Broadcast rights and competitive balance in European soccer.

Peeters, Thomas

Introduction

European soccer has seen two major financial developments over the last decade. First, the Champions League has emerged as a supranational competition, and second, the value of media rights has risen immensely. In the 2008-2009 season, the Champions League distributed almost 600 million Euros in prize money among 32 participating clubs (UEFA figures). On top of this, participating clubs are able to generate extra income through ticket sales and they enjoy surplus media exposure. Meanwhile, revenues from TV rights sales have become (one of) the mainstay(s) of club finances in all European leagues. In the top leagues, they now far exceed 500 million Euros per season (see TV Sports Markets, 2008 for further details). The potentially dramatic impact of these developments on the European soccer industry raises questions about such issues as the financial stability of clubs, player wages, and competitive balance. These concerns have resulted in a lively debate among policymakers and people in the industry. A crucial issue in this debate is the manner in which TV rights are sold. Two rivaling sales systems have emerged: one individual, the other collective. Under the individual system, each club owns and sells the rights to its home games; under the collective system, the league monopolizes all the rights, sells them as a package and subsequently distributes the revenues among the clubs. In certain cases some of the individually selling clubs have voluntarily opted to pool their rights without forcing the other teams to participate in the collective agreement. This has lead to a third "mixed" regime. Opponents of the collective system point at the potentially negative welfare effects of allowing a cartel or monopolist to operate in the TV rights market. Forrest et al. (2004) provide econometric evidence that collective sales have indeed lead English clubs to restrict the number of televised matches below the competitive level. Clubs even selected an output level below that of a profit-maximizing monopolist. Forrest et al. (2004) identify this as the result of inefficient cartel behavior. On the other hand, the main argument to justify a collective system is that it may encourage solidarity and therefore better maintain the competitive balance (tension) within a league. The European Commission (2007) expresses the following view on the issue in its White Paper on Sports:

"The application of the competition provisions of the EC Treaty to the selling of media rights of sport events takes into account a number of specific characteristics in this area [...] While joint selling of media rights raises competition concerns, the Commission has accepted it under certain conditions. Collective selling can be important for the redistribution of income and can thus be a tool for achieving greater solidarity within sports."

In other words, the Commission recognizes that collective selling might be an important tool for achieving solidarity and maintaining competitive balance, yet possibly at the expense of reduced competition. Cave and Crandall (2001) state this possible trade-off even more explicitly:

"Do restrictive agreements among a league's teams reflect a desire to increase the "competitive balance" of the league or are they simply a means to limit competition and reduce sports fans' choices?"

In this paper, I investigate exactly this question and, more generally, which other factors might impact on competitive balance. I obtain four main findings. First, I show that the choice for the collective sales system has no positive impact on competitive balance. Second, I demonstrate that the UEFA Champions League has a detrimental effect on competitive balance in national competitions. Third, I reveal the important impact of the distribution of drawing power (i.e., the size of a club's local market) on competitive balance. Finally, I present evidence that a larger domestic market enhances the seasonal competitive balance in a soccer league.

These findings shed new light on competition policy in respect of the sports industry. I find no support for the major argument in favor of collective sales of TV rights. This suggests that competition authorities should be very cautious when granting leagues permission to monopolize TV rights. Other arguments are required to justify the practice. Second, the European soccer association, UEFA, might want to rethink its solidarity policy. While UEFA claims to safeguard solidarity between clubs and countries, its own Champions League is destroying competitive balance in member countries. A serious overhaul of the manner in which Champions League prize money is distributed may offer a solution. Finally, my results support the notion that smaller countries should strive to create leagues covering a larger market, such as a Scandinavian or a Benelux League. Such mergers would not only expand the target market of the league, but also improve the drawing power distribution. Therefore, there is potentially a twofold positive effect on competitive balance.

Related Literature

The link between competitive balance and revenue redistribution has been a prominent topic in the sports economics literature ever since the emergence of the field. Since the dramatic increase in broadcast rights revenues, this specific source of club income has come to the attention of several authors. Cave and Crandall (2001) describe the legal and constitutional background of broadcast rights sales in the United States and Europe. Important differences between both continents lead them to conclude that efficiency and competition problems are less severe in the United States than in Europe. This may be seen as an extra motivation to more urgently look into the European sports industry, as is done in the present paper. Palomino and Sakovics (2004) rationalize the existence of different distribution practices for broadcast revenues. Their analysis also points at the importance of institutional differences across the Atlantic. More importantly, however, they show how a league may face a trade-off between competitive balance and attracting star players, when designing a distribution system for broadcast revenues. Falconieri et al. (2004) perform a social welfare analysis of both the individual and collective system and identify conditions under which a collective system might be preferable. Noll (2007) sketches the demand and supply for televised sporting contests. From this analysis he concludes that centralized sales of broadcast rights are harmful for consumers and unlikely to increase competitive balance. Finally, Kesenne (2009) argues that decisions on sales mechanisms and the manner in which revenues are distributed should be considered separately. He succeeds in demonstrating that an individual sales mechanism followed by performance-based sharing is the best tool for preserving competitive balance in a league of profit-maximizing teams.

Apart from these theoretical contributions, some empirical research has been conducted into competitive balance in relation to broadcast rights and the Champions League. Andreff and Bourg (2006) perform an empirical analysis of the big five soccer leagues, (1) but struggle with a lack of financial data. This obliges them to restrict the analysis to six country-years. They conclude that the individual system leads to a worse competitive balance than the collective system. Frick and Prinz (2004) produce a dataset on the survival probabilities of newly promoted teams in a diverse sample of European soccer leagues. Using this dataset they show that the extent of revenue sharing has no effect on the strength of new teams. This result implies collective sales do little to increase competitive balance in soccer. A final contribution by Pawlowski et al. (2010) looks at a variety of competitive balance measures in the big five leagues. They identify the Champions League prize money distribution as an important factor that might disturb competitive balance.

I contribute to the aforementioned literature by integrating factors identified in previous studies in one model. I therefore construct a panel dataset on seasonal and championship competitive balance in 32 European leagues, covering a period of 10 years, from 2000-01 to 2009-10. This allows me to use econometric tools that are not available for smaller sample sizes. The inclusion of smaller countries leads to greater variation in the participation in the UEFA Champions League, which is necessary for a clear analysis of its impact. On top of this, it enables me to analyze the effect of market size on competitive balance. Contrary to the findings of Andreff and Bourg (2006), but in line with Frick and Prinz (2004), I find empirical support for the theoretical skepticism regarding collective sales of media rights in European soccer. I further provide supporting evidence for the result of Pawlowski et al. (2010), as I find Champions League participation to diminish competitive balance.

Model Specification

In this section I explicitly draw the relationship between the club revenue distribution, talent distribution, and competitive balance of a league in a simple model. These relationships implicitly underlie the reasoning on most measures taken to protect competitive balance. Modeling them explicitly allows deriving the factors that explain competitive balance variation among European leagues. Figure 1: Factors that Impact on Competitive Balance

[FIGURE 1 OMITTED]

As Figure 1 graphically depicts, sporting results heavily depend on talent investments. In general, more talented teams win more often. However, in any sports contest chance factors have an important role to play and therefore neither team is ever certain of a win. The extent to which chance factors are important depends on the format of the contest teams engage in. Some sports contest formats create larger advantages for more talented teams than others (see Groot, 2004 for more details). Therefore, competitive balance (CB), which is essentially the distribution of sporting results among clubs, depends on the distribution of playing talents (TD) and on the contest format (CF). Some measures which aim to protect competitive balance in sports, such as the introduction of a playoff system, are essentially contest format changes.

Clubs which enjoy large revenues have more potential to buy talented (i.e., expensive) players. As a result, the distribution of playing talents (TD) is strongly driven by the revenue distribution (RD) in a league. However, clubs have to buy talent on the labor market for players. Hence, player labor market conditions (PLM) may affect the extent to which high revenues effectively lead to the acquisition of talented players. Player labor markets have been a popular target for measures aimed at protecting competitive balance. Famous examples are salary caps and drafts of talented youngsters in American major leagues and the transfer system in European soccer. On the other hand, it is clear that revenue sharing measures (RS), such as collective sales of broadcast rights, try to impact on competitive balance by equalizing the revenue distribution in a league. Other factors that impact the distribution of revenues in a league are the distribution of local market sizes (drawing power (DP)) and the existence of performance- related rewards (prize money (PM)). An unequal distribution of drawing power and heavily performance-related rewards lead to a more unequal distribution of revenues and consequently a low level of competitive balance. The simple model which arises from these basic relationships is given by:

(1) CB = f(TD,CF)

(2) TD = g(RD,PLM)

(3) RD = h(DP,RS,PM)

As a large part of the variables in (1) to (3) are none quantifiable in nature (e.g., contest format), this model is not extremely useful for econometric estimation. However, the present paper examines differences in competitive balance levels by relating them to differences in protective measures taken by leagues. So, it makes sense to rephrase (1) to (3) into differences. The advantage of doing this is that most factors that are hard to quantify simply do not differ among European soccer leagues. This yields:

(4) d(CB) = f(d(TD),d(CF))

(5) d(TD) = g(d(RD),d(PLM))

(6) d(RD) = h(d(DP),d(RS),d(PM))

Expressions (4) to (6) still pose serious problems for data gathering and estimation. Especially reliable data on revenues and talent investments are hard to come by for most European soccer leagues. This issue may be resolved by equating a reduced form expression from the system (4) to (6), which is characterized as:

(7) d(CB) = j(d(CF),d(PLM),d(DP),d(RS),d(PM))

The Variable and Data Collection section outlines a procedure that was designed to calculate a measure for drawing power distribution that prevents feedback effects and endogeneity between d(DP) and d(CB).

Measuring Competitive Balance

Competitive balance in European soccer has various dimensions. Although other classifications are possible (e.g., Groot, 2004), Szymanski (2003) distinguishes between three main categories: match, seasonal, and championship competitive balance. Match level competitive balance refers to the uncertainty of outcome in a particular fixture during the season. Seasonal competitive balance relates to the amount of tension in the competition as a whole. A good seasonal competitive balance is found when the higher-ranked teams have a small points lead over the lower-ranked teams. Championship competitive balance concerns the degree of domination in a league over multiple seasons. If the same teams continuously achieve the top places at the end of the season, this type of competitive balance is low. In this paper I consider only seasonal and championship competitive balance.

Many authors have proposed measures of competitive balance (e.g., Humphreys, 2002). To gauge seasonal competitive balance, I rely on two different measures which both rescale the standard deviation of winning percentages (where a draw is worth half the points of a win). The first one builds on the insights of Quirk and Fort (1992). They introduced the idea to rescale the actual standard deviation in a league using the "ideal" standard deviation given the number of teams. This ideal deviation may be calculated by assuming an underlying probability distribution, i.e., the binomial distribution in this case. This measure will be referred to as QF. Second, I calculate NAMSI, a measure first proposed by Goossens (2006). NAMSI compares the actual standard deviation with the standard deviation under the worst possible competitive balance. Both measures allow for comparisons between leagues with a different number of teams. As a measure, NAMSI offers two advantages. First, there is no need to introduce any kind of probability distribution for the win percentages. Second, one can easily establish what the worst competitive balance is, while the "ideal" competitive balance is often a topic of dispute.

To measure championship competitive balance, I follow Dobson and Goddard (2001), who apply a points system on the basis of top-three, -four, or -five classifications. Under the top-three system the championship winner is awarded 3 points, the runner-up 2 points, and the third-placed team 1 point. The top-four system awards 4 points to the champion, 3 to the runner-up, and so forth. By adding up these points over several seasons, I obtain a distribution of scores among teams in the league. Subsequently, I calculate a gini coefficient on this cumulative distribution. If the number of teams in the league has changed during the sample period, I calculate the gini using the average number of teams over the seasons considered, controlling for the number of years in which the league consisted of this number of teams. This measure is referred to as gini3 for the top-three system, gini4 for the top-four, and gini5 for the top-five.

Another important issue in competitive balance measurement is how to deal with the presence of playoff systems. Under a playoff system, teams first play a qualifying stage. Only the best-placed teams advance to a second stage to decide the championship or, as the case may be, qualification for the European competitions. If a playoff system is applied, I calculate championship competitive balance on the basis of the playoff results. For seasonal competitive balance, this is impossible in most cases. Therefore, I make use of the results of the preliminary round to establish QF and NAMSI. If the first stage of the competition is played in different groups, this approach is again not feasible. In such cases, winning percentages of the final stage are taken into account.

Measuring Drawing Power Distribution

In this section, I propose a procedure for determining the distribution of drawing power in a league, as no suitable measure was found in the existing literature. First, I take the average attendance over the period 2001-2010 for each club. By relying on averages over the entire period, I eliminate the effect of exceptionally good or bad seasons for any given club. I calculate two different averages. In the first measure (dp1) only seasons in the top tier league are taken into account. In the second procedure (dp2) I also include seasons spent at the second level, whenever these data are available. The omission of any seasons played at a lower level is inspired by the fact that, during such seasons, the club in question will inevitably have played fixtures against commercially less interesting opponents. The inclusion of these seasons therefore holds a danger of the club's drawing power being underestimated. On the other hand, if clubs have on average better results at the lower level they may draw more spectators in the second tier. This might lead to an underestimation of the drawing power of relegating teams, when second-level seasons are excluded.

Subsequently, I examine for each season which clubs participated in the top league. From the averages for these clubs, I calculate the standard deviation. As I wish to obtain a measure that allows comparison between leagues, I divide by the average attendance in the league over the sample period. Thus,

[dp.sub.it] = [stdev.sub.it] [[Club Average.sup.01-10](attendance)]/[League Average.sup.01-10](attendance)

The subscript i refers to a certain league and t indicates a season. It is clear that a lower value of these measures indicates a more equal distribution of drawing power. The procedure aims at avoiding endogeneity in two ways. First, season averages might be influenced by the amount of tension during the season (i.e., seasonal competitive balance). However, by taking averages over the entire period, I eliminate this individual seasonal effect. Second, a tenser competition over the entire period could still impact positively across averages. This is resolved by dividing by the league averages for the entire period.

This measure of drawing power can only change through promotion and relegation of clubs. This corresponds to reality, as in practice this is the most important factor impacting on drawing power in the short run. An important drawback of relying on attendance data is that this approach may be biased by capacity constraints. Large clubs might tend to sell out more readily, which would mean their drawing power is effectively underestimated. As data on stadium capacities are not available in all leagues over the entire sample period, it is not possible to control for this bias in a systematical manner in the estimation results.

In order to check the robustness of the estimation results with respect to the chosen measure for drawing power I will estimate three distinct models. Under (A) no measure of drawing power enters the model. In a panel setting this means that the drawing power distribution is added to the fixed or random effect. This produces consistent estimates if the distribution of drawing power is relatively stable over time. In a cross-section model it may lead to omitted variable bias if the drawing power distribution is correlated heavily with the other variables. Model (B) contains the drawing power measure based solely on the seasons in the top league. Model (C) includes the measure which also considers seasons spent at the second level. If models (A)-(C) produce homogeneous estimates of the other model parameters, they are robust with respect to the chosen measure of drawing power. This would imply that the possible bias from capacity constraints has not changed the results.

Variable and Data Description

In order to go from (g) to an expression that may be used in econometric estimation I first examine which factors of (g) actually fluctuate among leagues. For each of these factors I then introduce (a) variable(s) to quantitatively characterize this difference.

Firstly, the league format may differ among leagues, either because they have a different number of clubs or because they have different types of competition formats. QF, NAMSI and gini3-5 are constructed in a way as to allow comparison if leagues differ in terms of number of teams, so no additional corrections are required in this respect. In terms of competition format European soccer leagues differ in two important respects, playoff systems and relegation rules. The presence of playoff systems may impact on competitive balance, because they could lead to changes in the effort teams choose to put into the qualifying stage. One might, for example, expect top teams to settle for second-stage qualification and not go all out for first place. In that case they would obtain a smaller points lead and the seasonal balance would be higher. I introduce three dummy variables to take this into account: po takes value 1 if a country had a conventional playoff system in a given season; po2 takes value 1 if a country had a competition involving two groups; efpo takes value 1 if a conventional playoff system was in place to decide on qualification for the European cup competitions, but not on the title race. Relegation rules on the one hand provide an incentive to smaller teams to make an effort to avoid relegation. On the other hand a very fierce relegation system may prevent clubs from firmly establishing themselves in the first division. In that sense it may deter investments in talented players. To control for this effect I introduce the variable relperc. It represents the percentage of teams relegated to the second tier, where clubs forced to play a relegation playoff are counted as half.

Secondly, player labor market conditions might affect competitive balance. However, the 1995 Bosman Ruling confirmed free movement of players between European countries and clubs. This led to the abolition of restrictions on the number of foreign players (such as the "3-plus-2 rule") and at once marked the end of the traditional transfer system. Consequently, the player labor market is no longer a discriminating factor among European leagues. In practice, all clubs operate on a single unified European player market. Therefore, I introduce no variables to account for differences in player labor market conditions.

Thirdly, I need to take due account of how the distribution of drawing power may vary across leagues. As explained in section four, a procedure was developed aiming to measure the drawing power distribution. Three different versions of the model are estimated. Under version (A) none of the drawing power measures is used. Version (B) contains the variable dp1, which is the measure containing only attendance data from the first division. Finally (C) includes dp2, the drawing power measure containing all available attendance data, both from the first and second division. The attendance data used to calculate dp1 and dp2 were obtained from European Soccer statistics. (2)

Next, the extent of revenue sharing in the league may be important. The only kind of revenue which is commonly shared in European soccer (as opposed to certain major leagues) is broadcast revenue. To introduce the effects of TV rights management in the cross-sectional model, I use a dummy variable br that takes value 1 for the collective model and 0 for the individual model. In the panel model I distinguish between three systems. Collective sales are the bottom line case. When pure individual sales are present, the dummy ind takes value 1. A mixed system where some clubs pool their rights, but not all participate is indicated by the dummy mix. (3)

The most important distributor of prize money in European football is the UEFA Champions League. Therefore, it is necessary to determine to what extent domestic competitions are affected by this Champions League. To this end I introduce two variables, the number of teams qualifying for the group stage of the tournament, clteams, and the total prize money in billions of Euros earned by teams from the league, clinc. The relevant timeframe for this variable runs from 2000 to 2009, as it is the prize money earned in the previous season that determines financial power in the league. In order to calculate clteams and clinc I made use of data provided by UEFA.

Finally, the size of the leagues' domestic market is introduced as a control variable. This is achieved by including the logarithm of GDP lgdp of a given country (or region) into the regression analysis. All GDP data are taken from the World Bank. In the case of the United Kingdom, these were split up using data from the Office for National Statistics. A larger market size might be beneficial to competitive balance for at least three reasons. First, I expect to see a higher quality of play across the league, so large teams will enjoy far fewer walkover victories. A related issue is the more secure financial situation of the smaller clubs, which tend to be threatened with financial insolvency more often in smaller leagues. Finally, the effects of prize money from the Champions League might be less disturbing to competitive balance in a larger market, as budgets will generally be greater in such a league.

The chosen sample period runs from the season starting in 2000 until the season starting in 2009. This firstly reflects the availability of attendance date from European Soccer Statistics. Further, it convenes to a period in which the Champions League competition format was not changed. This means that throughout 32 teams were allowed to participate in the group phase, with a maximum of four per country. In order to calculate NAMSI and Gini coefficients for this period, I made use of data from the RSSSF archives. (4)

I have at my disposal a fairly limited number of cross-sections (32). Further increasing the sample size turned out to be infeasible for four reasons. Firstly, the attendance data for Israel, Ireland, and Turkey were incomplete, so I choose not to calculate dp1 and dp2 using them. Secondly, some leagues were split up during the sample period (e.g., Serbia and Montenegro) and were consequently dropped from the sample. Thirdly, I have not included leagues that serve an exceptionally small market (e.g., Liechtenstein, Andorra, ...), as they are in most cases not employing professional players. Finally, I choose not to include leagues of a lower than premier league level, even though I have data for some of these leagues. In secondary leagues clubs compete for promotion, instead of a title or participation in European competitions. Furthermore, they can never be influenced by Champions League prize money and broadcast rights revenues are far less significant. A final issue is the presence of "parachute payments" in several leagues. In such an arrangement recently relegated clubs still receive a share of the collective TV revenues. All of these features make second tier leagues poorly comparable to the premier leagues.

Cross-Sectional Model

In this section, I estimate a cross-sectional model of championship competitive balance. For seasonal competitive balance I estimate a panel data model in the next section.

Table 1 gives an overview of the data employed in this model. Since the sample size is limited, the usual caveats apply and therefore I choose to limit the number of explanatory variables in the cross-section analysis. I introduce only one variable for playoffs, which takes value 1 if a playoff system was in place in the majority of the seasons 2001 to 2010. The dichotomous br variable represents the choice of TV rights management system. In order to avoid multicollinearity between both Champions League measures I only feed the clteams variable into the model. The results when only introducing clinc are exactly the same. All continuous variables take their average value over the sample period. The equations I estimate for gini3 are then given by:

(A) gini[3.sub.i] = [[beta].sub.0] + [[beta].sub.1][br.sub.i] + [[beta].sub.2][clteams.sub.i] +[[beta].sub.4][relperc.sub.i] + [[beta].sub.5][po.sub.i] + [[beta].sub.6][jlgdp.sub.i] + [u.sub.i]

(B) gini[3.sub.i] = [[beta].sub.0] + [[beta].sub.1][dp1.sub.i] + [[beta].sub.2][br.sub.i] + [[beta].sub.3][clteams.sub.i] + [[beta].sub.5][relperc.sub.i] + [[beta].sub.6][po.sub.i] + [[beta].sub.7][lgdp.sub.i] + [u.sub.i]

(C) gini[3.sub.i] = [[beta].sub.0] + [[beta].sub.1][dp1.sub.i] + [[beta].sub.2][br.sub.i] + [[beta].sub.3][clteams.sub.i] + [[beta].sub.5][relperc.sub.i] + [[beta].sub.6][po.sub.i] + [[beta].sub.7][lgdp.sub.i] + [u.sub.i]

The same equations are estimated for both gini4 and gini5. Table 2 depicts the OLS estimation results for the cross-sectional regressions, with heteroskedasticity robust standard error estimates provided in parentheses. I find no significant violations of OLS conditions in the data. As far as the individual coefficients are concerned, I find that the estimates are fairly consistent over different gini coefficients. Leagues which display a more equal distribution of drawing power are significantly more balanced (i.e., have a lower gini). The same is true when fewer teams participate in the Champions League. All other variables fail to be significant across models in a consistent way. This means that playoffs or relegation systems have no clear impact on the championship balance in a league. Most importantly, the use of collective sales appears not to increase the balance. The estimated coefficient is significant at the 0.1 level in only two cases, but has a positive sign. If anything, collective sales seem to erode the championship balance. Leaving out the drawing power measure, as is done in model (A), leads the R-squared to decrease substantially. The estimated coefficients of other variables do not drastically alter in models (B) or (C). This indicates that the results do not suffer from possible bias in the dp measures.

Panel Data Model

In this section, I estimate a panel data model for seasonal competitive balance (measured by NAMSI and QF). Table 3 provides an overview of the panel dataset.

I have at my disposal a balanced panel of 32 cross sections over a period of 10 years. At this point I include all variables mentioned previously, as the number of observations allows for identification. Using this dataset I conduct a regression analysis of the following models for NAMSI:

(A)[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

(B) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

(C) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

The same models are estimated using QF as the dependent variable. When estimating a panel model, it is necessary to control for unobserved heterogeneity between cross sections by introducing a fixed or random effect (variable [[gamma].sub.i] in (j)). An F-test on the joint significance of the estimated fixed effects in all models unequivocally confirms the presence of unobserved heterogeneity. Consequently, pooled OLS estimation is inconsistent. I therefore apply fixed or random effects estimators. A necessary condition for these estimators to be consistent is strict exogeneity between the independent variables and the individual effects. In order to test for this, I introduced the lead of the explanatory variables in the within estimator model. As I found none of the coefficients of the lead observations to be significant, I assume the strict exogeneity condition to have been met.

Table 4 and 5 give the estimation results for two procedures, the fixed and random effects estimator with robust standard errors. I find more coefficients to be significant and a higher R-squared when moving from fixed to random effects. On top of this, the fixed effects procedure cannot identify both the ind and mix variable at the same time. In general the estimated coefficients appear to be similar over all estimation procedures. This is tested more formally by the Hausman test statistic, presented in Table 6. The difference between fixed and random effects are found not to be systematic except in model (B) for NAMSI and (A) for QF. Therefore, preference is given to the random effects estimator in all other cases, as this is the more efficient procedure.

Over all models the estimation results again show that the impact of pooling broadcast rights is insignificant or even harmful to competitive balance. The negative effect of Champions League participation is confirmed by the positive coefficient of clinc. This result is consistently significant either at the 10-, 5- or 1-percent level over the majority of all estimations. I also find that a more unequal distribution of drawing power worsens seasonal competitive balance, an even more robust result. The random effects results reveal the positive effect of a large market size (measured by lgdp) for seasonal competitive balance. The only important difference between the NAMSI and QF model lies in the effect of playoff systems. In the QF model I find confirmation of the hypothesis that a championship playoff increases competitive balance in the preliminary round, as indicated by the significantly negative coefficient of po. The two other playoff variables remain insignificant. In the NAMSI model this effect is not significant, but it appears that introducing a competition in two groups worsens the balance in the final round. In general, it is clear that the estimated coefficients across models (A)-(C) are fairly homogeneous in Tables 4 and 5. This indicates again that the estimates are robust with respect to the way drawing power was measured. Introducing a measure of drawing power distribution, however, strongly raises the R-squared in all models.

Conclusions, Future Research, and Policy Recommendations

In this paper I applied an empirical approach to analyze factors impacting on seasonal and championship competitive balance in European soccer. I first constructed a dataset on competitive balance covering 32 leagues over a 10-year period. Subsequently, I introduced a procedure to assess the distribution of drawing power in a sports league. Using these data, I first found that the choice for either a collective or an individual sales mechanism for media rights has no positive impact on competitive balance. Secondly, I found evidence of the detrimental effect of Champions League participation on competitive balance. A third observation was that the leagues with a more equal distribution of drawing power show higher levels of competitive balance. Finally, a larger domestic market may also increase the seasonal competitive balance.

These results have important policy implications for the sport industry. First, they do not offer support for one of the major arguments in favor of collective sales of broadcast rights. This should inspire caution in competition authorities when granting leagues permission to monopolize TV rights. Other arguments are required to justify collective sales from a social welfare point of view. Secondly, UEFA should reconsider the way it organizes the Champions League. If UEFA wants to enhance solidarity and competitive balance in the soccer industry, distributing immense amounts among only the richest clubs would seem a rather counterproductive policy option. A final policy recommendation on the basis of my analysis is that expanding the markets of individual leagues through mergers might lead to an increased seasonal competitive balance in the merged league, as compared to the constituting national leagues.

The empirical exercise performed in this paper raises important questions and creates numerous opportunities for future research. One such question is why collective sales and redistributing revenues of broadcast rights have not aided competitive balance in European Soccer. Kesenne (2009) sheds some theoretical light on this issue, though, as he himself concedes, a lot of work remains to be done. A crucial issue in this discussion is, in my view, the nature of the distribution schemes adopted by collectively selling leagues. On the other hand, it is necessary to examine whether other arguments in favor of collective sales can justify the practice. A welfare economic analysis, such as performed by Falconieri et al. (2004), seems to be the way forward in addressing this topic. A final question is whether the results obtained in this paper also apply to other sports leagues, such as the North American major leagues.

References

Andreff, W., & Bourg, J. F. (2006). Broadcasting rights and competition in European football. In C. Jeanrenaud & S. Kesenne (Eds.), The economics of sport and the media (pp. 37-70). Cheltenham, UK: Edward Elgar Publishing Ltd.

Cave, M., & Crandall, W. (2001, February). Sports rights and the broadcast industry. The Economic Journal, 111, F4-F26.

Commission of the European Communities. (2007). EU White Paper on Sports.

Dobson, S., & Goddard, J. (2001). The economics of football. Cambridge, UK: Cambridge University Press.

Falconieri, S., Palomino, F., & Sakovics, J. (2004). Collective versus individual sale of television rights in league sports. Journal of the European Economic Association, 2(5), 833-862.

Forrest, D., Simmons, R., & Szymanski, S. (2004). Broadcasting, attendance and the inefficiency of cartels. Review of Industrial Organization, 24, 243-265.

Frick, B., & Prinz, J. (2004). Revenue sharing arrangements and the survival of promoted teams: Empirical evidence from the major European soccer leagues. In R. Fort & J. Fizel (Eds.), International Sports Economics Comparisons (pp. 141-156). Westport, CT: Praeger.

Goossens, K., (2006). Competitive balance in European football: Comparison by adapting measures: National measures of seasonal imbalance and top 3. Rivista di diritto ed economia dello sport, 2(2), 77-122.

Groot, L. (2008). Economics, uncertainty and European football. Cheltenham, UK: Edward Elgar Publishing Ltd.

Hausman, J. A., & Taylor, W. E. (1981, November). Panel data and unobservable individual effects. Econometrica, 9(6), 1377-1398.

Kesenne, S. (2009). The impact of pooling and sharing broadcast rights in professional team sports. International Journal of Sport Finance, 4(3), 211-218.

Noll, R. (2007, July). Broadcasting and team sports. Scottish Journal of Political Economy, 54(3), 400-421.

Palomino, F., & Sakovics, J. (2004). Inter-league competition for talent vs. competitive balance. International Journal of Industrial Organisation, 22, 783-797.

Pawlovski, T., Breuer, C., & Hovemann, A. (2010). The club's performance and the competitive situation in European domestic football competitions. Journal of Sports Economics, 11, 186-202.

Quirk, J., & Fort, R. (1992). Pay dirt: The business of professional team sports. Princeton, NJ: Princeton University Press.

Szymanski, S. (2003). The economic design of a sporting contest. Journal of Economic Literature, 41(4), 1137-1187.

TV Sports Markets. (2008). European TV & Sports Rights 2008. London, UK: TV Sports Markets Ltd.

Endnotes

(1) The big five leagues are the five dominant soccer leagues in Europe, i.e., England, France, Germany, Italy, and Spain.

(2) These data are retrievable online at http://www.european-football-statistics.co.uk

(3) Spanish competition authorities forced soccer clubs to sell their rights individually from 1997 onwards. Some clubs have however kept pooling, such that a mixed system has come into effect. In Portugal the competition council ended collective bargaining between the league and broker firm Olivedesportos on TV rights for summaries in 1997. This marked the definitive introduction of an individual system. Since then, however, SportTV, a private pay-tv channel, has succeeded in obtaining a monopoly position by buying TV rights from all premier league clubs. The Italian league adopted an individual selling system, following the passing of a law by the Italian Parliament on March 29, 1999. In the 2010-11 season Italian clubs move back to collective sales. This season falls just outside the scope of this study. Law 2725/1999, also voted in 1999, established that Greek clubs are the sole owners of the broadcast rights to their home games. As in Italy, this law resulted in the introduction of an individual sales mechanism. This system came into effect after the last collective agreement expired in 2001. Recently the Greek and Italian leagues have made efforts to return to a collective sales mechanism. These efforts have resulted in a mixed system from the 2009-10 season onwards, because two clubs (Olympiakos and Xanthi) refused to participate.

(4) The RSSSF is the Rec. Sport. Soccer Statistics Foundation. More information on this organisation and the full archives can be found at http://www.rsssf.org.

(5) A negative value for the Hausman test value indicates that the conditions for performing the test are not satisfied in this case. We therefore should stick to the FE estimates here.

Author's Note

I would like to thank Stefan Kesenne, Jan Bouckaert, Stefan Szymanski, Stephanie Leach, Stijn Rocher, and three anonymous referees for help and useful suggestions. This paper further benefited from comments received upon its presentation at seminars of the University of Antwerp, at the European Conference on Sports Economics (Sorbonne, Paris, 2009), and at the Flemish Economic Society conference (LUC, Hasselt, 2009). I gratefully acknowledge financial support from the Flanders Research Foundation (FWO) and the University of Antwerp TOP research program. Any remaining errors are my own.

Thomas Peeters

University of Antwerp

Thomas Peeters is a doctoral researcher in the Faculty of Applied Economics. His research interests include industrial organization, contest theory, and the economics of sport.

Table 1: Cross-Sectional Dataset Summary

Variable Observations Mean Stdev Min Max

gini3 32 0.686 0.113 0.461 0.835
gini4 32 0.627 0.122 0.416 0.806
gini5 32 0.563 0.128 0.332 0.767
br 32 0.875 0.336 0 1
dp1 32 0.562 0.210 0.314 1.136
dp2 32 0.600 0.213 0.324 1.184
clteams 32 0.947 1.204 0 3.8
relperc 32 0.146 0.031 0.083 0.214
po 32 0.094 0.296 0 1
lgdp 32 25.365 1.555 23.026 28.303

Table 2: Estimation Results for Cross-Sectional Model (robust std.
errors in parentheses)

 gini3 gini4

VAR (A) (B) (C) (A) (B)

br 0.000634 0.0796 * 0.0809 * -0.00467 0.0776
 (0.0473) (0.0422) (0.0441) (0.0528) (0.0475)
dpi 0.212 *** 0.221 **
 (0.0727) (0.0844)
dp2 0.201 ***
 (0.0717)
clteams 0.089 *** 0.094 *** 0.096 *** 0.090 *** 0.095 ***
 (0.0250) (0.0231) (0.0235) (0.0239) (0.0227)
relperc 0.264 0.277 0.330 0.451 0.464
 (0.541) (0.479) (0.484) (0.610) (0.550)
lgdp -0.0287 -0.0292 -0.0320 * -0.0256 -0.0262
 (0.0193) (0.0182) (0.0182) (0.0194) (0.0185)
po 0.0613 ** 0.00293 0.00974 0.0490 -0.0119
 (0.0248) (0.0354) (0.0344) (0.0311) (0.0450)
Cons 1.284 ** 1.109 ** 1.167 ** 1.125 ** 0.942 *
 (0.463) (0.456) (0.455) (0.470) (0.476)
Obs 32 32 32 32 32
R-sq 0.493 0.589 0.581 0.482 0.570

 gini4 gini5

VAR (C) (A) (B) (C)

br 0.0764 -0.00032 0.0826 0.0814
 (0.0502) (0.0562) (0.0494) (0.0525)
dpi 0.223 **
 (0.0860)
dp2 0.203 ** 0.205 **
 (0.0844) (0.0865)
clteams 0.097 *** 0.095 *** 0.100 *** 0.102 ***
 (0.0230) (0.0227) (0.0222) (0.0224)
relperc 0.517 0.549 0.562 0.616
 (0.555) (0.638) (0.555) (0.565)
lgdp -0.0290 -0.0228 -0.0234 -0.0262
 (0.0183) (0.0190) (0.0186) (0.0184)
po -0.00308 0.0273 -0.0341 -0.0252
 (0.0433) (0.0297) (0.0368) (0.0357)
Cons 1.007 ** 0.969 ** 0.785 0.850 *
 (0.473) (0.457) (0.485) (0.479)
Obs 32 32 32 32
R-sq 0.558 0.519 0.602 0.591

* significant at 0.1 level ** significant at 0.05 level
*** significant at 0.01 level

Table 3: Panel Dataset Summary

Variable Observations Mean Stdev Min Max

namsi 320 0.483 0.109 0.201 0.869
QF 320 1.715 0.403 0.754 3.220
ind 320 0.091 0.182 0 1
mix 320 0.034 0.288 0 1
dp1 320 0.562 0.209 0.251 1.167
dp2 320 0.600 0.212 0.271 1.205
clinc 320 0.015 0.027 0 0.129
clteams 320 0.947 1.274 0 4
relperc 320 0.146 0.044 0 0.375
po 320 0.103 0.305 0 1
efpo 320 0.025 0.079 0 1
po2 320 0.006 0.156 0 1
lgdp 320 25.357 1.543 22.715 28.369

Table 4: Panel Data Model Results Namsi (robust std errors
in parentheses)

VARIABLES Fixed Effects

namsi (A) (B) (C)

ind 0.00720 -0.00591 -0.0147
 (0.0236) (0.0232) (0.0233)
mix

dp1 0.332 **
 (0.147)
dp2 0.466 ***
 (0.137)
clinc 0.797 0.763 0.778 *
 (0.489) (0.472) (0.448)
clteams -0.00992 -0.0129 -0.0127
 (0.00805) (0.00807) (0.00787)
relperc -0.150 -0.132 -0.137
 (0.109) (0.111) (0.111)
lgdp -0.0512 -0.0410 -0.0451
 (0.0339) (0.0337) (0.0324)
po -0.00545 -0.00951 -0.0115
 (0.0191) (0.0183) (0.0181)
po2 0.0289 0.0466 * 0.0502 *
 (0.0284) (0.0281) (0.0287)
efpo -0.00581 0.000412 0.00335
 (0.0274) (0.0270) (0.0269)
COnstant 1.801 ** 1.357 1.370 *
 (0.861) (0.870) (0.823)

Observations 320 320 320
Overall R-sq 0.230 0.305 0.259

VARIABLES Random Effects

namsi (A) (B) (C)

ind 0.0390 -0.0420 -0.0546
 (0.0369) (0.0404) (0.0410)
mix -0.0128 -0.0670 -0.0741
 (0.0494) (0.0459) (0.0460)
dp1 0.223 ***
 (0.0629)
dp2 0.247 ***
 (0.0626)
clinc 0.866 ** 1.125 *** 1.182 ***
 (0.427) (0.413) (0.409)
clteams -0.00661 -0.0102 -0.00997
 (0.00742) (0.00746) (0.00737)
relperc -0.153 -0.137 -0.136
 (0.0998) (0.0992) (0.0972)
lgdp -0.0423 *** -0.0396 *** -0.0426 ***
 (0.0101) (0.00948) (0.00935)
po -0.00122 -0.0125 -0.0136
 (0.0157) (0.0153) (0.0152)
po2 0.0295 0.0417 ** 0.0411 **
 (0.0213) (0.0195) (0.0193)
efpo 0.0133 0.0104 0.00974
 (0.0284) (0.0270) (0.0271)
Constant 1.569 *** 1.383 *** 1.435 ***
 (0.255) (0.248) (0.242)

Observations 320 320 320
Overall R-sq 0.265 0.362 0.356

* significant at 0.1 level ** significant at 0.05 level
*** significant at 0.01 level

Table 5: Panel Data Model Results QF (robust std
errors in parentheses)

VARIABLES Fixed Effects

QF (A) (B) (C)

ind 0.0431 -0.0121 -0.0398
 (0.0876) (0.0849) (0.0856)

mix

dp1 1.401 ***
 (0.518)

dp2 1.762 ***
 (0.508)

clinc 2.946 * 2.804 * 2.874 *
 (1.745) (1.685) (1.623)

clteams -0.0334 -0.0459 -0.0441
 (0.0304) (0.0305) (0.0301)

relperc -0.330 -0.251 -0.278
 (0.427) (0.429) (0.429)

lgdp -0.123 -0.0803 -0.100
 (0.139) (0.136) (0.132)

po -0.188 ** -0.205 ** -0.211 **
 (0.0889) (0.0849) (0.0845)

po2 -0.173 -0.0985 -0.0927
 (0.129) (0.102) (0.101)

efpo -0.0336 -0.00743 0.000993
 (0.101) (0.0990) (0.0991)

Constant 4.898 3.025 3.266
 (3.545) (3.461) (3.328)

Observations 320 320 320

Overall R-sq 0.123 0.201 0.179

VARIABLES Random Effects

QF (A) (B) (C)

ind 0.0834 -0.282 ** -0.324 **
 (0.132) (0.137) (0.140)

mix -0.0741 -0.321 ** -0.345 **
 (0.170) (0.155) (0.156)

dp1 0.998 ***
 (0.215)

dp2 1.067 ***
 (0.215)

clinc 3.456 ** 4.936 *** 5.152 ***
 (1.533) (1.479) (1.471)

clteams -0.0239 -0.0405 -0.0386
 (0.0285) (0.0287) (0.0284)

relperc -0.503 -0.511 -0.499
 (0.411) (0.407) (0.400)

lgdp -0.125 *** -0.117 *** -0.130 ***
 (0.0391) (0.0344) (0.0341)

po -0.163 ** -0.213 *** -0.216 ***
 (0.0699) (0.0664) (0.0668)

po2 -0.163 -0.102 -0.107
 (0.122) (0.0883) (0.0900)

efpo 0.0233 0.0124 0.0110
 (0.105) (0.0993) (0.0998)

Constant 4.939 *** 4.223 *** 4.475 ***
 (0.988) (0.891) (0.875)

Observations 320 320 320

Overall R-sq 0.155 0.321 0.321

* significant at 0.1 level ** significant at 0.05
level *** significant at 0.01 level

Table 6: Hausman Test Results

Hausman test NAMSI QF

 (A) (B) (C) (A) (B) (C)

Chi sq. (8) 13.79 17.79 14.99 -11.02 (5) 4.25 10.35
Prob > Chi sq. 0.087 0.038 0.091 NA 0.895 0.323