The spatial dimension of the French private rental markets: Evidence from microgeographic data in 2015

Abstract

This article draws on data collected by local rental observatories in 12 French urban units in 2015 to analyze the spatial dimension of hedonic rental prices in the private rental market through (i) the spatial heterogeneity between urban units and (ii) the wide variety of contextual and locational characteristics (socio-economic, environmental (dis)amenity, and accessibility) and flexible specifications to capture their potential non-linear influence on rent. Based on a joint test of equality of coefficients across all urban units, we find that hedonic prices differ for 75% of the characteristics, thereby justifying a detailed analysis of heterogeneity. Lyon, Nice, and Paris taken individually are the urban units with the most specific valuations of housing characteristics and socio-economic characteristics. Our analysis reveals that housing characteristics, median income, and distance to the center are clearly the variables with the most heterogeneous effects on hedonic prices.

Keywords

Private rental markets hedonic models spatial heterogeneity

Introduction

Although one in five homes in France is privately rented (Chodorge and Pavard, 2016), our understanding of how this market operates remains fragmentary and is often restricted to specific geographical areas (e.g. Prandi and Moumouni, 2014 for Paris) or the regular reports published by the Observatoire des Loyers de l’Agglomération Parisienne. The creation in 2013 of a network of local rental observatories (LROs) collecting standardized data from individuals and real-estate professionals has opened up new opportunities for further research. The network affords fresh insight into the functional diversity of France’s local rental markets and the implications of that diversity for the future of local housing policies.

Drawing on a large cross-sectional analysis of data collected in 2015 by 12 LROs covering a total of more than 148,000 homes (5.69% of the total private rental housing units in our study area), we examine the spatial aspect of rental prices based on two considerations.

The first consideration is spatial heterogeneity in the valuation of the intrinsic and contextual attributes of housing, which we ascertain by comparing the hedonic prices of the same attribute across our 12 urban units. For each attribute, we test for (i) the joint equality of coefficients across all 12 urban units and (ii) the equality of coefficients for all pairs of urban units taken two by two. Although our study does not indicate how households choose among urban units, it does provide better knowledge of the heterogeneity of housing preferences and can therefore help policymakers better adjust the supply of public goods and better manage externalities in their area.¹ Hedonic modeling reveals preferences, especially by estimating demand for goods that are not explicitly traded on markets.

The second spatial consideration arises from the wide variety of explanatory variables that we have constructed to finely characterize the distance of housing to transport infrastructure and facilities as well as environmental (dis)amenities. Although, since the seminal model of Alonso-Muth-Fujita (Fujita, 1989), urban microeconomics considers accessibility to city-center jobs as the main determinant of property values, housing services and access to urban amenities are also major determinants (Brueckner et al., 1999). We also include flexible specifications (Hastie and Tibshirani, 1990) in our regression models to capture the fact that some explanatory variables, particularly distances, may have a non-linear influence on rents.

By integrating space by way of these two considerations into the estimation of a hedonic model based on LRO data, we estimate the hedonic prices of the contextual (socio-economic, accessibility, and amenities) characteristics of properties. Most applications of the hedonic approach using French data concern real-estate transactions (Ahamada et al., 2007; Baumont and Legros, 2013; Choumert and Travers, 2010; Décamps and Gaschet, 2013; Desponds and Bergel, 2014; Fritsch, 2007; Maslianskaïa-Pautrel and Baumont, 2016; Maurer et al., 2004; Poulhès, 2018; Travers et al., 2008, 2009, 2013). In the French rental market, Marchand and Skhiri (1995) and Prandi and Moumouni (2014) confine themselves to hedonic prices of the intrinsic characteristics of properties. Ayouba et al. (2019) were the first to exploit the LRO database and to compare hedonic prices across urban units. Unlike Ayouba et al. (2019), we perform a more thorough analysis of inter-urban unit heterogeneity and omit the analysis of intra-urban unit heterogeneity between center and periphery. In addition to considering more urban units and a larger set of explanatory variables, we build our analysis of hedonic price heterogeneity on tests of joint equality of coefficients between all urban units and tests of equality of coefficients for all pairs of urban units taken two by two. Our specification also differs from Ayouba et al. (2019) in that it introduces fixed effects by municipality and flexible modeling of the non-linear impact of floorspace through a B-splines function. It should be noted that Ayouba et al. (2020) also drew on this dataset but focused on the consequences of Airbnb on rental prices. Many articles using foreign data have applied a hedonic model to the rental market but with a focus on a particular city: Bala et al. (2014) for Brussels, Brunauer et al. (2010) for Vienna, Efthymiou and Antoniou (2013) for Athens, Hanink et al. (2012) for China, Hoshino and Kuriyama (2010) for Tokyo, and Löchl and Axhausen (2010) for Zurich. Valente et al. (2005) analyze a set of eight urban areas but they include housing variables only, whereas we consider a large number of contextual variables, and they do not provide tests of equality of coefficients. In addition, our analysis covers nearly 150,000 properties, versus their 4570.

From an econometric viewpoint, in order to ensure reliable statistical inference, we take into account both spatial autocorrelation and heteroscedasticity, which are often to be found in the residuals of hedonic models (Anselin and Lozano-Gracia, 2009). We estimate our hedonic models with fixed effects by municipality using Ordinary Least Squares (OLS) and the Spatial Heteroscedasticity and Autocorrelation Consistent (SHAC) robust statistical inference method for the presence of heteroscedasticity and spatial autocorrelation of unknown forms (Kelejian and Prucha, 2007). The joint test of equality of coefficients across all urban units shows that hedonic prices differ for 75% of the characteristics, thus justifying a detailed analysis of heterogeneity. Furthermore, while the impact of some variables, such as floorspace is consistently important across urban units, our analysis reveals that Lyon, Nice, and Paris taken individually are the urban units with the most specific valuations of housing characteristics and socio-economic characteristics compared to the other urban units in the sample.

This article is organized as follows. The following section presents data on the housing characteristics of properties and variables constructed to capture the socio-economic and physical environment of properties. The next section describes the hedonic approach as well as our general estimation strategy. Then, in the next section, we present the results for the 12 LROs and then test for the equality of coefficients. We conclude in the last section.

Data

Geographical scope

The geographical scope has been determined by data availability on rents and structural characteristics of properties. In addition, given the small number of single-family detached houses in the sample and in order to work on a homogeneous market segment, we consider collective properties only in the analysis. Twelve local rental observatories shared the data from their 2015 collection campaign with us. Perimeters of collection of data are very variable, corresponding rarely to the boundaries of urban areas as defined by National Institute of Statistics and Economic Studies (INSEE) zonings.² Nevertheless, most LRO perimeters are included within the urban unit, i.e. the central area containing more than 1500 jobs, inside the large urban area. Only two LROs (Nantes and Rennes) collected a substantial amount of data from suburban rings. To work on a homogeneous definition, we use only the observations included in the urban units of large urban areas, excluding those located in suburban rings and other categories of zoning space in urban areas. In total, our scope of study corresponds to the following urban units: Paris, Lyon, Marseille–Aix-en-Provence, Lille, Nice, Toulouse, Bordeaux, Nantes, Strasbourg, Rennes, Montpellier, and Bayonne. Figure S1 shows the geographical contours of the observations provided by the LROs, i.e. the municipalities for which at least one observation is available. This sample provides a wide diversity of situations in terms of population and tourist attractiveness.

Table 1 displays for each LRO the number of observations and their share in the total number of properties in the private rental sector of the area concerned. LROs seek to ensure their data are representative by collecting data from private individuals (landlords and tenants) and professionals (real-estate agencies, notaries, institutional investors, and property managers) in a standardized manner in the area. Although the sample is not stratified, data collection complies with strict rules laid down by a scientific committee to prevent statistical bias, such as representativeness in terms of management (direct versus delegated), sufficient data coverage to calculate statistical aggregates, and goals in terms of collection to be achieved by area, type of habitat and number of rooms.³ In total, we have 148,543 observations representing approximately 5.7% of the private rental stock in the areas considered. Their number varies greatly from one urban unit to another: from more than 21,000 for the urban unit of Paris against just 5185 for Bayonne. Sampling rates also differ widely and are generally lower when the housing stock is large: it is only 1.73% for Paris but exceeds 13% for Nice, Montpellier, and Bayonne; elsewhere it ranges from 5.2% to 12.29%. Figure S1 indicates these shares at the municipality level for each urban unit.⁴

Table 1.

Number of observations and share in the total private rental housing stock of the samples used in the estimates.

Urban units	Number of observations in the sample	In % of the total number of properties in the private rental sector of the area
Bayonne	5185	15.58
Bordeaux	7424	5.20
Lille	7884	6.75
Lyon	18,413	8.09
Marseille–Aix-en-Provence	12,815	5.87
Montpellier	11,735	13.44
Nantes	10,241	11.57
Nice	19,643	13.76
Paris	21,441	1.73
Rennes	7058	12.29
Strasbourg	9276	11.00
Toulouse	17,428	10.13
Total	148,543	5.69

Source: LROs, 2015; Census, INSEE, 2015.

Rent data and housing characteristics

The dependent variable is the monthly rent without charges, accurate as of the day on which the survey was completed and expressed in Euros. Data concerning the period of construction, the landlord type (direct or delegated), the usable floorspace, average room floorspace⁵ and the time spent by the tenant in the property are available for all LROs. A full description of the rent data and housing variables used is given in Table 2, with the corresponding descriptive statistics in Table S1a–c.

Table 2.

Description of housing characteristics.

Variable	Description	Source	Scale
RENT	Monthly rent excluding charges, as of date of survey (€)	LRO	Home
SURFACE	Usable floorspace (m²)	LRO	Home
DAT_CONS	Period of construction (1: before 1946; 2: 1946–1970; 3: 1971–1990; 4: 1991–2005; 5: after 2005) according to the LRO classification	LRO	Home
DAT_ENTR	Time spent by the tenant in the property (years)	LRO	Home
LANDLORD_TYP	Landlord type (1: delegated; 2: direct)	LRO	Home
AVROOM_SIZE	Average room size (m²)	LRO	Home

The database also provides the geographical coordinates of the properties but with some limitations. Indeed, to keep the data confidential, the geographical coordinates of the homes surveyed have been scrambled. We detail the procedure and its consequences in part C of the supplementary material.

Location variables

In addition to the housing characteristics of properties, we construct a large set of variables to describe the accessibility of local services and facilities for the residents of the rental properties, along with the socio-economic context and the presence of environmental amenities or inconveniences in the immediate vicinity. These variables are listed in Table 3, the same table with the source of the data is in Table S2 and the corresponding description and descriptive statistics are in Tables S3–S5. Variables accounting for the socio-economic context of the neighborhoods in which rental properties are situated are lagged when possible by at least two years in our estimates to limit possible endogeneity bias due to simultaneity.

Table 3.

Location variables.

Name	Description	Scale	Date
Socio-economic characteristics
PROP_SOC	Percentage of social housing	IRIS	2013
PROP_OWN	Percentage of homeowners	IRIS	2013
PROP_SEC	Percentage of second homes	IRIS	2013
PROP_STU	Percentage of students	IRIS	2013
PROP_WORK	Percentage of manual workers	IRIS	2013
UNEMPLOY1564	Unemployment rate in 15–64 age group (%)	IRIS	2013
IMMIG	Percentage of immigrants	IRIS	2013
INCOME	Median income per consumption unit (€)	IRIS	2011
MENT_BREV_PU	Honors rate at nearest public junior high school (%)	Home	2015
MENT_BREV_PR	Honors rate at nearest private junior high school (%)	Home	2015
SRU	=2 if municipality is subject to the Solidarité et Renouvellement urbain (SRU) Law; 1 otherwise	Municipality	2015
DIST_ZUS	Distance to nearest “designated priority neighborhood” (km)	Home	2015
Accessibility to services and jobs
JOBS_0015	Number of jobs accessible within 15 minutes	Municipality	2013
DENS_COM	Density of commercial facilities (shops/hectare)	IRIS	2015
DIST_TRAM_METRO	Distance to nearest tram or metro station (km)	Home	2015
NB_BUSSTOP_500	Number of bus stops within 500 meters	Home	2015
DIST_CENT	Distance to nearest large town (pop. of 50,000+, distance in km)	Home	2014
Amenities
DIST_ROAD	Distance to nearest major road link (km)	Home	2013
DIST_RAIL	Distance to nearest train line (km)	Home	2013
DIST_BUSPARK	Distance to nearest large business park (over 5 ha) (km)	Home	2013
DIST_WATER	Distance to nearest water body (km)	Home	2013
DIST_COAST	Distance to coast (km)	Home	2009

Note: IRIS is the smallest census tract unit available in the French census and contains about 2000 individuals.

Method: The hedonic model

Our empirical analysis is based on a hedonic model (Rosen, 1974), which breaks down the properties into their constituent characteristics. Under the assumptions of agents’ atomicity, perfect information of buyers and sellers, and quasi-linearity of the buyers’ utility functions, the hedonic method can be used to estimate an implicit price for the individual attributes of a property based on its total price. The estimation of these implicit prices is the first step in the hedonic analysis and, as in most papers focusing on housing prices or rents, we confine ourselves to this first step to analyze the implicit prices of each characteristic and their spatial variability. Formally, consider a private rental market with n individual properties, each possessing k (j = 1, …, k) characteristics. The rent for a given property i (i = 1, …, n) is considered to be the combination of the prices associated with its k characteristics, noted x_j (j = 1, …, k). The estimated model is as follows, in a semi-logarithmic form

\ln y = X β + ε

(1)

where

\ln y

is the (n, 1) vector for observations of rent prices (in € excluding charges), expressed in logarithms, and X is a dimensional matrix (n, k) for observations of the explanatory variables. In order to mitigate the omitted bias problem, we include a large set of variables to capture the variability of property rents while limiting multicollinearity using appropriate diagnostics. ε is a (n, 1) vector of random terms, the properties of which are discussed later. The vector β of dimension (k, 1) is the vector of unknown parameters to be estimated corresponding to the vector of hedonic prices and that can be interpreted as semi-elasticities. In order to account for the fact that the impact of some variables on rents may be non-linear, we apply B-spline functions (Hastie and Tibshirani, 1990) to some of the variables. These functions consist in estimating piecewise polynomial functions with smoothing constraints on the knots. They require the number of knots and the order of the splines to be exogenously chosen. In the empirical application, we choose third-order polynomials as they allow enough flexibility to capture non-linear impacts of some variables, while remaining in a parametric framework. In a hedonic framework, this strategy has been also applied by, inter alia, Zheng (1991) or Bao and Wan (2004).

Assuming that the error terms ε are identically and independently distributed, model (1), with or without variables expressed as B-splines, can be estimated using OLS. Nevertheless, this hypothesis is rarely valid in practice for hedonic models: the error terms often reveal a form of heteroscedasticity (i.e. non-constant variance) and/or spatial autocorrelation (Anselin and Lozano-Gracia, 2009). When error terms are heteroscedastic and/or spatially autocorrelated, the OLS estimators remain consistent but are no longer efficient.⁶ To tackle these problems, we applied a SHAC estimator for the variance–covariance matrix of error terms, which remains consistent even in the presence of heteroscedasticity and spatial autocorrelation of unknown forms. This method avoids the need for assumptions regarding the forms taken by heteroscedasticity and spatial autocorrelation of error terms (Kelejian and Prucha, 2007). As a nonparametric function, it necessitates the choice of a kernel function and bandwidth that we specify below.

Estimation results of the determinants of rents for the 12 LROs

Specification and estimation method

We estimate the following model

\ln ({RENT}_{i}) = \sum_{j = 1}^{12} α_{j} \times D_{j} + \sum_{j = 1}^{12} γ_{1 j} {f (STRUC}_{i} \times D_{j}) + \sum_{j = 1}^{12} γ_{2 j} {f (SOC_ECO}_{i} \times D_{j}) + \sum_{j = 1}^{12} γ_{3 j} {f (AME}_{i} \times D_{j}) + \sum_{j = 1}^{12} γ_{4 j} {f (ACCESS}_{i} \times D_{j}) + \sum_{j = 1}^{12} γ_{5 j} {f (MUN_DUMMIES}_{i} \times D_{j}) + ε_{i}

(2)

where

D_{j}

is the jth urban unit dummy variable, with j in alphabetical order (j = 1 for Bayonne and j = 12 for Toulouse), using R (2005).

STRUCT

is the vector for the following housing characteristics: landlord type (

LANDLORD_TYP

), time tenant has been in the property (

DAT_ENTR

), period of construction (

DAT_CONS

), and average size of rooms (

AVROOM_SIZE

). The variable for total floorspace (

SURFACE

) is integrated as splines in order to account for its potential non-linear effects on rent prices (

RENT

SOC_ECO

is the vector representing socio-economic variables.

AME

and

ACCESS

represent the vectors for the variables “amenities” and “accessibility”. While the same set of socio-economic variables is included for each city, the selection of the accessibility and (dis)amenity variables is city specific. In other words, vectors

AME

and

ACCESS

may be composed differently in different urban units. Tables S6 and S7 list the amenity and accessibility variables used for each LRO. These variables measure very localized effects and not all of them need to be introduced for all urban units. We therefore keep only those variables that best explain the variability of rents while avoiding multicollinearity by applying a tolerance threshold of 5 for the Variance Inflation Factor measure. In addition, for each urban unit, we test the need to integrate all the variables expressed as distances in splines up to order 3 as most economic relationships are detectable at this order. If the use of splines proves irrelevant, we introduce the variable in level in the specification.

α

and

γ_{j l} (l = 1, 2, 3, 4, 5, 6 and j = 1, \dots, 12)

are the parameter vectors to be estimated.

MUN_DUMMIES

is composed of municipality dummies of dimension M – 1 (M is the number of municipalities in the sample, one municipality is taken as the reference in the estimation to avoid collinearity with urban unit fixed effects) controlling for (unobservable) variables affecting all rents equally within municipalities. In Paris, Lyon, and Marseille, we use “arrondissement” instead of municipality dummies. As a robustness check, we also fit models without these dummy variables but including variables measured at the municipality level.

ε

is the vector of the error terms, the properties of which are given in the Method: The Hedonic Model section.⁷

To estimate the SHAC variance–covariance matrix, we use a Parzen window. The argument of this function is a distance matrix and the result is a variance–covariance matrix. The distance matrix is calculated as follows: (i) we construct a distance matrix using the geographical coordinates for rented properties,(ii) we then apply a Parzen window with a threshold distance of 250 m ⁸ meaning that we assume that properties more than 250 m apart have no effect on each other. The further apart two properties are, the lower the weighting assigned in the variance–covariance matrix. In order to avoid a situation where properties only interact with themselves, we remove from our initial database (for each LRO) all properties with no neighbors within a 250 m radius. Table S8 shows the number of observations for each LRO before and after pruning, when all the observations are included and when only the new tenancy agreements are included. This procedure results in the loss of a relatively small number of observations, from 0.26% in Strasbourg to 3.97% in Paris.

Estimation results

The quality of fit amounts to 0.866, so the variables incorporated in the specification account for a substantial part of the variance observed in rent prices.

We present graphically the estimated coefficients for the specification with FE in the OLS-SHAC version to facilitate comparison between LROs. The detailed estimation results and the comparison with the coefficients estimated for three other specifications (OLS without FE, OLS-SHAC without FE, and OLS with FE) are presented in Tables S10–S13. With respect to housing variables, the results show that the impact of floorspace is not linear, as the coefficients of the spline functions are significant (Table S10), with a relatively small curve (Figure S2) for each LRO.

Figure 1 shows that properties built during the period 1946–1970 (reference modality) are the least valued everywhere, except in Nice. This effect is not surprising since the period 1946–1970 saw the large-scale construction of huge apartment blocks, generally considered to be of mediocre quality and generating relatively high charges, such as heating costs. However, this phenomenon does not apply in Bayonne, Marseille, and Nantes. Properties built since 1990 are valued more highly than those built before 1946 right across the board: these properties are well-equipped and comfortable, which means that they fetch relatively high prices when rented. Overall, these results are consistent with those of previous studies carried out, among others, in the Bordeaux agglomeration (Décamps and Gaschet, 2013) and in Angers (Travers et al., 2013) for real-estate markets or in Paris (Prandi and Moumouni, 2014) for the private rental sector. As for the other housing factors, the length of time a tenant has been in a property, measured by the variable ( $DAT_ENTR$ ), has a significant and negative impact on rent prices in all LROs except Rennes, where the effect is not significant. An old lease is a lease signed at a time when rent was generally lower. Landlords often take advantage of tenancy changes to raise the rent on their properties. The effect of this variable is particularly strong in Paris and Bordeaux. The coefficient assigned to average room size ( $AVR O O M_SIZE$ ) is significant and negative across all LROs. The number of available rooms is thus relatively high: for the same total floorspace, there is greater demand for three-room than two-room properties. Finally, the effect of the “landlord type” variable ( $LANDLORD_TYP$ ) is globally positive: compared to delegated management, direct housing management is associated with higher or at least equivalent rent, with the exception of Strasbourg.

Figure 1.

Estimated coefficients for housing variables. Note: The colors correspond to the estimated effects: red (negative and significant effect), green (positive and significant effect), and gray (non-significant effect). The values on the y-axis are interpreted as the percentage change in rent/m² following an increase of one unit in the explanatory variable for the quantitative variables or with respect to the reference category for the qualitative variables.

With respect to socio-economic variables (Figure 2 and Table S11), the proportion of social housing within an IRIS unit (PROP_SOC), when significant, has a positive effect on rents, which may seem counter-intuitive. Nevertheless, Nguyen (2005), in a literature review focusing on the United States, shows that low-rent housing has a complex effect on real-estate values in the neighborhood and may even positively affect real-estate values under some conditions. Moreover, since 2013 French law requires large cities to have at least 25% of social housing, which makes this variable less discriminatory in France. In our specifications without municipal fixed effects, we control for the SRU law, i.e. whether the municipality is constrained in terms of social housing. Being subject to constraints in terms of social housing significantly increases rental prices in half of LROs, except Rennes where the sign is negative. However, controlling by this variable does not affect the positive effect of social housing in the specifications without fixed effects.

Figure 2.

Estimated coefficients for socio-economic variables. See note to Figure 1.

When significant, the proportion of home ownership (PROP_OWN) within an IRIS has a negative impact on rental prices in all LROs except Lille: a strong presence of homeowners is more likely to characterize residential neighborhoods less sought after by tenants. A high proportion of second homes (PROP_SEC), typical of tourist areas, pushes up rents in Paris, Toulouse, Nice, and Lille. As expected, the “income” variable (INCOME) has a positive effect on rent prices across all LROs. The effect is most pronounced in Bayonne and non-significant in Bordeaux. The distance of properties from the nearest “designated priority neighborhood” (DIST_ZUS) has a linear but insignificant effect on rental prices in eight urban units, the only significant effect being in Rennes. For Montpellier, Strasbourg, and Paris where this variable has a non-linear effect on rental prices, we depict the impact on rents in Figure S3. All in all, it seems that proximity to such zones is unpopular. We note globally, as in Décamps and Gaschet (2013), that the immediate neighborhood of these areas is rather devalued. This result is also in line with those obtained by Desponds and Bergel (2014) who, based on the example of six Parisian inter-municipal groups, find a similar negative effect on real-estate prices, with proximity effects felt up to 750 m away, more markedly for apartments than for houses. The share of manual workers (PROP_WORK), frequently used (together with those of managers and higher intellectual professions) to analyze the social differentiation of urban spaces, has a negative and significant effect in most urban units.

In addition, the performance of schools only has a significant effect in public junior high schools (MENT_BREV_PU), in Nice, Strasbourg, and Lille. Thus, like Poulhès (2018) with the example of the Paris real-estate market, the quality of schools also tends to be capitalized in the rents in the private rental sector. The share of students (PROP_STU) is only significant (and positive) in Lille, Montpellier, and Marseille. Finally, the other socio-economic variables introduced in the estimates show more nuanced results. The share of immigrants (IMMIG) has a positive and significant effect in Paris, Strasbourg, Nice, and Bordeaux but a negative effect in Toulouse and Marseille while the unemployment rate (UNEMPLOY1564) has a negative impact on rents in Toulouse and Marseille and, counterintuitively, a positive impact in Marseille, Toulouse, and Bayonne.

The hedonic prices of accessibility variables are presented in Figure S4 and Table S12. The distance to the center of the urban unit (DIST_CENT) has a linear effect on rent levels in Lyon, Montpellier, Lille, Paris, Bordeaux, and Toulouse. For these cities, the further away the housing is from the center of the urban unit, the less it is valued. Nantes and Rennes are the only urban units where this variable has a non-linear effect on rent levels (Figure S5). Other accessibility variables are more rarely significant. The effect of the proximity to the bus network (NB_BUSSTOP_500) on rents is significant only in the urban unit of Lyon, where surprisingly it is negative. The tram or metro station (DIST_TRAM_METRO) is perceived as a disamenity in Lyon and Strabourg. It should also be noted that a high commercial density in the IRIS (DENS_COM) has a positive and significant effect in Paris, Rennes, Toulouse, Lille, and Nantes, whereas it is negative in Strasbourg.

Concerning the amenity/disamenity variables (Figure S6 and Table S13), proximity to a major road (DIST_ROAD) and an industrial and/or commercial zone (DIST_BUSPARK) are rather perceived by households as disamenities given the associated nuisances (mainly noise). These infrastructures outweigh the benefits they provide, particularly in terms of accessibility. In terms of proximity to a major road, Travers et al. (2013) find similar results in the property market of Angers, Décamps and Gaschet (2013) on that of the Bordeaux agglomeration and on real-estate prices in suburban Dijon. The effect of proximity to a railway (DIST_RAIL) is significant in Bordeaux only. Bodies of water increase rental prices in Bayonne, Strasbourg, Rennes, and Lille. Proximity to the coast is highly valued in Marseille.

Our results are robust to alternative specifications (OLS with and without FE, SHAC without FE), as changes in coefficients are minor. We then proceed to a more detailed analysis of the heterogeneity in hedonic prices across urban units by testing (i) the joint equality of coefficients for each variable (Table 4) and (ii) the equality of coefficients for all pairs of urban units taken two by two for each variable. For the latter tests, we use the Bonferroni correction, which is one of the most conservative corrections for multiple comparisons. Table 4 shows that the joint equality of coefficients across all urban units is rejected for all variables except the proportion of social housing, the honors rate at the nearest private junior high school, distances to the nearest body of water, the nearest large town, and the nearest large business park, as well as—when relevant—distances to the coast and the nearest tram or metro station. The null hypothesis is therefore rejected in 75% of cases, confirming the relevance of the spatial analysis of hedonic prices.

Table 4.

Results of equality tests.

	P-value of the joint test	% of cases where the test of equality between pairs of urban units is not significant
Urban unit dummy	***	69.7
SURFACE	***	36.36
DAT_CONS1	***	66.67
DAT_CONS3	***	89.39
DAT_CONS4	***	72.73
DAT_CONS5	***	63.64
DATE_ENTR	***	37.88
LANDLORD_TYP	***	62.12
AVROOM_SIZE	***	71.21
PROP_SOC	NS	96.97
PROP_OWN	*	96.97
PROP_SEC	***	87.88
UNEMPLOY1564	**	95.45
PROP_STU	**	95.45
PROP_WORK	**	95.45
IMMIG	*	98.48
INCOME	***	84.85
DENS_COM	**	96.97
DIST_CENT	***	74.24
MENT_BREV_PU	***	87.88
MENT_BREV_PR	NS	100
DIST_ROAD	***	95.45
DIST_RAIL	NS	100
DIST_BUSPARK	NS	100
DIST_WATER	NS	100
DIST_ZUS	**	95.45
DIST_COAST	NS	100
NB_BUSSTOP_500	*	96.43
DIST_TRAM_METRO	NS	100

Mean		85.09
Percentage of cases where the test is not significant	24.14

Note: * significant at 10%, ** significant at 5%, *** significant at 1%.

When focusing on the pairs of urban units, we show that the three urban units with the most different hedonic prices relative to others are Lyon, Nice, and Paris (Table 5). For the urban unit of Lyon (Table S14d), the variables that are the most differently valued in rental prices in comparison to others are landlord type, time spent in the property, floorspace, period of construction, distance to the center, and honors rate at the nearest public junior high school. The urban unit of Nice (Table S14h) stands apart mainly by the valuation of housing characteristics, distances to the center and to major roads, as well as honors rate at the nearest public junior high school. In the urban unit of Paris (Table S14i), the valuation of floorspace, time spent in the property, periods of construction, median income, proportion of students, number of bus stops, and distance to the center have the most different valuations. Finally, housing characteristics, median income, and distance to the center are clearly variables that have the most heterogeneous effect on hedonic prices.

Table 5.

Results of the tests of equality of coefficients between pairs of urban units.

	Percentage of cases where the test of equality of coefficients is not significant
Bayonne	85.42
Bordeaux	86.33
Lille	83.22
Lyon	78.48
Marseille	86.72
Montpellier	87.84
Nantes	88.05
Nice	79.82
Paris	80.53
Rennes	82.34
Strasbourg	86.90
Toulouse	81.17
Mean	83.90

Conclusion

This article emphasizes the need to explicitly integrate the spatial dimension of the data in the analysis of hedonic rent prices for the private rental market in France. We show that hedonic prices vary widely from one urban unit to another, especially those pertaining to contextual characteristics. These results indicate that the way housing attributes are valued by tenants is heterogeneous in the metropolitan areas. Our joint test of equality of coefficients between all urban units also reveals that Lyon, Nice, and Paris taken individually are the most specific urban units in terms of valuations of housing characteristics and socio-economic characteristics. Direct housing management and recent periods of construction have forced rents up in Lyon much more than elsewhere. The time spent in the property holds rents down most in Paris. Nice stands out in having the lowest valuation for the period of construction. As such, our study can feed the design of space-based policies, helping with the design of new public policies tailored to the specific requirements of different areas, such as housing subsidies (creating or revising zoning criteria upon which subsidies, particularly tax breaks, depend) or direct intervention to manage rent prices (rent controls, housing benefits, approved rent agreements, or lease renewals). For future research, it would be interesting to go a step further and analyze, besides the heterogeneity between urban units, the potential diversity within each urban unit by distinguishing market segments depending for instance on the number of rooms in housing units. This would make it possible to test whether the effects of the various variables on rents and their contribution to the coefficient of determination are different between the defined segments within each urban unit.

Supplemental Material

sj-pdf-1-epb-10.1177_2399808320977877 - Supplemental material for The spatial dimension of the French private rental markets: Evidence from microgeographic data in 2015

Supplemental material, sj-pdf-1-epb-10.1177_2399808320977877 for The spatial dimension of the French private rental markets: Evidence from microgeographic data in 2015 by Kassoum Ayouba, Marie-Laure Breuillé, Camille Grivault and Julie Le Gallo in Environment and Planning B: Urban Analytics and City Science

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors acknowledge funding from “Ministère de la Cohésion des Territoires”.

ORCID iD

Julie Le Gallo

Notes

Kassoum Ayouba is a researcher in economics at INRAE. He received his PhD from the University of Lille in 2015 and had a post-doctoral position at “CESAER” in Dijon prior to joining the joint research unit “Territoires” located in Clermont-Ferrand (France). His research interests lie in the fields of spatial economics, production economics, and applied econometrics ..

Marie-Laure Breuillé is researcher in economics at INRAE. She received her PhD in economics from Paris X-Nanterre in 2006 and had a post-doctoral position at the WZB in Berlin prior to joining CESAER. Her main area of specialization is public finance. Her research focuses on fiscal interactions among (same-tier and multi-tier) jurisdictions, with a focus on competition to attract mobile tax base, and more recently on the relationship between taxation and the housing market. She has published in leading international journals, including Journal of Publics Economics and Journal of Urban Economics.

Camille Grivault currently holds the position of engineer at AgroSup Dijon. He previously held different positions as GIS engineer at the University of Burgundy, CNRS and Inra and has a work experience in local authority responsible for urban planning. From the 2010s, he began his collaboration with economists, in particular on issues with a spatial dimension (dissemination of innovative agricultural practices, urban densification, and housing markets) and developed his expertise in GIS science, spatial data analysis, and visualization.

Julie Le Gallo received her PhD from University of Bourgogne in 2002 and is full professor of economics at AgroSup Dijon, University of Bourgogne Franche-Comté (France). Her research interests lie in the fields of spatial econometrics, land and housing markets, and economic geography. She has published over 70 articles in peer-reviewed scientific journals including American Journal of Agricultural Economics and Journal of Urban Economics, 20 book chapters including review chapters in econometrics handbooks. She is a co-editor of the peer-reviewed journals Spatial Economic Analysis and Papers in Regional Science.

References

Ahamada

Flachaire

Lubat

(2007) Prix des logements et autocorrélation spatiale: une approche semi-paramétrique. Économie Publique/Public Economics 20: 131–145.

Anselin

Lozano-Gracia

(2009) Spatial hedonic models. In: Mills

Patterson

(eds) Palgrave Handbook of Econometrics: Volume 2, Applied Econometrics. Berlin: Springer-Verlag, pp.1214–1250.

Ayouba

Breuillé

Grivault

, et al. (2019) Hétérogénéité spatiale des prix hédoniques des appartements du marché locatif privé en France. Revue Française D’économie 34(2): 203–247.

Ayouba

Breuillé

M-L

Grivault

, et al. (2020) Does Airbnb disrupt the private rental market? An empirical analysis for French cities. International Regional Science Review 43(1–2): 76–104.

Bala

Peeters

Thomas

(2014) Spatial issues on a hedonic estimation of rents in Brussels. Journal of Housing Economics 25: 104–123.

Bao

HXH

Wan

ATK

(2004) On the use of spline smoothing in estimating hedonic housing price models: Empirical evidence using Hong Kong data. Real Estate Economics 32(3): 487–507.

Baumont

Legros

(2013) Nature et impacts des effets spatiaux sur les valeurs immobilières. Revue Économique 64(5): 911–950.

Brueckner

Thisse

J-F

Zenou

(1999) Why is Central Paris rich and downtown Detroit poor? An amenity-based theory. European Economic Review 43(1): 91–107.

Brunauer

Lang

Wechselberger

, et al. (2010) Additive hedonic regression models with spatial scaling factors: An application for rents in Vienna. The Journal of Real Estate Finance and Economics 41(4): 390–411.

10.

Charlot

Hilal

Schmitt

(2009) La périurbanisation renforce-t-elle la ségrégation résidentielle urbaine en France? Espace Populations Sociétés 1: 29–44.

11.

Chodorge

Pavard

(2016) Le parc locatif privé et ses bailleurs en 2013. ANIL.

12.

Choumert

Travers

(2010) La capitalisation immobilière des espaces verts dans la ville d’Angers: Une approche hédoniste. Revue Économique 61: 821–836.

13.

Décamps

Gaschet

(2013) La contribution des effets de voisinage à la formation des prix du logement. Revue Économique 64(5): 883–910.

14.

Desponds

Bergel

(2014) “La mauvaise reputation”. Distance aux ZUS et prix de l’immobilier: Application d’un modèle hédonique à six intercommunalités franciliennes. Cybergeo.

15.

Efthymiou

Antoniou

(2013) How do transport infrastructure and policies affect house prices and rents? Evidence from Athens, Greece. Transportation Research Part A: Policy and Practice 52: 1–22.

16.

Fritsch

(2007) Tramway et prix des logements à Nantes. Espace Géographique 36(2): 97–113.

17.

Fujita

(1989) Urban Economic Theory: Land Use and City Size. Cambridge: Cambridge University Press.

18.

Hanink

Cromley

Ebenstein

(2012) Spatial variation in the determinants of house prices and apartment rents in China. The Journal of Real Estate Finance and Economics 45(2): 347–363.

19.

Hastie

Tibshirani

(1990) Generalized Additive Models. London: Chapman & Hall.

20.

Hoshino

Kuriyama

(2010) Measuring the benefits of neighbourhood park amenities: Application and comparison of spatial hedonic approaches. Environmental and Resource Economics 45(3): 429–444.

21.

Kelejian

Prucha

(2007) HAC estimation in a spatial framework. Journal of Econometrics 140: 131–154.

22.

Löchl

Axhausen

(2010) Modeling hedonic residential rents for land use and transport simulation while considering spatial effects. Journal of Transport and Land Use 3(2): 39–63.

23.

Marchand

Skhiri

(1995) Prix hédoniques et estimation d’un modèle structurel d’offre et de demande de caractéristiques [une application au marché de la location de logements en France]. Économie & Prévision 121(5): 127–140.

24.

Maslianskaïa-Pautrel

Baumont

(2016) Environmental spillovers and their impacts on housing prices: A spatial hedonic analysis. Revue D’économie Politique 126(5): 921–945.

25.

Maurer

Pitzer

Sebastian

(2004) Hedonic price indices for the Paris housing market. Allgemeines Statistisches Archiv 88(3): 303–326.

26.

Nguyen

(2005) Does affordable housing detrimentally affect property values? A review of the literature. Journal of Planning Literature 20(1): 15–26.

27.

Poulhès

(2018) From Latin quarter to Montmartre investigating Parisian real-estate prices. Annals of Economics and Statistics 130: 39–68.

28.

Prandi

Moumouni

(2014) Le modèle hédonique d’estimation des loyers de l’Olap. Revue Foncière 2: 7–11.

29.

Rosen

(1974) Hedonic prices and implicit markets: Product differentiation in pure competition. Journal of Political Economy 82: 34–55.

30.

Travers

Appéré

Larue

(2013) Évaluation des aménités urbaines par la méthode des prix hédoniques: une application au cas de la ville d’Angers. Economie et Statistique 460(1): 145–163.

31.

Travers

Bonnet

Chevé

, et al. (2009) Risques industriels et zone naturelle estuarienne: une analyse hédoniste spatiale. Economie & Prévision 4: 135–158.

32.

Travers

Nassiri

Appéré

, et al. (2008) Evaluation des bénéfices environnementaux par la méthode des prix hédonistes: une application au cas du Littoral. Économie & Prévision 4: 47–62.

33.

Valente

Gelfand

, et al. (2005) Apartment rent prediction using spatial modeling. Journal of Real Estate Research 27(1): 105–136.

34.

Zheng

(1991) Metropolitan spatial structure and its determinants: A case study of Tokyo. Urban Studies 8: 87–104.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

2.13 MB