The myth of declining competitive balance in
the UEFA Champions League group stage

Abstract

According to previous studies, competitive balance has significantly declined in the UEFA Champions League group stage over the recent decades. Our paper introduces six alternative indices for measuring ex ante and ex post competitive balance in order to explore the robustness of these results. The ex ante measures are based on Elo ratings, while the ex post measures compare the group ranking to reasonable benchmarks. We find no evidence of any trend in the competitive balance of the UEFA Champions League group stage between the 2003/04 and 2023/24 seasons.

“Different measurements are of different use, and all lines of research into competitive balance have, to date, proven quite instructive. To ignore this is to forgo important insights into the behavior of competitive balance.”

Keywords: competitive balance; football; measurement; sports economics; UEFA Champions League

MSC class: 62P20, 91B82

JEL classification number: L11, Z20, Z21

1 Introduction

One of the most prestigious football tournament, the UEFA Champions League, has been organised in the same format over 21 seasons, which is fundamentally changed from 2024/25. The reform explicitly aims to improve competitive balance by replacing the traditional group stage with a league phase (UEFA,, 2022)—and can be successful in this respect according to recent research (Gyimesi,, 2024).

Competitive balance has two interpretations: ex ante and ex post. Ex ante competitive balance is experienced before the matches are played and is connected to suspense: the consumers hope to see an exciting game where winning and losing are far from being predetermined (Richardson et al.,, 2023). On the other hand, ex post competitive balance is related to surprise. According to Ely et al., (2015), it is more surprising if the current events are greatly different from the previous events.

Recent studies have found a robust decline in both the ex ante (Triguero-Ruiz and Avila-Cano,, 2023) and ex post (Ramchandani et al.,, 2023; Triguero-Ruiz and Avila-Cano,, 2023) competitive balance of the UEFA Champions League group stage over the last decades, which might explain its reform. Our paper aims to check the robustness of these findings because we think that there exist alternative but reasonable measures of competitive balance.

In order to quantify ex ante competitive balance, Triguero-Ruiz and Avila-Cano, (2023) consider UEFA club coefficients as a measure of team strength. However, since they do not reflect most of the available information pertinent to predicting future performances, it would be better to use a more accurate rating such as Elo points (Csató,, 2024).

Ex post competitive balance is usually quantified by the Herfindahl–Hirschman index (HHI) based on the shares of each team from the total points awarded (Avila-Cano et al.,, 2023; Owen and Owen,, 2022; Owen et al.,, 2007). However, the clubs do not necessarily have appropriate incentives in the last round(s) (Chater et al.,, 2021; Csató et al.,, 2024). Kendall and Lenten, (2017) and Guyon, (2020) provide several examples where such a situation has led to (tacit) collusion or even tanking. Therefore, we will consider only the group ranking, which solely determines the subsequent matches of the teams.

Based on the above arguments, two indices of ex ante and four indices of ex post competitive balance will be constructed in order to implement a proper sensitivity analysis. In contrast to previous results, we find no evidence for any decline in the competitive balance of the UEFA Champions League group stage between the 2003/04 and 2023/24 seasons. Consequently, if UEFA has chosen the barely used incomplete round-robin design of the UEFA Champions League with its inherent risks (Devriesere et al.,, 2024) because of the worsening trend in competitive balance, the decision-makers might have been misled.

The paper has the following structure. Our measures of competitive balance are introduced in Section 2 together with a description of the underlying data. Section 3 provides and discusses the results, while Section 4 offers concluding remarks.

2 Methodology and data

The UEFA Champions League was organised in the same format between the 2003/04 and 2023/24 seasons. The group stage contained eight groups of four teams each. The top two teams qualified for the Round of 16, where the group winners were matched with the runners-up, the third-placed teams were transferred to the second most prestigious UEFA club competition, and the fourth-placed teams were eliminated.

In order to ensure the balancedness of the groups, the 32 teams were allocated into four pots. Until the 2014/15 season, the seeding was determined by the UEFA club coefficients except for assigning the titleholder to the first pot (Csató,, 2021). The first pot consisted of the titleholder and the champions of the highest-ranked associations from the 2015/16 season (Dagaev and Rudyak,, 2019), and the UEFA Champions League titleholder, the UEFA Europa League titleholder, as well as the champions of the highest-ranked associations from the 2018/19 season (Csató,, 2020).

Sections 2.1 and 2.2 motivate and define the proposed indices for ex ante and ex post competitive balance, respectively. Section 2.3 provides an example of their calculation, while Section 2.4 gives information about our data.

2.1 Measures of ex ante competitive balance

Triguero-Ruiz and Avila-Cano, (2023) start to quantify ex ante competitive balance by calculating the shares of the clubs from UEFA club coefficients. These shares are used to compute HHI for each group, which gives the distance to competitive balance (DCB) index (Dubois,, 2022; Scelles et al.,, 2022; Triguero Ruiz and Avila-Cano,, 2019) as follows:

$$\mathit{DCB}=\sqrt{\frac{\mathit{HHI}-\mathit{HHI}_{\min}}{\mathit{HHI}_{\max}-\mathit{HHI}_{\min}}}.$$

UEFA club coefficient is the official measure of team strength used by UEFA, the number of points gained by a team in UEFA club competitions over the last five years (Dagaev and Rudyak,, 2019). Furthermore, there is a lower bound based on the five-season association coefficient, which might be relevant for emerging clubs without meaningful experience at the international level (Csató,, 2022). The rankings derived from the UEFA club coefficients determine seeding in the UEFA club competitions.

However, UEFA club coefficients suffer from a major shortcoming: they do not use the majority of matches played by the clubs since all games in the national leagues and cups are omitted. Home advantage does not count either. Unsurprisingly, Csató, (2024) finds robust evidence that a variant of the Elo method provides much higher accuracy in terms of explanatory power for the UEFA Champions League. Following Csató, (2024), we use the Football Club Elo Ratings as a measure of team strength (Football Club Elo Ratings,, 2024). According to its formula, the expected probability that team $i$ wins against team $j$ is:

$$W_{ij}=\frac{1}{1+10^{-\left(R_{i}-R_{j}\right)}},$$ (1)

where $R_{i}$ and $R_{j}$ are the Elo ratings of the two teams, respectively. Note that $W_{ji}=1-W_{ij}$.

To update the Elo ratings, the expected winning probability $W_{ij}$ is compared to the actual result $Q$ (1 for win, 0.5 for draw, 0 for loss):

$$\Delta R_{i}=20\left(Q-W_{ij}\right).$$

Since $\Delta R_{i}=-\Delta R_{j}$, the sum of Elo ratings does not change after any game. Finally, home advantage is taken into account by increasing the Elo difference in (1). The home advantage parameter is not fixed in advance but dynamically updated to guarantee that it converges to the actual effect of home advantage (Football Club Elo Ratings,, 2024).

Groups of four teams imply six pairs of clubs. For each pair, the winning probability of the stronger team—a value between 0.5 and 1—is computed according to (1). The six values are added to get $\mathit{UCB}_{1}^{A}$, which is normalised similarly to the idea behind DCB:

$$\mathit{CB}_{1}^{A}=\frac{\mathit{UCB}_{1}^{A}-\left(\mathit{UCB}_{1}^{A}\right)_{\min}}{\left(\mathit{UCB}_{1}^{A}\right)_{\max}-\left(\mathit{UCB}_{1}^{A}\right)_{\min}}=\frac{\mathit{UCB}_{1}^{A}-3}{6-3}=\frac{\mathit{UCB}_{1}^{A}}{3}-1.$$

$\mathit{CB}_{1}^{A}\in\left[0,1\right]$ is our first index of ex ante competitive balance, for which a lower value is preferred as it indicates a higher (expected) uncertainty in the group matches.

However, if only two teams qualify from a group of four, a group is usually considered a harsh “group of death” with three strong teams, independent of the strength of the fourth team (Guyon,, 2015; Laliena and López,, 2019). Thus, it makes sense to focus on the three strongest teams and derive $\mathit{UCB}_{1}^{A}$ from the three pairs implied. Our second index of ex ante competitive balance, normalised to lie between 0 and 1 is

$$\mathit{CB}_{2}^{A}=\frac{\mathit{UCB}_{2}^{A}-\left(\mathit{UCB}_{2}^{A}\right)_{\min}}{\left(\mathit{UCB}_{2}^{A}\right)_{\max}-\left(\mathit{UCB}_{2}^{A}\right)_{\min}}=\frac{\mathit{UCB}_{2}^{A}-1.5}{3-1.5}=\frac{2\mathit{UCB}_{2}^{A}}{3}-1.$$

Following Triguero-Ruiz and Avila-Cano, (2023), the ex ante measures of competitive balance are averaged for the eight groups in each season.

2.2 Measures of ex post competitive balance

Both Ramchandani et al., (2023) and Triguero-Ruiz and Avila-Cano, (2023) use HHI based on the points share for each club in the group to compute ex post competitive balance. The underlying assumption is that the teams have appropriate incentives to collect points even though their qualification depends on the group ranking rather than the number of points. In the history of football, numerous examples exist when a team was satisfied with a (moderate) loss if it was sufficient to achieve its objectives (Guyon,, 2020; Kendall and Lenten,, 2017). This consideration could be especially important in the UEFA Champions League group stage since the clubs usually play in their domestic leagues both on the previous and subsequent weekends, which creates a powerful incentive to rest the best players and field a younger squad to gain experience if the game is stakeless with respect to the rank of the team.

As an illustration, consider Group G in the 2018/19 UEFA Champions League. Before the last matchday, Real Madrid already won the group, and Roma was guaranteed to finish in the second place. In the last round, Real Madrid suffered a shocking 3-0 defeat against CSKA Moscow after fielding a fully rotated squad, which was its first European home loss by more than two goals (Bell,, 2018). Roma also lost against Viktoria Plzeň despite its home win of 5-0 in the second round.

Table 1: MIndices of ex post competitive balance

Measure	Ex ante ranking	Are group winners and runners-up distinguished?
$\mathit{CB}_{1}^{P}$	Pot allocation	✔
$\mathit{CB}_{2}^{P}$	Pot allocation	✗
$\mathit{CB}_{3}^{P}$	Elo rating	✔
$\mathit{CB}_{4}^{P}$	Elo rating	✗

Based on the arguments above, we compute the Kendall rank correlation coefficient (Kendall,, 1938) between an ex ante ranking and the final group ranking. Since each pot contained one team from each pot, pot allocation can be regarded as a reasonable ex ante ranking of the clubs in each group. Thus, the ex ante ranking is given by either the pot allocation or the ranking derived from the Football Club Elo Ratings. In addition, it can be said that a discordant pair at the top of the final group ranking should not be taken into account because both the group winner and the runner-up qualify for the Round of 16. Therefore, four alternative measures of ex post competitive balance are defined as shown in Table 1.

Following Ramchandani et al., (2023) and Triguero-Ruiz and Avila-Cano, (2023), the ex post measures of competitive balance are averaged for the eight groups in each season.

2.3 An illustrative example

The four teams of Group C in the 2023/24 UEFA Champions League are Napoli (drawn from Pot 1; Elo rating: 1911), Real Madrid (Pot 2; 1917), Braga (Pot 3; 1677), and Union Berlin (Pot 4; 1757). The pairwise differences between their Elo ratings are 6, 234, 154, 240, 160, 80, respectively, thus

$$\mathit{UCB}_{1}^{A} = \frac{1}{1+10^{-6/400}}+\frac{1}{1+10^{-234/400}}+\frac{1}{1+10^{-154/400}}+\frac{1}{1+10^{-240/400}}+$$

$$+\frac{1}{1+10^{-160/400}}+\frac{1}{1+10^{-80/400}}=4.138.$$

Analogously, the differences between the three highest Elo ratings are 6, 154, 160, respectively, thus

$$\mathit{UCB}_{2}^{A}=\frac{1}{1+10^{-6/400}}+\frac{1}{1+10^{-154/400}}+\frac{1}{1+10^{-160/400}}=1.932.$$

The corresponding normalised values are:

$$\mathit{CB}_{1}^{A}=\frac{\mathit{UCB}_{1}^{A}}{3}-1=0.379\qquad\text{and}\qquad\mathit{CB}_{2}^{A}=\frac{\mathit{2UCB}_{2}^{A}}{3}-1=0.288.$$

The final group ranking is Real Madrid, Napoli, Braga, Union Berlin. Compared to the pot allocation, one discordant pair (Napoli and Real Madrid) exists, hence $\mathit{CB}_{1}^{P}=1-2/6=2/3$. This discordant pair affects the first two places in the group ranking, hence $\mathit{CB}_{2}^{P}=1-0/5=1$. Compared to the ranking implied by Elo ratings, there is one discordant pair (Braga and Union Berlin), hence $\mathit{CB}_{3}^{P}=1-2/6=2/3$. This discordant pair affects not only the first two places in the group ranking, hence $\mathit{CB}_{4}^{P}=1-2/5=3/5$.

2.4 Data

We consider all seasons between 2003/04 and 2023/24 when the basic structure of the UEFA Champions League has not changed. The pot allocation and group rankings are readily available from several sources; we have used Wikipedia after cross-checking with the official UEFA site.

Football Club Elo Ratings on a given date can be downloaded from api.clubelo.com/YYYY-MM-DD. In contrast to UEFA club coefficients, the Elo rating of a team changes during the season. Thus, the Elo ratings on 1 September are used for each season because the group stage is usually played between September and December. As usual, the Elo ratings are rounded to the nearest integer. Then Group B in the 2020/21 UEFA Champions League contains two teams (Shakhtar Donetsk and Borussia Mönchengladbach) with the same value (1766), which should be broken for indices $\mathit{CB}_{3}^{P}$ and $\mathit{CB}_{4}^{P}$. The tie is decided by the higher value before rounding.

3 Results and discussion

Figure 1: Evolution of competitive balance in the UEFA Champions League group stage
Notes: The seasons are denoted by their first year when the group stage is played.
The thick dashed red line shows the linear trend. None of them are significant even at a 25% significance level, see Table 2.

Figure 1 shows the evolution and the linear trend of all competitive balance measures. The linear trend has been obtained by estimating the following regression model:

$$\mathit{CB}_{i,t}=c+\alpha t+\varepsilon_{t},$$

where $\mathit{CB}_{i,t}$ is the $i$th competitive balance measure and $\varepsilon_{t}$ is the error term in season $2003\leq t\leq 2023$ (every season is denoted by its first year when the group stage is played). Furthermore, $c$ is the intercept and $\alpha$ is the coefficient of the season.

The values of $\mathit{CB}_{1}^{A}$ are always higher than $\mathit{CB}_{2}^{A}$, that is, competitive balance seems to be worse if all teams playing in the group are considered rather than only the three strongest teams. Consequently, the difference between the strengths of the weakest team and one of the three strongest teams is usually higher than the difference between the strengths of two of the three strongest teams. Regarding ex post competitive balance indices, their means are around 0.5 and even the smallest seasonal average is above 0.2. Therefore, both pot allocation and Elo ranking are better predictors of the final group ranking than a random permutation as expected. Group G in the 2019/20 UEFA Champions League is an interesting outlier, where the final ranking is the opposite of the ranking derived from the pot allocation.

Table 2: Linear regression models for competitive balance measures in the
UEFA Champions League group stage between the 2003/04 and 2023/24 seasons

Measure	Period	Intercept $c$	Coefficient $\alpha$	$R^{2}$	$p$-value
$\mathit{CB}_{1}^{A}$	2003/04–2023/24	$-$3.5418	0.001960	0.061	0.2784
$\mathit{CB}_{2}^{A}$	2003/04–2023/24	$-$2.5841	0.001447	0.050	0.3297
$\mathit{CB}_{1}^{P}$	2003/04–2023/24	$-$5.0573	0.002760	0.014	0.6047
$\mathit{CB}_{2}^{P}$	2003/04–2023/24	$-$7.5424	0.004026	0.029	0.4574
$\mathit{CB}_{3}^{P}$	2003/04–2023/24	$-$2.9361	0.001732	0.006	0.7460
$\mathit{CB}_{4}^{P}$	2003/04–2023/24	$-$1.5176	0.001039	0.002	0.8583

Figure 1 may suggest a small drop in competitive balance over the recent two decades since the slope of the linear trend is always positive. However, none of them are significant according to Table 2. The $p$-values are much higher than the ones found by Triguero-Ruiz and Avila-Cano, (2023), which is below 0.02 for ex ante and approximately 0 for ex post competitive balance. Unsurprisingly, the values of $R^{2}$ are never above 0.1, in stark contrast to the results of Ramchandani et al., (2023) and Triguero-Ruiz and Avila-Cano, (2023).

Table 3: Linear regression models for ex ante competitive balance measures in the
UEFA Champions League group stage between the 2003/04 and 2023/24 seasons

Measure	Period	Intercept $c$	Coefficient $\alpha$	$R^{2}$	$p$-value
$\mathit{CB}_{1}^{A}$	2003/04–2014/15	$-$22.6219	0.01146	0.50	0.01034
$\mathit{CB}_{1}^{A}$	2014/15–2023/24	24.3567	$-$0.01186	0.75	0.00124
$\mathit{CB}_{2}^{A}$	2003/04–2014/15	$-$15.7854	0.00802	0.31	0.05861
$\mathit{CB}_{2}^{A}$	2014/15–2023/24	12.6759	$-$0.00611	0.48	0.02656

Nonetheless, Table 3 presents that some trends can be found in ex ante competitive balance. In particular, it declined until the 2014/15 season and improved between the 2014/15 and 2023/24 seasons. This pattern is less robust but still exists if only the three strongest teams are considered in all groups.

How can these divergent conclusions be explained? Although both Ramchandani et al., (2023) (28 seasons from 1992/93 to 2019/20) and Triguero-Ruiz and Avila-Cano, (2023) (19 seasons from 1999/2000 to 2017/2018) study different samples, this cannot be the main reason. According to Ramchandani et al., (2023), the reduction in ex post competitive balance is even more evident between the 2002/03 and 2019/20 seasons, where they report $R^{2}=0.49$ for the linear regression (unfortunately, the $p$-value is missing). Analogously, the five-year averages of the DCB index continuously increase (the competitive balance worsens) in both the ex ante and ex post settings (Triguero-Ruiz and Avila-Cano,, 2023).

Our ex ante competitive balance measure is based on the more accurate Elo rating. Triguero-Ruiz and Avila-Cano, (2023) use the UEFA club coefficients, which have likely become more concentrated among the leading clubs without a parallel dominance in the Elo ratings. A potential reason might be that the best clubs in the top leagues—except England, where the set of competitors is wider—can somewhat relax in their domestic leagues now since the number of guaranteed slots in the UEFA Champions League group stage has increased from two to four for the four highest-ranked leagues. Therefore, they are able to increasingly focus on international competitions.

Regarding ex post competitive balance, we compare the final group ranking to a benchmark provided by either the pot allocation or the Elo ranking. Since the HHI does not require such a benchmark, a worsening ex post competitive balance may be covered if our benchmark becomes a more accurate predictor over time. However, this is unlikely. Even though the seeding system has been reformed as discussed in Section 2, it has rather increased uncertainty (Corona et al.,, 2019; Dagaev and Rudyak,, 2019). Similarly, the differences in the Elo ratings have not increased (Figure 1), thus, it would be difficult to assume that the quality of Elo ranking has changed. On the other hand, the clubs may focus in the group stage more strongly on their number of points than previously, possibly due to the increasing financial incentives provided by the UEFA. This leads to a higher HHI but a lower probability of losing stakeless games. Then the earlier studies have only mixed a favourable trend (teams do not lose their stakeless matches) with declining ex post competitive balance. The hypothesis may be tested in the future by an empirical investigation of stakeless games, which can be identified by the approach of Csató et al., (2024).

4 Conclusions

UEFA has chosen a risky strategy by fundamentally reforming the well-established group stage of its major tournament, the UEFA Champions League, from the 2024/25 season. According to the existing literature using convenient measures, UEFA should have acted effectively because of the notable drop in both the ex ante and ex post competitive balance of the group stage over the last two decades. Our paper has proposed six alternative but reasonable indices to check the robustness of these results. Crucially, no evidence is found for a declining competitive balance between the 2003/04 and 2023/24 seasons. Therefore, we urge researchers and analysts to consider a broader set of competitive balance measures in the future before proposing powerful policy interventions.