Abstract
Hosting a mega-event is a costly activity of short duration. Still, cities frequently compete to become host of all types of events. This paper examines the effect of staging the largest and most important sporting event in the world, the Summer Olympic Games, on the host city. Applying a difference-in-differences methodology, we analyse the population size of Olympic cities, candidate cities and other large cities in host and candidate countries over the period from 1860 to 2010. We find that, following the Games, host cities do not experience a measurable increase in population size relative to cities in the control group. On the contrary, to the extent that any effect of hosting the Games is identifiable, our results indicate that being awarded the Summer Olympics has a negative impact on cities.
Introduction
Cities (and countries) often compete fiercely to host institutions and mega-events. Location decisions are typically made only after an extensive ‘beauty contest’ among various candidates. For instance, when in January 2012 the United Nations called for expressions of interest to host one of its programmes, the Global Water Operators’ Partnership Alliance, three cities offered bids, including the commitment to provide substantial financial funding. 1 For the World Expo 2020, an international exhibition, the Bureau International des Expositions had the choice of four candidate cities. 2
Empirically, the benefits of host functions seem to be well established for institutions. Ades and Glaeser (1995), for instance, find that capital cities tend to be disproportionately large; they estimate that a country’s main city is, on average, 42% larger if it is also the capital city of the country. Nitsch (2003) argues that the pattern of persistence in Vienna’s excessive size after the break-up of the Austro-Hungarian Empire may partly be explained by its new role as a seat of international organisations.
In contrast, for events with limited duration, the evidence on the (net) effects on host cities is more controversial. While some studies indicate a positive impact, they often focus on individual events and tend to predict effects ex ante; many of these analyses are commissioned.
More generally, any empirical assessment of the effects of events on host cities has to deal with a number of (potential) issues. For one thing, mega-events affect cities along various dimensions. While some effects may be easily quantifiable, such as the number of visitors and tourists, others are more difficult to identify, such as the effects of hosting events on the local labour market. Most importantly, effects on intangibles, such as a city’s (long-term) reputation, seem to be hard to quantify at all. Another important issue is to properly control for the costs. Hosting costs not only include the pecuniary expenses for managing the event. 3 Often, events require considerable upfront investments in facilities and infrastructure, typically followed by expenses for long-term operation, maintenance and rebuilding. 4 Moreover, events may involve noticeable non-pecuniary costs for local inhabitants; examples include increased security measures, temporary road closures and congestion. Finally, when quantifying the impact of mega-events, a reasonably large sample of events should be analysed. While case studies may be insightful, events are typically characterised by strong host-specific components (especially because applicants aim to distinguish themselves from competitors in the selection process). As a result, estimates of average effects should be derived from a broad cross-section of events.
In this paper, we analyse the effects of hosting one of the world’s largest international events, the Summer Olympic Games. The Summer Olympics are a multi-sport event, featuring a large number of competitions (currently about 300) in a wide variety of sports. Organised by the International Olympic Committee (IOC), the Games are typically held every four years over two to three weeks.
For our purposes, the Summer Olympics provide, apart from size and importance of the event, a number of useful features. First, the Games are staged in a single city. While some competitions may be held at outside locations, the event itself essentially takes place in a spatially concentrated area for which effects may be properly identified; quantification seems to be more difficult, in contrast, for mega-events hosted at multiple locations, such as the FIFA World Cup or the Tour de France. Second, host cities of Summer Olympics are large urban areas. For these locations, often the capital city of a country, relevant indicators are readily available; data are much harder to obtain, in contrast, for an analysis of events at small and remote places, such as some host locations of the Winter Olympic Games or the Formula One Grand Prix. Third, the Games have a long history. The first edition of the Summer Olympics (in the modern era) dates back more than 100 years to the 1896 Games of Athens, allowing to identify the long-term impact of hosting the event. Moreover, because of the relatively low frequency of the Olympics, with Rio de Janeiro hosting the Games of the XXXI Olympiad in 2016, the total number of events appears manageable. Similarly, the notable variation in host cities (across countries and continents) seems particularly favourable for empirical analysis. Fourth, cities regularly compete to host the event. While the steady stream of applicant cities illustrates the general interest in hosting the Olympic Games, the group of cities with an unsuccessful bid forms a useful control group with which the performance of Olympic cities can be reasonably compared.
We contribute to the literature along various lines. Our key innovation is to empirically explore the overall impact of holding the Games. Instead of focusing on a specific feature associated with hosting the mega-event, we are interested in an all-encompassing cost-benefit analysis for the host city. Specifically, we argue that a location’s relative change in population size is a useful indicator of its overall attractiveness. 5 In addition, we analyse a sample which covers the full list of Olympiads. While most previous studies assess individual events, we aim to quantify the average effect across all hosted Summer Olympics.
In order to identify the impact of the Summer Olympics on host cities, we apply a time-shifted difference-in-differences approach that compares the population size of host cities before and after the ‘treatment’ (i.e. having staged the Games) with that of a control group of cities. Our analysis is based on a unique, self-compiled data set that covers 452 cities from 1860 through 2010. City population sizes are obtained from various (often national) sources; this comes at the cost that we are typically forced to use administrative (not economic) city definitions as the spatial unit of analysis. Previewing our main results, we find that the population size of Olympic cities, if anything, tends to decline after having hosted the Games relative to that of other large cities in the country or defeated candidate cities.
Background and literature
Why do cities apply to host the Olympic Games? According to the Olympic Charter, the IOC itself defines its role, in a very general fashion, ‘to promote a positive legacy from the Olympic Games to the host cities and host countries’. 6 In practice, the expectations of local authorities may be similarly opaque. While the motivation to host the Games is probably influenced by many factors, two reasons seem to be of particular importance across all applicants. On the one hand, cities aim to benefit from the tangible improvements associated with hosting the Olympics; the required large investments in facilities and infrastructure are widely expected to promote urban development. On the other hand, cities hope for strong intangible benefits, such as increased international recognition. As the IOC notes in a marketing report, ‘the Olympic Games is uniquely popular amongst everyone, regardless of age, sex, income or nationality’; 7 Glaeser et al. (2001) emphasise the role of local amenities to consumers for city growth.
The positive effects of staging the Games are balanced by the costs for host cities, which are similarly diverse. Some costs, such as direct expenditures on facilities, may be easily quantifiable. Others are hard to identify at all. These costs may, for instance, be difficult to measure (such as environmental damages), or they may apply over the very long term. 8
In our empirical analysis, instead of capturing the diverging effects individually, we focus on city population as a summary measure of a city’s overall attractiveness. Conceptually, we consider a situation in which the distribution of population across cities is initially in equilibrium and assume that holding the Olympic Games affects the quality of life in the host city (in potentially many different ways). Households, which seek to maximise utility, respond to this change in the spatial distribution of amenities in a Tiebout-like process through their locational choices. As a result, the distribution of population adjusts until spatial equilibrium is attained.
Consequently, we make use of one of the fundamental concepts in urban economics, compensating differentials. According to this conceptual framework, any shift in the perceived attractiveness of a location directly translates into a change in demand. If the supply of land is inelastic (e.g. for neighbourhoods within cities), the changing desirability of locations is primarily mirrored in prices (or, more precisely, capitalised into land rents). 9 For cities, however, we follow the argument of Henderson (1974) that a change in the utility level across cities leads to an adjustment in population size. 10
Our comprehensive empirical approach allows dealing with a number of issues that typically arise in economic impact studies of sporting events. For instance, a frequent concern is, as noted before, the proper identification of the various (dis)amenities associated with the event. 11 The analysis is further complicated by difficulties in isolating the individual effects, taking also offsetting factors into account (such as substitution or crowding out effects). As a result, studies rarely aim to assess the overall effect of events on cities, instead highlighting selected issues only. Examples include, among others, the analysis of the Olympic effect on investment, employment, tourism and property prices. 12
In addition, we deviate from previous work along at least two other dimensions. First, we explore the full history of Summer Olympic Games. Some studies focus on only individual episodes; others analyse a selected sample of events, often with a strong emphasis on experiences from the USA. Second, we explicitly deal with the issue that host cities are likely to be fundamentally different from other cities. By analysing also the performance of unsuccessful candidate cities, we are able to identify the host effect separately from a possible bid effect.
Most closely related to our work, Billings and Holladay (2012) examine the effect of hosting the Olympics on city population size, per capita income and trade openness. In contrast with our analysis, however, which spans one and a half centuries, they limit their attention to the period from 1950 to 2005. More notably, they use a restricted sample of international cities with a population of at least 750,000 in 2007, such that their analysis may be subject to selection bias. Their application of a propensity score matching approach implies that the performance of host cities is related to that of cities with similar characteristics based on selected covariates. Reassuringly, despite these differences, Billings and Holladay’s (2012) results are qualitatively similar to ours.
Methodology and data
Empirical strategy
To analyse the effect of hosting the Summer Olympic Games on city size, we apply a difference-in-differences methodology. Specifically, we compare the population size of Olympic cities (our treatment group) before and after the Games with that of other cities (our control group) over the same period. 13 Moreover, since Olympic cities host the Games in different years, we treat the timing of the before–after comparison as Games-specific; that is, we formally employ a time-shifted difference-in-differences specification which is centred on the date that each Olympic city hosts the Games. 14 In sum, we estimate equations of the following general form:
where PopSizect is the log population size of city c at time t; OlyCityc is a dummy variable which takes the value of one when a city is a member of our treatment group of Olympic host cities and zero otherwise; PostOlympicst>τ(i) is a time-varying dummy variable which is Games-specific and takes the value of one for periods after Games i have been hosted and zero otherwise; ηc and λt are comprehensive sets of city-specific and time-specific fixed effects; and εct is the error term. 15
We consistently include a comprehensive set of city dummies to account for time-invariant city-specific features (other than the Olympic city status). 16 We also often include a full set of time dummies to control for common trends in the population size of cities, unrelated to the timing of the Olympic Games. In practice, however, we modify and extend the baseline difference-in-differences specification in various ways to assess the robustness of our results.
The coefficient of interest to us is γ, the effect of the Olympic Games on the performance of host cities relative to a control group of other cities (which are otherwise hopefully identical). To the extent that hosting the Olympics measurably affects the city population size, this coefficient takes values significantly different from zero.
Data description
At the heart of our empirical analysis is a newly compiled panel data set, comprising the population size of major cities in applicant countries for more than a century. Our data set is constructed as follows. For each edition of the Olympic Games, we collect the populations of the 15 largest cities in each of the countries that bid to host the Olympics (and also have been awarded official candidate status from the IOC 17 ). While this selection automatically includes the host city (i.e. the city which receives the ‘treatment’ of staging the event), it easily allows for the construction of two reasonable ‘control’ groups. A within-country comparison explores the development of the host city relative to other cities in the country. Alternatively, the performance of the host city is related to that of other candidate cities (whose bid for the Olympics was not chosen by the IOC). In total, our sample comprises data for 452 cities from 26 countries; Appendix Table A1 lists the host and candidate cities. 18
List of Olympic and candidate cities.
Source: IOC Olympic Studies Centre.
Having identified the relevant sample of cities, we aim to track the population size of these cities over the period from 1860 to 2010 in 10-year intervals. 19 Data are not always available, however, for all cities at all times, for various reasons. For instance, Canberra, the capital city and one of the largest urban areas in Australia, was established only in 1913, thereby effectively reducing (part of) the control group for the 1956 Games of Melbourne and the 2000 Games of Sydney. As a result, our analysis is based on an unbalanced sample.
The data are collected by hand from various national sources, typically annual statistical yearbooks, mainly for three reasons. First, our analysis covers a period of more than 150 years for which only few comprehensive statistics are available. The first modern Olympic Games were held in Athens in 1896 so that our sample of city population sizes stretches back to 1860 to reasonably capture trends in the population size of cities before the Games. Second, by design, our sample consists of a fixed number of cities per country while international statistics typically cover only cities above a certain population threshold. At an extreme, these comprehensive sources (such as the United Nations’ World Urbanization Prospects) may even exclude particularly small applicant cities, such as Lausanne, Switzerland. Finally, only (time-varying) contemporary sources allow properly identifying the top cities in the national city size distribution at the time of the bid, thereby avoiding potential selection bias.
The use of national statistics also has possible drawbacks as data may not be directly comparable, both across countries and over time. In fact, conceptually, the analysis of consistently defined functional urban areas would have been desirable. For within-country analyses, however, it may be argued that inconsistencies in the definition of urban areas across sources from different countries are of only limited relevance. Also, national statistical yearbooks typically provide city populations for previous years, using the same geographic definition of the urban area, for purposes of comparison, thereby allowing the identification of methodological changes and irregularities. Cross-country comparisons benefit from the international practice to regularly take a population census at least every 10 years, typically at the turn of the decade. Still, to ensure robustness of the results, we also examine harmonised city population data from the United Nations. Detailed sources are listed in Appendix Table A2.
List of sources.
Note: As standard references, we use the national statistical yearbooks. Only additional references are listed here.
Empirical results
We begin our empirical analysis by reviewing the population size of Olympic cities relative to other cities in the (host) country. Table 1 presents the benchmark estimation results. Each column tabulates the estimates from a separate regression, gradually expanding the regression specification from the left to the right of the table.
The Olympics’ effect on city size using cities of host countries as control group.
Notes: OLS estimation. Dependent variable is the log of city population size. Robust standard errors in parentheses. **, * and # denote significant at the 1%, 5% and 10% level, respectively.
In a first exercise, we examine a parsimoniously specified version of equation (1), in which β is set equal to 0. In this specification, plain year dummies control for the variation in the average population size of cities in our sample over time. By ignoring the time-shifted character of our analysis, which requires a separation of the sample into pre- and post-episodes, this specification gets close to a standard difference-in-differences specification; see, for instance, Redding and Sturm (2008) for a recent application on city growth.
As shown, the estimated coefficient on the Olympic city dummy is positive and statistically significant, although city-specific fixed effects already control for long-run differences in city size. Taken literally, this finding indicates that cities which enter our sample both in the treatment and in the control group turn out to be more dominant in the national city size distribution at the time when they actually host the Games and, therefore, are part of our treatment group. More notably, the γ coefficient on the variable of interest, the Olympic city×Post Olympics interaction, is economically small and statistically indistinguishable from zero, such that no measurable change in the population size of host cities relative to other cities in the sample is identifiable. Taken together, the estimation results suggest that the Summer Olympics are awarded to disproportionately large cities, but holding the event has no measurable long-term effect on the population size of host cities.
In column 2, we relax our assumption that β equals 0, thereby effectively including an additional regressor. In fact, part of the explanation for our empirical (non-)finding may be sample selection bias, which is not properly controlled for in a conventional difference-in-differences setting. Specifically, early Olympic Games mainly took place in European countries where the population growth dynamics of (large) cities has generally slowed over time and city size stabilised, while later editions of the Summer Olympics were partly held in countries with strong urban growth and a rapid expansion of some initially small urban areas before the Games (such as Korea and China), with plain year dummies picking up a mix of these effects. Therefore, for γ to be an unbiased estimate of the effect of the Olympics, we must control for the (event-specific) difference in the population size of cities before and after the Olympics in our sample. The results in column 2 provide evidence that is in line with our intuition. The estimate of β is negative and economically and statistically significant, indicating that cities in the host country indeed experience a relative decline in city size after the Games. However, this pattern is not specific to Olympic host cities; it applies, in similar fashion, to other large cities in the country. The parameter estimate of γ, in contrast, remains largely unaffected by this perturbation, indicating, again, that there is no fundamental difference in the size performance of Olympic cities and their national peers after the Games.
In the next two columns, we extend the set of fixed effects. The inclusion of Olympics fixed effects captures event-specific idiosyncrasies. For instance, the 1904 Games were initially awarded to Chicago, but then shifted to St. Louis to coincide with the World’s Fair. Also the 1908 Games were re-located: Having been originally scheduled to be held in Rome, London stepped in as host after the eruption of Mount Vesuvius in 1906. The fixed effects also control for all other (unobserved) heterogeneities related to individual Olympics (or, for that matter, host countries), such as, for instance, differences in the definition and collection of national city population data. The use of Olympics-year fixed effects provides an even stronger test, capturing data variations within countries over time. Reassuringly, for both extensions, our baseline results turn out to be reasonably robust. As before, the estimated coefficient on our variable of interest is indifferent from zero at any conventional level of statistical significance, implying that there is no lasting shift in the population size of host cities, relative to other large cities in the country, from staging the event.
In our difference-in-differences model, we compare hosts with non-hosts. For non-host cities, it may be reasonable to start with an exploration of other urban areas in the country that typically face a similar economic, geographic and institutional background as the host city. However, there is another equally plausible control group of cities: candidate cities which have submitted a bid to host the Games but have not been awarded the event by the IOC. Following this approach, Table 2 tabulates results when we use rival (but unsuccessful) candidate cities instead of other national cities as control group. We apply a similar set of estimation specifications as before such that the table presents exact analogues.
The Olympics’ effect on city size using candidate cities as control group
Notes: OLS estimation. Dependent variable is the log of city population size. Robust standard errors in parentheses. **, * and # denote significant at the 1%, 5% and 10% level, respectively.
Reviewing the results, city and year fixed effects seem to capture differences in city population size remarkably well, both between host and candidate cities and over time. In fact, there is no further systematic variation in city size picked up by the Olympic city dummy and the past Olympics dummy; the estimated coefficients on these variables are close to zero and insignificant statistically. The estimated coefficient on the interaction term, in contrast, is consistently and often significantly negative, indicating that host cities fall behind in population size relative to non-host cities after the Olympics. After having been awarded the Games, the population size of successful bidders relative to unsuccessful bidders decreases by about 0.18% per decade. A legacy of debt from staging the mega-event may be one of various reasonable explanations for this finding.
We have performed extensive sensitivity analysis, modifying our empirical set-up along various lines. For instance, in one modification, we perform a placebo test, which examines the growth performance of unsuccessful bidding cities relative to non-bidders from the same country. Table 3 reports the results, presenting estimates for the same set of specifications as before. Interestingly, the coefficient of interest takes a significantly positive value, indicating that, in contrast to our findings for Olympic cities, applicant cities tend to become more dominant on a national scale after the Olympics. As a result, our findings suggest that candidate cities outperform their peers in terms of population size as long as they do not host the event.
The bidding effect on candidate city size using cities in candidate countries as control group.
Notes: OLS estimation. Dependent variable is the log of city population size. Robust standard errors in parentheses. **, * and # denote significant at the 1%, 5% and 10% level, respectively.
Next, we pool our samples and examine the effect of holding the Games, in addition to the effect of submitting a bid. That is, we treat host cities as a sub-group of all applicant cities. Table 4 provides estimates of the permanent effects for various specifications. As before, we find that cities which decide to apply for the Games generally tend to gain in population size. However, this effect is much weaker for host cities. The estimated coefficients on the Olympic city×Post Olympics interaction consistently take negative values, although none is significantly different from zero.
The Olympics’ effect on city size using cities of host and candidate countries as control group.
Notes: OLS estimation. Dependent variable is the log of city population size. Robust standard errors in parentheses. **, * and # denote significant at the 1%, 5% and 10% level, respectively.
In a final perturbation, we explore the robustness of our results for city population data taken from a different source. Instead of compiling data individually country by country, we use (harmonised) information from a standard reference, the United Nations’ World Urbanization Prospects: The 2014 Revision. The UN provides population data for urban agglomerations with more than 300,000 inhabitants (at the time of the revision) over the period from 1950 to 2030 in five-year intervals.
Table 5 presents the results. Reassuringly, our baseline results turn out to be reasonably robust. For instance, in line with our earlier findings, the estimates imply that the Summer Olympics are awarded to relatively large cities; there is also evidence of a relative decline in the population size of cities in host countries after the Games. More notably, however, for a national control group of cities based on current population size (that is, a sample that covers cities as ranked by their size in the year 2014), we also observe that host cities experience a particularly strong decline in the population size after the event. The estimates of γ are consistently and, with one exception, significantly negative, indicating that host cities also tend to lose in dominance vis-à-vis their national peers. We conclude that sample selection is indeed an issue of major relevance for our analysis.
The Olympics’ effect on city size using cities of host countries as control group.
Notes: OLS estimation. Dependent variable is the log of city population size. Robust standard errors in parentheses. **, * and # denote significant at the 1%, 5% and 10% level, respectively.
Conclusions
Hosting a mega-event is a costly activity of short duration. Still, cities frequently compete to become host of all types of events, from sports and music to trade fairs, political summit meetings, and even academic conferences.
This paper examines the effect of staging the largest and most important sporting event in the world, the Summer Olympic Games, on the host city. Applying a difference-in-differences methodology on a newly compiled data set, we analyse the population size of Olympic cities, candidate cities and other large cities in host and candidate countries over the period from 1860 to 2010.
We find that, following the Games, host cities do not experience a measurable increase in population size relative to cities in the control group. On the contrary, to the extent that any effect of hosting the Games is identifiable, our results indicate that being awarded the Summer Olympics has a negative impact on cities. An obvious extension of our research is the analysis of why (and when) do cities fail to benefit from hosting the mega-event, with the analysis of consistently defined spatial units being a reasonable starting point; we leave this issue for future research.
Footnotes
Acknowledgements
We thank two anonymous referees and participants who attended presentations at Tinbergen Institute, ERSA (Bratislava), Technische Universität Darmstadt, the Olympic Legacies conference in London, VfS (Münster) and WEAI (Seattle) for helpful comments. Olinda Papst provided excellent research assistance.
Funding
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.