localization economies and firm productivity: evidence...
TRANSCRIPT
Localization Economies and Firm Productivity:Evidence from Football Teams in São Paulo, Brazil
Brad R. Humphreys∗
West Virginia University
Amir B.Ferreira Neto†
Florida Gulf Coast University
Preliminary DraftDo not cite or circulate without permission
Abstract
Agglomeration economies clearly affect firms in urban areas. Interestingly, the ex-isting literature on outcomes in professional sports largely ignores these effects. Weanalyze variation in sports team productivity and the agglomeration of teams acrossleagues and cities in Campeonato Paulista an annual football competition played in SãoPaulo state in Brazil. Our results show that localization and urbanization positivelyaffect team success. These results help to shed more light on why teams in larger citiescontinuously enjoy more success than those isolated in smaller markets.Keywords: Agglomeration economies, urbanization economies, sports league out-comes, football, BrazilJEL Classification: R12, Z21
∗Department of Economics, 1601 University Ave., PO Box 6025, Morgantown, WV 26506-6025, USA;Email: [email protected]†Lutgert College of Business, 10501 FGCU Blvd. S., Fort Myers, FL 33965; Email: aborgesfer-
1
1 IntroductionAgglomeration economies clearly affect firms in urban areas. Firms in specific industriestend to cluster together, and economic activity tends to cluster in small spatial areas. Firmsin cities are more productive because of the spatial concentration of other firms in theirindustry and because of the urban spatial concentration of overall economic activity andpopulation. A large body of empirical evidence shows that both localization economies andurbanization economies affect firms in a number of industries across a large number of cities.
Marshall (1895) first introduced the idea of agglomeration economies or agglomerationexternalities. Firms who locate close to each other enjoy localization economies due to:sharing of intermediate inputs, job-market pooling and matching, and knowledge spillovers.Glaeser et al. (1992) develop the MAR (Marshall (1895) – Arrow (1962) – Romer (1986))model to explain localization from knowledge spillovers. Another source of agglomerationeconomies are urbanization economies which originates in the work of Jacobs (1969). Jacob’sargument opposes Marshall’s in that diversification represents a key ingredient to fosterinnovation and growth. Moreover, Jacobs (1969) argues that urban scale, i.e., density andthe level of local demand also affect firm’s outcomes. A large body of empirical literatureexamines the impact of localization and urbanization economies (Moomaw, 1988; Henderson,2003; Viladecans-Marsal, 2004; Duranton and Overman, 2005; Devereux et al., 2007; Grootet al., 2014; Galliano et al., 2015), but there is no conclusive answer in terms of which effectdominates.
Urbanization economies clearly play an important role in the performance of teams inprofessional sports leagues. The standard textbook economic model of sports league out-comes emphasizes that teams play in home markets (cities) of different sizes, and that teamproductivity increases with the size of the home market (Fort and Quirk, 1995). This modelpredicts that teams in larger cities will be more successful than teams in smaller cities; alarge body of empirical research confirms this prediction.
Interestingly, the existing literature on outcomes in professional sports largely ignoreslocalization economies. Standard textbook models of sports league outcomes do not includeclustering of teams in cities and do not address any impact this might have on outcomes liketeam success. This appears to be a curious omission, since sports teams clearly cluster inlarge cities. Three MSAs in the US1 – New York, Los Angeles and Chicago – have more thanfive professional football, basketball, baseball and hockey teams and another 10 MSAs havefour. The greater London metro area currently has 12 teams playing in the top two footballleagues in the UK and a large number of top-level football teams in the German Bundesligaare concentrated in the Rhein-Ruhr region.
The literature on sport and localization is thin and recent. Most of this literature focuseson broadcast viewership or team pricing decisions. Mills et al. (2016); Mondello et al. (2017)discuss how shared markets impact broadcast viewership in Major League Baseball (MLB)and the National Football League (NFL), finding evidence supporting localization effects inprofessional baseball broadcasts and evidence of local media cannibalization in professionalfootball broadcasts. Driessen and Sheffrin (2017) analyze how industry concentration im-
1From a North American perspective, one can also think about Minor Leagues in baseball and ice hockeythat concentrate in specific regions.
2
pacts location decisions for race car drivers and golfers, finding that tax preferences are animportant determinant of clustering for golfers, but agglomeration reduces this effect on thelocation of race car drivers, who tend to cluster in a small area in North Carolina. Hen-rickson (2012) analyzes pricing decisions made by professional sports teams in US cities andfinds evidence of spatial competition in pricing. None of these papers analyze the impact oflocalization on team success. To the best of our knowledge this is the first paper analyzingthe impact of both localization and urbanization economies on team’s level of success.
Identifying how localization and urbanization externalities influence team outcomes re-quires variation in the concentration of teams in cities over time. Team relocation representsone way to generate such variation, but relocation of teams is a rare event in US leagues andelsewhere. However, unlike professional sports leagues in the US, football leagues in Europeand South America use a promotion-relegation structure which generates variation in thedegree of localization of teams in different leagues/divisions as some teams are promotedand others relegated at the end of every season.2
We analyze variation in team productivity and the agglomeration of teams across leaguesand cities in Campeonato Paulista (CP), an annual football competition played by teamsin São Paulo state in Brazil. São Paulo state is an interesting setting for the analysisof localization and team sports outcomes. It is roughly the size of the United Kingdom(248,222 km2 versus 243,610 km2) and has a similar population to Spain (45,149,486 versus46,347,576). São Paulo state contains three cities with a population of more than one million(including São Paulo with a population of 11 million) and another six cities with populationof more than 500,000.
Campeonato Paulista is a good setting for this analysis because it has a promotion andrelegation structure that generates variation in localization and also has several institutionalfeatures that help mitigate some possible empirical issues. For instance, the CP competitionoccurs over the January through May period, which does not coincide with the primaryfootball hiring season in Europe (June-September). Moreover, because it is a regional league,although some teams may hire international players, most of the talent will be from within thestate. This would be a bigger concern in the case of Europe, as players have free movementacross European Union members. São Paulo has the oldest state football competition in thecountry and is the wealthiest state in Brazil.
We collect data on Campeonato Paulista outcomes from 2007 to 2018. We focus ontwo types of outcomes: short-term and long-term outcomes. Short-term outcomes are thoseduring season-year: win-loss ratio, goal difference, and points scores. Long-term outcomesare proxied by the average league-level played and by an Elo score we developed.
We also collect data on time-varying characteristics of the municipalities such as employ-ment and value added in production by industry. We calculate four indexes of concentration:Herfindahl-Hirschman Index (HHI) of the concentration of teams in each microregion overtime, the raw concentration index developed by Ellison and Glaeser (1994, 1997) (EGI), theraw concentration measure developed by Maurel and Sédillot (1999) (MSI), and the shareof teams within 50km of its official address. We proxy for urbanization economies using
2In all parts of the world except the US and Canada, the immensely popular eleven player ball and goalsport is called football. In the US and Canada, this sport is called soccer. In keeping with the worldwideconvention, we refer to this sport as football throughout this paper.
3
population density and median wage.Our results show that localization and urbanization has no effect team’s short-term suc-
cess. However, we find that localization economies, have heterogenous effects on the longterm success. While being surrounded by better teams is associated with better long-termsuccess (Elo points) it does decrease the probability of a team to be in the higher level di-visions. On the other hand, the concentration mid-level teams has no effect on team’s longterm success. However, lower division teams impacts negatively the likelihood of successmeasured by the league played.
The implications of these results for sports leagues are two-fold. First, in the US context,with increasing movement of teams across cities, and renewed discussion of league expansion,these results can inform teams about how their location decision may affect their long-runlevel of success. In the European and Brazilian context, these results help to shed more lighton why teams in larger cities continuously enjoy more success than those isolated in smallermarkets.
In an urban economics context, team success is a clean measure of firm productivity. Inother settings, firm productivity could increase because of external factors like localizationand urbanization economies, but could also reflect benefits from internal scale economies;estimates of productivity increases from localization could, in part, reflect unobservableinternal scale economies. Football teams cannot put more than eleven players on the pitchat one time, and capital plays a very small role in team success. So more successful teams aremore productive with the same level of labor and capital inputs as less successful teams. Thelocalization effects estimated in this context are unlikely to be contaminated by unobservableinternal scale economies.
The remainder of the paper is as follows: section 2 discusses the institutional details ofCampeonato Paulista; section 3 presents our conceptual model; section 4 describes the dataand econometric model used; section 5 analyses our results; and section 6 concludes.
2 The Campeonato PaulistaThe Campeonato Paulista (CP), commonly called Paulistão, is the professional footballcompetition in the state of São Paulo in Brazil. This is the oldest and most traditional statecompetition in the country. Its first competition dates to 1901; 102 different teams competedin it. As a brief historic background, the CP was founded by Charles Miller in 1901 andthe players competing in the CP became professionalized in 1933. In this year, the CPmade history as the first professional football tournament in Brazil. In 1941, the FederaçãoPaulista de Futebol (FPF), the football confederation in São Paulo state, was founded andsince then is the organization responsible for Paulistão. In the next two sections we discusssome of the institutional details of the CP, such as the league/division structure and detailsof team formation, management, and player transactions.3
3The terms “league” and “division” are used interchangeably throughout this paper to refer to variouslevels of competition in the CP. “League” is widely used to describe tiers of domestic European footballcompetitions. For example: the Premier League, the English Football League Championship, League One,and League Two in England; Bundesliga 1, Bundesliga 2, and Bundesliga 3 in Germany; Ligue 1, Ligue 2,and Ligue 3 in France.
4
2.1 Division Structure and Organization
Since 1901, the CP has used several competition formats and included a variable number oftiers of competition and numbers of teams in each tier. CP is a promotion-relegation leaguewith an annual schedule. Currently, there are four tiers or leagues in the CP: Série A1, A2,A3 and Segunda Divisão (Série A4 henceforth). Série A1 is the top league. Série A1, A2and A3 matches are played between January and May; Segunda Divisão matches are playedbetween April and November. Below we summarize4 each division’s rules and structure ineach of the years in the sample.
Série A1 contains 20 teams. Four of them are relegated to Série A2 at the end of eachseason. Between 2007 and 2016 the Série A1 season had four different formats. From 2007to 2010, there were three stages in the competition. In the first stage, every team would playeach other and the four teams with the fewest points would be relegated; the top four teamswould play in the second stage knockout round, with the first place team facing the fourthplace team and the second place team facing the third place team. The winners would faceeach other in the third and final stage.
From 2011 to 2013, the first stage format was maintained, but 8 teams qualified for thesecond stage knockout round. From 2014 on, in the first stage the teams were divided into4 groups of 5 teams, and the top two teams in each group faced each other in the secondknockout round. The bottom four teams at the end of the first stage were still relegatedSérie A2 in all these years. In 2017, CP distributed around R$ 12 million Brazilian Reais inprize money to teams, compared to R$ 10 million in 2016.
Série A2 also used several formats in sample period. From 2007 to 2009 and from 2012 to2013, the 20 clubs would play each other in a first stage double round-robin competition andthe 4 worst teams would be relegated Série A3. The top 8 clubs would advance to a secondstage of group play. These 8 clubs were divided into two groups of four and played a secondround-robin competition. The winner of each group would then play for the championshipand the top two teams in each group would be promoted to Série A1.
In 2010, the format was similar to the one described above, however there was no finalmatch between the second-stage group winners. In 2011 the format was similar to the 2007-2009 period, but in the first round the 20 clubs were divided into 2 groups of 10. In 2014and 2015 there was a single round with teams facing each other once. The four top teamswere promoted to Série A1 and the top team was the champion, while the four worst wererelegated.
In 2016 the format was quite different. There were 4 rounds of play. In the first roundthe 20 clubs played each other once and the four bottom clubs were relegated. The eightbest advanced to a second knockout round. The four remaining teams in the semi-finals werepromoted to Série A.
Série A3 had the most consistent format. In all the analyzed seasons 20 teams played adouble round-robin stage with the top 4 promoted to Série A2 and the bottom 4 relegatedto the Segunda Divisão (Série A4). In 2016, however, 6 teams were relegated instead of 4.From 2007 to 2016, with exception of 2011, the 20 teams faced each other. The 8 best teamsof were then divided into two groups in the second round. The teams played group memberstwice and the groups’ winner faced each other on the finals. The two best teams in each
4These are based on the information provided by FPF at http://futebolpaulista.com.br/Home/
5
group of the second round were promoted to Série A2. The difference in 2011 is that in thefirst round the teams were divided into two groups of 10 playing each other twice (withingroup only).
Segunda Divisão (Série A4) is the league with the most different formats The leaguecontracted from 48 teams in 2007 to 32 teams in 2016. Overall the league format can besummarized as follows. The first round would have teams divided into 4 to 7 groups. In thesecond round 12 to 24 teams would be divided into 4 or 6 groups and play each other. In thethird round there would be either 2 groups of 4 teams or a knockout-style final stage. Theforth round would be wither the semi-finals for cup-style format or else the final between thewinners of each group. The fifth-round, when in place was the final in the cup-style format.The best 4 teams – those in the semi-finals or else the two best in each group in the roundprior to the final – were promoted to Série A3. An extra feature of this league is that, eventhough this is the only access league to Série A3, players must be under 23 years-old.
As in any promotion and relegation system, the quality of play, the talent of players, andattendance increase in higher divisions. Série A1 contains the best teams, the most talentedand highly paid players, and plays before the largest crowds. Série A2 contains the next bestteams, and so on.
Although there are costs associated with promotion, such as hiring extra talent, thebenefits from promotion should, on average, exceed the costs. For instance, promoted teamshave access to a larger prize pool and attract larger crowds to games. Also, promotionincreases team visibility, which should help reduce search costs for talent and increase theirrevenues from sponsors. Lastly, if indeed there are location economies by league, there shouldbe indirect benefits to teams as well.
2.2 Team Formation, Management and Player Movement
The process of football team formation in Brazil generally occurred long ago; few teamshave been formed in recent times (Lima, 2002). In brief, soccer teams in Brazil were formedby social (athletic) clubs, neighborhoods, and as a part of industrial unions. Therefore,Brazilian clubs are not privately owned and do not have publicly traded shares. Insteadteams are controlled by members associated with social or neighborhood clubs. This meansthat football teams in Brazil never relocate to different towns or regions5.
In terms of team management, Amorim Filho and Silva (2012) summarize the evolutionof Brazilian soccer team management and the transition to professional management of clubsand leagues. The biggest change occurred in 1998 with the passage of the “general law ofsports”, or the so-called “Pelé Law” (Law n. 9615/98). This law required that all Brazilianfootball clubs become business enterprises. It also eliminated club “ownership” of athletes,who now could negotiate contracts with any club, creating free agency in Brazilian football.
Even though clubs became business enterprises after 1988, there still exist critics of thissystem. In general, teams have a Board of Directors called “Conselho Deliberativo”. TheBoard will be composed of two types of directors: those who are elected by teams’ associatesand meritorious directors. The meritorious directors are permanent members of the board
5There is some anecdotal evidence of a few clubs being bought by private firms and relocated. These arevery rare occurrences, especially among traditional teams.
6
who obtain this position after making a significant monetary contribution to the team. TheBoard of Directors elects a president who runs the club for a three-to-four year period.Members of the Board also participate in the team’s management in different functionalareas such as marketing, finance, etc.
Amorim Filho and Silva (2012) identify two important features of standard firms thatare not present in the soccer business in Brazil. First, there is little accountability for poormanagement, such as persistent financial losses. Second, given that there is no payment toBoard members who run the team’s day-to-day operations, this creates rent-seeking behaviorby the board members who may act in their own self interest by tying themselves to playersrights or embezzling money.
In terms of player movement, there are two periods in which most player transactionsoccur: December and January, at the end of the season; and between May and July whentransactions in Europe occur. Note that between February and May it is possible for teamsto hire and register new players in the CP. However, these transactions are rare, as the bulkof transactions happen before the season starts and after the state championship ends, asthis coincides with the beginning of the national league and the European hiring season.
Lastly, differently than some American leagues such as the NBA and NFL which enforce acap on expenditures and have a draft system to maintain a competitive balance in the league,this is not the case for Brazilian soccer. This is also true in Europe, in which although thereis some cap on expenditures, there is no draft system in place. As a consequence, powerhouseclubs are common in soccer leagues since they are able to hire more talent through an increasein expenditures.6
3 Conceptual ModelFirms experience both internal and external benefits from locating in cities, compared tofirms not located in cities (Quigley, 1998). These benefits include scale economies or indivis-ibilities with firms that can only be exploited by growing large, shared inputs in productionwith other co-located firms, reductions in transaction costs from better matching betweenfirms and employees, and related decreases in search costs. These external benefits can stemfrom city size, typically called urbanization externalities, and from the presence of othersimilar or related firms, called localization externalities. Rosenthal and Strange (2004) findthat doubling urban size increases firm productivity by about 5%, a substantial benefit fromurbanization. While a large number of studies confirm the importance of localization exter-nalities in a number of different industries, no previous literature has applied localization tofirms in the sports industry.
In order to understand the effect of localization on outcomes in sports leagues, we extendthe standard “two team” model of league outcomes developed by Fort and Quirk (1995) toinclude localization economies In the standard two team model, teams produce wins usingplaying talent and managerial talent and sell these wins to fans. For simplicity, the modelincludes only two teams, identified by subscripts i and j, although the results generalize in
6In Brazil, there is a large concern with the excessive expenditure to hire players that lead clubs to havea high debt level. Several programs have been put in place by the Brazilian government to help clubs reducetheir debts.
7
a straightforward way to the case of N teams (El-Hodiri and Quirk, 1971; Fort and Quirk,1995). Following the standard approach in the literature, assume that team i plays in a largecity where both urbanization and localization effects are strong and team j plays in a smallcity where agglomeration effects are weak.
Teams i and j maximize profits by choosing their level of success, proxied by team winningpercentage (W ). Teams’ revenues are a function of team success W and the team’s marketrevenue potential (M). The team’s market revenue potential (M) is assumed to be a functionof urbanization economies (u) and localization externalities (a) that will be different acrosseach city where teams i and j locate. Mi(ui, ai) > Mj(uj, aj) by assumption, because teami plays in a large city. The urbanization and localization economies (or externalities) shouldhave a positive effect on each team’s market revenue generating potential. For instance, largeand more dense cities would create a larger market for the team to sell wins in. Also, urbanareas have higher median wages (Glaeser and Maré, 2001; Glaeser, 1998; Glaeser et al., 2001)which should positively impact M .
The overall impact of localization externalities on M is ambiguous. For example, thepresence of several teams in the same city implies the market will be split between teams,which can reduce the each team’s individual market potential. However, because thereare more than one team, a rivalry between teams is likely to emerge. This rivalry wouldincrease the market potential, as customers would become more passionate about their teams,increasing the importance of teams in consumers’ utility functions. For example, one canlook at football teams in London and Buenos Aires, where there are several teams withintense rivalries and passionate fans.
For São Paulo, there is not only municipal level rivalries as well as regional or state levelones. The municipality level rivalry stems from neighborhoods and even socio-demographicsof the average supporter. On the regional level, however, these rivalries emerge from theinteraction in state and national competitions. Some anecdotal evidences are: in São Paulomunicipality the powerhouse teams, like São Paulo, Palmeiras, and Corinthians that havedifferent historic formation and their identification; in Campinas, the main two teams usedto have side-by-side private stadiums; in a state level, the powerhouse teams in São Pauloversus Santos, a team located in around 46mi from São Paulo and that team that showcasedPelé.
Teams’ production costs include a fixed cost (Fi) and the marginal cost of playing talent(p), which can be interpreted as the salary paid to players of all players are homogenousin ability. If a team operates in an area in which agglomeration economies exist, it shouldbenefit from shared intermediate inputs, labor market pooling and matching, and knowledgespillovers; these are defined as localization economies (a), which are internal to teams butexternal to a team’s “plant”. These localization economies should have an effect on the team’scosts, in terms of either the marginal price of talent or fixed costs. Therefore, we can writeeach team i’s objective function as
maxWi
Πi = Ri(Wi,Mi(ai, ui))− Fi − pi(ai)Wi (1)
and team j’s objective function as
maxWj
Πj = Rj(Wj,Mj(aj, uj))− Fj − pj(aj)Wj (2)
8
where a captures localization economies, u capture urbanization economies, and p(a) is themarginal cost of talent, which depends on localization. Localization economies imply thatthe marginal cost of talent is lower for the team playing in a dense urban area, so p′(a) < 0.u captures the effect of urbanization economies, which affect revenues but not labor costs.
By definition, as both teams are playing in the same league, then the so-called “addingup constraint” holds; i.e., the sum of the team’s winning percents equals one. In addition,the equilibrium in this market will be achieved where the marginal revenue (MR) of team iequals the marginal revenue of team j.
We are primarily interested in the relationship between the agglomeration economies(urbanization and localization) and team success or win percents. Thus, we can use theimplicit function theorem and the first order conditions of profit maximization to obtaincomparative statics between equilibrium winning percent and agglomeration. For simplicity,we assume that only team i will enjoy agglomerative effects as a consequence of playing in alarge city. Job market pooling and matching should make it easier for team i to hire talent.Moreover, the knowledge spillover from informal nonmarket interaction between players andcoaches should also increase the productivity of such players at any marginal price (Glaeseret al., 2000). Therefore, we expect a negative relationship between agglomeration and themarginal cost of playing talent. On the other hand, because there is no agglomerationeconomies for team j, we should expect a fixed price for talent in this case. As assumedby El-Hodiri and Quirk (1971) and Fort and Quirk (1995), total team revenue is strictlyconcave in W . We also assume revenue is strictly concave in M . Subscripts represent thepartial derivatives. The relevant comparative static terms are
∂Wi
∂ai= −(RWa ∗Ma − pa)
RWW
≷ 0 (3)
∂Wj
∂aj= −(RWa ∗Ma − pa)
RWW
= 0 (4)
∂Wi
∂ui= −(RWu ∗Mu)
RWW
> 0 (5)
∂Wj
∂uj= −(RWu ∗Mu)
RWW
= 0. (6)
The comparative statics generate an unambiguous prediction regarding the impact ofurbanization economies. In the presence of urbanization economies (equations 5 and 6) likethose described by Jacobs (1969), a team playing in a large city has a larger equilibriumwinning percent and a team playing in a smaller city does not enjoy such benefits. Inequilibrium, teams playing in large cities will be more productive, and more successful, thanteams playing in small cities. However, in terms of localization economies (equations 3 and4), as described by Marshall (1895), because we cannot clearly sign the relationship betweenthe localization externality and market revenue potential, there is no clear prediction fromour conceptual model. Our empirical analysis will help to shed some light in this issue.
One nice feature of the two team model is the possibility of a graphical solution, in termsof the determination of equilibrium winning percentages in a league of profit maximizingteams. Figure 1 shows the standard league equilibrium in the two team model, point E1
9
with urbanization, reflected in the assumption that Mi > Mj and no agglomeration effectson input prices (pj = pi = p1). In the graphical solution to the two team model, team j’ssuccess, Wj is shown as increasing from left to right and team i’s success is shown increasingfrom right to left. The adding up constraint means that any league outcome must occur ona vertical line in the (W × $) space.
Figure 1: League Equilibrium With and Without Agglomeration
6$j 6$i
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.01.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0
��������������
MRi@@@@@@@@@@@
MRj
E1
p1 = MRjE1 MRi
E1 = p1
pj(aj) = MRjaj
MRiai
= pi(ai) < pj(aj)
Team i plays in a large city with larger urbanization externalities than team j, so teami’s MR curve is higher than team j’s MR curve. In equilibrium, both teams maximize profitsand face the same price of talent. In equilibrium, MRj = MRi = p1 so both teams aremaximizing profits. However, because team i enjoys the benefits of urbanization, and hasmore fans to sell wins to, and potentially a higher willingness to pay for wins due to higherurban wages, Wi > Wj in equilibrium. The league equilibrium outcomes are unbalanced andteam i is always more successful than team j.
If agglomeration effects reduce the marginal cost of acquiring playing talent for team i,because of shared input pools and knowledge spill overs, then team i’s marginal cost of talentis lower than team j’s (pi(ai) < pj(aj)). With no exchange of talent between team i andteam j, the league outcome is pushed to the left or point E1. Team i acquires even moretalent, and Wi increases while team j, facing a higher marginal cost of talent, acquires lesstalent, decreasing Wj. The effects of agglomeration with no exchange of talent between theteams makes the competitive imbalance in the leagues worse, as team i becomes strongerand team j becomes weaker. In this case, agglomeration increases the equilibrium level ofsuccess of teams playing in dense urban areas relative to those playing in smaller, less densecities.
If the teams can exchange players, and the pool of players in the agglomerated area isrelatively small and fixed, then team i will sell players to team j at a price ps = MRj
aj>
Pi(aj) and earn a profit. The league will eventually return to the original equilibrium E1,where team i is still more successful than team j, but the league competitive balance isimproved, relative to the outcome under agglomeration and no player exchange. However,many leagues place restrictions on the exchange of players for cash, which would reduce thetendency for the league to return to E1. Also, it could be that players will be unwilling to
10
move to smaller cities with fewer urban amenities, even with the higher salary paid by teamj.
Even under player exchange, if the pool of players in the agglomerated area is largeand can expand due to demographics, as new cohorts of children mature enough to behired by teams, and the agglomerative effects are permanent, then the league outcome tothe left of point E1 could be permanent. In this case, teams that can take advantage ofthe localization benefits would be permanently better than teams located in areas with nolocalization benefits.
If the effect of agglomeration is to reduce team’s fixed costs such that Fi < Fj, theallocation of talent across teams is unaffected because both teams face the same marginalcost of acquiring talent, p1. However, team i will be more profitable than team j, since teami faces a lower total cost of producing any level of team success. In this case, team i willstill be more successful than team j in equilibrium, because team i still enjoys the benefitsof urbanization, and has more, higher income fans to sell wins to than team j.
If localization effects increase the marginal revenue of teams located in the large urbanarea, then the marginal revenue curve for these teams, MRi, will shift up farther than itwould under only urbanization effects. This would increase the competitive imbalance inthe league, as the teams in large urban areas would use the increased revenue from thelocalization economies to purchase even more talent. Localization economies could increasemarginal revenues in a large urban area by fostering more, and stronger rivalries among fansof different teams, generating more, and more intense derby matches between specific pairsof teams. Derby matches would be less likely to occur among spatially differentiated teams,since their fans would not have a chance to interact to the same degree as fans in large urbanareas.
4 Data and Econometric Approach
4.1 Data
To analyze the impact of agglomeration of teams on productivity we use data from twosources. First, we scrape data from the Federação Paulista de Futebol (FPF), the organiza-tion that manages football leagues in São Paulo state. The FPF website contains informationfor each of the four leagues in the CP for the 2007 to 2018 seasons. We also obtained infor-mation on characteristics of each team competing in the CP, such as if they field an under 17team, if they are considered a development team by the Confederação Brasileira de Futebol(CBF), whether the team plays in a private or public stadium and if this stadium is sharedwith another team, and their geolocation (official address). Figures 2 to 5 shows the dis-tribution of teams across municipalities of the state of São Paulo for each division by year.The variation shown in this figure is the one we explore in our empirical strategy.
Success Measures
We define short-term success of a team by its performance in the league it competes ineach season. We use three measures of short-term success: win-loss ratio, goal differential,
11
and points scored. The win-loss ratio is calculated as the ratio between win and losses,disconsidering any draws. Because some teams had unbeaten season, we add unit to bothwins and losses. The goal differential is calculated by subtracting the number of goals againstfrom the number of goals scored. Lastly, the number of points is the points scored in theseason. CP follows the international rules and have the following points break down: threepoints if the team wins, one point for draws and zero if it loses. For seasons in which therewas additional rounds in the form of knockout, we consider the result of the match. In otherwords, if a team wins it gets three points, however if there was a tie, which led to penaltykicks, we consider the tie result.
We define long-term success by two proxies. The first is the average league played by theteam. This is accomplished by estimating an ordinal logit model in which the dependentvariable is the league level. In the result session we present the marginal effect of an increasein concentration of teams and how that affects the league level teams are playing.
The second, and preferred, long-term success measure is Elo rating for each season.Elo rating are commonly used to rank team in several sports (Gásquez and Royuela, 2016;Lehmann and Wohlrabe, 2017). The Elo rating system is a relative skill ranking based ongame-by-game outcome. To properly calculate the Elo ranking it is necessary to have initialassignment of values. We use the 2007 season as benchmark and assign the value of 1500 toevery team that has not played in that year in CP. For each team that has played we add 3points per position in the ranking, i.e., the last ranked teams initial Elo is 1503, the secondlast 1506, and so on.
To calculate the new Elo ranking we use the package elo in R. The Elo calculationfollows the algorithm described below. Let r(i) be the current Elo rating of each team i,where i = 1, 2. First, we compute the transformed rating (R(i)) as:
R(i) = 10r(i)/400 (7)
Then, we calculate the expected score for each team i
E(i) =R(i)∑iR(i)
(8)
For each match we define team’s i score (S(i)) as 1, if it wins, 0.5 if there is a draw and 0 ifit loses. Then, we calculate the updated Elo-rating (rN) for each team (i) as:
rN(i) = r(i) +K ∗ (S(i)− E(i)) (9)
where K is a factor of how each match impacts a team’s rating. In this paper we use thevalue of 30.
These measures of both long-term and short-term success can also be interpreted asmeasures of team productivity. Each team fields only eleven football players, and the level ofcompetition and playing talent increases across leagues from Segunda Divisão at the bottomto Série A1 at the top. Teams of eleven football players in Série A1 are more productivethan teams of eleven football players in Série A4 because they play in a league with othermore talented players and successful teams.
Table 1 provides the summary statistics for the success measures. Panel A presents theshort-term success measures and Panel B the long-term ones. Notice that some teams have
12
negative points. This happened because they were punished by the FPF with losses of pointsfor disrespecting the rules of the league in a season.
Concentration and Urbanization Measures
As laid out in our conceptual model, we are interested in two types of variables: onetype that captures the localization economies and another type that captures urbanizationeconomies. Regarding localization effects, we use the Herfindahl-Hirschman Index (HHI), atraditional measure of market concentration in industrial organization, formally defined foreach municipality as
HHIl =∑m
s2ml (10)
where m is the municipality, l is the league and s is the share of total league seasons.One critique regarding the HHI is that it implicitly assumes regions have the same area.
Therefore, we also calculate the raw concentration measures proposed by Ellison and Glaeser(1997) and Maurel and Sédillot (1999) that take the area into account. Overall, localizationmeasures should take into account geographical concentration and industrial concentration(Duranton and Overman, 2005; Ellison and Glaeser, 1997; Maurel and Sédillot, 1999). How-ever, the industry concentration in the CP set up is determined by size of each league,which is determined by FPF and is fixed for all but A4 division. Thus, any variation in thelocalization measure would come from changes in the geographical concentration.
We follow Maurel and Sédillot (1999) and use the raw concentration measure in Ellisonand Glaeser (1994) (EGI). In Ellison and Glaeser (1997) the raw concentration measure isequivalent to the working paper version and is defined as EGIl1997 =
∑i(sml − (xm)2). This
index is
EGIl =∑m
(sml − xm)2
1− x2m(11)
where sm is share of teams in league (l) in microregion m, and x is the share of the totalnumber of teams in microregion m. Ellison and Glaeser (1997) show that his index isfunctionally related to the HHI, but also accounts for a counterfactual spatial distributionof firms based on a random distribution over space (a “dartboard” approach). We also usethe raw concentration measure developed by Maurel and Sédillot (1999) (MSI)
MSIl =∑m
s2ml − x2m1− x2m
(12)
where sm and x are defined the same way. This differs from the Ellison and Glaeser (1997)approach in the exact functional form used, and not in the variables used (the fraction offirms in each geographic area and the fraction of total area accounted for by each geographicarea).
In the case of football outcomes, it does not make sense to expand the concentrationmeasure to calculate the γ concentration measure based on labor inputs contained in both,
13
because the outcome of a season’s competition depends on contests played between footballteams of eleven players each, thus we would not get any additional information.
Lastly, we calculate the concentration of teams per league in a 50km radius from theteam’s offical address in the FPF website. Note that Duranton and Overman (2005) developan alternative measure of localization based on the distance between each establishment/firmpair. In our set-up this location is less important, as teams have headquarters in differentlocation from the stadium which they play. Also, some teams may share stadiums, especiallyif these are public provided. Given the historic formation of teams in Brazil, as described inSection 2, the actual location of each team, should be random conditional on the concentra-tion of teams and the location of sports teams in the state. Nevertheless, the concentrationof team based on distance from the team headquarter provides extra robustness check forour results.
In terms of urbanization, we focus on population density and median wage in a microre-gion. The population density is a straightforward measure of urbanization, as it capturesthe ratio of population to area. The median wage variable severs two purposes. As argued insection 3 the purchase power of the supporters is of importance. However, more importantly,given the the spatial equilibrium approach, wages plus amenities should offset the housingcosts to ensure the indifference across the space (Roback, 1982; Glaeser, 2007).
Table 2 presents descriptive statistics for each raw location and urbanization measures.Panel A presents the localization measures, i.e, HHI, EGI and MSI, which were standardizedto make the results more intuitive. Panel B presents the urbanization measures, that is,density and median wage.
To control for time varying factors in each municipality, we collect data on employment,value added in production in the agriculture, government, manufacturing, and services in-dustries, along with the total number of establishments and population in each municipality.These variables come from SEADE, a independent public agency sponsored by São Paulostate, and RAIS, an annual government report on Brazil’s formal industries. Panel C intable 2 presents their descriptive statistics.
4.2 Econometric Approach
We estimate the following equation for the determination of team success or productivityto assess the impact of agglomeration economies (localization and urbanization) on teamoutcomes
Stdmy = β0 + β′1LOCmy + β′2URBmy +Xmyβ4 + µt + µd + µm + µy + εtdmy (13)
where Stdmy is the measure of football success of team t, in division d, in municipality m, inyear/season y; LOCmy is the localization measure per division d for each municipality m inyear/season y; URB is a vector of urbanization variables including median wage and popula-tion density; X is a vector of control variables including the total number employment, totalestablishments (firms), population and value added by sector (agriculture, manufacturing,service, and government) in each municipality. µt, µd, µm, µy are teams, division, munici-pality and year/season fixed effects; εtdmt is the idiosyncratic error term. This is assumedto be mean zero and possibly heteroscedastic across teams. One important feature of CP is
14
that teams only play teams from their own division in each season. Therefore, variation insuccess generated by matches between teams of very different strength is less of a concernin this setting.
The parameters of interest are β1 and β2, which capture the effects of localization and ur-banization economies, respectively, on team productivity. To make our results more intuitivewe standardize our measures of localization and urbanization.
To identify the impact of localization economies on team productivity, we exploit thepromotion and relegation system in Campeonato Paulista to generate exogenous variation inagglomeration, in terms of the fraction of teams in each league playing in each municipality.We expect that teams in better leagues to interact among themselves more than teams inlower leagues.
One possible concern that can emerge is the presence of the so-called yo-yo teams. Yo-yoteams are frequently promoted and relegated, bouncing back and forth between two leagues.We argue that these teams are not a concern for our results because their promotion andrelegation is exactly the variation we exploit. The localization and urbanization economiesenjoyed by these teams may as well explain why they cannot build on their success and endup being relegated; or else, can keep some level of success which helps them being promoted.For our analysis to be biased, the only source of variation in success would have to comefrom yo-yo teams, which is not the case.
5 ResultsOur main results are presented in Tables 3 to 7. Column (1) presents models using the HHIas localization measure; column (2) uses the Ellison and Glaeser (1997) raw concentrationmeasure (EGI) as localization measure; column (3) uses the Maurel and Sédillot (1999) rawconcentration measure (MSI) as localization measure; and column (4) the concentrationof teams in a 50km radius. As explained above, we standardize the HHI, EGI, MSI andurbanization proxy variables for each year in order to generate comparable results.
First we focus on the short-term success. Table 3 presents the results for Win-Loss ratio,Table 4 the Goal Difference results and Table 5 for the number of points. According to theconceptual model presented in section 3 we should expect that teams located in denser andricher municipalities as well as those with a concentration of better teams to perform better.However, the results show that neither localization nor urbanization externalities affect theshort-term success of teams.
The exception is for the number of points scored by teams. Teams that are located inmunicipalities with a one standard deviation more Division 1 teams score on average around2 points less, as shown by the HHI and MSI, only, which is the opposite of what we expect.There can be two explanations for this results that are not mutually excludable: (i) teamsin Division 1 (D1) are better teams and can “steal” the best talent from other teams makingthem weaker; (ii) since there is a a zero-sum feature to each season, if teams located nearthese D1 are also D1 teams, then they would underperform and score less points.
Now we turn our attention to the long-term success measures. Table 6, shows the resultsfor the Elo rankings we created. This is our preferred measure of success. As predictedby the conceptual model, teams enjoy localization spillovers, that is, teams with D1 teams
15
closer by are on average more successful than other teams. One standard deviation increasein the number of D1 teams increase the Elo ranking between 14 (EGI) and 19 (HHI) points.There is no statistical significant effect for the concentration of D2 and D3 teams, but thereis weak evidence that the concentration D4 teams can harm the long-term success of teamslocated in the same municipality as these teams.
Note that, different than predicted in section 3, density negatively impact the long-termsuccess of teams. Consistently across the four models estimated, one standard deviationincrease in density decreases the Elo ranking in about 55 points. One possible explanationfor this is that although density increases the likelihood of more teams to emerge in theseareas, they are on average worse teams. The bigger and more traditional teams would beable to enjoy the benefits of having more talent around, but these worst teams would beunable to attract supporters and talent to be as competitive. Anecdotally, teams in Brazilare more likely to partner with teams outside their main area of influence in trading players,loaning players, among others.
Lastly, Table 7 presents the ordered probit estimations considering the division played.7Because in these models the order is important we recode the seasons such that Division 1is the highest (4) and Division 4 is the lowest (1). Panel A shows the marginal effects forplaying Division 4, Panel B for Division 3, Panel C for Division 2 and Panel D for Division1.
Focusing on the urbanization measures we notice that density is not statistically differentfrom zero in any model. As for Median Wage, it is only statistically significant for the MSImodel, and it has heterogenous effects. While the marginal effect of median wage is negativeon the probability of being in Division 4, it is positive for Divisions 1 to 3. This corroboratesthe idea that urbanization is important to more teams to emerge and be participate incompetition and be successful, by attracting more and better talent as well as supportersand sponsors.
Now we turn our attention to the localization measures. First we notice some heteroge-nous effects regarding these results. Similar to the urbanization measure, the results arestatistically for the MSI and concentration by distance measures. Overall they are not sta-tistically significant for the HHI and EGI measures, with exception of the effect of Division1 teams on the likelihood of playing Division 3, which are negative.
Focusing then on the MSI and the Concentration of teams in a 50km radius measure, theresults suggest that the concentration of D4 teams negatively affects the success (Divisionplayed) of other teams as it decreases the likelihood of playing in Divisions 1 to 3, butincreases the likelihood of playing in Division 4. The concentration of Division 3 teams haveno effect on other teams. As for the concentration of Division 2, there is evidence that itdecreases the likelihood of playing in Divisions 1 to 3, but no effect on playing division 4.Lastly, the concentration of Division 1 teams have positive effects on the likelihood of playingDivision 4, but it has negative effects on the probability of playing Divisions 1-3.
At first glance, these results may seem to contradict the theoretical model, however,given the zero-sum nature of each league and the inability of free mobility across leaguesthese results suggest in fact corroborate the idea that the team playing in higher divisionsare able to secure more supporters and talent, and in some sense maintaining a status quo
7We present the marginal effects only, but the full results are available upon request.
16
of teams in the state. This in fact, is observed in Brazil, that is, there are just a few cases ofteams that were able to grow change their importance. Note that, if this was not the caseas in the Elo rankings, than there is clear spillover effects of concentration of better teams.
Taken together, then, these results show that agglomeration economies – both in terms oflocalization and urbanization are important to explain teams success. However, the resultssuggest that these results are heterogenous depending on the quality of teams, in the caseof localization spillovers, as well as there is contrasting evidence for urbanization spillovers.While there is evidence that density has negative effect on the success of teams, median wageshave positive effects. Nevertheless, the agglomeration economies of soccer teams impact onlytheir long-term success.
6 ConclusionThis paper evaluates the importance of localization and urbanization economies for explain-ing the productivity of firms. In particular, we focus on Campeonato Paulista (CP), playedin São Paulo state, Brazil, and exploit its promotion and relegation structure to identifychanges in team concentration across municipalities by league. The conceptual model con-struct makes clear predictions about the impact of urbanization economies on team success,but uncertain predictions for localization economies.
Using data for Campeonato Paulista between the years 2007 and 2018, we find thatlocalization economies positively impact team long-term productivity/success but has noeffect on their short-term success. Urbanization economies also only affects the long termsuccess. While there is some evidence that median wage positively contributes to long termsuccess, density has a negative effect on team’s success. Moreover, we find a heterogenouseffect in terms of the quality of the league.
There are several implications from these results. First, this is the first paper, to ourknowledge, to investigate the importance of localization in sport leagues. Second, althoughin Brazil team movements from city to city never occur, we show that changes in league havean important impact on team’s level of success. Therefore, not only the presence of moreteams plays a role in the success of other teams, but also the quality of the “neighboring”teams.
These results allow us to speculate about what may happen in U. S. leagues if entry/exitwere to be allowed. These results suggest that we should expect that larger markets wouldattract more teams, and these teams would benefit from localization economies, makingthem more successful. The effect of urbanization economies alone, assuming new entrantscannibalize the existing sports market by taking fans from existing teams, implies that theentrance of new teams would reduce the success all teams in a large urban area. Also, wewould not expect to see new franchises enter in isolated smaller cities, as they would notenjoy the benefits of either urbanization or localization economies.
ReferencesAmorim Filho, M. H. and Silva, J. A. F. (2012). A gestão de clubes de futebol ? regulação,
modernização e desafios para o esporte no Brasil. Revista Interesse Nacional.
17
Arrow, K. (1962). The economic implications of learning by doing. Review of EconomicStudies, 29(3):155–173.
Devereux, M., Griffith, R., and Simpson, H. (2007). Firm location decisions, regional grantsand agglomeration externalities. Journal of Public Economics, 91(3-4):413–435.
Driessen, G. A. and Sheffrin, S. M. (2017). Agglomeration, tax differentials, and the mobilityof professional athletes. Public Finance Review, 45(2):283–302.
Duranton, G. and Overman, H. (2005). Testing for localization using micro-geographic data.Review of Economic Studies, 72(4):1077–1106.
El-Hodiri, M. and Quirk, J. (1971). An economic model of a professional sports league.Journal of Political Economy, 79(6):1302–1319.
Ellison, G. and Glaeser, E. (1994). Ageographic concentration in U.S. manufacturing indus-tries: a dartboard approach. NBER Working Paper 4840.
Ellison, G. and Glaeser, E. (1997). Ageographic concentration in U.S. manufacturing indus-tries: a dartboard approach. Journal of Political Economy, 105(5):889–927.
Fort, R. and Quirk, J. (1995). Cross-subsidization, incentives, and outcomes in professionalteam sports leagues. Journal of Economic Literature, 33(3):1265–1299.
Galliano, D., Magrini, M.-B., and Triboulet, P. (2015). Marshall’s versus jacobs’ exter-nalities in firm innovation performance: The case of french industry. Regional Studies,49(11):1840–1858.
Glaeser, E., Kallal, H. D., Scheinkman, J., and Shleifer, A. (1992). Growth in cities. Journalof Political Economy, 100(6):1126–52.
Glaeser, E. L. (1998). Are cities dying? The Journal of Economic Perspectives, 12(2):139–160.
Glaeser, E. L. (2007). The economics approach to cities. Working Paper 13696, NationalBureau of Economic Research.
Glaeser, E. L., Henderson, V., and Inman, R. P. (2000). The future of urban research:Nonmarket interactions [with comments]. Brookings-Wharton Papers on Urban Affairs,pages 101–149.
Glaeser, E. L., Kolko, J., and Saiz, A. (2001). Consumer city. Journal of Economic Geogra-phy, 1(1):27–50.
Glaeser, E. L. and Maré, D. C. (2001). Cities and skills. Journal of Labor Economics,19(2):316–342.
Groot, S. P., de Groot, H., and Smit, M. (2014). Regional wage differences in the netherlands:Micro evidence on agglomeration externalities. Journal of Regional Science, 54(3):503–523.
18
Gásquez, R. and Royuela, V. (2016). The determinants of international football success: Apanel data analysis of the elo rating*. Social Science Quarterly, 97(2):125–141.
Henderson, J. V. (2003). Marshall’s scale economies. Journal of Urban Economics, 53(1):1–28.
Henrickson, K. E. (2012). Spatial competition and strategic firm relocation. EconomicInquiry, 50(2):364–379.
Jacobs, J. (1969). The economy of cities. Random House, New York.
Lehmann, R. and Wohlrabe, K. (2017). Who is the ‘journal grand master’? a new rankingbased on the elo rating system. Journal of Informetrics, 11(3):800 – 809.
Lima, M. A. (2002). As origens do futebol na Inglaterra e no Brasil.http://www.klepsidra.net/klepsidra14/futebol.html.
Marshall, A. (1895). Principles of Economics. MacMillan; London.
Maurel, F. and Sédillot, B. (1999). A measure of the geographic concentration in frenchmanufacturing industries. Regional Science and Urban Economics, 29:575–604.
Mills, B. M., Mondello, M., and Tainsky, S. (2016). Competition in shared markets andmajor league baseball broadcast viewership. Applied Economics, 48(32):3020–3032.
Mondello, M., Mills, B. M., and Tainsky, S. (2017). Shared market competition and broadcastviewership in the national football league. Journal of Sport Management.
Moomaw, R. L. (1988). Agglomeration economies: Localization or urbanization? UrbanStudies, 25(2):150–161.
Quigley, J. M. (1998). Urban diversity and economic growth. The Journal of EconomicPerspectives, 12(2):127–138.
Roback, J. (1982). Wages, rents, and the quality of life. Journal of Political Economy,90(6):1257–1278.
Romer, P. (1986). Increasing returns and long-run growth. Journal of Political Economy,94(5):1002–37.
Rosenthal, S. S. and Strange, W. C. (2004). Evidence on the nature and sources of agglom-eration economies. Handbook of Regional and Urban Economics, 4:2119–2171.
Viladecans-Marsal, E. (2004). Agglomeration economies and industrial location: city-levelevidence. Journal of Economic Geography, 4(5):565–582.
19
Figures and Tables
Figure 2: Team Distribution for Division 1
20
Figure 3: Team Distribution for Division 2
21
Figure 4: Team Distribution for Division 3
22
Figure 5: Team Distribution for Division 4
23
Table 1: Descriptive Statistics for Success Variables
Statistic N Mean St. Dev. Min Max
Panel A: Short-term outcomesPoints 1,194 26.272 14.038 −12 70Wins 1,194 7.264 4.225 0 22Loss 1,194 7.265 2.676 0 19Goal Difference 1,194 −0.003 13.554 −70 47
Panel B: Long-term outcomesElo 1,183 1,635 142 1,217 2,087Division 1,194 2.821 1.162 1 4
24
Table 2: Descriptive Statistics
Statistic N Mean St. Dev. Min Max
Panel A: Localization measuresHHI 1,194 0.010 0.021 0.0004 0.122EGI 1,194 0.003 0.007 0.000 0.051MSI 1,194 0.007 0.017 −0.025 0.109
Panel B: Urbanization measuresDensity 1,194 1,871.116 3,045.096 18.230 13,535.460Median Wage 1,102 22.558 12.045 8.334 79.284
Panel C: Other VariablesNumber of Firms 1,102 27,426 72,379 517 306,156Population 1,194 1,137,926 2,956,281 16,979 11,753,659Agriculture Value Added 1,016 527 519 0 3,684Manufacture Value Added 1,016 65,270 145,619 106 668,601Service Value Added 1,016 320,298 908,629 1,153. 4,760,035Public Administration Value Added 1,016 30,456 79,150 368 393,752Total Employment 1,102 454,751 1,273,247 3,007 5,308,401
25
Table 3: Short-Term Success: Win-Loss ratio
Dependent variable: Win-Loss ratio
(HHI) (EGI) (MSI) (50km)
Division 1 −0.104 −0.121 −0.115 −1.116(0.102) (0.109) (0.106) (0.974)
Division 2 −0.041 −0.032 −0.009 −0.455(0.073) (0.085) (0.072) (1.048)
Division 3 −0.031 −0.020 −0.013 −0.255(0.097) (0.115) (0.082) (0.864)
Division 4 −0.008 −0.014 0.002 −1.296(0.068) (0.053) (0.094) (0.878)
Median Wage 0.148 0.149 0.154 0.141(0.309) (0.308) (0.308) (0.300)
Density −0.044 −0.050 −0.051 −0.042(0.263) (0.266) (0.263) (0.288)
R2 0.385 0.385 0.385 0.386
Clustered standard error in parentheses. N = 1,102;∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01. Control variables: valueadded for agriculture, manufacturing, services and gov-ernment, number of firms and formal employment. Team,division, municipality and year fixed effects are includedin all regressions.
26
Table 4: Short-Term Success: Goal Difference
Dependent variable: Goal Diff.
(HHI) (EGI) (MSI) (50km)
Division 1 −0.117 −0.115 −0.159 −18.551(1.225) (1.228) (1.225) (16.317)
Division 2 −0.755 −0.622 −0.403 −19.184(0.852) (1.110) (0.976) (16.283)
Division 3 −0.218 −0.253 −0.168 −10.551(1.673) (1.661) (1.163) (12.853)
Division 4 −0.050 −0.172 0.014 −20.582(1.027) (0.845) (1.370) 14.785)
Median Wage 1.342 1.370 1.417 1.293(5.094) (5.070) (5.090) (4.905)
Density −0.125 −0.285 −0.240 −0.950(3.673) (3.678) (3.675) (3.777)
R2 0.383 0.383 0.383 0.385
Clustered standard error in parentheses. N = 1,102;∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01. Control variables: valueadded for agriculture, manufacturing, services and gov-ernment, number of firms and formal employment. Team,division, municipality and year fixed effects are includedin all regressions.
27
Table 5: Short-Term Success: Points
Dependent variable: # Points
(HHI) (EGI) (MSI) (50km)
Division 1 −1.998∗∗ −1.372 −1.854∗ −6.860(0.957) (0.960) (0.963) (16.634)
Division 2 −0.420 0.269 0.573 −8.525(0.820) (1.172) (0.991) (17.567)
Division 3 0.735 −0.630 −0.542 −7.294(1.528) (1.465) (1.110) (12.695)
Division 4 −0.064 0.200 0.359 −11.376(1.119) (0.944) (1.498) (14.195)
Median Wage 1.768 1.687 1.789 1.312(4.406) (4.446) (4.434) (4.268)
Density 1.249 1.642 1.501 0.944(3.144) (3.122) (3.117) (3.328)
R2 0.371 0.370 0.371 0.370
Clustered standard error in parentheses. N = 1,102; ∗p<0.1;∗∗p<0.05; ∗∗∗p<0.01. Control variables: value added for agri-culture, manufacturing, services and government, number offirms and formal employment. Team, division, municipalityand year fixed effects are included in all regressions.
28
Table 6: Long-Term Success: Elo
Dependent variable: elo score
(HHI) (EGI) (MSI) (50km)
Division 1 18.993∗∗∗ 14.379∗∗∗ 17.611∗∗∗ −74.645(5.598) (5.434) (5.648) (101.276)
Division 2 2.837 3.649 4.477 −67.405(4.349) (5.714) (5.144) (89.077)
Division 3 0.959 0.197 1.525 −28.343(7.670) (7.486) (5.470) (68.376)
Division 4 −7.833 −3.086 −10.727∗ −140.010(4.934) (3.807) (6.125) (89.402)
Median Wage 38.035 39.098 37.990 39.061(30.142) (30.798) (30.452) (30.540)
Density −55.577∗∗ −55.583∗∗ −55.186∗∗ −62.977∗∗(27.367) (28.218) (27.856) (27.906)
R2 0.844 0.843 0.844 0.843
Clustered standard error in parentheses. N = 1,092; ∗p<0.1;∗∗p<0.05; ∗∗∗p<0.01. Control variables: value added for agriculture,manufacturing, services and government, number of firms and formalemployment. Team, division, municipality and year fixed effects areincluded in all regressions.
29
Table7:
Long
-Term
Success:
Averag
eDivision
Dep
ende
ntva
riab
le:
Div
isio
n
(HHI)
(EGI)
(MSI)
(50k
m)
(HHI)
(EGI)
(MSI)
(50k
m)
Pan
elA:M
argina
lEffe
ctson
Pr(Division=
4)Pan
elB:M
argina
lEffe
ctson
Pr(Division=
3)Division1
0.808
0.87
80.31
7∗∗∗
0.89
2Division1
-0.542∗−0.591∗∗−0.269∗∗∗
−0.595∗∗∗
(0.806
)(0.831
)(0.079
)(0.263
)(0.287
)(0.283)
(0.066)
(0.181)
Division2
−0.00
20.06
4−0.00
80.55
2Division2
0.00
1−0.043
0.007
−0.368∗∗
(0.025
)(0.066
)(0.011
)(0.264
)(0.016
)(0.028)
(0.009)
(0.179)
Division3
0.00
10.00
3−0.00
10.18
8Division3
−0.00
1−0.002
0.001
−0.126
(0.031
)(0.031
)(0.011
)(0.246
)(0.021
)(0.021)
(0.009)
(0.165)
Division4
0.09
90.06
00.84
9∗∗∗
0.27
0Division4
−0.06
6−0.040
−0.719∗∗∗
−0.180
(0.103
)(0.064
)(0.117
)(0.315
)(0.041
)(0.027)
(0.102)
(0.211)
Med
ianWag
e−0.18
5−0.19
6−0.11
9∗∗
−0.09
3MedianWag
e0.12
40.132
0.100∗∗
0.062
(0.208
)(0.209
)(0.054
)(0.091
)(0.091
)(0.090)
(0.045)
(0.060)
Den
sity
−0.04
4−0.10
1−0.06
1−0.10
6Den
sity
0.02
90.068
0.052
0.070
(0.159
)(0.184
)(0.072
)(0.142
)(0.103
)(0.110)
(0.061)
(0.095)
Pan
elC:M
argina
lEffe
ctson
Pr(Division=
2)Pan
elD:M
argina
lEffe
ctson
Pr(Division=
1)Division1
−0.22
8−0.24
9−0.04
3∗∗∗
−0.23
4∗∗∗
Division1
−0.03
8−0.038
−0.006∗∗∗
−0.064∗∗∗
(0.437
)(0.468
)(0.013
)(0.070
)(0.085
)(0.084)
(0.002)
(0.019)
Division2
0.00
1−0.01
80.00
1−0.14
4∗∗
Division2
0.000
−0.003
0.000
−0.039∗∗
(0.007
)(0.035
)(0.001
)(0.069
)(0.001
)(0.006)
(0.000)
(0.019)
Division3
0.00
0−0.00
10.00
0−0.04
9Division3
0.00
00.000
0.000
−0.013
(0.009
)(0.009
)(0.002
)(0.064
)(0.001
)(0.001)
(0.000)
(0.017)
Division4
−0.02
8−0.01
7−0.11
4∗∗∗
−0.07
1Division4
−0.00
5−0.003
−0.015∗∗∗
−0.019
(0.054
)(0.033
)(0.021
)(0.082
)(0.010
)(0.006)
(0.003)
(0.022)
Med
ianWage
0.05
20.05
60.01
6∗∗
0.02
4Med
ianWag
e0.00
90.008
0.002∗∗
0.007
(0.104
)(0.108
)(0.008
)(0.024
)(0.020
)(0.019)
(0.001)
(0.007)
Den
sity
0.01
20.02
90.00
80.02
8Den
sity
0.00
20.004
0.001
0.008
(0.049
)(0.070
)(0.010
)(0.038
)(0.009
)(0.012)
(0.001)
(0.010)
Stan
dard
errorin
parenthe
ses.
N=
1,09
2;∗ p<0.1;∗∗p<
0.05
;∗∗∗p<
0.01
.Mun
icipalityfix
edeff
ects
areinclud
edin
allregressions.Fo
rthis
analysis
Division1is
code
das
4,Division2as
3,Division3as
2an
dDivision4as
1,representing
that
Division1is
thehigh
estpo
ssible
outcom
ean
dso
on.
30