators for each player from the last season. In order to reduce this multitude of variables without sacrificing complexity, we carried out a factor analysis. We developed 10 performance/factor indices corresponding to the categories mentioned above. As part of the data adjustment, we excluded goalkeepers (and thus also the performance indicators goalkeeping and advanced goalkeeping statistics) due to their significantly different performance function compared to outfield players (Schneemann and Deutscher 2017; Watanabe, Wicker, and Yan 2017; Weimar and Scharfenkamp 2019).

In contrast to previous research (e.g. Deutscher and Büschemann 2016; Frick 2011, 92), we do not use proxies to operationalize the salary of football players. We use the actual metrically available salary data provided by the website Capology.com. Following Berri et al. (2023), we argue for the superiority of our dataset due to the primary source of the salary data. Capology.com receives information on salary data directly from insiders, player agencies or the clubs themselves, i.e. from sources that are often directly involved in salary negotiations.

4.2 Descriptive Statistics

Players can represent themselves in salary negotiations and negotiate their own salary directly with the club. Another option is to be represented by relatives. However, as shown in Table 3, the majority of players (more than 85 %) are represented by a player agency. Just under 4 % are represented by family members and the remaining 11 % represent themselves. However, this is common practice in professional sports and it is the rule rather than the exception. Player agencies can be identified not only in football, but in a wide variety of sports (Mason and Duquette 2005).

Table 4 provides summary statistics. The average player is 25.6 years old, 1.82 m tall and a right-footed defender. We distinguish 114 different nationalities of players. In 42 % of our cases, the nationality of the league does not correspond to the nationality of the player. One agency manages a total portfolio (in terms of the cumulative market value of all players represented by the agency) of €1.9 billion. The average player represented by a player agency has a market value of €2.67 million. In total, we consider 111 different clubs from five different leagues in our data set.

On average, the gross annual salary of a professional football player in the top five leagues in Europe, excluding bonuses, is €2.4 million. A simple look at the mean and the standard deviation of the salary implies its skewedness. Generally, and in different fields of economics the dispersion of a variable is outlined by extreme values and outliers presenting heavy tails. Such anomalies are especially true in markets from the entertainment industry (Krueger 2005, 17–21; Malmendier and Tate 2009; Rosen 1981, 845–858) as in sports, music or the movie industry, because here “a relatively small number of people” (Rosen 1981, 845) – so called superstars – dominate the income distribution due to exceptional talent. Since football players do certainly fall into the entertainment category, some kind of superstars (e.g. Messi, Neymar, Mbappé) are responsible for the bulk of income. It illustrates that the salaries are heavily tailed to the right. Or put differently, one of the main findings from the descriptive statistics is that a relatively small number of players account for a disproportionally large share of all player salaries analyzed in our sample. Figure 1 shows that specifically this form of long tailed data displays a deviation of the Gaussian standard normal data distribution and should be taken into consideration when explaining the drivers of income. In order to estimate unbiased coefficients and the fact that the impact of the independent variables on salaries might be asymmetrical at different points of the distribution function the literature typically proposes two reliable strategies. One opportunity is quantile regression, an extension of the linear regression since the mean does not describe the whole distribution assuming that the effect of the explanatory parameters is not equal on each point of the function (e.g. Berri, Farnell, and Simmons 2022; Fort, Lee, and Oh 2019).^[3] The other opportunity is to model the natural logarithm of individual wages which solves the problem of mitigating the outlier and long tail concern and makes the interpretation of the estimated coefficients more convenient. For this reason, we logarithmize the dependent variable salary, as suggested in previous research (e.g. Battré, Deutscher, and Frick 2008; Berri et al. 2023; Frick 2011, 94), which is illustrated in Figure 2.

Figure 1:

Distribution function of the absolute gross annual salaries.

Figure 2:

Distribution function of the logarithmized gross annual salaries.

4.3 Econometric Findings

We start the empirical analysis with a (player) fixed-effects (FE) model series to test for individual heterogeneity. We conducted the Hausman and Breusch–Pagan tests and came to the conclusion that the FE model delivers the most robust results. The FE model is best suited to control for unobserved, time-invariant player characteristics (e.g. talent, personality) that could potentially impact the salary. We start the model series with the representation model dummy and gradually include the player agency interaction term, the individual and team performance and control for the league as well as the club. To improve the robustness of the coefficients we clustered the standard errors. The estimation model has the following general form:

The independent variable of the model is the logarithmized salary of the ith soccer player at time t. β ₀ is the intercept of the model. The respective representation model at time t for player i is included in the model as a dummy variable by β ₁ RMD _it . The representation model can have the three variants self-representation, relatives or player agencies. β ₂(PAD × X _it ) comprises the interaction term of the dummy variable for player agency and the number of managed players in the agency as a multiplier. ∑ k = 3 10 β k I P k i ( t − 1 ) represents the individual performance of the ith player in period t-1. The individual performance consists of eight performance indices (defence, goal and shot creation, general statistics, passing, passing types, playing time, possession, shooting), which were calculated from 147 individual key statistics as part of the factor analysis. With ∑ m = 11 13 β m T P m i t − 1 we include the team performance in t-1, which includes the variables rank, goals and goals against. In addition, we control for the league (Ref.: Bundesliga) with β ₁₄ LeagueD _it and for the club (Ref.: Bayern Munich) with β ₁₅ ClubD _it . ε _it denotes the error term in the model (Table 5).

Starting with our first model just including the representation variables where we use the player agency as reference and controlling for the league as well as the club, we observe a highly significant impact for both the representation by relatives and the self-representation. Players who are represented by family members earn on average 22 % less than if they had been represented by an agency. For the self-representation model, we observe a comparable effect, whereby players who represent themselves earn around 8.5 % less on average. The second model additionally includes the player agency interaction term. We observe a highly significant effect for this term and at the same time the effects of the representation model become insignificant. Although the coefficient is very small, it indicates that there is a positive correlation between the number of players managed by an agency and the salary. This could be caused by the fact that agencies with more players in their portfolio can actually negotiate better or have more experience, which means that the players receive a slightly higher salary. Therefore, H2 must be rejected.

In the third model we include the individual performance factors respectively indices which were calculated for the last season. The fourth model additionally contains the team performance variables. Following previous research (e.g. Hall et al. 2019; Scarfe, Singleton, and Telemo 2021), we argue that both individual and team performance in the last season determine today’s salary. There is thus a time lag of one year between the independent and the dependent variable, which is advantageous from an econometric point of view as the problem of simultaneity is avoided and also beneficial from an interpretative point of view as we thus come slightly closer to the causal analysis of the effect of performance on salary (Thrane 2019, 1,054–1,055). The wealth of individual performance variables increases the complexity of the models, as 147 performance indicators for each player are included, but makes it more difficult to interpret the results. It should be noted that there are consistently negative significant effects for the defence performance index. This indicates that defensive playing qualities are given less weight in the salary structure compared to more offensive skills such as shooting or playmaking. All models show a significant positive effect of playing time on salary which is plausible. This means that continuous playing time is an indicator of a player’s importance and has a direct impact on the negotiating position. The significant positive effect of the variable goals against requires further explanation. In the fifth model, we applied the interquartile range method to identify the outliers of our data set which are commonplace in sport industries, where the winner-takes-it-all phenomenon exacerbates their occurrence and potential biases. The only difference compared to the previous models is that the passing index now has a significant impact on salary, which means that passing performance seems more important in the mid-salary segment. The most surprising fact, however, is that despite numerous model specifications, there is no significant difference between relatives, the self-representation model and the player agency regarding a player’s salary. Accordingly, it does not matter in salary negotiations whether a player represents himself, is represented by family members or by a player agency. This means that we reject H1, that self-representation is the most lucrative representation model in terms of salary.

Above, we have shown the 237 changes of representation models that were identified in our study. With regard to causality and in order to substantiate the validity of our results, the question now arises, which determinants affect the probability of a change from one representation model to another. For this purpose, we first summarized the models of self-representation and representation by relatives. Subsequently, we created a dummy variable that can take the values 1 for player agency and 0 for non-player agency. In the following, we estimate a logit model and include individual characteristics of the players in order to illustrate the impact of these variables on the probability of changing (Table 6).

An increase in age by one year increases the probability of being represented by a player agency by 7.60 %. An increase in squared age, on the other hand, reduces the probability by 0.14 %. Both effects are highly significant (both p-values: 0.000). The non-linear effect of age was anticipated, considering the parabolic relationship between the two variables. This could be due to the fact that young players in particular tend to prefer a player agency at the start of their career expecting to benefit from their network and expertise. However, this effect diminishes with increasing age. The increase in height by one cm also increases the probability of choosing the agency’s representation model (0.16 %; p-value: 0.002). The importance of height can be interpreted as an indicator that physical attributes, which are essential in football, also affect the likelihood of choosing professional representation. Taller players may be considered more valuable because they fit better in certain positions and are superior to smaller players, which affects their bargaining power. We cannot find any significant effects with regard to footedness. Captains are about 2.10 % less likely to be represented by an agency (p-value: 0.081). This could be due to the fact that a captain is already a self-confident leader who takes responsibility. They might already have a more secure position in the team that is less dependent on transfers or negotiations. Furthermore, we did not find a significant effect when the nationality of the player matched the nationality of the league. This indicates that agencies make their decisions independently of nationality. The lack of such an effect could imply that the globalization of football ensures that national or cultural differences are less important when it comes to representation. Being a striker reduces the probability of being represented by a player agency by 1.62 % compared to being a defender (p-value: 0.050). One possible interpretation would be that strikers are often in the spotlight and are less reliant on agency to attract attention due to their exposed role on the pitch. Defenders, who may have less obvious moments of success, may be more likely to rely on the negotiating skills of agencies to strengthen their position and negotiate contracts.