Housing prices and crime perception Paolo Buonanno, Daniel

Housing prices and crime perception
Paolo Buonanno, Daniel Montolio &
Josep Maria Raya-Vílchez
Empirical Economics
Journal of the Institute for Advanced
Studies, Vienna, Austria
ISSN 0377-7332
Empir Econ
DOI 10.1007/s00181-012-0624-y
1 23
Your article is protected by copyright and
all rights are held exclusively by SpringerVerlag. This e-offprint is for personal use only
and shall not be self-archived in electronic
repositories. If you wish to self-archive your
work, please use the accepted author’s
version for posting to your own website or
your institution’s repository. You may further
deposit the accepted author’s version on a
funder’s repository at a funder’s request,
provided it is not made publicly available until
12 months after publication.
1 23
Author's personal copy
Empir Econ
DOI 10.1007/s00181-012-0624-y
Housing prices and crime perception
Paolo Buonanno · Daniel Montolio ·
Josep Maria Raya-Vílchez
Received: 14 December 2010 / Accepted: 21 June 2012
© Springer-Verlag 2012
Abstract In this article, we combine data from the housing market with data from
a victimization survey to estimate the effect of crime perception on housing prices in
the City of Barcelona from 2004 to 2006. Using dwelling data and a hedonic price
model (using both OLS and quantile regressions), in the first stage, we estimate the
shadow price of the location of dwellings. In the second stage, we analyse the impact
of crime perception, after controlling for other district characteristics such as local
public spending and immigration, on this locational valuation. After accounting for
the possible endogeneity of crime and housing prices, our findings suggest that crime
exerts relevant costs beyond its direct costs. Indeed, a one standard deviation increase
in perceived security is associated with a 0.57 % increase in the valuation of districts.
Moreover, in districts perceived as being less safe than the average for the City of
Barcelona, houses are highly discounted. Less safe districts have on average a valuation that is 1.27 % lower.
Keywords Housing prices · Crime perception · Security perception ·
Hedonic prices
P. Buonanno
Dipartimento di Scienze Economiche, University of Bergamo,
Via dei Caniana 2, 24127 Bergamo, Italy
D. Montolio (B)
Facultat d’Economia i Empresa, University of Barcelona and Barcelona Institute of Economics,
Av. Diagonal 690 Torre 4 Planta 2, 08034 Barcelona, Spain
e-mail: [email protected]
J. M. Raya-Vílchez
Departament d’Economia i Empresa, Escola Universitària del Maresme,
University Pompeu Fabra, c/ Trias Fargas 25, 08005 Barcelona, Spain
123
Author's personal copy
P. Buonanno et al.
JEL Classification
K42 · R21 · R31
1 Introduction
Crime and the fear it generates are among the most important determinants of individual welfare and of expected returns on many economic activities. In particular,
robbery, theft, breaking and entering and the fear of these crimes inflict many direct
and indirect costs on city residents, including the monetary value of property stolen
or damaged; insecurity, anxiety and lack of safety; and an impact on property values.
Several studies have sought to quantify the welfare and social costs associated with
crime (see Soares 2010 for a critical review). Anderson (1999) calculates the social
costs of crime for the US at $1 trillion, while according to UK Home Office estimates,
the consequences of crime against individuals and households account for £25 billion
of the £60 billion total cost of crime (Brand and Price 2000). Recently, Detotto and
Vannini (2010) have quantified the cost of crime in Italy at e38 billion.
A high crime rate is strongly and negatively associated with neighbourhood quality,
having a marked impact on the prices homebuyers are willing to pay for a house. In
other words, as crime is perceived as detrimental, individuals may be discouraged
from buying a house and this behaviour is, in turn, reflected in the market property
price. Moreover, as Gibbons (2004) notes, the fear of crime through its indirect effect
on housing prices may also “inhibit local regeneration and catalyse a downward spiral
in neighbourhood status”.
In this article, we combine district level data from the housing market and a victimization survey in the City of Barcelona from 2004 to 2006 to estimate the effect
of crime perception on housing prices. In the first part of the article, we propose a
two-step estimate using both OLS and quantile regressions to quantify the impact of
crime perception on housing values. Using dwelling data and a hedonic price model,
in the first stage, we estimate the shadow price of the location of dwellings. In the
second stage, we analyse the impact of crime perception, after controlling for other
area characteristics such as local public spending and immigration, on this locational
valuation.
As the relationship between housing prices and crime is likely to be endogenous
(see discussion below) in the second part, we test the robustness of our previous results
by providing instrumental variable (IV) estimates.
The literature has made extensive use of hedonic price models to quantify the value
of spatial differences, for example, in education (Black 1999), in transport facilities,
amenities and air pollution (Chay and Greenstone 2005) on housing prices. In this
framework, the seminal study by Griliches (1971) popularised models of this type,
while Rosen (1974) gave them a theoretical framework and established how heterogeneous products are a compound of different characteristics.1 The marginal implicit
1 Hedonic models are based on the modern theory of consumer choice, that is, consumers extract utility
not from the product itself but from its characteristics (see Lancaster 1966). Goodman (1998) and Colwell
and Dilmore (1999) point out that it was Court (1939) who first proposed the hedonic price methodology.
Griliches (1971) popularised this methodology; nevertheless, from 1941 to 1971 other papers also used the
hedonic methodology (see Tinbergen 1951).
123
Author's personal copy
Housing prices and crime perception
price of such characteristics can be estimated by means of a model that explains the
price of a product given these characteristics. This hedonic price model has been widely
applied, and the housing market is no exception.2 The earliest attempt to isolate the
effect of crime on housing value, made by Thaler (1978), found that a one standard
deviation increase in the crime rate caused a reduction in the price of a house of 3 %.
Qualitatively similar results have been obtained by Hellman and Naroff (1979) and
Rizzo (1979). More recently, Lynch and Rasmussen (2001) drew the conclusion that
homes are highly discounted in high crime areas in Jacksonville (Florida), while in
Atlanta an additional crime per acre per year in a given census tract has the effect
of reducing house prices by around 3 % (Bowes and Ihlanfeldt 2001). As Gibbons
(2004) notes, most of the previously cited contributions do not deal with the potential
endogeneity of crime rates in a property value model. Gibbons (2004), however, pays
careful attention to identification issues and proposes two alternative instruments for
crime: one based on spatial lags of the crime density and the other based on the distance to the nearest alcohol licensed premises. His findings show that crimes such as
vandalism, graffiti and arson have a significant negative impact on housing prices,
while burglaries have no measurable effect on housing prices.
As Gibbons (2004) stresses, estimating the effect of crime rates on housing prices is
empirically challenging. First, it is likely that the estimated housing price-crime gradient is significantly biased because of omitted variables. In our analysis, we control for
area level characteristics, including the immigration rate and local public spending.
Second, in many studies identification relies on the inclusion of a wide set of control
variables at the household and area levels. Even after controlling for other determinants
of housing prices, the distribution of the crime rate across areas could be correlated
with the error term, that is, the set of controls could neglect some time-varying, possibly unobserved factors that are also correlated with housing prices. In this study,
we seek to tackle this issue. First, we exploit the structure of our data (pooled crosssection for the period 2004–2006) by estimating models that allow for district and year
fixed effects. Second, we employ a set of IVs that helps us identify the causal effect
of crime perception on housing values. More specifically, we use data at a district
level of the victimization rate 20 years ago and the share of youth aged between 15
and 24. Moreover, instead of using official crime statistics, which are usually plagued
by underreporting problems, we rely on victimization data. As noted in MacDonald
(2002) and Gibbons (2004), official crime statistics tend to significantly understate
the true incidence of crime and more troubling report rates vary over time and space.
Thus, in line with Gibbons (2004), we rely on victimization data.
Our findings suggest that crime exerts relevant costs beyond its direct costs. Indeed,
a one standard deviation increase in perceived security is associated with a 0.57 %
increase in the valuation of a district. Moreover, in districts perceived as being less
safe than the average for the City of Barcelona, houses are highly discounted. In these
less safe districts, houses are valued on average at 1.27 % lower.
2 For instance, Palmquist (1984), Mendelsohn (1984), Ihlanfeldt and Martínez-Vázquez (1986), Bartik
(1987), Mills and Simenauer (1996), to cite a few. Examples of the estimation of hedonic price models in
the Spanish housing sector include, among others, Peña and Ruiz-Castillo (1984) and Garcia et al. (2010).
123
Author's personal copy
P. Buonanno et al.
The article is structured as follows. The next section presents the empirical and
methodological framework. Section 3 discusses our data sources. Section 4 presents
the results and discusses their interpretation. Finally, Sect. 5 concludes.
2 Methodology
The estimation of the effect of crime perception on housing prices for the City of
Barcelona is performed using a two-stage approach (see Garcia et al. 2010). First,
by means of a hedonic price model, we estimate the hedonic price of the location of
dwellings. Second, we test whether crime perception affects the estimated locational
valuation. This two-stage approach is applied sequentially, first to OLS regressions
and then to quantile regressions where the same set of conditional percentiles (10, 25,
50, 75 and 90 %) is used in the two stages.
Hedonic first stage
In the hedonic first stage, we estimate a regression model that captures the effect of
the housing characteristics and district location on dwelling price:
Pricei =
pH,k Hi,k +
k
p j Di, j + u i ,
(1)
j
where Pricei is the natural logarithm of the price per square metre of a dwelling i,
Hi,k represents the vector of physical characteristics k of the dwellings, Di, j is the
fixed effect of district j in which dwelling i is located, pH,k and p j are, respectively,
the hedonic prices of physical characteristics k and the hedonic price of the location
j (district in our case), while u i is an error term.
In our analysis, we employ a pooled cross-section of dwellings for the period from
2004 to 2006, thus assuming that the effect of the location characteristics can vary
over time. We can rewrite Eq. (1) as:
Pricei,t =
k
pH,k Hi,k,t +
p j,t Di, j,t + u i,t ,
(2)
j
where subscript t indicates the period of time. Now Di, j,t is the fixed effect of district
j in which dwelling i is located in each year (t). By estimating Eq. (2), we obtain an
estimate of pH,k which indicates the effects of the physical characteristics of dwellings
on their price. Moreover, we obtain the estimate of the locational effects or, in other
words, the district hedonic prices, p j,t , which are time-variant. We estimate Eq. (2)
by means of OLS and a quantile regression approach to obtain robust (to outliers)
estimates of the main characteristics of interest.3
3 We employ the quantile regression approach because the hedonic prices of the dwellings’ characteristics
may vary depending on the price of the dwelling itself. The quantile regression approach allows us to
estimate the impact of the explanatory variables along the distribution of housing prices.
123
Author's personal copy
Housing prices and crime perception
Hedonic second stage
The crucial aspect of our analysis is to estimate the determinants of the district hedonic
price ( p j,t ). Thus, we propose a hedonic second stage estimation procedure as follows:
p j,t = γ CRIME j,t + β X j,t + τt + η j,t ,
(3)
where the district hedonic price ( p j,t ), estimated for each quantile in the hedonic
first stage, depends on crime perception (CRIME j,t, ), other time-variant characteristics (X j,t ), year fixed effects (τt ) and an error term (η j,t ). Introducing time fixed
effects allows us to take into account the annual and common increase in prices due
to the overall situation of the housing market in the City of Barcelona over the period
2004–2006. The same two-stage procedure is applied to the OLS regressions.
We can interpret the estimated parameter for CRIME, γ , as the effect of crime
perception on dwelling hedonic price having first controlled for the physical characteristics of the dwellings, other time-variant characteristics of the district (local public
spending and immigration rate) and time fixed effects.
3 Data
The database consists of information collected for 1,653 dwellings sold in Barcelona
by a real estate agency over the period 2004–2006 (617 in 2004, 693 in 2005 and 343
in the first half of 2006).4 For each dwelling, we have detailed information on physical
characteristics, while its postal code allows us to accurately locate the district in which
it stands. Therefore, we organise the entire dataset at a district level. Table 1 presents
the location of the dwellings contained in our dataset in the ten districts of the City of
Barcelona. Note from Table 1 that, in this sample, the distribution of dwellings across
districts only falls below 5 % in two cases (Les Corts and Gràcia) while the rest of the
districts account, on average, for around 10 % of the observations.
For each dwelling, we have information on its price per square metre, which represents our dependent variable in the hedonic first stage estimation, floor area in square
metres, number of bedrooms (not including kitchen, living room and bathroom), availability of a lift, type of kitchen (independent or otherwise), the floor of the building on
which it is located and the age of the dwelling. Moreover, we have a dummy variable
indicating the year the dwelling was sold. Table 2 presents the descriptive statistics
of the variables used. According to these, an average dwelling in our sample has a
kitchen, a lift (in 52.15 % of the cases), three bedrooms, is located mainly between the
first and fourth floors of the building, is 51.4-year old, and has an average floor area
of around 64 m2 .
We obtained our crime perception variables from the Barcelona City Council victimization survey, which contains data at a district level. The survey, conducted every
4 The real estate agency, which remains unnamed for reasons of privacy, is one of the largest in the Spanish
housing sector.
123
Author's personal copy
P. Buonanno et al.
Table 1 Number of dwellings by district
Districts
Freq.
%
Cum.
District I (Ciutat Vella)
198
11.98
11.98
District II (Eixample)
240
14.52
26.50
District III (Sants-Montjuïc)
360
21.78
48.28
19
1.15
49.43
54.51
District IV (Les Corts)
District V (Sarrià-Sant Gervasi)
84
5.08
District VI (Gràcia)
72
4.63
58.86
District VII (Horta-Guinardó)
100
6.05
64.91
District VIII (Nou Barris)
143
8.65
73.56
District IX (Sant Andreu)
226
13.67
87.24
District X (Sant Martí)
Total
211
1, 653
12.76
100
100
year by Barcelona City Council, is representative at the district level and on average
includes 5,000 individuals aged 16 or over.5
Districts are an aggregation of neighbourhoods and, as such, we might lose some
heterogeneity in the crime data (which might, in principle, be of a neighbourhood
or even a street level nature). Unfortunately, there are no crime data for the City of
Barcelona (be it survey or officially recorded data) at the neighbourhood level. Nevertheless, we should stress that the City of Barcelona is organised at a district level and
that this is a relevant level for decision taking. For instance, unlike neighbourhoods,
districts have a political representative (Regidor de districte) and a Council (Consell
Municipal de districte) with powers to report and propose plans, programmes, budgets
and urban planning that affect the district as well as powers to distribute the budget
assigned to them. Moreover, some preliminary results (not reported here but available
upon request) show that the main heterogeneity (as regards the level of education of
the population and income levels) is between districts and not inside districts.6
5 The survey, as regards both the sample and the analysis, is predominantly urban. The sample is, therefore,
stratified by the population in each district and is defined broadly (as the victim status is not universal).
Specifically, the sample comprised more than 5,000 telephone interviews each year. This allows a maximum
error of 1.15 %, and with respect to the district sub-samples, 2.1 % (the margin of error was calculated for
a significance level = 0.05 and p = q = 0.5). The fieldwork was carried out by a specialized company
by means of random telephone interviews. The fieldwork begins the first working day after the Christmas
holidays and is conducted over a 10-week period.
6 We test the statistical difference of the education variable at the district level with respect to the same
variable at the city level. We find statistically significant differences (at 1 % level) for districts 1, 2, 4, 5, 6
and 8; that is, 60 % of the districts are heterogeneous with respect to the city. We also tested the statistical
difference of the education variable at the neighbourhood level with respect to the same variable at the
district level. In this case we find that neighbourhoods 1, 2, 3, 5, 23, 24, 26, 32, 36, 37 have different educational values than those of the district they belong to. In other words, only 31.25 % of the neighbourhoods
(from a total of 32 neighbourhoods for which data are available) are heterogeneous with respect to their
district. In terms of the income level variable, districts 1, 3, 4 and 8 (40 %) are statistically different from
the city while only neighbourhoods 1, 2, 4, 5, 10, 15, 28 and 32 (25 %) are different with respect to their
district.
123
Author's personal copy
Housing prices and crime perception
Table 2 Descriptive statistics
Variable
Freq.
%
Dwelling variables
Kitchen
0
176
10.65
1
1, 477
89.35
Lift
0
791
47.85
1
862
52.15
Number of bedrooms
1
109
6.59
2
447
27.04
3
840
50.82
4
226
13.67
5
24
1.45
6
5
0.30
7
2
0.12
15
0.91
Floor num
−1 or less
0
138
8.35
1
336
20.33
2
253
15.31
3
244
14.76
4
263
15.91
5
177
10.71
6
85
5.14
7
70
4.23
8
33
2.00
9
18
1.09
10
7
0.42
11 or more
Variable
14
0.84
Mean
Std. dev.
Min.
Max.
Obs.
Floor
64.16
20.64
18
187
N = 1, 653
Age
51.43
31.38
1
266
N = 1, 653
Price
3.63
0.95
0.86
10
N = 1, 653
6.0
District variables
Security (0,10)
0.51
4.7
6.7
N = 50
Between
0.50
4.9
6.5
n = 10
Within
0.16
5.6
6.5
t=5
0.09
0.76
1.15
N = 50
Overall
Crime perception
Overall
0.88
123
Author's personal copy
P. Buonanno et al.
Table 2 continued
Variable
Mean
Std. dev.
Min.
Max.
Obs.
Between
0.09
0.78
1.09
n = 10
Within
0.03
0.79
0.93
t=5
District population
Overall
159.1
Between
Within
50.9
81.6
265.5
N = 50
53.1
82.6
262.5
n = 10
1.4
155.7
162.1
t=5
21.93
52.44
153.0
N = 50
21.88
60.85
138.3
n = 10
6.42
64.27
94.72
t=5
Local public spending
Overall
80.04
Between
Within
Immigration rate
0.091
0.070
0.43
N = 50
Between
0.093
0.092
0.39
n = 10
Within
0.019
0.090
0.17
t=5
0.077
0.041
0.348
N = 50
Between
0.079
0.047
0.324
n = 10
Within
0.011
0.076
0.134
t=5
Overall
0.139
Non-OECD immig. rate
Overall
0.107
Note Individual dwelling data include kitchen, floor number, lift, number of bedrooms, floor area, age and
price. Data at a district level include security, crime perception, district population (expressed in 1,000s
inhabitants), local public spending (expressed in euros per capita), immigration rate and non-OECD immigration rate
Victimization surveys attempt to bypass the underreporting problems that typically
affect the official crime statistics made available by the police. As such, these surveys
provide crime statistics that are closer to reality. Moreover, victimization surveys have
advantages in that they reveal a range of crimes that are not so well reported to, or
recorded by, the police. They also contain additional information on the nature of these
crimes and include the respondents’ views.
From the victimization survey, we use two variables as proxies for crime perception
at the district level. The first variable (security) measures the level of security in the
district. Survey respondents were asked to rate the level of security perceived in their
district on a scale from 0 (lowest, very unsafe) to 10 (highest, very safe).
The second variable is crime perception and is constructed as the ratio between the
average perceived security in the city as a whole7 and the level of security perceived in
the district. Hence, values greater (lower) than 1 indicate that the district is perceived
as being less (more) safe than the rest of the city.
The difference between the level of security in the district and that in the rest of
the city is caused by the different experiences that might have occurred in the two
7 Average city security is measured in the same way as the level of security perceived by respondents in
the rest of the city from 0 (lowest, very unsafe) to 10 (highest, very safe).
123
Author's personal copy
Housing prices and crime perception
territories. These experiences are determined by appreciations of space and people.
The district is known and neighbours recognize people and places, and thus collective appropriation is possible. However, the rest of the city is perceived as unknown
territory, where the appropriation of space and, especially, of human relations can
encounter greater difficulties.
In principle, we would relate high levels of criminality at a district level with low
values of security and values of crime perception higher than 1.
Our dataset includes other control variables at district level such as local public
spending and the district immigration rate. With respect to local public spending and
as Oates (1969) points out “if, as the Tiebout model suggests, individuals consider the
quality of local public services in making locational decisions, we would expect to find
that, other things being equal (including tax rates) across communities, an increased
expenditure (per capita) should result in higher property values”. As shown in Garcia
et al. (2010), this hypothesis holds in the case of the City of Barcelona and local public
spending has a direct impact on housing values. Therefore, we perform the hedonic
second stage regression with data on local public expenditure for the City of Barcelona,
obtained from the Gaseta Municipal, published by the City Council. This publication
contains information on the composition of the local budget. The time span covers the
relevant period and allows us to introduce lags of this variable into the regression to
test the effect of previous spending on current housing prices.8 We use an aggregate
measure of Local public spending which includes spending on personnel, purchases of
goods and services, current transfers and real investment that includes street maintenance and cleaning, waste management, building improvements, the creation of sports
and entertainment areas and parks and garden conservation. District public spending
is not directly devoted to crime prevention. Crime prevention spending is made at the
City Council level and is devoted mainly to local police (Guardia Urbana) and to
governmental areas for which we do not have a direct assignment to districts.9 On
average, this aggregate measure accounts for 86 % of the total budget dedicated to
districts, which in turn represents, on average, 12 % of the total budget of Barcelona
City Council. We use district population to obtain this variable in per capita terms.
A further variable that could influence the locational valuation of a given dwelling
is the rate of immigration present in a district. During the period under analysis there
was a major increase in the immigration rate in Spain. This was especially true in
Barcelona and the Catalan region. Data from the municipal register in Spain for 2005
show that 69 % of registered foreigners resided in four regions (22 % in Catalonia,
21 % in Madrid, 15.5 % in Valencia and 11.5 % in Andalusia) and an additional 10 %
in the two island regions (6 % in the Canary Islands and 4 % in the Balearic Islands).10
The immigration rate is constructed as the total number of immigrants per district with respect to the total district population. We also construct the non-OECD
8 We consider public spending to require a certain amount of time to become “effective” or perceived by
citizens.
9 For robustness checks, we also introduced each type of local public expenditure individually in the
regressions and obtained basically the same conclusions as for the aggregate measure.
10 The same data show how, within Catalonia, the City of Barcelona (and its surrounding area) has the
largest share of recent immigrants to the Catalan region.
123
Author's personal copy
P. Buonanno et al.
Table 3 Hedonic first stage results for districts
Variable
OLS
Area in square metres −0.0065
Q10
Q25
Q50
Q75
Q90
−0.0080
−0.0072
−0.0068
−0.0064
−0.0051
(−21.4)*** (−13.3)*** (−14.7)*** (−20.9)*** (−18.9)*** (−7.65)**
Age
−0.0007
−0.0006
(−3.57)*** (−1.05)
Lift
Kitchen
Floor
−0.0001
−0.0005
−0.0008
−0.0011
(−0.70)
(−1.90)*
(−3.48)**
(−4.4)***
0.1253
0.1273
0.1210
0.1208
0.1231
0.1080
(11.2)***
(5.81)***
(9.07)***
(11.1)***
(11.3)***
(6.10)***
−0.0757
−0.0451
−0.0375
−0.0849
−0.1147
−0.1063
(−4.98)*** (−1.78)*
(−1.87)*
(−3.83)*** (−9.63)*** (−4.1)***
0.0152
0.0181
0.0160
0.0112
0.0095
0.0150
(7.35)***
(5.03)***
(7.63)***
(5.39)***
(4.76)***
(4.97)***
0.0153
0.0267
0.0156
0.0138
0.0128
−0.0114
(2.17)*
(1.52)
(1.47)
(2.19)**
(1.59)
(−0.78)
R2
0.518
0.3108
0.3585
0.3803
0.3922
0.8707
N
1,653
1,653
1,653
1,653
1,653
1,653
Number of bedrooms
Note Dependent variable: price per square metre (log). t values in parentheses. *, ** and *** indicate
significant at 10, 5 and 1 % level, respectively. The full set of dummies for district and year (29) are not
reported but available upon request. Regressions are performed including the full set of interactions terms
between each district and each year. Huber-White standard errors used for OLS estimates
immigration rate to distinguish those districts that receive more immigration from
countries not belonging to the OECD.
4 Results
4.1 Baseline estimates
As discussed above, we employ a two-stage procedure in order to assess the impact of
crime perception on housing values. In the hedonic first stage, we estimate the hedonic
housing price on the basis of a dwelling’s physical characteristics, the year and district
fixed effects, while in the hedonic second stage, we use the hedonic housing price at
the district level as our dependent variable.
Our hedonic first stage results are presented in Table 3. The first column shows the
OLS estimates while the quantile regressions are presented in columns 2–6. Both OLS
and quantile estimates are consistent with the main findings in the literature on hedonic
housing prices. Specifically, an additional bedroom increases prices by 1.53 %, while
the availability of a lift and a higher floor location is associated with an increase in
value of 12.53 and 1.52 %, respectively. Older dwellings are marginally cheaper than
new ones. Finally, an open kitchen as opposed to an independent kitchen reduces
housing prices by 7.57 %.
For the quantile regressions, we perform a “normal location test” (Buchinsky
1997) to determine whether the estimated parameters for each quantile differ from
one another statistically. We reject the null hypothesis that the coefficients of the
123
Author's personal copy
Housing prices and crime perception
Table 4 Normal location test
for the quantile first stage
estimation
Variable
F-statistic
p value
Area in square metres
3.04
0.02
Age
4.13
0.00
Lift
0.20
0.94
Kitchen
4.01
0.00
Floor
2.08
0.08
Number of bedrooms
1.44
0.22
Table 5 Hedonic second stage OLS estimations
Model 1
Security
Crime perception
Local public spending
Immigration rate
Model 2
Model 3
Model 4
–
–
0.1525
0.2124
(2.68)**
(3.16)***
–
–
−1.3759
(−2.44)**
(−3.02)***
0.0010
0.0005
0.0005
0.0009
(0.41)
(0.39)
(0.65)
(0.76)
1.0737
–
1.1685
–
(3.31)***
Non-OECD immigration rate
−0.8845
–
(3.13)***
1.6045
–
(3.55)***
1.1911
(3.44)***
N
29
29
29
29
Time dummies
Yes
Yes
Yes
Yes
R2
0.6387
0.6557
0.6236
0.6458
Note t values in parentheses. *, ** and *** indicate significant at 10, 5 and 1 % level, respectively. The
dependent variable is district hedonic prices ( p j,t ) obtained from the OLS estimation in the first stage (see
column 1, Table 3). Huber-White standard errors used
different quantiles are equal for surface area, age, kitchen and floor (see Table 4).
Note that the results from the quantile regression indicate that at higher percentiles of
the price per square metre distribution, the impact of floor area is greater (less negative) while the impact of age and kitchen is lower (more negative). Specifically, an
additional square metre of floor area reduced dwelling prices from 0.80 % (percentile
10) to 0.51 % (percentile 90), while one additional year in the age of the dwelling is
associated with a decrease in value from 0.05 % (percentile 50) to 0.11 % (percentile
90). An open kitchen as opposed to an independent kitchen reduces housing prices
by between 3.75 % (percentile 25) and 11.47 % (percentile 75). Finally, the impact of
floor presents no clear pattern.
Having obtained the hedonic housing price at district level, we next estimate the
impact of our crime perception variables ( security and crime perception) on district
hedonic prices ( p j,t ). Table 5 shows the hedonic second stage OLS estimates, while
the hedonic second stage quantile regressions are presented in Tables 6 and 7. As mentioned in the previous section, our estimating equation includes the following controls
at district level: total immigration rate (non-OECD immigration rate), local public
spending per capita and year fixed effects.
123
Author's personal copy
P. Buonanno et al.
Table 6 Hedonic second stage quantile estimations for security
Q10
Security
Local public spending
Non-OECD immigration rate
Q25
Q50
Q75
Q90
0.1606
0.2017
0.2047
0.1707
0.2717
(2.05)**
(2.79)***
(3.29)***
(2.50)**
(3.08)***
0.0007
−0.0003
0.00001
−0.0002
0.0020
(0.43)
(−0.22)
(0.02)
(−0.16)
(1.11)
1.2315
1.3794
1.5385
1.4810
2.2369
(2.33)**
(2.83)***
(3.67)***
(3.21)***
(3.76)***
N
29
29
29
29
29
Time dummies
Yes
Yes
Yes
Yes
Yes
Pseudo-R 2
0.5697
0.5667
0.7041
0.6359
0.6490
Note t values in parentheses. *, ** and *** indicate significant at 10, 5 and 1 % level, respectively. The
dependent variable is district hedonic prices ( p j,t ) obtained from the quantile estimation in the first stage
(see columns 2–6 in Table 3). Results do not change significantly when replacing non-OECD immigration
rate with immigration rate
Table 7 Hedonic second stage quantile estimations for crime perception
Q10
Crime perception
Local public spending
Non-OECD immigration rate
Q25
Q50
Q75
Q90
−1.1002
−1.329
−1.2877
−1.1900
−1.6479
(−2.10)**
(−2.74)**
(−3.02)***
(−2.63)**
(−2.64)**
−0.0016
0.0006
0.0001
−0.0009
0.0009
(−0.70)
(2.29)
(0.06)
(−0.47)
(0.32)
1.8796
1.5278
1.7346
2.0464
2.1304
(2.10)**
(1.84)*
(2.37)**
(2.64)**
(1.99)*
N
29
29
29
29
29
Time dummies
Yes
Yes
Yes
Yes
Yes
Pseudo-R 2
0.5737
0.5674
0.6894
0.6446
0.6079
Note t values in parentheses. *, ** and *** indicate significant at 10, 5 and 1 % level, respectively. The
dependent variable is district hedonic prices ( p j,t ) obtained from the quantile estimation in the first stage
(see columns 2–6 in Table 3). Results do not change significantly when replacing non-OECD immigration
rate with immigration rate
In the case of the quantile regressions, we use the design matrix bootstrap method
to obtain estimates of the standard errors for the coefficients in quantile regression
(Buchinsky 1995).11
Our findings show that crime variables exert a sizeable effect on housing prices.
Security, which measures the perceived level of security in the district, has a positive
impact on the district hedonic price: a one standard deviation increase in security leads
to an average increase in the locational valuation of dwellings of between 0.55 and
11 Based on a Monte Carlo study, Buchinsky (1995) recommends the use of this method as it performs
well for relatively small samples (such as ours) and it is robust to changes in the bootstrap sample size
relative to the data sample size. More importantly, the design matrix bootstrap method is valid under many
forms of heterogeneity. This method of bootstrap performs well even when the errors are homoscedastic.
123
Author's personal copy
Housing prices and crime perception
0.76 %.12 Crime perception, which is the ratio between the average perceived security
in the city and the level of security perceived in the district, is negatively correlated
with the district hedonic price. A one standard deviation increase in crime perception, i.e. the district being considered less safe than the average, leads to a consistent
reduction in district valuation of between 7.38 and 11.6 %.
In both cases, the impact of security and crime perception on district valuation
is stronger in the higher quantiles of the district hedonic prices distribution, that is,
insecurity causes a larger price reduction in the more highly valued areas of Barcelona.
As for the control variables, we find the expected sign for local public spending
(positive), as in Garcia et al. (2010), although in our case it is not statistically significant for any of the estimations. The results for the immigration variables (immigration
rate and non-OECD immigration rate) are in line with those obtained by González and
Ortega (2009), who estimate that the influx of immigrants in Spain between 1998 and
2008 increased house prices by about 52 %.13 Our result seems to support the argument
that new immigrants increase district valuation, through a demand effect for housing
in the City of Barcelona (either directly as homeowners or indirectly as renters).
A growing body of literature (Bianchi et al. 2012; Bell et al. 2010; Nunziata 2011)
has explored the effects of immigration on crime rates, although empirical research to
date has failed to reach a consensus on this relationship. Our findings suggest that the
demand effect of immigration on housing prices offsets the possible negative effect of
immigration acting through the crime channel.
Finally, note that the impact of the non-OECD immigration rate on district valuation is stronger in the higher quantiles of the district hedonic prices distribution, that
is, the non-OECD immigration rate causes a larger price increase in the more highly
valued areas of the City of Barcelona.
4.2 Robustness checks: IV estimates
Even after controlling for other determinants of housing prices and for time fixed
effects, crime perception could still be correlated with the error term for several reasons. First, we can expect cheaper housing to attract individuals with a higher propensity for crime (i.e. lower income individuals); on the other hand, more expensive
houses may attract criminals by offering greater expected payoffs from delinquent
behaviour. Moreover, as Gibbons (2004) discusses, we can expect certain specific
housing characteristics to affect both prices and crime rates. For instance, large windows or secluded gardens make a residential area attractive to both burglars and house
buyers, while poorly maintained property will tend to attract vandals and a low market
price. For all these reasons, it is extremely difficult to identify the causal effect of crime
on housing prices. Thus, we need to rely on IVs that are highly correlated with the
crime rate and uncorrelated with the error term. In order to address the endogeneity
12 Note that a one standard deviation increase in security is rather small (0.07 on an index that ranges from
1 to 10). Alternatively, the results can be quantified as follows: a one-point increase in the security index
leads to an average increase in the locational valuation of dwellings of between 15.2 and 21.2 %.
13 They perform their analysis at the province level and report that the average Spanish province received
an influx of immigrants equal to 17 % of the initial working-age population.
123
Author's personal copy
P. Buonanno et al.
issue we propose two IVs at district level: (i) the victimization index 20 years ago
(1983–1987) and (ii) the percentage of youth aged between 15 and 24.
The former, the victimization index 20 years ago, resembles the historical IVs
widely used in the economic literature (Tabellini 2008; Buonanno et al. 2009). Should
crime variables be affected by time dependency, this would cast certain doubts on the
validity of our instruments. In other words, it may be the case that the crime rate in
the eighties was higher in the same neighbourhoods than today, implying lower house
values. It is worth noting that in 1992 Barcelona hosted the Olympic Games and the
organisation of this event brought about major urban changes. As a consequence, many
neighbourhoods were renewed and their social composition changed significantly.14
So, even if we are unable to provide compelling evidence regarding the validity of the
exclusion restriction, we can conclude that the “historic” victimization index in the
neighbourhoods is unlikely to have been affected by time dependency and, as such, is
a good potential candidate to serve as an instrument.
Our second instrument, the percentage of youth aged between 15 and 24, is a well
recognized factor in many types of crime (Levitt and Lochner 2001; Buonanno and
Montolio 2009). Even if a priori, we cannot exclude the possibility that the share of
youth may affect housing values through channels other than that of crime (i.e. demand
effect), this appears very unlikely in a city like Barcelona where young people tend to
live with their parents for a longer time than in other EU countries.15
Once equipped with these IVs, in Table 8, we proceed to examine the causal relationship between housing prices and crime depicted by the OLS estimates. The IV
first stage regression confirms the robustness of our instruments. Both the past victimization index and the share of youth are strongly significant and present the expected
sign. The IV estimates show the relevance of the instruments. The F-statistic of the
regressions is equal to 17.47 and 16.14, which is well above the lower bounds indicated
by the literature for weak instruments (Bound and Holzer 2000; Stock and Yogo 2002).
Moreover, the validity of our instruments is confirmed by the Hansen’s J-test for overidentifying restrictions, which fails to reject the null hypothesis that the instruments
are valid in all cases.
Overall, the IV estimates are quantitatively consistent with the OLS results
confirming all our previous findings. However, the crime perception coefficient
(column 2) is smaller in magnitude with respect to OLS estimates, while it remains
unchanged for security (column 1). In the case of crime perception, the IV results indicate that if a district becomes less safe than the average (by one standard deviation),
this causes a reduction in the district valuation of 1.27 %.16
14 As Essex and Chalkley (1998) claim: “The Barcelona Games of 1992 is probably the best example of
the role of the Olympics as a catalyst for urban change and renewal.”
15 As documented, among others, by Becker et al. (2010), youth emancipation in Europe has declined
in recent years. Co-residence rates for men aged 25–29years in Spain were above 60 % with a sustained
upward trend in the last two decades, which indicates we can expect even higher co-residence rates for
youngsters aged between 15 and 24.
16 That is, a one-point increase in the crime perception index leads to an average decrease in the locational
valuation of dwellings of 14.21 %.
123
Author's personal copy
Housing prices and crime perception
Table 8 IV estimations
(1)
(2)
IV first stage
Past victimization index
−0.0439
(-4.97)***
−28.3046
22.2019
(−4.06)***
(2.97)***
F-stat
17.47
16.14
Sargan
0.258
0.824
0.1584
–
Youth 15–24
Note t values in parentheses,
except for F-stat and Sargan test
for which statistics are reported.
*, ** and *** indicate
significant at 10, 5 and 1 % level,
respectively. The dependent
variable is district hedonic prices
( p j,t ) obtained from the OLS
estimation in the first stage (see
column 1, Table 3). Huber-White
standard errors used
0.0367
(4.48)***
IV second stage
Security
(2.05)**
Crime perception
–
−0.1421
29
29
(−1.88)*
N
Other controls
Yes
Yes
Time dummies
Yes
Yes
5 Conclusions
In this article, we have combined dwelling data from the housing market with crime
victimization data at the district level for the City of Barcelona for the period 2004–
2006. Employing a two-stage hedonic price model, we have assessed the effect of
crime perception on housing prices. In our analysis, we draw on victimization data
as opposed to the official statistics usually employed in studies of this type. This has
enabled us to avoid any potential underreporting and to consider a “true” crime rate.
Moreover, we have carefully dealt with any possible problems of endogeneity of crime
perception in the valuation of districts.
The results presented here consistently indicate that crime perception negatively
affects housing prices (based on the hedonic price of the location of dwellings), while
district security is associated with a higher housing value via its impact on locational
valuation. These results are robust to the inclusion of district and year fixed effects
and district level characteristics, such as the district immigration rate (total and nonOECD) and local public spending. Results for the immigration variables (which were
significant in our estimates) are in line with those obtained by González and Ortega
(2009) and seem to point to a “call effect” of immigrants who push up housing prices.
Our estimates suggest that crime exerts relevant effects beyond its direct costs. In
particular, a one standard deviation increase in the perceived security of the neighbourhood is associated with a 0.57 % increase in the valuation of the district. Our findings
shed light on the importance of individual perception on crime on individual and private welfare. As stressed in a recent contribution by Cook and MacDonald (2011)
policy makers should take into account the role of private individuals in determining the effectiveness of the criminal justice system and the quality and availability of
123
Author's personal copy
P. Buonanno et al.
criminal opportunities. Their analysis of the 30 Los Angeles BIDS (Business Improvement Districts) shows that BID expenditures have been very effective in reducing crime
and have generated a social benefit about 20 times higher than the private expenditures.
To this extent, private action aimed at improving the security and safety of districts
may have some indirect effect other than the direct reduction in crime rate. In particular, the social benefit could be even greater if we consider the effect that crime
reduction has on property values.
Our approach, however, presents a number of limitations. We treat hedonic price
estimates, as is usual practice in the literature, as if they were the true hedonic prices.
This assumption is acceptable when very large samples are used in the first stage
regression (as was the case here), since it can be assumed that estimation errors are
negligible compared with other sources of variability (for instance, the small sample
size available for the hedonic second stage estimation). Yet, when only small samples are available for the first stage estimation, the proposed approach requires further
refinements to provide suitable (asymptotically correct) standard errors for the second
stage estimators, especially in the case of IV and quantile regression estimates.
Acknowledgments We would like to thank the Editor and two anonymous referees for helpful and detailed
comments that have significantly improved the initial version of this article. Any remaining errors are our
sole responsibility. Daniel Montolio gratefully acknowledges financial support from Grant 2009SGR102
of the Catalan Regional Government (Generalitat de Catalunya).
References
Anderson DA (1999) The aggregate burden of crime. J Law Econ 42:611–642
Bartik TJ (1987) The estimation of demand parameters in hedonic price models. J Polit Econ 95:81–88
Becker SO, Bentolila S, Fernandes A, Ichino A (2010) Youth emancipation and perceived job insecurity of
parents and children. J Popul Econ 23:1175–1199
Bell B, Fasani F, Machin S (2010) Crime and immigration: evidence from large immigrant waves. CEP
Discussion Papers dp0984, Centre for Economic Performance, LSE
Bianchi M, Buonanno P, Pinotti P (2012) Do immigrants cause crime?. J Eur Econ Assoc (forthcoming)
Black SE (1999) Do better schools matter? Parental valuation of elementary education. Q J Econ 114:
577–599
Bound J, Holzer H (2000) Demand shifts, population adjustment, and labour market outcomes. J Lab Econ
1:20–54
Bowes DR, Ihlanfeldt KR (2001) Identifying the impacts of rail transit stations on residential property
values. J Urban Econ 50:1–25
Brand S, Price N (2000) The economic and social costs of crime. Home Office Research Study n◦ . 217
Buchinsky M (1995) Estimating the asymptotic covariance matrix for quantile regression models a Monte
Carlo study. J Econ 68:303–338
Buchinsky M (1997) Recent advances in quantile regression models: a practical guideline for empirical
research. J Hum Resour 33:88–126
Buonanno P, Montolio D (2009) Juvenile crime in Spain. Appl Econ Lett 16:495–500
Buonanno P, Montolio D, Vanin P (2009) Does social capital reduce crime?. J Law Econ 52:145–170
Chay KY, Greenstone M (2005) Does air quality matter? Evidence from the housing market. J Polit Econ
113:376–424
Colwell F, Dilmore G (1999) Who was first?. An examination of an early hedonic study. Land Econ 75:
620–626
Cook PJ, MacDonald J (2011) Public safety through private action: an economic assessment of BIDS. Econ
J 121:445–462
Court AT (1939) Hedonic price indexes with automotive examples in “the dynamics of automobile demand”.
General Motors, New York 98–119
123
Author's personal copy
Housing prices and crime perception
Detotto C, Vannini M (2010) Counting the cost of crime in Italy. Working Paper CRENoS, No. 2010-13
Essex S, Chalkley B (1998) Olympic games: catalyst of urban change. Leis Study 17:187–206
Garcia J, Montolio D, Raya JM (2010) Local public expenditures and housing prices. Urban Study 47:
1501–1510
Gibbons S (2004) The costs of urban property crime. Econ J 114:F441–F463
González L, Ortega F (2009) Immigration and housing booms: evidence from Spain. IZA Discuss Paper
4333
Goodman CA (1998) Andrew court and the invention of hedonic price analysis. J Urban Econ 44:291–298
Griliches Z (1971) Hedonic price indexes for automobiles: an econometric analysis of quality change.
NBER 73:173–196
Hellman DA, Naroff JL (1979) The impact of crime on urban residential property values. Urban Study
16:105–112
Ihlanfeldt KR, Martínez-Vázquez J (1986) Alternative value estimates of owner-occupied housing: evidence
on sample selection bias and systematic errors. J Urban Econ 20:356–369
Lancaster KJ (1966) A new approach to consumer theory. J Polit Econ 74:132–157
Levitt SD, Lochner L (2001) The determinants of juvenile crime. NBER Chapters in: Risky behavior among
youths: an economic analysis, pp 327–374
Lynch AK, Rasmussen DW (2001) Measuring the impact of crime on house prices. Appl Econ 33:1981–
1989
MacDonald Z (2002) Official crime statistics: their use and interpretation. Econ J 112:F85–F106
Mendelsohn R (1984) Estimating the structural equations of implicit markets and household production
functions. Rev Econ Stat 66:673–677
Mills ES, Simenauer R (1996) New hedonic estimates of regional constant quality house prices. J Urban
Econ 39:209–215
Nunziata L (2011) Crime perception and victimization in Europe: does immigration matter? CSEA Working
Paper 04/2011
Oates WE (1969) The effects of property taxes and local public spending on property vales: an empirical
study of tax capitalization and the tiebout hypothesis. J Polit Econ 77:957–971
Palmquist RB (1984) Estimating the demand for the characteristics of housing. Rev Econ Stat 66:394–404
Peña D, Ruiz-Castillo J (1984) Distributional aspects of public rental housing and rent control policies in
Spain. J Urban Econ 15:350–370
Rizzo MJ (1979) The effect of crime on residential rents and property values. Am Econ 23:16–21
Rosen S (1974) Hedonic prices and implicit markets: product differentiation in pure competition. J Polit
Econ 82:34–55
Soares R (2010) Welfare costs of crime and common violence: a critical review. PUC Working Paper Series
no. 581
Stock JH, Yogo M (2002) Testing for weak instruments in linear IV regression. NBER: 284
Tabellini G (2008) Presidential address institutions and culture. J Eur Econ Assoc 6:255–294
Thaler R (1978) A note on the value of crime control: evidence from the property market. J Urban Econ
5:137–145
Tinbergen J (1951) Some remarks on the distribution of labour incomes. Int Econ Pap 1:195–207
123