Journal of Sports Economics Market Competition and ...
Transcript of Journal of Sports Economics Market Competition and ...
Article
Market Competition andThreshold Efficiency inthe Sports Industry
Young Hoon Lee1, Hayley Jang1,and Sun Ho Hwang1
AbstractPrevious studies of competitive balance (CB) have analyzed the variations in gameoutcomes that is the second moment of winning percentage. This article differs in twoaspects. First, it analyzes market competition with respect to efficiency. Second, itanalyzes efficiency distribution (in particular, the third moment of the efficiency dis-tribution and the efficiency bound). It also suggests the efficiency bound as a newmeasure of market competition. By applying stochastic frontier models to a panel dataset of the European football leagues (English Premier League, Spanish La Liga, ItalianSerie A, and German Bundesliga), it derives important implications. First, the minimumefficiency estimates seem to reveal information relevant to market competition andadditional information from the conventional CB measure. Second, differences in theshapes of the efficiency distribution curves across the four leagues imply that the fourleagues faced different environments in terms of market competition.
Keywordsmarket competition, minimum efficiency, competitive balance, stochastic frontiermodels, football
Introduction
A joint product is a peculiar characteristic that is unique to the sports industry and
cannot be observed in other industries. The two sports teams (firms) are able to
1 Department of Economics, Sogang University, Seoul, Korea
Corresponding Author:
Young Hoon Lee, Department of Economics, Sogang University, Baekbeom-ro, Mapo-gu, Seoul 121-742,
Korea.
Email: [email protected]
Journal of Sports Economics1-20
ª The Author(s) 2014Reprints and permission:
sagepub.com/journalsPermissions.navDOI: 10.1177/1527002514556719
jse.sagepub.com
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
produce an indivisible product when they play each other in the same field. The
importance of competitive balance (CB) of playing talents among teams is driven
by this factor because two teams with equal strength generally produce a tight
match, which fans prefer. Neale (1964) called this the ‘‘league standing effect,’’
and said, ‘‘The closer the standings, and within any range of standings the more
frequently the standings change, the larger will be the gate receipts’’ (p. 3). There-
fore, studies on CB occupy a significantly large portion of the literature of sports
economics (e.g., Fort & Quirk, 1995; Noll, 1988; Szymanski, 2003). Fort and
Maxcy (2003) categorized the study of CB into the analysis of CB itself and the
relationship between CB and fan demand. The former study covers measurements
of CB, the impact of regulation changes on CB, and time-series analysis of CB,
while the latter study tests for uncertainty of outcome hypothesis.
This article focuses on the former study and approaches the study of CB in a
different way from previous studies. All of the previous studies examined the ulti-
mate outcome of matches, winning percentage, and the dispersion of wins across
member teams when they measure CB. That is, they analyzed CB by using the sec-
ond moment of winning percentage. For example, the relative standard deviation
of actual win percentage to the idealized standard deviation (RSD) is the second
moment of win percentage. However, the higher moments may include valuable
information relevant to CB. A positive skewness (the third moment) of win per-
centage implies that there are a few extremely good teams and the number of teams
with below 0.5 is greater than half of the total members, while a negative skewness
implies that there are a few teams with an extremely low win percentage and the
number of teams with above 0.5 is greater than half of the total members. The rem-
edy will be different depending on the sign of the skewness. In the case of positive
skewness, CB is worsened because of the teams with extremely high levels of
talent, meaning that a remedy would be most effective when it focuses on those
top teams. The luxury tax adopted in major league baseball may be an example
of this. On the other hand, imposing a high salary floor may be a legitimate remedy
for enhancing CB in cases of negative skewness.
This article differs from the previous studies in two aspects. First, it analyzes
market competition by examining technical efficiency. Second, it examines not only
the second moment of efficiency but also other characteristics of the distribution, in
particular, the threshold efficiency (efficiency bound). There have been empirical
studies on the relationship between market competition and productivity (De
Loecker, 2011; Dunne, Klimek, & Schmitz, 2008; Holmes & Schmitz, 2010; Matsa,
2011; Syverson, 2004). They presented empirical evidence that increased competi-
tion led to productivity improvement. In particular, Syverson (2004) pointed out the
importance of the efficiency bound in the way the market competition influences the
distribution in the market since low-productivity firms cannot survive when market
competition is tight. He presented two distributions of productivity: one for plants in
markets above median competition and another for plants in markets below median
competition.1 As expected, the mean (SD) of productivity in the high-competition
2 Journal of Sports Economics
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
market is higher (smaller) than the mean (SD) in the low-competition market. In
addition, the distribution of the high-competition market is skewed toward the left
(a negative skewness). That is, the minimum productivity is higher in the high-
competition market than in the low-competition market.
The minimum efficiency may also be informative in the sports industry. It can
be interpreted as the threshold efficiency for survival. League structure, the
strength of revenue sharing, and governance structure may influence market com-
petition. In an open league with relegation and promotion, the low-tier teams must
make their best efforts to build up efficiency. They are generally ‘‘low-budget
teams’’ and have to figure out an efficient way to win and to stay in the first divi-
sion. On the other hand, the low-tier teams in a closed league do not face the risk of
being thrown out of their league every season. The minimum efficiency is expected
to be different, based on league structure. Sports teams in East Asia are generally
supported by major corporations, while sports teams in North America and Europe
operate as independent businesses. It is expected that there is less competition in
East Asian sports leagues, since financial losses are subsidized by parent compa-
nies. The governance structure may influence the competition level and the min-
imum efficiency.
Recently, several studies in the literature of stochastic frontier models (SFMs)
have shown interest in the minimum level of efficiency (Almanidis, Qian, &
Sickles, 2014; Feng & Horrace, 2012; Hwang & Lee, 2014; Lee & Lee, 2013).
Almanidis, Qian, and Sickles (2014) and Lee and Lee (2014) focused on threshold
inefficiency. They implicitly perceived the argument of Syverson (2004) and spec-
ified their models, reflecting the fact that inefficient firms cannot survive in the
real world and, thus, there should be a bound of inefficiency. For example, Alma-
nidis et al. assumed the one-sided distribution of inefficiency as the tail-truncated
normal or the doubly truncated normal distribution. Incorporating threshold inef-
ficiency into SFMs not only improves the explanatory power of regression but
also allows the parameterization of the threshold inefficiency.
This article introduces the minimum efficiency as an alternative measure of
market competition. It is understood that this measure is not directly related to the
level of game outcome uncertainty but is closely related to market competition
that ultimately influences the balance of playing talents across member teams.
It applies the SFMs with the minimum efficiency to the panel data set of the
European football leagues (English Premier League, Spanish Primera Liga, Italian
Serie A, and German Bundesliga). We adopt both the semi-parametric and para-
metric approaches to compare efficiency distributions of the four leagues. The
study of Schmidt and Sickles (1984) is regarded as demonstrating the semi-
parametric approach since it does not impose any parameterization on the ineffi-
ciency distribution and the threshold efficiency is not considered in the model.
By applying this model, we estimate the efficiency of all individual teams
and compare the efficiency distributions of the four leagues. Hwang and Lee
(2014) constitute the extension to Lee and Lee (2014). This model imposes the
Lee et al. 3
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
group-specific minimum efficiency on their model specification and then is used
in our study to compare minimum efficiency among the four leagues. This is the
parametric approach, as it assumes a distribution of inefficiency and there are para-
meters defining the distribution.
The main purpose of this article was not to compare the market competition
levels of the four European football leagues in depth. Instead, it proposes an alter-
native measure of market competition and evaluates its validity by the applied
study. This article found that there are reasonably large variations in the minimum
efficiency across leagues and that the shape of the efficiency distribution is not the
same as the shape of the win percentage distribution. Moreover, the rank of the min-
imum efficiency among the four leagues is also not the same as the rank of RSD, the
conventional CB measure. We believe that our introduction of the alternative measure
of market competition and the empirical findings contribute to the literature of CB
since they nest information that is not found in previous studies on CB.
This article is organized as follows. In the second section, we introduce both the
semi-parametric and parametric SFMs. Data sources and the descriptive statistics are
discussed in the third section. Empirical results and their discussion are presented in
the fourth section. Finally, the fifth section contains our concluding remarks.
Econometric Models
In order to examine the distribution of efficiency, we utilize two different SFMs. The
first SFM is developed by Schmidt and Sickles (1984), which has been used widely.
This model assumes the time invariance of efficiency and is a semi-parametric
approach since it does not impose any distributional restriction on efficiency.
Another SFM is by Hwang and Lee (2014); this model focuses on the estimation
of efficiency bound and comparison of efficiency bounds across markets or groups.
It assumes a uniform distribution of technical inefficiency. A production function
and a parameter of inefficiency bound can be estimated simultaneously by the
method of moments or the maximum likelihood estimation (MLE), while the effi-
ciency of an individual firm is calculated by the conditional expectation that was
originally proposed by Jondrow, Lovell, Materov, and Schmidt (1982).
The SFMs were proposed by Aigner, Lovell, and Schmidt (1977, ALS) and
Meeusen and van den Broeck (1977). The SFM is of the form
yi ¼ aþ x0ibþ vi � ui; i ¼ 1; . . . ;N ; ð1Þ
where yi is a log of output. xi is a K-dimensional vector of exogenous variables and
b is the K-dimensional vector of parameters to be estimated. The vi represents an
uncontrollable disturbance term. The ui is a technical inefficiency (more spe-
cifically, logged output loss due to technical inefficiency) term and ui � 0. The
disturbances ui and vi are distributed independently of each other and of the
4 Journal of Sports Economics
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
regressors. It is often assumed that vi is distributed normally with zero mean and
variance of s2v . ALS assumed ui is distributed either with a half normal density
or with an exponential density. Various SFMs arose by assuming different one-
sided distributions of ui. For example, Stevenson (1980) and Greene (1990)
assumed a truncated normal distribution and a gamma distribution, respectively.
In panel data setting, Schmidt and Sickles (1984) imposed the following produc-
tion function.
yit ¼ aþ x0itbþ vit � ui ¼ x0itbþ ai þ vit; i ¼ 1; . . . ;N ; t ¼ 1; . . . ; T ; ð2Þ
where ai ¼ a� ui. Equation 2 is exactly the same as the conventional panel data
model with individual effects. Thus, it can be estimated by the fixed effects (the
within estimation) and random effects (the generalized least squared (GLS) estima-
tion) treatments. They proposed the following max operator2 for the estimation of
technical inefficiency, ui and we will examine the distribution of the estimates.
ui ¼ a� ai; a ¼ maxi ai: ð3Þ
Recently, Almanidis et al. (2014) and Lee and Lee (2014) developed SFMs by
focusing on efficiency bound. Lee and Lee assume ui to be distributed uniformly
with a parameter y, that is, ui � Uð0; yÞ. Therefore, the probability density function
of ui is given by
f uið Þ ¼1
yIð0;yÞ uið Þ; ð4Þ
where Ið0;yÞðuiÞ is an indicator function and y > 0. This means that the support is
defined by the parameter y. The parameter y is the upper bound of the distribution
of ui, and we may call it the inefficiency threshold.3 Because the inefficiency
threshold represents the maximum level of inefficiency or minimum level of effi-
ciency, the parameter y identifies not only the dispersion of inefficiency but also
the degrees of competition. Hwang and Lee (2014) extended this model in order
to allow researchers to be able to compare efficiency bounds across markets or
across groups.
Hwang and Lee (2014) assumed the time invariance of inefficiency in the same
way as Schmidt and Sickles (1984), and thus Equation 2 is their regression equa-
tion. They also assumed ui to be distributed uniformly but allow for a different
parameter of y in different markets. That is, the probability density function of
ui is given by
fjðuiÞ ¼1
yj
Ið0;yjÞðuiÞ; i 2 Gj; j ¼ 1; . . . ;G; ð5Þ
where an individual firm i belongs to market j. This model allows for market-
specific inefficiency bounds. Hwang and Lee (2014) derived the probability density
function of ei ¼ vi � 1T ui, where 1T is a T dimensional vector of ones as
Lee et al. 5
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
fjðeiÞ ¼1
yj
1
ð2pÞðT�1Þ=2sT�1V
ffiffiffiffiTp exp � 1
2
ðei � 1T�eiÞ0ðei � 1T�eiÞs2
V
� �
Fyj þ �ei
sV=ffiffiffiffiTp
� �� F
�ei
sV=ffiffiffiffiTp
� �� �;
ð6Þ
where �ei ¼ ð1=TÞP
eit and F is a cumulative standard normal distribution.
This model can be estimated by the MLE and the method of moments. It is well
known that the maximum likelihood estimators are asymptotically efficient, but
both methods provide us with consistent estimators and our panel data sample is
not large.4 Therefore, this article utilizes the method of moments as explained
below. After estimating the production function (2) by either the within or GLS
estimator, we can estimate the variance of vi and yj by the following equations that
are derived by moments.
s2v ¼
T
T � 1m2 ð7Þ
y2j ¼ 12 m�2;j �
T
T � 1m2
� �for i 2 Gj; ð8Þ
where m2 is the sample second moment of the error term in the within trans-
formed production function, m2 � E½e2it� and ei � MT yi � Xibð Þ5 and m�2;j is the
demeaned sample second moment of the error term in the production function,
m�2;j � E½e�it � Eðe�itÞ�2
and e�it � yit � a� x0itb for i 2 Gj. Finally, the inefficien-
cies of individual firms are estimated by the conditional expectation method
of Equation 9:
EjðuijeiÞ ¼�sVffiffiffiffi
Tp F
yj þ �ei
sV=ffiffiffiffiTp
� �� F
�ei
sV=ffiffiffiffiTp
� �� ��1
fyj þ �ei
sV=ffiffiffiffiTp
� �� f
�ei
sV=ffiffiffiffiTp
� �� �� �ei:
ð9Þ
Data and Descriptive Statistics
We constructed a panel data set of four major European football leagues (English
Premier League, Spanish Primera Liga, Italian Serie A, and German Bundesliga)
over three seasons (2006-2007, 2008-2009, and 2010-2011). The source of our
data is www.transfermarkt.com, which has been used in many sports economics
studies (Battre, Deutscher, & Frick, 2009; Bryson, Frick, & Simmons, 2013; Frick,
2007). Production functions need to be estimated in order to obtain technical effi-
ciency estimates. The output of a soccer team is wins or points, and playing talents
6 Journal of Sports Economics
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
are inputs. The website evaluates the market value of players based on their
collection of experts’ and supporters’ opinions. Torgler and Schmidt (2007) inves-
tigated the correlation between players’ effective reported salaries and estimated
market values as provided by Transfermakt.de6 and found a relatively strong pos-
itive correlation (.735). Finally, we constructed a panel data set with 234 observa-
tions for 105 teams. The promotion and relegation system makes it unbalanced.
Table 1 shows descriptive statistics for our sample data set. We calculated total
points in a season as 3 points for a win and 1 point for a draw. Bundesliga has a
smaller number of matches in a season than the other three leagues, and this dif-
ference results in lower average points. Thus, we use average points per match
in our regression. It is inferred that Bundesliga and Primera Liga are more
balanced than the Premier League and Serie A since the SDs of win percentage and
points in the former are smaller than those in the latter. On the other hand, Primera
Liga had the most variations in playing talents across member teams, while
Bundesliga had the least variations.
Figure 1 reveals the kernel density function of the winning percentage of each
league over the three seasons.7 It is clear that all four distributions are skewed to
the right. There are a few extremely outperforming teams in each league. Therefore,
the number of teams with a winning percentage below 0.5 is larger than that with
above 0.5. These positively skewed distributions are evidence that analysis on the
higher moments can provide additional information. For example, it is likely that a
few outstanding teams are the main source of competitive imbalance. Then, reme-
dies focusing on the few teams may be effective if a league decides to enhance CB.
Table 1. Descriptive Statistics.
Variable Observations Mean SD Maximum Minimum
Premier League Win% 60 0.50 0.18 0.93 0.20Total points 51.90 15.51 90.00 28.00Players valuea 1.59 1.16 4.30 0.33
Primera Liga Win% 60 0.50 0.16 0.94 0.23Total points 52.67 14.73 96.00 28.00Players valuea 1.25 1.22 6.02 0.29
Serie A Win% 60 0.50 0.18 0.97 0.17Total points 51.9 15.70 97.00 24.00Players valuea 1.11 0.85 3.46 0.23
Bundesliga Win% 54 0.50 0.16 0.82 0.22Total points 47.00 12.91 75.00 26.00Players valuea 0.89 0.56 3.09 0.26
Total Win% 234 0.50 0.17 0.97 0.17Total points 50.97 14.87 97.00 24.00Players valuea 1.22 1.02 6.02 0.23
Note. aMillion Euro.
Lee et al. 7
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
Empirical Results
Equation 2 is our production function for the four European football leagues. The
dependent variable, yit, is the winning percentage (Win%) or points per game. The
inputs in sports industry are playing talents and Value is our proxy variable for
playing talents. Since a quadratic form is adopted, xit ¼ fValue;Value2g.8 In the
manufacturing industry, various specifications of production have been studied
and typical functions, such as Cobb–Douglas and Translog production functions,
have been discovered. However, the specification of sports production function
has not been studied seriously yet. Our strategy was to estimate both linear and
logged production functions and then compare the estimates in order to check
robustness upon functional specification. Because of the joint product characteris-
tic of the sports industry, the outputs of a sports team (wins) are determined not
only by a given team but also by rival teams. An increase in the amount of playing
talent of a sports team does not necessarily guarantee increases in wins. The num-
ber of wins in a season may decrease if other teams hire more playing talents. It is
inferred that relative playing strength would be a more legitimate input variable.
Therefore, we make the playing talent variable, Value, to be a relative variable
by dividing it by league average playing talent.
Table 2 reveals the estimation results of Schmidt and Sickles (1984). The four
different specifications of production function are estimated. The winning percent-
age is a dependent variable in the first and second columns and Value (a relative
measure of playing talent) is included as a linear form in the first column and as a
Figure 1. Kernel Density Function of Winning Percentages.
8 Journal of Sports Economics
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
logged form in the second. The dependent variable is points per game (Points) in
the third and fourth columns while the linear specification of production function
is adopted in the third column and the logarithmic specification in the fourth col-
umn. The four different specifications of production function generate similar esti-
mation results and the explanatory powers are more or less equal to each other. For
example, comparing the two Win% equations, the coefficient estimate of Value2 is
negative, but the coefficient estimate of (ln Value)2 is positive. However, both esti-
mates in the first and second columns produce diminishing marginal products of
Value. The last row presents the Hausman test results and the null hypothesis that
the random effects specification is correct cannot be rejected. Therefore, both
within and GLS estimates are consistent, but GLS estimates are more efficient.
We will discuss the GLS estimates of the Win% regression equation with the
logged Value hereafter.
Figure 2 displays the kernel density functions of technical efficiency estimates
for the four different football leagues that are calculated by the max operator of
Equation 3. That is, this distribution is obtained by the semi-parametric approach
of Schmidt and Sickles (1984). It is somewhat surprising that the distributions in
Figure 2 are distinctly different from those in Figure 1, which contains winning
percentage. The winning percentage distributions shown in Figure 1 were skewed
to the right, and all four distributions of different leagues are reasonably similar to
each other. However, the distributions of efficiency are noticeably different across
leagues and positive skewness can be found only in the English Premier League
according to Figure 2. There are a few extremely efficient teams in the Premier
League and the median efficiency is lower than the mean efficiency. On the other
Table 2. Production Function Estimates.
Variables Win% Win% Points lnPoints
Panel 1: Within estimationValue 0.206* (2.85) 0.472* (2.79)Value2 �0.013 (�0.80) �0.025 (�0.62)lnValue 0.191* (4.59) 0.299* (4.29)(lnValue)2 0.072* (2.02) 0.093 (1.56)
R2 .828 .828 .829 .826Panel 2: GLS estimation
Value 0.304* (11.78) 0.667* (11.36)Value2 �0.039* (�5.15) �0.075* (�4.38)ln Value 0.214* (23.11) 0.344* (19.63)(ln Value)2 0.040* (3.12) 0.028 (1.21)
Constant 0.253* (15.66) 0.527* (60.84) 0.827* (22.45) 0.345* (21.18)R2 .825 .825 .827 .822Hausman test 2.776 [.250] 2.041 [.360] 1.984 [.371] 2.913 [.233]
Note: The numbers given in parenthesis are t values and those given in square brackets are p values.*5% significance level.
Lee et al. 9
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
hand, the German Bundesliga has a symmetric distribution of efficiency estimates,
while the distributions in Primera Liga and Serie A are likely to be skewed to the
left. In particular, there are several extremely inefficient teams in Serie A. Table 3
also reveals descriptive statistics of the efficiency estimates obtained by the semi-
parametric approach. The mean efficiency is the highest in Primera Liga, whereas
it is the lowest in Serie A. The SD is the smallest in Primera Liga, in which the
distribution is skewed to the left the most. The estimation results of Primera Liga
(high mean, small variance, and negative skewness of efficiency) are consistent
with Syverson’s (2004) argument relevant to market competition. That is, our esti-
mation results with the semi-parametric approach imply that Primera Liga faces
the tightest competition among the four leagues. The Premier League seems to
be located on the opposite side. Its mean efficiency is located in the middle, but
the variance of efficiency is the largest and the skewness is positive, implying that
Figure 2. Kernel Density Function of Efficiency Estimates: the Semi-parametric Approach.
Table 3. Descriptive Statistics: Efficiency Estimates of Individual Teams by the Semi-Parametric Approach.
League Mean SD Maximum Minimum Skewness
Premier League 0.814 0.064 1.000 0.718 0.773Primera Liga 0.820 0.045 0.891 0.734 �0.359Serie A 0.809 0.056 0.943 0.674 �0.249Bundesliga 0.813 0.052 0.898 0.715 �0.004
10 Journal of Sports Economics
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
a reasonably large number of inefficient teams can survive in the first division.
Therefore, it is inferred that the Premier League has the least tight competition
among the four different leagues.
Turning our attention to the efficiency estimates of the parametric approach, the
minimum efficiency of each league can be estimated by this approach. The mini-
mum efficiencies are estimated by the method of moments of Equation 8 and the
estimates are presented in the first column of Table 4. The means of efficiency esti-
mates with the semi-parametric approach and the conventional measure of CB and
RSD are presented in the second and third columns, respectively. Let’s compare
the minimum efficiency estimates with the mean efficiency estimates first and then
compare those with the RSDs. From this comparison, we may be able to see how
the implications of minimum efficiency estimates are different from those of the
mean efficiency and the conventional CB measure, which have been examined
in previous studies. Comparing the mean efficiency estimates driven by the
semi-parametric approach, Primera Liga has the highest minimum efficiency
(0.721) and mean efficiency (0.820) and both measures suggest that Primera Liga
has the tightest market competition. However, there are a few minor disparities
between the parametric and semi-parametric approaches. For example, the compe-
tition level of Bundesliga is the lowest with respect to minimum efficiency but is in
the middle with respect to the mean efficiency across the four leagues.
Bundesliga and Primera Liga have smaller RSDs than the Premier League and
Serie A. Thus, it is inferred that Bundesliga and Primera Liga are more balanced
than the other two leagues with respect to RSD. In other words, Bundesliga has the
tightest competition in the football field. According to the minimum efficiency
estimates, the tightest competition exists in Primera Liga, whereas the least com-
petition is in Bundesliga and this empirical finding contradicts the results of RSD.
The Premier League and Serie A have the same level of minimum efficiency and
their RSDs are rather similar to each other. In summary, the market competition
implied by the minimum efficiency is generally similar to that by the RSD, but
there are also some contradictory implications. The findings, that the minimum
efficiency estimation draws some different implications with regard to market
competition from the mean efficiency or the conventional CB measure, are not
revelatory since CB is concerned with game outcome, while efficiency distribution
Table 4. The Minimum Efficiency Estimates: The Parametric Approach.
MinimumEfficiency
Mean of Efficiency(the Semi-Parametric Approach) RSD
Premier League 0.697 0.814 2.244Primera Liga 0.721 0.820 1.960Serie A 0.697 0.809 2.256Bundesliga 0.679 0.813 1.889
Lee et al. 11
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
concerns not only game outcome but also input usages. The different implications
may be empirical evidence that minimum efficiency can reveal additional informa-
tion relevant to market competition.
Figure 3 displays the kernel density functions of technical efficiency estimates
for the four different football leagues that are calculated by the conditional expec-
tation of Equation 9. Note that the conditional probability density function fjðuijeiÞis not the same as the uniform density function even though ui is assumed to be
uniform. The kernel density functions in Figure 3 are somewhat different from
those in Figure 2. For example, the distributions in Figure 3 are located in a nar-
rower range. However, the main findings in Figure 2 can be observed in Figure 3
too. The Premier League has the efficiency distribution skewed to the right,
whereas the distribution of Primera Liga is skewed to the left. Based on the values
of skewness, it is inferred that Primera Liga faces the tightest competition, while
the Premier League has the least tight competition.
Turning attention to the efficiency of individual teams, Table 5 reports the most
efficient teams and the least efficient teams of each league based on the efficiency
estimates of the parametric approach. Reading FC and Manchester United are the
top two teams in efficiency ranking of the Premier League, while Charlton Athletic
and Newcastle United are the least efficient teams. Considering the fact that
Reading FC appeared in the Premier League only once during our sample periods,
the league standing is not directly equal to the efficiency ranking. Real Sporting
de Gijon in Primera Liga is another example that is generally a low-tier team in
league standing but is placed high in the efficiency ranking. Reading FC and Real
Figure 3. Kernel density function of efficiency estimates: the parametric approach.
12 Journal of Sports Economics
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
Sporting de Gijon would have limited budgets and low payrolls in their leagues;
thus, they are unable to rise to the top in league standings even though they operate
their teams in technically efficient ways.
Table 6 (Table 7) selects the top (bottom) five teams from each league based on
average league standing during the sample period. It also reveals their efficiency
rankings and efficiency estimates. Generally speaking, highly ranked teams were
relatively efficient but not necessarily. For example, all top teams in Bundesliga
except for Bayern Munchen are very efficient. On the other hand, FC Barcelona
and Real Madrid, which are perennial winners in Primera Liga, have relatively low
Table 5. The Most Efficient and the Least Efficient Teams.
LeagueBestEfficiency
Second-BestEfficiency
BestInefficiency
Second-BestInefficiency
Premier League Reading FC(.961)
Manchester United(.927)
Charlton Athletic(.735)
NewcastleUnited (.739)
Primera Liga Malaga CF(.890)
Real Sporting deGijon (.881)
Real Betis (.756) UD Almerıa(.759)
Serie A FC Empoli(.932)
FC Genus (.879) AS Bari (.718) ACR Messina(.730)
Bundesliga Dortmund(.896)
Kaiserslautern (.894) Monchengladbach(.725)
Bremen (.733)
Note. The numbers given in parenthesis are efficiency estimates.
Table 6. Top 5 Teams in Average League Standings and Efficiency Estimates.
TeamRank
EfficiencyRank Team
TeamRank
EfficiencyRank Team
Premier League Primera Liga
1 2 Manchester United (.961) 1.3 17 FC Barcelona (.807)2.3 10 FC Chelsea (.838) 1.7 23 Real Madrid (.770)3.7 8 FC Liverpool (.845) 4.3 8 Sevilla FC (.852)4 5 Arsenal FC (.853) 4.3 15 Valencia CF (.814)6 18 FC Everton (.786) 4.7 4 Villarreal CF (.866)
Serie A Bundesliga
1.3 3 Inter Mailand (.877) 3 20 Bayern Munchen (.766)2.7 13 F.C. Milano (.820) 5.3 4 VfB Stuttgart (.872)4.5 20 Juventus (.797) 5.3 8 Leverkusen (.844)4.7 19 AS Roma (.799) 5.3 1 Dortmund (.896)6 4 Lazio Rom (.866) 6 3 Nurnberg (.880)
Note. The numbers given in parenthesis are efficiency estimates.
Lee et al. 13
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
efficiency. This implies that these two teams won many championships with large
payrolls and, thus, there is a lot of room to save on payrolls while still being able
to win championships if they can enhance their efficiency. Compared with Table 7,
it is clear that the average efficiency rankings of the top teams are higher than those
of the bottom in league standings. The bottom five teams in league standings are
also found to be inefficient technically. This implies that teams are placed in the
bottom of league standings not only because of small budgets but also because
of inefficient management of playing talent.
Table 8 also examines the technical efficiencies of the bottom teams in league
standings as a way to compare surviving teams in the first division with relegated
teams during the sample period. Five teams from each group are selected based
on average league standing. That is, the five surviving teams have the lowest league
standings on average among teams that remained in the first division during the sam-
ple period and the five relegated teams are also located on the lowest average league
standing among teams that were relegated at least once during the sample period.
Generally, efficiency rankings of surviving teams are better than those of relegated
teams even though there are several exceptions. For example, West Ham United and
Newcastle United in the Premier League are the exceptions since they had remained
in the first division in all three seasons, but their efficiency rankings are 24 and 26,
respectively. The average ranking of the five teams from each group suggests that
the surviving teams are more efficient than the relegated teams, since the survivors’
average rankings are higher than the relegated teams’ average rankings in the Pre-
mier League, Primera Liga, and Serie A. Bundesliga is the exception in that the
Table 7. Bottom 5 Teams in Average League Standings and Efficiency Estimates.
TeamRank
EfficiencyRank Team
TeamRank
EfficiencyRank Team
Premier League Primera Liga
20 25 FC Watford (.741) 20 24 Gimnastic de Tarragona (.763)19 9 FC Blackpool (.842) 19 22 Hercules CF (.777)19 27 Charlton Athletic (.735) 19 18 CD Numancia (.803)18 20 Birmingham City (.776) 18 21 Celta de Vigo (.782)18 12 Sheffield United (.827) 17 26 Real Betis (.756)
Serie A Bundesliga
20 27 ACR Messina (.730) 18 17 FC St. Pauli (.779)20 28 AS Bari (.718) 17 21 Karlsruher SC (.762)19 23 Brescia (.770) 17 14 Aachen (.801)19 26 Ascoli (.740) 16.3 24 Monchengladbach (.825)18.5 21 U.S. Lecce (.786) 15 15 Bielefeld (.788)
Note. The numbers given in parenthesis are estimated efficiency.
14 Journal of Sports Economics
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
Tab
le8.
Effic
iency
Com
par
ison
ofth
eR
emai
ned
Tea
ms
inth
eFi
rst
Div
isio
nan
dth
eR
eleg
ated
Tea
ms
toLo
wer
Div
isio
n.
Rem
ained
Tea
mEffic
iency
Ran
kR
eleg
ated
Tea
mEffic
iency
Ran
kR
emai
ned
Tea
mEffic
iency
Ran
kR
eleg
ated
Tea
mEffic
iency
Ran
k
Pre
mie
rLe
ague
Pri
mer
aLi
ga
Wig
anA
thle
tic
15
FCW
atfo
rd25
Get
afe
CF
9G
imnas
tic
de
Tar
rago
na
24
Wes
tH
amU
nited
24
FCBla
ckpool
9O
sasu
na
14
CD
Num
anci
a18
New
cast
leU
nited
26
Char
lton
Ath
letic
27
Dep
ort
ivo
11
Her
cule
sC
F22
Bla
ckburn
Rove
rs13
Bir
min
gham
City
20
Mal
lorc
a6
Cel
tade
Vig
o21
Bolton
Wan
der
ers
7Sh
effie
ldU
nited
12
Ath
letic
Clu
b16
Rea
lBet
is26
Ave
rage
rank
17.0
18.6
Ave
rage
rank
11.2
22.2
Seri
eA
Bundes
liga
Chie
vo18
AS
Bar
i28
Monch
engl
adbac
h24
FCSt
.Pau
li17
Cag
liari
9A
CR
Mes
sina
27
Fran
kfurt
22
Kar
lsru
her
SC21
Sam
pdori
a24
Bre
scia
23
Wolfs
burg
19
Aac
hen
14
Cat
ania
8A
scoli
26
Bre
men
22
Bie
lefe
ld15
Udin
ese
15
U.S
.Le
cce
21
Han
nove
r5
Cott
bus
10
Ave
rage
rank
14.8
25.0
Ave
rage
rank
18.4
15.4
Not
e.Fi
vete
ams
with
the
low
est
aver
age
leag
ue
stan
din
gar
ese
lect
edfr
om
each
group.
15 at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
survivors’ average efficiency ranking is 18.4, but the relegated teams have a higher
average ranking of 15.4. In particular, the gaps between the average efficiency
rankings between the surviving teams and the relegated teams are large in Primera
Liga (11.2 vs. 22.2) and Serie A (14.8 vs. 25.0). These results are consistent with
Syverson’s (2011) assertion which is called Darwinian selection or reallocation
effect. Darwinian selection is the process in which inefficient firms fall behind and
then the market moves toward more efficient firms and the efficiency threshold
rises. The empirical results of Primera Liga are consistent with the reallocation
effect, as the least efficient teams exited the market (the first division), as seen
in Table 8, and the minimum efficiency is the highest among the four leagues in
Table 4, and its efficiency distribution is skewed to left the most in Table 3. In
Serie A, the survivors’ average efficiency is also significantly higher than the rele-
gated teams’ average efficiency, so this implies that inefficient ‘‘firms’’ fell out of
the market. Table 3 shows that the efficiency distribution of Serie A is skewed to
the left, implying that the exits raised the efficiency bound. The empirical results
of the Premier League and Bundesliga are also consistent with the reallocation
effect. As shown in Table 3, the efficiency distribution of Bundesliga is more or
less symmetric and that of the Premier League is skewed to the right, implying that
the efficiency bounds are low. The minimum efficiency estimates in Table 4 con-
firm this implication. Bundesliga has the lowest minimum efficiency and the Pre-
mier has the second lowest. The potential reason can be found in Table 8. In these
two leagues, ‘‘firms’’ pushed out of the market are not more significantly ineffi-
cient than surviving ‘‘firms.’’
Conclusions
This article aimed to contribute to sports economics literature by developing an
alternative measure of market competition. Unlike previous studies, which have
focused on game outcome, winning percentage, and its second moment for analyses
of CB, we examined (1) the efficiency distribution (in particular, efficiency thresh-
old) and (2) the third moments (skewness) of the distribution. We adopted SFMs for
this purpose. The first SFM of Schmidt and Sickles (1984) has been used in various
empirical studies, and we analyzed the distribution of efficiency estimates obtained
by this model. The second SFM of Hwang and Lee (2014) was only developed
recently and has not been applied yet. The model allows for estimation of the effi-
ciency bound in a market and comparison of the efficiency bounds across different
markets in a sample. This is an alternative measure of market competition.
The empirical results that we obtained by applying the two SFMs to the unba-
lanced panel data set of the four major European football leagues permit us to
derive several remarkable implications. First, the minimum efficiency estimates
seem to reveal information relevant to market competition, additional to the infor-
mation provided by the mean efficiency or the CB measure. For example, the
16 Journal of Sports Economics
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
distributions of efficiency estimates in the four leagues are strikingly different
from those of winning percentage. Second, the skewness and shapes of the effi-
ciency distribution curves are significantly different across the four leagues. This
implies that the four leagues faced somewhat different environments in terms of
market competition. This is a bit surprising because the player movement rulings
are uniform across European football leagues. There may be league-specific or
country-specific characteristics influencing market competition. Third, we found
that the reallocation effect, which has been observed in previous productivity stud-
ies of manufacturing industry (Foster, Haltiwanger, & Syverson, 2008; Syverson,
2011), is also confirmed in our empirical exercise of European football leagues. A
market is more efficient and has a higher efficiency threshold as inefficient teams
are drawn out of the market through the relegation system. In our empirical results,
this reallocation effect could be clearly observed. Primera League, in which inef-
ficient teams were driven out of the first division, could achieve a higher efficiency
threshold as well as a higher average efficiency. On the other hand, Bundesliga,
which failed to throw inefficient teams out of the first division, could not achieve
a high-efficiency bound. Fourth, the large variations in technical efficiency across
leagues and across different member teams imply that there is a lot of room to save
in the cost of playing talents. Lago, Simmons, and Szymanski (2006) argued that
overinvestment in players is one of main sources of financial crisis in European
football leagues. In other words, the overinvestment can be remedied if a team was
able to enhance its managerial efficiency.
With respect to future studies with our market competition measure, we would
like to address the following suggestions. First, more exact specifications of pro-
duction function would improve accuracy of efficiency estimates. As workforces
can be divided into skilled or unskilled workers, playing talents in a sport team can
be categorized based on skills. For example, Lee and Berri (2008) divided total
playing talents by positional playing talents (guards, small forwards, and big men)
in their analysis of basketball production. Second, studies on determinants of mar-
ket competition in sports industries would help in deriving practical policy impli-
cations. The league structure, regulations regarding revenue sharing, governance
structure, and regulations regarding foreign players, among other factors, may
influence market competition and, thus, efficiency threshold.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, author-
ship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship,
and/or publication of this article: This work was supported by the National Research Founda-
tion of Korea Grant funded by Korean Government (NRF-2013S1A3A2053312).
Lee et al. 17
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
Notes
1. Specifically, Syverson (2004) divided markets by density. Consumers’ substitutability
increases in a market with high density, and thus it can be inferred that firms in markets
above median density face tight competition.
2. For the properties of this max operator, refer to Kim, Kim, and Schmidt (2007).
3. The efficiency threshold can be calculated by e�y.
4. Olson, Schmidt, and Waldman (1980) compared the small sample properties of the max-
imum likelihood estimator (MLE) and the method of moments (corrected ordinary least
squared) estimator in the model of Aigner, Lovell, and Schmidt through the Monte Carlo
simulation. They found their mean biases and root mean squared errors (RMSEs) are sim-
ilar to each other, but RMSEs of the method of moments estimator are smaller than those of
MLE when the sample size is smaller than 200.
5. MT is the within transformation matrix defined as MT ¼ IT � ð1=TÞ1T 10T .
6. Transfermakt.de is the German version of transfermarkt.com.
7. The kernel density functions of points are analogous to Figure 1. They are available upon
request.
8. The cubic and quartic specifications of production function are often found in the labor
economics literature. However, adding them in our regression did not increase explanatory
power noticeably.
References
Aigner, D. J., Lovell, C. A. K., & Schmidt, P. (1977). Formulation and estimation of sto-
chastic frontier production function models. Journal of Econometrics, 6, 21–37.
Almanidis, P., Qian, J., & Sickles, R. C. (2014). Stochastic frontier models with bounded
inefficiency. In R. Sickles & W. C. Horrace (Eds.), Festschrift in honor of Peter Schmidt
(pp. 47–81). New York, NY: Springer.
Battre, M., Deutscher, C., & Frick, B. (2009). Salary determination in the German Bundesliga:
A Panel Study. Stellenbosch, In No 0811, IASE Conference Papers. International Associ-
ation of Sports Economists.
Bryson, A., Frick, B., & Simmons, R. (2013). The returns to scarce talent footedness and
player remuneration in European soccer. Journal of Sports Economics, 14, 606–628.
De Loecker, J. (2011). Product differentiation, multiproduct firms, and estimating the impact
of trade liberalization on productivity. Econometrica, 79, 1407–1451.
Dunne, T., Klimek, S., & Schmitz, J. (2008). Does foreign competition spur productivity?
Evidence from post WWII US cement manufacturing (Working Paper). Minneapolis,
MN: Federal Reserve Bank Minneapolis.
Feng, Q., & Horrace, W. C. (2012). Alternative technical efficiency measures: Skew, bias and
scale. Journal of Applied Econometrics, 27, 253–268.
Fort, R., & Maxcy, J. (2003). Comment: Competitive balance in sports leagues: An introduc-
tion. Journal of Sports Economics, 4, 154–160.
Fort, R., & Quirk, J. (1995). Cross-subsidization, incentives, and outcomes in professional
team sports leagues. Journal of Economic Literature, 23, 1265–1299.
18 Journal of Sports Economics
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
Foster, L., Haltiwanger, J., & Syverson, C. (2008). Reallocation, firm turnover, and efficiency:
Selection on productivity or profitability? The American Economic Review, 98, 394–425.
Frick, B. (2007). The football players’ labor market: Empirical evidence from the major
European leagues. Scottish Journal of Political Economy, 54, 422–446.
Greene, W. H. (1990). A gamma-distributed stochastic frontier model. Journal of Econo-
metrics, 46, 141–163.
Holmes, T. J., & Schmitz, J. A. (2010). Competition and productivity: A review of evidence.
Annual Review of Economics, 2, 619–642.
Hwang, S. H., & Lee, Y. H. (2014). Stochastic frontier models for comparison of threshold
efficiency (Working Paper). Seoul, South Korea: Research Institute for Market Economy,
Sogang University.
Jondrow, J., Lovell, C. A. K., Materov, I. S., & Schmidt, P. (1982). On the estimation of
technical inefficiency in the stochastic frontier production function model. Journal of
Econometrics, 19, 233–238.
Kim, M., Kim, Y., & Schmidt, P. (2007). On the accuracy of bootstrap confidence intervals
for efficiency levels in stochastic frontier models with panel data. Journal of Productiv-
ity Analysis, 28, 165–181.
Lago, U., Simmons, R., & Szymanski, S. (2006). The financial crisis in European football an
introduction. Journal of Sports Economics, 7, 3–12.
Lee, Y. H., & Berri, D. (2008). A re-examination of production functions and efficiency
estimates for the national basketball association. Scottish Journal of Political Economy,
55, 51–66.
Lee, Y. H., & Lee, S. (2014). Stochastic frontier models with threshold efficiency. Journal of
Productivity Analysis, 42, 45–54.
Matsa, D. A. (2011). Competition and product quality in the supermarket industry. The Quar-
terly Journal of Economics, 126, 1539–1591.
Meeusen, W., & Van den Broeck, J. (1977). Efficiency estimation from Cobb–Douglas
production functions with composed error. International Economic Review, 18,
435–444.
Neale, W. C. (1964). The peculiar economics of professional sports. The Quarterly Journal of
Economics, 78, 1–14.
Noll, R. G. (1988). Professional basketball. Stanford University studies in Industrial Eco-
nomics Paper No. 144. Stanford, CA: Stanford University.
Olson, J. A., Schmidt, P., & Waldman, D. M. (1980). A Monte Carlo study of estimators of
stochastic frontier production functions. Journal of Econometrics, 13, 67–82.
Schmidt, P., & Sickles, R. C. (1984). Production frontiers and panel data. Journal of Business
and Economic Statistics, 2, 367–374.
Stevenson, R. E. (1980). Likelihood functions for generalized stochastic frontier estimation.
Journal of Econometrics, 13, 57–66.
Syverson, C. (2004). Market structure and productivity: A concrete example. Journal of
Political Economy, 112, 1181–1222.
Syverson, C. (2011). What determines productivity? Journal of Economic Literature, 49,
326–365.
Lee et al. 19
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from
Szymanski, S. (2003). The economic design of sporting contests. Journal of Economic
Literature, 41, 1137–1187.
Torgler, B., & Schmidt, S. L. (2007). What shapes player performance in soccer? Empirical
findings from a panel analysis. Applied Economics, 39, 2355–2369.
Author Biographies
Young Hoon Lee, PhD, is a professor in the Department of Economics at Sogang University.
His research fields include econometrics (panel data models and productivity analysis) and
sports economics.
Hayley Jang, is a graduate student in the Department of Economics at Sogang University. Her
research fields include applied microeconomics and sports economics.
Sun Ho Hwang is a graduate student in the Department of Economics at Sogang University.
His research field includes econometric theory.
20 Journal of Sports Economics
at PENNSYLVANIA STATE UNIV on May 17, 2016jse.sagepub.comDownloaded from