The Impact of the London Olympics Announcement on Property Prices

The Impact of the London Olympics Announcement
                                          on Property Prices
                                               Georgios Kavetsos∗

                                              Cass Business School
                                             City University, London

    Draft version: 10/2009. Please do not quote without permission.


          This study estimates the impact of London’s successful 2012 summer Olympics bid on property
      prices. Using a self-constructed dataset of a sample of property transactions, we estimate a semi-
      logarithmic hedonic model for Greater London. Applying a difference-in-differences estimator, we
      find that properties in the host boroughs are sold between 2.1 and 3.3 percent higher, depending on
      the definition of the impact area. A similar investigation based on radius rings suggests that properties
      up to three miles away from the main Olympic stadium sell for 5 percent higher. The impact on host
      boroughs leads to an estimated overall property price increase of £1.4 billion, having substantial social
      and financial implications for existing residents.

JEL classification: H0; L83; R53
Keywords: Property Prices; Hedonic Model; Olympic Games

1    Introduction
Over recent decades cities and countries have attempted to redefine themselves by staging a major sporting
event, such as the Olympic Games. To successfully win the bid to stage the event, the candidate city usually
needs to provide new sports facilities of the highest standard. In order to justify public resources diverted
for the construction of stadiums, proponents carefully promote the potential positive effects the event will
have. New jobs, income benefits and a lasting legacy are terms frequently advanced, even though academic
research appears to be more sceptical about the economic impacts of staging mega events (see Siegfried
and Zimbalist (2000)).
    Previous research argues that regeneration of an area hosting a major event may lead to the gentrification
of the neighbourhood (Hiller, 1998). For example, since Barcelona was awarded the 1992 Olympics, the
Games have been argued to have been a major determinant of residential property price increases in the
city. From 1986 to 1993 residential property values are estimated to have increased by 250-300 percent
   ∗ Cass Business School, 106 Bunhill Row, EC1Y 8TZ, U.K. E-mail: Tel: +44 (0)20 7040

8647. The author wishes to thank Giles Atkinson, Roger Bivand, Paul Cheshire, Peter Dawson, Carolyn Dehring, Pantelis
Koutroumpis, Stefan Szymanski and Tommaso Valletti for comments and suggestions on earlier drafts of this paper.

(McKay and Plumb, 2001). Following similar evidence, the aim of this paper is to study an indirect impact
staging a sports mega event might have; namely, the impact the announcement of London’s successful
Olympic bid has on residential property prices. In estimating the impact of the announcement, we assume
that the property market, like any other market, reacts to new available information. To the extent that this
information is unexpected we can draw some interesting conclusions on the relation between major events
and property prices.
    Based on a self-constructed database, a semi-logarithmic hedonic model for properties within Greater
London is estimated. Controlling for dwelling specific and neighbourhood specific characteristics we es-
timate a significant price premium one has to pay in order to reside in the Olympic host boroughs. The
empirical evidence of this paper is augmented by studying the Olympics impact based on property dis-
tances from the main Olympic Stadium. We find that each additional mile away from the stadium decreases
property prices, where a final analysis based on consecutive three-mile radiuses suggests that property
prices mostly impacted are those within the three-mile ring. The impact decreases for properties in the
three-to-six and six-to-nine miles rings, although smaller both in terms of size and significance, and is not
significant for the remaining, more distant, radius rings.
    The rest of this study is structured as follows. Section 2 outlines the literature on the economic impact of
sports facilities and their potential impact on property prices. The case study of the London 2012 Olympic
Games is presented in section 3. Section 4 describes the data and methodology used. Section 5 presents the
estimated results and addresses some issues regarding policy. Section 6 concludes.

2    Sports Facilities and Property Prices
The price of a property can be viewed as a function of a bundle of physical attributes. A variety of location
and neighbourhood attributes, such as the proximity to transport services, schools, parks, industrial estab-
lishments, the regional crime, noise and pollution levels, etc, naturally appear to have significant impacts
on price (see Li and Brown (1980)). Not surprisingly, it is the variety of neighbourhood characteristics such
as the ones above that enrich the literature examining the impact certain factors have on property values.
For example, Espey and Lopez (2000) and Lipscomb (2003) study the impact airport proximity, and its
associated noise; Chay and Greenstone (2005) and Anselin and Le Gallo (2006) measure the impact of air
quality; Portnov et al. (2005) study the effect of environmental conditions and maintenance/expansion of
properties; Gibbons and Machin (2003) and Cheshire and Sheppard (2004) examine the effect of schools;
Davis (2004) measures the impact a regional increase in a health related risk; Irwin (2002) measures the
impact of open space, whereas Gibbons and Machin (2005) and Hess and Almeida (2007) examine the
impact of transport innovations and distance to rail transit stations on house prices, to name a few.
    There are scarce empirical investigations on the impact sports stadiums have on property prices. Until
recently, most of the related literature mainly focused on the impact of parks and golf courses (e.g. Cromp-
ton, 2000, 2005, 2007). These studies show that a golf course is not unintentionally designed, as proximate
residents are willing to pay a price premium to have such an amenity near by. Nevertheless, the latest
stadium construction boom and the fierce competition observed over the last couple of decades between
countries and cities to host a major sports event or franchise drew the attention of researchers on the impact
these venues have on proximate property prices.
    In what appears to be the first study on this topic, Carlino and Coulson (2004) relate a professional sports
franchise with a non-excludable public good, meaning that people derive utility through its presence in the

form of civic pride. The argument then is that (a) people might wish to relocate to the area in proximity
to the stadium, thus driving rent prices up, and (b) the inflow of manpower will generate labour abundance
in the area, thus driving wages down. Using a semi-logarithmic model of US monthly rent data they find
that the presence of a National Football League (NFL) franchise increases annual rents by approximately
8 percent in cities and about half that amount in the wider metropolitan area. Supporting their hypothesis,
they find that wages in NFL host areas decrease by approximately 2 percent, though the effect is statistically
    In Carlino and Coulson’s (2004) study it is not entirely clear though whether the property price increase,
as captured by increases in rents, is due to the existence of the stadium or the major league franchise. Ac-
knowledging this issue, Tu (2005) provides evidence that the construction of a brand new stadium, the
FedEx Field in the US, increased values of proximate properties. More precisely, he finds that, prior to its
completion, properties proximate to the stadium were sold at a discount compared to other more distant
ones. After completion of the stadium though, the price differential between proximate and non-proximate
properties was significantly decreased, implying that the stadium had a positive impact on residential prop-
erty values.
    Dehring et al. (2007) study differs from Tu’s (2005) only in that it examines the impact the construction
announcement of a potential stadium might have on the price of proximate properties. A basic assump-
tion here then is that property markets react to new relevant information, the effects of which are evident
on property price movements. Examining a series of announcements of stadium construction to host the
NFL’s Dallas Cowboys they first find that announcements associated with a positive probability of stadium
construction in Dallas had a significantly positive impact on property values in Dallas City and a negative,
though insignificant, impact on the remaining County. This result is not surprising given that Dallas County
would be contributing towards the stadium subsidy without sharing potential benefits associated to its prox-
imity. The pattern of signs reversed when the construction project was cancelled, though both coefficients
are insignificant. Second, the impact of the decision to build the stadium in Arlington, Texas, instead, was
surprisingly associated with a negative, though insignificant, impact on property values- even though the
construction project was given the green light after a positive vote by local citizens.
    On a similar investigation on European grounds, Ahlfeldt and Maennig (2007) estimate the impact
two sports stadiums, the Velodrom and Max-Schmeling (MS) Arena, have on land values in Germany’s
capital, Berlin. Estimating the significance of various binary variables determining the distance between
the property and the stadium, they find that both sports venues induce a positive impact on land values up
to a two kilometre radius- although the closest proximity effect (0-1 km dummy) is insignificant for the MS
    Feng and Humphreys (2008) estimate the impact of two different sports facilities, the Nationwide Arena
and Crew Stadium in Columbus Ohio, on residential property values. Using cross-sectional data from that
area for the year 2000 they find that both facilities generate a positive externality in house prices in the
surrounding area.
    In an interview-based analysis, Davies (2006) provides some evidence that the City of Manchester
Stadium and the Millenium Stadium in Cardiff have a positive impact on residential properties in the sur-
rounding area, respectively. She notes, though, that the impact on commercial properties is mixed, as
consumers in Cardiff tend to avoid shopping on event days in an attempt to avoid congestion.
    From another perspective, Coates and Humphreys (2006) study precinct-level voting preferences re-
garding the decision to subsidise or not the construction or renovation of sports venues in Green Bay and

Houston, US. They find that precincts proximate to a newly proposed or existing facility cast a larger share
of ‘yes’ votes in the subsidisation referendum. This tendency in voting behaviour is possibly justified either
because residents observe an increase in their wealth, measured by an increase in the price of the property
they own or in increased business trade in proximate stores they own, or because they hold a higher degree
of fandom.

3     East London and the 2012 Olympics
London won the bid to host the 2012 summer Olympic and Paralympic Games on 6th July 2005. The initial
cost of the Games was estimated at £2.375 billion, of which £1.5 billion would come from the National
Lottery, £625 million from London’s council tax payers and £250 million from the London Development
Agency (LDA), a municipal agency in charge of London’s economic growth. However, following the
March 2007 revised estimates the National Lottery now has to provide an additional £675 million. To cover
the remaining costs, an additional £300 million will also be provided by the Mayor of London, whereas the
remaining amount (around £6 billion, including the original estimates required for the wider regeneration
of the area) will be provided by the government through the Exchequer (DCMS, 2007a, 2007b).
     The London Olympics will take place in one of the capital’s most deprived areas; the Lower Lea (or
Lee) Valley in East London. For nearly half a millennium this area has been associated with the industrial
sector, leading to 75 percent of the land having been contaminated before construction of the Olympic site
began.1 The Olympics will be hosted in five London boroughs: Greenwich, Hackney, Newham, Tower
Hamlets and Waltham Forest. With the exception of Greenwich, the remaining four Boroughs are located
within East London. The wider regeneration of the area was one of the main arguments stressed by the
London 2012 bidding team. Considering the scale of the regeneration projects, the London Olympics can
be compared to those undergone during preparations for the Barcelona 1992 Games.2 It is thus expected
that the Olympics will leave a lasting legacy in terms of new sports facilities, the conversion of the Olympic
village into housing units, and improved availability of open space areas (the Olympic park). Given the
proximity the Lower Lea Valley has to Central London and the improved transport infrastructure which
will be in place by 2012, there is little doubt that the regenerated area will leave the dynamics of property
demand unchanged (Davies, 2006; Poynter, 2006).3
     Importantly though, a potential negative externality also arises. As the area is one of the most deprived
in the capital, it has become the base of small businesses and residence of poor communities. Unless regen-
eration of the area is designed within the existing communities’ best interest, it is likely that is inflationary
pressures on the price of land and property will displace the poorest people currently residing there (Ryan-
Collins and Sander-Jackson, 2008). The Olympics have already caused a high disruption in the locality as
a Compulsory Purchase Order had to be issued in order for more than 200 local small and medium sized
businesses to relocate from the site planned for the Olympic Village.4 Potentially, this might also have had
an effect on local unemployment. Hence, the decision to bid and potentially host a major sporting event
may lead to “uneven economic and social outcomes within the city” (Jones, 2001:848).
     Finally, note that similar to Barcelona’s case, London’s East, especially Newham, was not lacking in-
    1 London 2012 website:
    2 Propertyprices in Barcelona increased following the Olympics. Importantly though, one must consider the substantial inflow
of monetary resources and investment fuelled by the European Commission at the time (Holt, 2005).
   3 A recent news article unveils a 15 percent average property price increase in parts of the East’s localities since the announce-

ment of the successful bid. At an extreme, average house prices in Hackney have risen by a reported 21 percent (Hill, 2007).
   4 See BBC News website:

vestment funds. Since the early 1990s, Newham had been going through regeneration plans in order for
the Channel Tunnel Rail Link terminal to be based in Stratford, whereas the Crossrail project would also
add to the transportation infrastructure of the area. Shortly thereafter, the region won a City Challenge and
a European Regional Development Fund grant; the former granting access to £37.5 million of British gov-
ernment funds distributed over a five year period, the latter granting access to £15 million over three years
(Florio and Edwards, 2001). Further local investment was enhanced by the relocation of many multinational
financial services companies previously based in the City of London to Canary Wharf, a location adjacent
to Stratford, in East London (Poynter, 2006). As many of these projects are still under way, isolating the
impact attributed solely to the Olympics is probably not a trivial task.

4      Data and Methodology
The aim of this study is to assess the impact the announcement of the 2012 successful Olympics bid has
on property prices in London. To address this question we consider the impact on the host boroughs and
the impact on consecutive radius rings drawn around the main Olympic stadium. This paper is the first to
empirically question the potential impact hosting a major sporting event has on property prices.

4.1     The Data
Although there appear to be several sources of property transactions and characteristics data in the US,
no single database exists yet containing similar information for properties sold in Britain. Lack of such a
database probably justifies the few hedonic models applied in the British housing market. For example,
Garrod and Willis (1992) study the impact of forestry on the prices of 1,000 properties across Great Britain
in 1988; Cheshire and Sheppard (1995) study the value of property amenities on 490 and 350 properties
based in Reading and Darlington, respectively, in 1984; Leech and Campos (2003) measure the impact of
secondary school admissions policy on property prices in Coventry in 2000. For the purposes of this study
the dataset had to be manually constructed from publicly available sources.
      The most crucial part of the data is gathered from MousePrice, which uses Land Registry data, the
official record dealing authority for property transactions in the UK. Although other web sources similar
to MousePrice contain Land Registry data, the former also contains additional characteristics data of the
property sold. These are the type of property (whether detached, semi-detached, terrace, or flat), the number
of bedrooms, the internal area of the property (in square meters), the year the property was built (and
consequently, whether the property was “sold as new”), and finally the ownership type of the property
(whether freehold or leasehold).
      Our data is retrieved from MousePrice by performing a general postcode search. Typing in a postcode,
the online database returns 200 random results, spanning the whole data range and property types. Note
that we only include properties for which all relevant information is available. No information on the
characteristics of properties was available for flats. A total of 113 postcodes exist for the Greater London
area.5 These are in turn allocated to six geographical boundaries: the North (22), North West (11), East
(18), South East (28), West (14) and South West (20) London. Properties in East Central and West Central
London are excluded, as these are mostly comprised of flats for which no data exists. Given our search
mechanism, we are able to self-construct a dataset containing more than 13,700 observations for 27 out
    5 The postcode search is performed using the number of the second part of the code as well. For example, for SW7 we perform

a separate search on SW7 0, SW7 1, SW7 2, ..., SW7 9.

of London’s 33 boroughs; our dataset does not contain properties located in the boroughs of Barking and
Dagenham, Harrow, Havering, Hillingdon and Sutton.
      Furthermore, we gather additional information from alternative sources in order to enrich our dataset.
We use MultiMap in order to get a measure of the distance of each property from its nearest tube station.6
We anticipate that transport links available to each property is an important characteristic of its price. Crime
statistics at the borough level are also included. These are available between April 2001- March 2007 from
the Office for National Statistics (ONS).7 The crime statistic used is the more general “violence against the
person” record, recalculated in percentage terms to make them comparable across boroughs.
      As mentioned previously, earlier studies have pointed out the significant impact of quality proximate
schools on property prices. Ideally, we would also like to control for school quality, however there are two
caveats associated with this task here. First, a combined measure of the quality of multiple schools needs to
be considered. This is because the net effect of the impact of school quality on property prices is arguably
captured by the weighted average of the quality of all proximate schools in a pre-determined distance from
each property in the sample. Second, although primary and secondary school league tables are available
mainly from newspapers, the latter provide relatively recent information and access to historical statistics
is limited. Thus, due to the dimensions of this study, covering a wide geographical area over a number of
years, gathering statistics on school quality is rather unfeasible.
      Nevertheless, an alternative approach is considered here. Following information from government
sources, we obtain a measure of the number of primary and secondary schools located within a one-mile
radius from each property.8 The rationale for this is simple. A relatively small radius ring is chosen to allow
the effects of close proximity, where we hypothesize that the larger the pool of proximate schools is, the
greater the probability a high quality one exists within the one-mile boundary.
      Importantly note that the age of the property is not a characteristic accounted for in the regressions.
Previous research stresses the significance a dwelling’s age has on price as a proxy for quality (see for
example Lee et al. (2005) and Rappaport (2007)). This information however was not available for the
overwhelming majority of the data. Also, leasehold properties are excluded from the regressions. The
justification for this is simple: the price of the property on offer crucially depends on the remaining years
on the lease, which is unobservable. Thus, we opt to exclude non-freehold properties from our estimations.

4.2     Methodology
4.2.1     A Hedonic Model for London

The methodology used in this study is that of a hedonic approach. This type of model treats the value of a
final good as the dependent variable, regressed upon the attributes, or components, of that good. The intu-
ition behind this approach is to obtain an inferential value of a non-market good (i.e. a characteristic), such
as an extra square meter on the property’s size, based on the price the observed good sells for (Crompton,
2001). As already mentioned, these can be both property specific and neighbourhood specific.
      Our estimation approach uses a semi-logarithmic functional form of the hedonic pricing model. Choice
of this functional form is not accidental. Using the Box and Cox (1964) transformation model, we estimate
  6 See Multimap’s website:
  7 See ONS Neighbourhood statistics website:
  8 Information is gathered by performing a postcode search using a governmental website:

the following by maximum likelihood:
                                                   (θ )                  (λ )
                                            price j = c + ∑ βi xi               +εj                       (1)

    where the dependent variable is the price of property j, transformed to:
                pθ −1 ,      f or θ 6= 0
    price = )      θ
               ln p   ,      f or θ = 0

    and the strictly positive explanatory variables, x, are transformed to:
            xλ −1 ,        f or λ 6= 0
    x = )       λ
           ln x      ,     f or λ = 0

    The interest from the Box-Cox regression of equation 1 lies with the estimated coefficients for lambda
(λ ) and theta (θ ). These estimates, shown in Table 1 below, are the optimal parameters to linearise the
relationship between the dependent and independent variables based on the Box and Cox transformations.

                                             λ                 1.071*** [0.072]
                                             θ                -0.180*** [0.011]
                                         Notes: Log Likelihood= -129,101.44. Stan-
                                         dard errors reported in brackets. The number
                                         of observations is 10,180. Estimation is per-
                                         formed for freehold properties only.

                            Table 1: Box-Cox Regression Estimates for θ and λ

    The Box-Cox representation nests the linear, log-linear and semi-log models (Maurer et al., 2004). If
θ = λ = 1, the suggested hedonic model is linear; if θ = λ = 0, the suggested model is log-linear. When
θ = 0 and λ = 1, the suggested model is semi-logarithmic. Hence, given the parameter estimates in Table 1,
estimation of the semi-log model is somewhat justified. Furthermore, previous empirical evidence suggests
that the use of a relatively simple functional form is more appropriate due to the lack of additional control
variables (Cropper et al., 1988). Applying the semi-log rather than the more complicated transformed model
based on the optimal parameter estimates has the added benefit of a simple and intuitive interpretation of
the estimated coefficients of the hedonic model. These will reflect the percentage change in price for a
one-unit change in the independent variable (Sirmans et al., 2005).
    The model applied here then is similar to Tu’s (2005), though as the construction of the facilities and
the surrounding Olympic Village has not been finalised, we assume property markets exhibit informational
efficiency and react to announcements, as in Dehring et al. (2007). In brief, investment reactions can be
summarised in three general scenarios: (a) announcement of staging the Olympics boosts property prices;
one wishes to reside in the area to share proximity benefits to the facilities, the park, improved transport
facilities, etc.; (b) fears of congestion on the aftermath of 2012 is an externality potentially leading to
decreased property values. This is certainly plausible as people might be commuting from other regions
to enjoy the scenery. An increased amount of visitors is likely to be associated with increasing noise and
pollution levels; (c) there is no significant change in the demand for housing- individuals are indifferent
whether to invest in the area, or not.

4.2.2         Hedonic Models based on Host Boroughs

Initially, we consider the impact on the five official host boroughs. These are the boroughs of Greenwich,
Hackney, Newham, Tower Hamlets and Waltham Forest. However, events hosted in Greenwich will take
place in existing and temporary facilities. Hence, apart from an improvement in transport infrastructure, this
borough is not directly related to the Lower Lea Valley regeneration project, which includes the construction
of the new sports facilities. To this extent we additionally consider only the boroughs of Hackney, Newham,
Tower Hamlets and Waltham Forest in our analysis.
       Three semi-logarithmic hedonic models are estimated in regards to the impact the Olympics might have
on properties within the host boroughs. We first wish to compare property price differentials between host
and non-host boroughs in both time periods; that is, before (t = 1) and after (t = 2) the announcement. This
is achieved by formulating a dummy variable, Host, equal to unity if the borough acts as the host of the
Games. Hence, for the pre- and post-announcement periods (Models 1 and 2, respectively) we separately
estimate:                                               n
                                     log p j,t=k = c + ∑ βi xi + δ0 Host j,t=k + ε j,t=k                                   (2)

       where k = 1, 2, p j is the transactions price of property j, xi is a set of property and neighbourhood
specific characteristics, and ε represents the error term.
       In Model 3 we model the impact of the announcement as a natural experiment, applying a difference-
in-differences (DD) methodology. Reliable estimates from this model arise in so far as winning the bid
to host the 2012 Olympics was fairly unexpected. Had the announcement been expected, property prices
would have arguably already incorporated the effect of hosting the Olympics. The best measure of prior
expectations regarding the Olympics in this context is betting odds. Prior to the announcement these were
1-4 for Paris and 11-4 for London. These translate into a 0.8 probability of victory for Paris and around
0.27 for London.9
       The DD requires a time period and group indicator (Wooldridge, 2002:129). A binary variable, Announcement,
equal to unity for the post-announcement period determines the time period. The host boroughs can be
thought as the treatment group, whereas the remaining non-hosts boroughs are the control group. The
DD estimator then is the interaction of the time and treatment variables, Host · Announcement; a dummy
variable equal to unity for the host boroughs in the post-announcement period only. The DD model is
summarised in equation 3 below.
                log p j,t = c + δ1 Announcementt + δ2 Host j + δ3 Host · Announcement jt + ∑ βi xi + ε j,t                 (3)

       Where, in its simplest form with the absence of any additional explanatory variables, the DD estimator

       δˆ3 = log pHost,Time=2 − log pHost,Time=1 − log pNonHost,Time=2 − log pNonHost,Time=1

       where log pq,t is the average log price of properties in host or non host boroughs, q, prior or following
the announcement at time t.
That is, it is the difference of means between the two time periods of the treatment group (i.e. hosts), less
the difference of means between the two time periods of the control group (i.e. non-hosts).
      9 See
         BBC News: A k − n betting odd is translated
into a probability according to k+n . Note also that probabilities add up to more than one, thus generating revenue for betting

4.2.3      Hedonic Models based on Proximity

Using a similar methodology, a potentially more rigorous analysis is employed based on the impact of the
announcement given the distance each property has from the main Olympic Stadium, denoted by Stadium.
This measure of distance is calculated using GoogleMap.10 Note that considering a gradient from the main
stadium is of course only an approximation of property proximity to the park and the sports facilities for
two main reasons. First, the value of some properties located around the perimeter of the park is likely
to be enhanced due to their view of the park alone. Second, even if the property is not adjacent to the
park, multiple entrances to the Olympic Park will presumably become available to the public once the 2012
Games are over, thus decreasing the distance between the property and the park’s boundaries. Regarding
the latter arguments though, such precise information is unavailable to the author as of date.
    However, the hypothesis that the degree of the impact might vary for properties on a certain radius
should not be overlooked. Using once again the coordinates of the main Olympic Stadium as the centre
point, we would ideally like to consider this possibility by estimating the impact on consecutive one-mile
radius rings. However, our sample does not contain enough observations to allow for such an analysis.
Given the number of available observations, consecutive three-mile radius rings are considered in this study,
covering all properties in our sample.11

    Estimations of all models use the neighbourhood and dwelling-specific attributes. In all estimations we
control for location effects at the borough level and for year of transaction effects. The former will capture
any location attributes correlated with price, not captured in our explanatory variables. The latter accounts
for the same unobserved effects at the annual level.
    Allocating a property sold to a specific year is not necessarily a trivial task. In the UK the norm for the
buyer is to purchase the property after applying for a mortgage; a process which, according to lawyers, can
take around 40-50 days on average. Even without the need to apply for a mortgage, a substantial period
of time is needed for the paperwork to be completed before the formal exchange of contracts. That is,
although the selling price of the property has been agreed between the seller and the buyer, the purchasing
fee recorded in our database is observed further in the future.
    Hence, similar to other studies (Dehring et al., 2007), we allow for a 45-day window for the exchange
of contracts to take place. This means that although a property has been legally sold on the 15th January
2004, for example, it is assumed that its price was determined on 1st December 2003. Crime statistics
per borough are allocated in a similar fashion. Hence, the post-event period begins for properties with a
transaction date of 21 August 2005, forty-five days following the announcement of the winner in Singapore
on 6th July 2005.
    Finally, as property values might be spatially correlated due to unobservable attributes common to
properties proximate to one another, we robustly estimate all models by clustering the error terms at the
Ward and year level.12 For a description of the variables used in this study see Table 2.
  10 See Google Map’s website:
  11 For example, for the 0-3 miles radius our sample contains 277 and 160 properties sold in the pre- and post-annoucement
periods, respectively. For the 3-6 miles radius, these figures are 1,105 and 595, respectively.
  12 Wards are divisions of Boroughs. Thus, properties within the same Ward are assumed to share most of the common neigh-

bourhood characteristics. The allocation of a property to its corresponding Ward was performed based on the dwelling’s postcode.
There are more than 9,400 ward-year clusters in our sample. Dehring et al. (2007) formulate similar clusters at the school level.

5      Results
We examine the data to detect properties with extreme attributes. Exclusion of outliers from the empirical
analysis is entirely justified in this context (see for example similar studies by Tu (2005), Coates et al.
(2006), Carlino and Coulson (2006) and Dehring et al. (2007)). Residences with more than six bedrooms,
greater than 400 sq. meters and less than 40 sq. meters, valued more than £2 million and less than £50,000
are subsequently excluded from all estimations. This accounts to the deletion of some 130 observations.
Recall that crime statistics are available only between April 2001 and March 2007, forcing us to discard all
property transactions before and after those dates. Therefore, regressions are based on a sample of 10,180

5.1     Results based on Host Boroughs
Table 3 reports the results on the three hedonic models for all five hosts. Model 1 suggests that the value
of a residential property within the host boroughs was lower than the corresponding value of a comparable
property placed outside the impact area prior to the Olympics announcement. Prior to the announcement,
properties in Greenwich, Hackney, Tower Hamlets and Waltham Forest sold for approximately 8.1 percent
lower (= exp (−0.084) − 1) than comparable properties in other boroughs.13
      Nonetheless, London’s successful Olympics bid did seem to have a substantial impact on property
values following the announcement as well. Post-announcement (Model 2), properties within the area
directly impacted from the Games are estimated to sell for 4.3 percent lower, compared to properties with
similar attributes in non-host areas. The size and significance of this coefficient indicates that hosting
the Olympics has a positive impact on property prices; the price differential between host and non-host
boroughs decreases by nearly 50 percent.
      Models 1 and 2, although presenting us with revealing results on the impact hosting a major sporting
event has on the London property market, fail to capture the evolution of prices in host and non-host borough
across both time periods- i.e. before and after the announcement. As discussed previously, the difference-
in-differences estimator measures the difference in average (log) prices between the two time periods for
hosts, less that for non-hosts (Model 3). In addition, the DD estimator is robust on the choice of the base
dummy for borough location. For example, if in Models 1 and 2 the Royal Borough of Kensington and
Chelsea is chosen for the base dummy, properties located within the five hosts are sold at a 41.2 percent
discount prior to the announcement and at a 35 percent discount after the announcement.
      In Model 3, the coefficient of Host (-0.079) measures the price differential between host and non-host
boroughs in the absence of the announcement, suggesting that over our entire sample host boroughs sell at
a 7.6 percent discount. The DD estimator, Host · Announcement, indicates however that properties in the
five host boroughs are actually sold for approximately 2.1 percent higher.
      Table 4 reports similar results for the four hosts; that is, for the main potential beneficiaries of the
Games in terms of infrastructural investment and renewal. The results on all models are similar and hold
the same interpretation. Notably though the estimate of the DD estimator in Model 3 is worth mentioning
in this case. Exclusion of Greenwich, a borough hosting the Games but not directly influenced by the
regeneration projects and construction of new facilities, leads to an increase in the DD estimator by nearly
a third compared to that of Table 3. Significant at the 1 percent level, properties in the four host boroughs
are now estimated to sell for approximately 3.3 percent higher.
   13 As noted by Halvorsen and Palmquist (1980), the percentage effect of the estimated coefficient, β say, of a dummy variable

of a semi-logarithmic model is 100 · [exp (β ) − 1].

The DD results of Tables 3 and 4 are in line with our prior expectations. That is, including an area
which, although acting as a host, is not gaining much Olympics specific related development only lessens
the magnitude of the impact.

5.2     Results based on Proximity
Finally, Table 5 reports the results based on the proximity variables. Two sets of results are presented here.
Column (1) reports the results based on the distance metric of each property from the main Olympic Sta-
dium, whereas column (2) reports results based on three-mile radius rings drawn around the main stadium.
For simplicity purposes we only report the results on the difference-in-difference type estimators; that is,
the ones corresponding to Model 3 above.
      The results under column (1) tell an interesting story. The coefficient of Stadium is positive and sig-
nificant, implying that over our sample size each additional mile away from the stadium increases property
prices by 0.2 percent. In essence, this translates to consumer unwillingness to reside in proximity to one of
Britain’s most deprived areas. Interestingly though, the estimated coefficient of the distance and announce-
ment variables, Stadium · Announcement, implies that following the announcement of the successful bid
each additional mile away from the main stadium decreases property prices by 0.4 percent.
      The results in (2) complement those in (1) by providing a more detailed analysis based on radius rings.
Here, note that R (i, j] denotes a radius ring which includes properties greater than i miles and less than, or
equal to, j miles around the stadium. The DD estimators of interest then are the interactions of the R (i, j]
variables with the Announcement variable; that is, properties within a specific radius and with a contract
exchanged after the announcement of the Olympics bid.
      According to this specification, prices of properties up to 3 miles away from the stadium significantly
increased by 5 percent. Furthermore, the impact is estimated to have a decreased effect of around 2 percent
for properties beyond 3 miles and up to and including 9 miles away from the stadium. Properties sold
beyond the 9 mile boundary do not seem to be significantly impacted; in fact the estimated coefficients on
the more distant radius rings are negative.
      The combined estimates of both specifications in columns (1) and (2) provide insightful results. Nev-
ertheless, a positively estimated and significant impact on properties up to nine miles away from the main
Olympic Stadium may be viewed with some scepticism. However, one should keep in mind that the main
Olympic Stadium is located fairly at the south of the Olympic Park, where the distance of the former from
the northern boundary of the Park is about three miles. Furthermore, note that the regeneration and ame-
lioration of the Lower Lea Valley (Olympic Park) will result in one of the largest urban parks in Europe.14
Hence, the persistence and size of the estimated impact on property prices is somewhat plausible.

      Overall, the estimated coefficients of the characteristics of properties appear to have the expected sign,
although for some the coefficient and degree of significance varies according to the model being estimated.
The number of bedrooms has a strongly significant impact on price, each bedroom adding approximately
5.3 percent to the price. The price is further positively correlated with the internal area of the dwelling;
the coefficient for the internal area is highly significant, but the effect on price is relatively small (each
additional square meter adds approximately 0.3 percent to the price). Note, though, that in accordance
to earlier studies (e.g. Palmquist (1984)) the price and internal area relationship is U-shaped; an increase
in a property’s internal area increases its price, though at a decreasing rate. A brand new property is
  14 See   London 2012 official website:

also associated with a price premium. The construction type of the property- whether detached or semi-
detached- also is highly significant. A detached property adds 12 percent to the selling price, whereas a
semi-detached only around half that amount, on average.
      On the neighbourhood characteristics, the further away the property is from its nearest tube station the
lower the price one has to pay- each additional mile decreases property values by 2.6 percent. The effect of
crime is similar, though about half that of the transportation facilities. The story on the impact of proximate
schools is not so straight forward. Although we find that the relationship between price and the number of
proximate primary schools within a one-mile radius is significantly positive, the same can not be said for
secondary schools. The latter appear to have an exact opposite, or in some cases insignificant, effect. In
any case though, note that the school metric included in these regressions is not very informative on school
quality. Future research might wish to determine whether there are negative externalities associated with
specific proximity to secondary schools, such as increased noise levels and potential teenage crime.15

5.3     Implications
The empirical approach followed in this paper has some implications for policy. The results suggest that
property prices in the East London area are subject to increase. Although this might be good news for
some property owners, price increases are likely to have devastating implications for those in lower income
bounds residing in the locality. Gentrification of an area potentially leads to social exclusion of the poorest
households. A failure to integrate the regeneration process to the needs of the local society does not actively
solve the problem of social deprivation in East London; it rather transfers it elsewhere.
      An interesting accounting implication based on our results is to estimate the overall monetary impact the
Olympics have on surrounding properties. The number of housing stock for the four main host boroughs
(excluding flats, which do not form part of our sample) is taken from the most recent UK 2001 Census
available from the Office for National Statistics. We then calculate the average property price per borough
and year following the Games announcement. Based on this calculation a freehold property in Hackney
sells, on average, for £450,017; in Newham for £234,557; in Tower Hamlets for £394,986; and in Waltham
Forest for £261,365. Based on the results of the DD estimation on the four host boroughs (Table 4), the
total monetary impact of the announcement is approximately £1.4 billion, calculated as the sum of a 3.3
percent increase on average property prices multiplied by the housing stock. This number accounts for
approximately 15 percent of the £9.3 billion cost for the 2012 London Olympic Games.

6      Conclusion
The impact sports facilities have on the surrounding property market is an under researched field. Following
a methodology similar to that of recent studies, this paper is the first to address the impact on property prices
hosting a major sporting event has. The case of the 2012 Olympics is considered where, more specifically,
we examine the impact the announcement of London’s successful bid has on residential property prices in
the host boroughs.
      The host boroughs are measured in two separate ways. First, we consider the official five boroughs
hosting the Games. Second, we consider only the four main host boroughs primarily linked with the
construction and regeneration projects; in the aftermath of 2012 the the sports venues and the Olympic
Park will lie within their borough boundaries. Irrespective of the way we define the hosts, i.e. the impact
    15 Regarding   all estimations, note that allowing for a 60-day window for the exchange of contracts produces similar results.

area, we find that the price deferential between host and non-host boroughs has decreased following the
successful Olympics announcement. A property with comparable attributes used to sell for significantly
less if located in the host area. The magnitude of the effect varies though depending on the choice of the
location base dummy.
    Subsequently, a difference-in-differences methodology is called for to control for price differentials
between host and non-host boroughs across the announcement periods. Based on the DD estimates the
price increase of properties located in the host boroughs, attributed to the Olympics, is about 3.3 percent.
Including the borough of Greenwich as a host decreases the impact to 2.1 percent.
    Furthermore, an alternative specification is applied based on the property distances from the main
Olympic stadium. Following the announcement, we find that each additional mile away from Lower Lea
Valley decreases property prices by 0.4 percent. In a final estimation we draw consecutive three-mile radius
rings centered at the main Olympic Stadium. In essence though, these are three-mile hemispheres expanded
over the western region of London. A DD methodology of distance based dummy variables indicates that
property prices within three miles from the stadium increase by 5 percent. The model also suggests a 2
percent increase on properties up to nine miles away, with an estimated insignificant impact for dwellings
located further away. These coefficients are of a similar magnitude to those estimated in other studies (e.g.
Tu (2005) estimates an approximately 5 percent property premium achieved following each phase of the
stadium development).
    One should however note that the impact we are measuring is partly influenced from a wider regenera-
tion strategy including the regeneration of the park and Lower Lea Valley, the construction of the facilities
and the development of transport links- notably the Channel Tunnel terminal in Stratford and the exten-
sion of the Docklands Light Railway- which have been shown to have a strong impact on property prices.
Another issue is the upward trend property prices were following over the last decade or so. Thus, as this
period is not a typical one, it is possible that the estimated effect on prices was based on irrational exu-
berance given the trend they had. Had the successful bid to host the 2012 Olympics been announced in a
market with lower prices, the Olympics might have had a lower effect.
    Admittedly, our self-constructed database contains a relevantly limited amount of dwelling and neigh-
bourhood characteristics compared to other hedonic studies, especially those based in the US housing mar-
ket. Additional property characteristics which could be controlled for are lacking here. Such could include
the property’s age, existence of a parking space, a fireplace, and the number of bathrooms, to name but a
few. Additional neighbourhood characteristics, whose statistics are unavailable for the UK at the borough
and year level could include pollution and noise levels. Though the amount of available information appears
to be satisfactory for the purposes of this study, through this paper we wish to call for the construction of
an official, detailed, property characteristics database for the UK market. A post-2012 study would be of
particular interest in order to determine whether the estimated impact on properties significantly alters once
the related construction and regeneration projects are completed.

Variable        Description
Price           The price the property was sold for, £.
lprice          The log of price.
Bedrooms        The number of bedrooms.
Area            The internal area of the property, square meters.
Area2           The internal area of the property squared.
New             Dummy variable, =1 if property is sold as new, =0 otherwise.
Terrace         Dummy variable, =1 if the property type is terrace, =0 otherwise.
Semi-Detached   Dummy variable, =1 if the property type is semi-detached, =0 otherwise.
Detached        Dummy variable, =1 if the property type is detached, =0 otherwise.
Primary         The number of mixed primary schools within a one-mile radius from the property.
Secondary       The number of mixed secondary schools within a one-mile radius from the property.
Tube            The property’s distance to the nearest tube station, miles.
Crime           The percentage of “violence against the person” at the borough level.
Announcement    Dummy Variable, =1 if property is sold post announcement, =0 otherwise.
Host            Dummy Variable, =1 if property belongs to a host borough, =0 otherwise.
Stadium         Distance of each property from the main Olympic Stadium, miles.
R (i, j]        Dummy Variable, =1 if property lies within an i- j radius ring centred at the main Olympic stadium,
                =0 otherwise.

                                    Table 2: Description of Variables

Model 1                      Model 2                       Model 3
                     Announcement                                                                            0.007       [0.006]
                     Host                         -0.084*** [0.009]            -0.044*** [0.013]            -0.079*** [0.008]
                     Host*Announcement                                                                       0.021*** [0.007]
                     Constant                      4.943*** [0.026]             5.195*** [0.044]             5.14*** [0.023]
                     Bedrooms                      0.051*** [0.003]             0.057*** [0.004]             0.053*** [0.003]
                     Area                          0.003*** [0.001]             0.003*** [0.001]             0.003*** [0.001]
                     Area2                        -0.001*** [0.001]            -0.001*** [0.001]            -0.001*** [0.001]
                     New                           0.049*     [0.028]           0.092*** [0.036]             0.052** [0.026]
                     Detached                      0.119*** [0.014]             0.114*** [0.016]             0.118*** [0.011]
                     Semi-Detached                 0.056*** [0.005]             0.056*** [0.006]             0.057*** [0.004]
                     Primary                       0.004*** [0.001]             0.005*** [0.001]             0.004*** [0.001]
                     Secondary                    -0.003*** [0.001]            -0.003*** [0.001]            -0.003*** [0.001]
                     Tube                         -0.024*** [0.003]            -0.031*** [0.004]            -0.026*** [0.002]
                     Crime                        -0.047*** [0.005]            -0.061*** [0.001]            -0.041*** [0.004]
                     Year Effects                        YES                           YES                           YES
                     Location Effects                    YES                           YES                           YES
                     Obs.                                7,033                        3,147                       10,180
                     R2                                  0.745                        0.762                          0.76
Notes: Regressions are Ordinary Least Squares. The dependent variable is lprice. Host boroughs are Greenwich, Hackney, Newham, Tower
Hamlets and Waltham Forest. The base dummy for Location is Brent. Terrace is the base dummy for the property’s construction type. Robust
standard errors, clustered at the ward-year level, are reported in brackets. *, **, *** denotes significance at the 10%, 5% and 1% level, respectively.

                          Table 3: Hedonic Model Estimation Results for the Five Host Boroughs

                                                       Model 1                 Model 2                            Model 3
                     Announcement                                                                            0.007       [0.006]
                     Host                         -0.075*** [0.001]            -0.03**     [0.013]          -0.071*** [0.008]
                     Host*Announcement                                                                       0.032*** [0.008]
                     Constant                      4.937*** [0.026]             5.119*** [0.047]             5.125*** [0.024]
                     Bedrooms                      0.051*** [0.003]             0.057*** [0.004]             0.053*** [0.003]
                     Area                          0.003*** [0.001]             0.003*** [0.001]             0.003*** [0.001]
                     Area2                        -0.001*** [0.001]            -0.001*** [0.001]            -0.001*** [0.001]
                     New                           0.053*     [0.029]           0.099*** [0.036]             0.057** [0.026]
                     Detached                      0.120*** [0.015]             0.115*** [0.016]             0.119*** [0.011]
                     Semi-Detached                 0.057*** [0.005]             0.058*** [0.006]             0.058*** [0.004]
                     Primary                       0.004*** [0.001]             0.004*** [0.001]             0.004*** [0.001]
                     Secondary                    -0.004*** [0.001]            -0.004*** [0.001]            -0.004*** [0.001]
                     Tube                         -0.019*** [0.003]            -0.021*** [0.004]            -0.019*** [0.003]
                     Crime                        -0.045*** [0.006]            -0.039*** [0.011]            -0.036*** [0.005]
                     Year Effects                        YES                           YES                           YES
                     Location Effects                    YES                           YES                           YES
                     Obs.                                7,033                        3,147                       10,180
                     R2                                  0.745                        0.765                        0.761
Notes: Regressions are Ordinary Least Squares. The dependent variable is lprice. Host boroughs are Hackney, Newham, Tower Hamlets and
Waltham Forest. The base dummy for Location is Brent. Terrace is the base dummy for the property’s construction type. Robust standard errors,
clustered at the ward-year level, are reported in brackets. *, **, *** denotes significance at the 10%, 5% and 1% level, respectively.

                        Table 4: Hedonic Model Estimation Results for the Four Host Boroughs

(1)                         (2)
             Announcement                          0.054*** [0.01]             -0.001         [0.011]
             Stadium                               0.002** [0.001]
             Stadium*Announcement                 -0.004*** [0.001]
             R (0, 3]                                                           0.052*** [0.017]
             R (3, 6]                                                           0.068*** [0.015]
             R (6, 9]                                                           0.022*        [0.013]
             R (9, 12]                                                          0.064*** [0.011]
             R (12, 15]                                                         0.029*** [0.011]
             R (0, 3]*Announcement                                              0.049*** [0.016]
             R (3, 6]*Announcement                                              0.020*        [0.012]
             R (6, 9]*Announcement                                              0.022** [0.011]
             R (9, 12]*Announcement                                            -0.001         [0.011]
             R (12, 15]*Announcement                                           -0.005         [0.011]
             Constant                              5.06*** [0.025]              4.817*** [0.03]
             Bedrooms                              0.052*** [0.003]             0.051*** [0.002]
             Area                                  0.003*** [0.001]             0.003*** [0.001]
             Area2                                -0.001*** [0.001]            -0.001*** [0.001]
             New                                   0.057** [0.026]              0.049*        [0.025]
             Detached                              0.119*** [0.011]             0.121*** [0.011]
             Semi-Detached                         0.06*** [0.004]              0.058*** [0.004]
             Primary                               0.004*** [0.001]             0.002*** [0.001]
             Secondary                            -0.004*** [0.001]            -0.001         [0.001]
             Tube                                 -0.021*** [0.003]            -0.027*** [0.003]
             Crime                                -0.038*** [0.005]            -0.015** [0.006]
             Year Effects                                   YES                         YES
             Location Effects                               YES                         YES
             Obs.                                        10,180                      10,180
             R2                                             0.76                      0.783
Notes: Regressions are Ordinary Least Squares. The dependent variable is lprice. The base dummy
for Location is Brent. Terrace is the base dummy for the property’s construction type. In column (2),
R(15, 18]   is the base dummy for the radius ring. Robust standard errors, clustered at the ward-year level,
are reported in brackets. *, **, *** denotes significance at the 10%, 5% and 1% level, respectively.

                  Table 5: Hedonic Model Estimation Results on Proximity


Ahlfeldt, G., and Maennig, W. (2007). “Impact of sports arenas on land values: Evidence from

  Berlin.” Working Papers 0703, International Association of Sports Economists, available online at: [Accessed June 2008].

Anselin, L., and Le Gallo, J. (2006). “Interpolation of air quality measures in hedonic house price models:

  Spatial aspects.” Spatial Economic Analysis, 1(1), 31–52.

Box, G., and Cox, D. (1964). “An analysis of transformations.” Journal of the Royal Statistical Society:

  Series B, 26, 211–243.

Carlino, G., and Coulson, N. (2004). “Compensating differentials and the social benefits of the NFL.”

  Journal of Urban Economics, 56(1), 25–50.

Carlino, G., and Coulson, N. (2006). “Compensating differentials and the social benefits of the NFL: Reply.”

  Journal of Urban Economics, 60(1), 132–138.

Chay, K., and Greenstone, M. (2005). “Does air quality matter? Evidence from the housing market.”

  Journal of Political Economy, 113(2), 376–424.

Cheshire, P., and Sheppard, S. (1995). “On the price of land and the value of amenities.” Economica,

  62(246), 247–67.

Cheshire, P., and Sheppard, S. (2004). “Capitalising the value of free schools: The impact of supply char-

  acteristics and uncertainty.” Economic Journal, 114, 397–424.

Coates, D., and Humphreys, B. (2006). “Proximity benefits and voting on stadium and arena subsidies.”

  Journal of Urban Economics, 59(2), 285–299.

Coates, D., Humphreys, B., and Zimbalist, A. (2006). “Compensating differentials and the social benefits

  of the NFL: A comment.” Journal of Urban Economics, 60(1), 124–131.

Crompton, J. (2000). “Designing golf courses to optimize proximate property values.” Managing Leisure,

  5(4), 192–199.

Crompton, J. (2001). “The impact of parks on property values: A review of the empirical evidence.” Journal

  of Leisure Research, 33(1), 1–31.

Crompton, J. (2005). “The impact of parks on property values: Empirical evidence from the past two

  decades in the United States.” Managing Leisure, 10(4), 203–218.

Crompton, J. (2007). “The role of the proximate principle in the emergence of urban parks in the United

  Kingdom and in the United States.” Leisure Studies, 26(2), 213–234.

Cropper, M., Deck, L., and McConnell, K. (1988). “On the choice of functional form for the hedonic price

  functions.” Review of Economics and Statistics, 70(4), 668–675.

Davies, L. (2006). “Sporting a new role?: Stadia and the real estate market.” Managing Leisure, 11(4),


Davis, L. (2004). “The effect of health risk on housing values: Evidence from a cancer cluster.” American

  Economic Review, 94(5), 1693–1704.

DCMS (2007a). “Explanatory memorandum to the payments into the Olympic Lottery Distribu-

  tion Fund etc., Order 2007.” Department for Culture, Media and Sport, UK, available online at: [Accessed January 2008].

DCMS (2007b). “Olympic funding: Revised memorandum of understanding between the Govern-

  ment and the Mayor of London.” Department for Culture, Media and Sport, available online at" [Accessed Jamuary 2008].

Dehring, C., Depken, C., and Ward, M. (2007). “The impact of stadium announcements on residential

  property values: Evidence from a natural experiment in Dallas-Fort Worth.” Contemporary Economic

  Policy, 25(4), 627–638.

Espey, M., and Lopez, H. (2000). “The impact of airport noise and proximity on residential property values.”

  Growth and Change, 31(3), 408–419.

Feng, X., and Humphreys, B. (2008). “Assessing the economic impact of sports facilities on residential

  property values: A spatial hedonic approach.” Working Papers 0812, International Association of Sports

  Economists, available online at: [Accessed August 2008].

Florio, S., and Edwards, M. (2001). “Urban regeneration in Stratford, London.” Planning Practice and

  Research, 16(2), 101–120.

Garrod, G., and Willis, K. (1992). “The environmental economic impact of woodland: A two-stage hedonic

  price model of the amenity value of forestry in Britain.” Applied Economics, 24(7), 715–728.

Gibbons, S., and Machin, S. (2003). “Valuing english primary schools.” Journal of Urban Economics, 53,


You can also read
NEXT SLIDES ... Cancel