1 Introduction

1Providing the means of a better knowledge of housing affordability is becoming increasingly important, for public policies and scholarly research. This data-paper presents and describes a consolidated, harmonized, internationally comparable database to quantify the impacts of the housing affordability crisis. This database is structured to promote some means to understand key issues in urban areas: social filtering processes, gentrification, accumulation, and socio-economic inequalities more generally. We do so by discussing a methodological framework to integrate neighborhood and local spatial data, structured with harmonized indicators, to examine and compare the unequal spatial patterns of housing affordability across Europe.

2The dataset derives from the ESPON Big Data for Territorial Analysis of Housing Dynamics 2018-19 applied research program. ESPON EGTC is a regional planning agency for the European Commission. In European larger cities decent and affordable housing is increasingly hard to get access to: the goal of the study is to inform the increased and unequal affordability gap at the local geographical level, with tools for comparison between cities, and within cities.

4The database has been structured so as to answer several issues altogether, that are the theoretical backbones of the research. One dimension of the problem is unequal access to housing opportunity with regard to income and wealth of households. Another dimension is that this gap has widened during the last decades. Since the 1990s, housing prices have on average increased faster than the income of residents and buyers in major post-industrial city-regions, but this trend is not ubiquitous and there are some exceptions. Second, since the 2008 Global Financial Crisis (GFC) more specifically, the affordability crisis seems to have followed unprecedented pathways of accumulation and vulnerability for households in OECD countries: general price inflation has maintained, but also with greater instability and volatility of local trends. A third issue between European cities is the variegated pathways housing regimes have followed since the GFC. The notion of housing regime can be defined as the systemic relationship between diverse housing systems, tenure and provision arrangements, and wider welfare regimes (Arbaci 2007 ; Tammaru et al. 2019). A decline of ownership has been observed in some countries (Byrne 2020) with an unprecedented growth of the private rental sector. In other countries, ownership has stabilized and other forms of access to housing have emerged, with new regulations of the rental sector as an alternative to social rental and owner-occupied housing, like in the Netherlands and France for instance (Wijburg 2019; Hochstenbach and Ronald 2020).

5Our research also copes with a data gap, i.e., a lack of harmonized spatial data to map affordability in Europe. Institutional, private, national or local data do exist. But these are not harmonized and inter-operable, not fit for comparison (Julliard and Gusarova 2019), and Eurostat does not provide local indicators on affordability. To cope with this data gap, the paper brings new insights on how (1) institutional data, such as transaction data, can be bridged with (2) unconventional data (“big data” harvested on line) to provide a cost-effective and harmonized data collection effort that can contribute to the analysis of affordability. We aggregated the data in various geographical levels of analysis (1km square-grid, municipalities, FUA), following a methodological framework to use local spatial data and to allow the comparison within cities (between neighborhoods) and between cities, and designed to allow reproducing in further studies. The methodology has been applied to 10 case-studies Functional Urban Areas (FUAs) covering 4 countries of the ESPON Area, among which one cross-border region: Geneva (Switzerland), Annecy-Annemasse, Avignon and Paris (France), Madrid, Barcelona and Palma de Majorca (Spain) and Warsaw, Łódź and Krakow (Poland). This database does not cover the entire set of FUAs, but has been designed as a proof of concept, to envision a wider study that could bring together experts from different countries in Europe, to build a more thorough analysis. Open code and transferable methods are documented and disseminated with the ESPON study, to allow further implementation.

6Section 2 provides a literature review, and discusses the conceptual contribution of the database in the wider context of a policy-oriented research on housing at a European level. Section 3 presents the methods used for the data collection: data sources, spatial coverage and case-studies, aggregation and harmonization procedures. It details the conceptual and theoretical models applied to conduct the data collection, and the methodological framework applied. We particularly insist on the effort to bridge the gap by harmonizing conventional and institutional data sources, with unconventional sources from the web ("big data"), to map affordability, along with the quality control and data validation procedures. Section 4 is dedicated to presenting a possible use of the dataset, by mapping and comparing local housing data and giving insights on the main findings from the data. The final section is dedicated to metadata.

2 Context, literature review and contribution of the database in housing research

7This section discusses the policy context in which the production of the database took place, and frames the issue of affordability and inflation, within the wider literature on housing regimes and inequality. We argue that the lack of harmonized local spatial data to map affordability in Europe is a major issue to better analyze and compare the situation at local geographical levels. The aim of the ESPON database is to present data and reproducible methods that contribute to filling this data gap.

2.1 Policy context : coping with inflation in Europe

8Housing prices have increased faster than the income of renters and buyers in major post-industrial city-regions, and real estate has become an important driver of socio-economic inequalities: “in 2014, approximately 7% of the EU-28 population faced the situation where housing costs accounted for more than half of their disposable income” (Eurostat, 2016). This forms the basis of our study. The policy-oriented broader reflection is to analyze the spatial dynamics of unequal local affordability, as framed by the Action Plan of the Partnership on Housing of the UE Urban Agenda pushes for improved knowledge regarding affordability of housing. We address the housing elements of European policies through one major problem: affordability, a concept defined as a gap between housing prices and households’ income (Tighe and Mueller 2012; Friggit 2017).

9Housing is a major policy issue regarding urban well-being, cohesion and sustainability in Europe. The study is framed by the Territorial Agenda of the European Union for 2020 and deals with the impacts of housing on territorial cohesion, and how to tackle with risks of exclusion. The Pact of Amsterdam (2016) recognizes housing as a key priority of the Urban Agenda for the European Union, because it affects social cohesion and stability in Europe, the Action Plan of the Partnership on Housing of the EU pushes for improved knowledge regarding affordable housing.

11This results in an unprecedented inflation: in 1985-2010 price-to-income ratio has increased from + 13% up to +28% in France, 44% in the UK, except in Germany and Japan (Aalbers 2016). In France, affordability of homeownership, i.e. real estate purchase power, calculated by dividing housing prices by level of wealth (income generally) has dropped at its lowest historical level (Friggit 2017), because of a +70% increase of the actual cost of real estate since 2000, while households benefited from lower interest rates and longer credit range to offset this price inflation and maintain purchase power. This is not, however, the case in every European country, as demonstrated in a study by Beresewicz (2015), where data show a decreasing trend on the stock market in Poland after 2008.

12This affordability problem spreads across markets, with unequal effects (Kemeny 2001; André and Chalaux 2018). The OECD Affordable Housing Database (AHD) characterizes the various conditions existing on European markets. Data show that in many European countries, price paid (through the average mortgage or rent burden) tops up the 25% of disposable income critical threshold, in Norway, Finland, Czech Republic, Sweden, the Netherlands, Greece, the UK, Denmark and Spain as for rents. For owner-occupied households, this is measured by means of the cost of mortgage, and it reaches the highest levels of disposable income in France, in Luxembourg, Ireland and Croatia. Among lower-income owners, covering all individuals with a net income below 50% of median income of total population, the situation shows critical threshold of debt (50%) in 14 countries of the OECD study (Oecd 2018). For renters, the worst situations are found in 9 countries among which Croatia, Greece, Spain, the UK and Portugal.

2.2 Literature review: affordability, housing regimes and inequalities

13Given the effects of price inflation on price-to-income gap in European cities, it is crucial to contextualize the issue within the different market regimes between European cities, as a large part of the literature considers a global shift in market regimes has occurred (Fernandez and Aalbers 2016). Affordability of housing has become, in this context more and more socially and politically embedded in welfare.

14This situation in European countries is linked with financial and macroeconomic parameters like monetary policies, credit affordability and fiscal policies targeting investment (national policies): 25 States in the OECD promote homeownership by means of subsidies and fiscal incentives to first-time property owners and/or fiscal incentives to offset the cost for individual buyers (André and Chalaux 2018). This shows how advanced economies have shifted toward an asset-based welfare model, yielding a regime linking an ideology of ownership, credit affordability and house price often subsidized by the State and local governments (Ronald 2008; Rolnik 2013). Topalov (1987) has well established how homeownership regimes shifted from a rentier-system to a credit-based massive homeownership system. Such a shift has been regulated by converging public policies (to increase household solvency and provide incentives to homeownership), the banking industry, market players’ strategies (among which developers have been preeminent), and preservation of assets strategies by households. Some scholars describe a path dependency-shift in almost every nation-state influenced by global World Bank policies (Rolnik 2013; Theurillat, Rérat, and Crevoisier 2015). Other studies suggest that housing regimes have also recently shifted towards a retargeting of investment, from owner-occupied housing to private lenders: recent studies showed that private sector rental has been preeminent in the restructuring of credit markets since the GCF (Byrne 2020).

15Trajectories are therefore highly dependent upon national frameworks, and the state enables households to act as investors, engaging through markets with prospects of future gains, while exposed to greater risks (volatility of price, loss of property values, risks of bankruptcy and foreclosures, etc.), raising questions of individual and systemic risks, and therefore vulnerabilities of households (Schwartz 2012).

16This has also been a gradual shift in welfare, across national models (Fernandez and Aalbers 2016). Housing asset-based welfare is described as a rising ideology of ownership across modern-industrialized societies and a path dependency in housing policy reforms (Malpass 2011), characterized market mechanisms driven by policy-measures, socio-ideological frameworks (Ronald 2008) and finance-led accumulation regimes (Boyer 2009). This holds for both owners and for private lenders, who are also often households developing investment strategies to recycle accumulated wealth. More specifically, theories link the high proportion of capital investment that housing represents for households and the ways welfare states are organized and reformed (Kemeny 2001), the pursuit of owner-occupancy being viewed as a superior form of tenure, while privatizing social housing (Van Gent 2010).

17The situation in Europe is very diverse in this respect. André and Chalaux (2018) used the OECD database to identify four groups among a set of OECD countries, showing how the dimensions of the affordability question are highly multifaceted, according to various market regimes:

A group named "Northern", as it covers mainly Northern Europe, including Germany, features extended private rental and generally high household debt. A "Western" group, which includes France and the United Kingdom, has higher home-ownership rates and more social housing. "Southern‐Central" and "Eastern" groups gather European countries, where outright home-ownership is prevalent, but housing conditions are poorer. (André and Chalaux 2018, p. 1)

18One policy implication is that no single recommendation can be issued at a European level regarding the affordability crisis, with regards to the diversity of market conditions in Europe. Generally speaking:

• For prospective homeowners: increased affordability gap leads to a socially-selective access to housing markets, yielding more spatial exclusion and increased social tensions. Real estate has become an important driver of socioeconomic inequalities, especially between owners and renters, with different trajectories of wealth and accumulation.

• The cost of ownership impacts rents, and also the availability of housing to let, especially because of increased private rental that are substitutes to the withdrawal of the State from public rental. The global financial and economic crisis also led to decreasing construction activities across Europe, in particular with respect to social housing, as many governments restrained their level of public spending.

• For owners, real estate has become a major component of household wealth. But local markets are also volatile. Housing prices are therefore unstable and contingent upon the market’s continuous restratification within and across neighborhoods. Real estate influences the local conditions through which household wealth is accumulated or lost (Le Goix et al. 2021).

2.3 Filling a data gap

19In such a context, the goal of the database is to detail the relationships between income, price and local market conditions in European FUAs. However, one major issue is the lack of harmonized spatial data to map and analyze affordability in Europe. There are plenty of institutional (tax, census), private (real-estate agents and websites) and national or local data (parcels, local tax rolls). These are not harmonized and easy to inter-operate.

20ESPON interest in the database has been policy oriented. However, data by OECD and Eurostat are disseminated respectively at the national and at the city levels, but the datasets are far from complete in terms of thematic and geographical objects available to accurately analyze housing dynamics. The Urban Audit survey (Eurostat) publishes some perception data on the conditions of housing, and annual statistic aggregates on price, type of structure and households’ income, with an average degree of completeness of 51%, along with missing data in several countries in Europe, including France. The OECD database characterizes affordability as national aggregates, and allows for comparisons to be made between countries (André and Chalaux 2018). To compare housing situations between European cities, Eurostat provides some indicators mainly derived from national censuses under the topic “living conditions”. It provides basic distinctions between houses and apartments, housing type (owner-occupied dwellings, households in social housing), as well as incomplete data on the average price for buyers and renters and average income. Eurostat also provides an overview of the distribution of housing conditions across the European urban hierarchy.

21Two shortcomings should be highlighted: (1) a lack of information regarding the definition and harmonization between databases; (2) the market segments and spatial coverages for which the aggregated information is relevant. These are common problems when harmonizing national census definitions and highlight the heterogeneity of national censuses led in each country in Europe. In short: some datasets are available, but on many respects, there is a need for ways to bridge and produce analysis at local level (Julliard and Gusarova 2019).

3 Methods

22This section describes the steps required to select data sources, define case studies, and data collection, harvesting and the methodology required to combine conventional and unconventional data sources. We describe harmonization and aggregation procedures, that are not only technical, but also conceptual. We finally discuss quality control and data validation procedures, and finally missing data handling, by means of interpolation. As detailed in Figure 1, the workflow required three major steps.

3.1 Data sources

23The data collection was organized using two categories of data sources (Figure 1). Conventional data sources are provided by traditional statistical offices through census for public, demographic use and policies. We gathered data from these providers mainly to inform households income, and we used local income by national statistical offices, and national income data at national level (Eurostat). We combine such classical indicators with unconventional data sources to describe housing market.

24Data were extracted from various platforms and sources and are often named “big data”. We define as “unconventional” data sources that have not been designed and/or documented to prepare measurements, geographical processing and socio-economic research. They often derive from administration or commercial purposes. We distinguish two kinds of unconventional data providers. Institutional data providers on the one hand (Chambre des Notaires in France, Wuestpartner in Switzerland). Such commercial providers distribute proprietary databases that derive from public records of property transactions. They provide information on real-estate transactions. On the other hand, we used scraped data from internet services, which gather advertisements by real-estate agents and private sellers (Leboncoin in France, Fotocasa in Spain): the database also aggregates information on real-estate offers. Such unconventional data is often viewed as interesting proxies to measure, and better understand spatial behaviors and territorial dynamics, and as a means of providing higher spatio-temporal resolution data when compared to institutional data sources (Gallotti et al. 2015), although findings suggests caution regarding the significance of such sources (Kitchin 2013). Prior to relying upon the unconventional data sources, it is important to assess their reliability, and if they provide accurate information when compared to the long established, statistically robust collection data. Some studies have addressed the robustness of IDS (Internet Data Sources) compared to more conventional data (Le Goix and Vesselinov 2014; Beresewicz 2015), so we also consider the issue of robustness of collected samples in this study (section 3.6).

3.2 Spatial coverage: 10 functional urban areas as case-studies

25Ten case studies have been selected, covering some of the variegated dynamics of housing market (gentrification process, tourism, housing crisis, etc.). Highlighting heterogeneous and contrasting situations has been deemed relevant to carry out a first international and comparative study on housing dynamics in Europe based on local indicators.

26The 10 case studies cover 4 countries of the ESPON Area as well as one cross-border region: Geneva-Annecy-Annemasse (Switzerland-France), Avignon and Paris (France), Madrid, Barcelona and Palma de Majorca (Spain) and Warsaw, Łódź and Krakow (Poland). They have been selected for their variety (size and urban dynamics), data availability, and local expertise on data sources. Because they cover cities with both high and low levels of owner-occupancy, the interest of the selection is to highlight a range of housing regimes in Europe, and also allows to consider how to fill data gaps, as on Figure 2 (Poland, for instance, was poorly documented in European databases).

27The selected case studies occupy various positions in the urban hierarchy: the spectrum ranges from 3rd tier cities functional urban areas (276,000 inhabitants for Annecy, 320,000 for Avignon or 670,000 for Palma de Majorca), up to larger cities (3.1 million inhabitants for Warsaw, 4.9 for Barcelona, 6.6 for Madrid) and includes one global city (11.9 M in Paris FUA). The Geneva-Annemasse-Annecy cross-border area offers an interesting setting in terms of unequal affordability, not yet studied with harmonized data.

28For these 10 case-studies, the aim is to provide statistical information on housing affordability at three territorial levels: at the FUA itself, and down to the LAU2 level (communes in France) and the 1km European reference grid level.

3.3 A survey of data sources for case studies

29To detail the procedure, we have first identified the availability of possible sources to analyze the local spatial structure of housing affordability in a selection of countries. This step has highlighted how difficult it would be to follow a data-driven approach in a European comparative perspective on housing and affordability. The goal has therefore been to include in the study various national situations regarding the availability of data, in order to test a methodology for data collection that could be extended to other contexts, in a future pan-European study.

• • 4 DVF has been released at the end of the project funding scheme, and has therefore not been included (...)
  • 5 As noted by Julliard and Gusarova (2019), many other sources are also available online : “property (...)

  France has a long-standing well-established tradition of nationally collecting statistical information on transactions. Given a costly fee (even for researcher of public institutions), transaction data are made available commercially by the Paris Chamber of Notaries database (BIEN). This database covers Ile-de-France real-estate transactions. For the rest of France, the same type of information are commercially made available by French notaries as PERVAL database. Such databases provide a wide range of information related to transactions: price, very detailed characteristics of the property, as well as extra information such as the socio-economic attributes of the seller and the buyer (age, employment category), or the contracted debt of the buyer. Data on property values declared to for land property taxation has been made available in 20194. This fiscal database (Demandes de valeurs foncières, DVF) is available free of charge. It however does not exactly cover transactions on the housing markets, but individual parcels and buildings known values. The problem being that one transactions is in many cases composed by several parcels (i.e. one apartment and one garage; one house, a piece of land, and sometimes a garage and an outbuilding). Major differences between transactions databases and DVF, and their possible use, has been thoroughly documented by Casanova et al. (2017); Casanova Enault, Boulay, and Coulon (2019). Furthermore, many online platforms coexist, that provide different entry-points as IDS. We used one of the most popular ones: Leboncoin.fr market-place advertises properties to buy and properties to let, either by real-estate agents or by individuals5. InsideAirbnb (harvested Airbnb data) is also an important data source available in Paris, Bordeaux and Lyon.

• In Poland, a variety of datasets are available: for some cities the transaction price can be obtained through notarial deeds, down to the address (Łódź), but this has not been the case in Warsaw or Krakow. Some datasets are also collected by the National Bank of Poland (LAU2 units). The online real-estate market place domiporta.pl is a widely used resource, that provides real-estate advertisement: it is a harvestable source for price asked and rent asked. Airbnb advertises properties in many cities, but the insideAirBnb harvesting platform is not available.

• Except aggregated price index and statistics (Instituto Nacional de Estadística, INE and Ministerio des Fomento, Registradores de España), Spain has no national transaction detailed local data widely and reliably available, although Registradores de España now release partial data. Prices are not released publicly, however. Some cities publish their own transactions data (Barcelonatran, for instance), otherwise price estimates are available through companies and private providers. It is therefore convenient in Spain to harvest data online: Fotocasa and Idealista are websites and interactive online platforms that collect property advertisements. Such tools provide property and rental prices. InsideAirbnb (harvested Airbnb data) is available for 6 Spanish cities: Madrid, Barcelona, Malaga, Mallorca, Sevilla and Valencia.

• Switzerland has official statistics on rents and new buildings (Census, Statistics Federal Office), but not on property prices. Several private companies collect and distribute property data for Switzerland, such as CIFI or Fahrländer Partner. In this project, we have used Wuest&Partner data sources, considered as the most accurate.

Table 1: Potential data sources available for case studies, a synthesis

Countries	Institutional data on transactions	Market place platforms	Census
(case study)	(land registry, notaries…)	(online data harvesting )
Switzerland		X	X
France	X	X
Spain	Some cities	X
Poland	Some cities	X

3.4 Data collection

3.4.1 Conventional institutional data: Eurostat and National Statistical Institutes indicators

32A first issue is data collection from the relevant providers to target harmonized indicators as outputs. Conventional census data are required to extract information at EU level on socio-economic characteristics of case studies and to collect data on income, for the construction of an estimation of real-estate affordability. Institutional unconventional data are also required to describe residential property markets.

33Two main categories of institutional data providers (“official statistics”) have been used: harmonized European statistics and national statistics. EU statistics (sources Urban statistics, Eurostat) have been used to characterize selected case-studies as regards to the other cities of Europe. Taking into account the availability of data, 20 indicators at Core City have been identified, against 14 only at FUA level. This allows to compare all the European cities and Functional Urban Areas. EU statistics include demographic indicators (age), households’ characteristics, information related to the employment (economy tertiary oriented or not) and other relevant factors to understand who lives in the cities. However, little information is available on housing. It is only possible to extract one item of the EU perception survey: “is it easy to find good housing in your city?”, which gives only a very rough qualitative assessment of affordability by European citizens. These indicators are provided in the data repository.

Table 2: Listing of Eurostat available indicators relevant for characterizing the housing market

ID	Indicator Name	Reference year	Reference
(in the data delivery)			(Eurostat table)
POP_2015	Total population	2015	“urb_cpop1”
POP024_2015	Share of population aged 0-24 years (%)	2015	“urb_cpop1”
POP2544_2015	Share of population aged 25-44 years (%)	2015	“urb_cpop1”
POP4564_2015	Share of population aged 45-64 years (%)	2015	“urb_cpop1”
POP65_2015	Share of population aged above 65 years (%)	2015	“urb_cpop1”
HOUSEHOLD_AREA_2011	Housing size (average size of dwellings)	2011	“urb_clivcon”
HOUSEHOLD_SIZE_2011	Average size of households	2011	“urb_clivcon”
SINGLE_HOUSEHOLD_2011	Share of single households (%)	2011	“urb_clivcon”
OWNED_DWELLINGS_2011	Share of owner-occupied dwellings (%)	2011	“urb_clivcon”
UNEMP_2014	Unemployment rate (%)	2014	“urb_clma”
EMP_INDS_2014	Share of employment in industry (%)	2014	“urb_clma”
EMP_HOTELS_2014	Share of employment in restauration, hotels and transports (%)	2014	“urb_clma”
EMP_RESTATE_2014	Share of employment in real estate activities (%)	2014	“urb_clma”
ST_HIGH_EDU_2011	Share of students in higher education (ISCED 5-6) (per 1000 persons in the population 18-24 years)	2011	“urb_ceduc”
WF_HIGH_EDU_2011	Proportion of popuation aged 25-64 years qualified at level 5 to 8 ISCED	2011	“urb_ceduc”
NIGHTS_2011	Total nights spent in tourist accommodation establishments per resident population	2011	“urb_ctour”
BEDS_2011	Number of available beds in tourist accomodation establishments per 1000 residents	2011	“urb_ctour”
HOUSING_EASY_2015	Share of persons answering "Strongly agree" to the question of the Urban Audit : is it easy to find good housing in your city ?	2015	“urb_percep”
HOUSING_DIFFICULT_2015	Share of persons answering "Strongly disagree" to the question of the Urban Audit : is it easy to find good housing in your city ?	2015	“urb_percep”

35LAU2 income, data provided by national providers, have also been used, especially to locally contextualize affordability. A classical caveat applies regarding income data. It is not recommended to compare local affordability, based on local income data, between cities of different countries. Indeed, the methodologies for income computation varies from one country to another: per capita or per households; before/after tax; with or without social benefits, etc. The statistical parameters for aggregated income also differ in institutional data, some as median, other are average income at LAU2 level. Nevertheless, local affordability indexes are still highly relevant to compare local affordability between cities of the same country (Avignon-Paris / Madrid, Barcelona-Palma de Mallorca / or Łódź-Warsaw-Krakow separately), the methodology of income calculation being generally harmonized at national level. Indicators available in the database are detailed section 3.5.3.3.2 Unconventional institutional data: transactions

36In some cases, in France, it has been possible to use property-level data. Two datasets have been used.

• One on the Paris region, provided under the Labex Dynamite agreement by Paris Notaries Services, a subsidiary of the Chamber of the Notaries. This sample contains transactions for the region and its suburbs, within the administrative limits of Ile-de-France (roughly 1 million rows), for year 1996 to 2012. All records contain information on the property amenities and pricing, and series of understudied interesting variable on sellers and buyers, such as age, sex, socio-economic status, national origin, place of residence, and some credit history related to the transaction. For a discussion on the interest and limitations of this dataset for local studies of housing inequality, see Le Goix et al. (2019a).

• In other French FUAs (Avignon, Annecy, Annemasse), PERVAL, a database of transactions provided by Les notaires de France has been purchased for the study. It provides properties information, based on individual transactions, with a data structure similar to BIEN database.

3.4.3 Unconventional real-estate data: harvesting

38Data harvesting, on real-estate advertisement websites and market-places platforms has been conducted following a procedure, generally using Python language Scrapy library, which can be summarized as follows:

1. Listing properties. Harvesting real estate online listings (property advertisements). It consists in getting the total number of ads and determining the pages to scrape after identification of the relevant tags syntax in the URL query, with a specific attention given to geographical tags and the type of offer (real estate offer / rental ; apartments and-or houses, etc). We then automatically collect all the adds included in all pages of the first query result. The first output is a list of URLs to be harvested (one URL by offer).

2. • 10 The cost and duration of the project allowed only for test drives and a few months of collection, a (...)

   Identifying all the relevant information for listed properties in step 1. This consists in preparing the script for each website to automatically fetch the data. It requires to harvest the webpage, and identify the interesting attributes/tags to get (price, number of rooms, surface, geographical location …). This is a tedious, very costly and time-consuming process that requires a lot of retro-engineering of webpages, with different data sources in each countries10.

3. Data cleaning. The most common errors are duplicate records (sometimes a real estate listing can be published several times), absence of location coordinates or inconsistencies in the information describing each advertised property (area, price, etc.). More inconsistencies are found in geocoding. This will be further discussed in the validation section.

3.5 Harmonization of the database, description of variables

39Data collection/cleaning and aggregation procedures are conducted so as to produce data that can be used in a multiscale perspective. Produced at the local fine-grain geographical city level (grid and LAU2), they are also aggregated up to the level of the entire FUAs. The indicators created are described below, and summarized Table 3.

40Local data produced contain the following information, where available:

• Official geometries identifiers (LAU2 and grid)

• Income data (municipal and national income data)

• Aggregated transaction data if available (rooms, surface, price, debt contracted)

• Aggregated web-scraped data for property sales (rooms, surface, price)

• Aggregated web-scraped data for property rental (rooms, surface, price)

43The indicators cover all the dimensions of the data collection, meaning real-estate transactions (Table 3a), real-transaction offers (Table 3b), the level of income of the population (Table 3c) and harmonized indicators (Table 3d, based on the variables detailed in Tables 3a,b and c). Metadata explicitly detail the indicator codes, labels, units of measure, reference year and data sources.

Table 3a: Indicators on real-estate transactions

Code	Label	Measure unit	Reference years
TRANS_NUMBER	Number of real estate transactions (apartments)	Number of transactions	2011-2012 (Paris), 2010 and 2014 (Avignon), 2017 (Polish case-studies), 2012 (Barcelona), 2010 and 2015 for Geneva cross-border area, no data for Madrid and Palma de Mallorca, Debt is only available for French case-studies.
PRICE_PAID	Price paid for properties	Euros
SURFACE	Surface of properties	Square meters
ROOMS	Number of rooms	Room
DEPT_SUM	Sum of debt contracted to buy a property	Euros

Table 3b: Indicators on real-estate offers

Code	Label	Measure unit	Reference year
OFFERS_NUMBER	Number of real estate offers	Number of offers	Data gathered from Web scraping on reference real-estate Websites in 2019.
OFFERS_PRICE	Advertized price for properties (real estate offers)	Euros
OFFERS_SURFACE	Surface of properties (real estate offers)	Square meters
OFFERS_ROOMS	Number of rooms	Room
RENT_NUMBER	Number of rental offers	Number of offers
RENT_PRICE	Advertized price for rental offers	Euros
RENT_SURFACE	Surface of rental offers	Square meters
RENT_ROOMS	Number of rooms (rental offers),	Room
RBNB_NUMBER	Number of Airbnb offer	Number of offer	2019 (Inside Airbnb)

Table 3c - Indicators from national statistical censuses and Eurostat

Code	Label	Measure unit	Reference year
TOT_P	Total population	Inhabitants	2011
INCOME_LOC	Average or median income (municipalities)	Euros	2016 for France, 2015 for Poland, 2011-2015 for France and Switzerland (average on time-period),
INCOME_NAT_D1	National Income, first decile	Euros	2015
INCOME_NAT_D5	National Income, median	Euros	2015
INCOME_NAT_D9	National Income, 9th decile	Euros	2015

45The most relevant ratios (Table 3d) are combined starting from basic indicators (Tables 3a,b & c).

• price/sq. meter (price paid, advertised price, rental), a standard indicator, but highly contingent to the local structure of housing;

• Three different affordability indices. Local affordability is a price-to-income ratio computed with local income. It describes to what extent a LAU2 is more or less affordable for household living in the LAU2, based on the local income distribution. National affordability answers a more general question: to what extent a given place is affordable for an average household in a given country. We also provide the difference between local and national affordability. This gives a better understanding of whether it is easier or not for a local household, as compared to another one moving out from any other place in the same country, to afford a property in the designated area (positive values), or not (negative values).

• A last index is referred to as the profitability index, calculated as a ratio between advertised price for property and advertised price for rental. This approach is best conceptualized as being a common approach that computes how much one landlord should invest in a rental property for 1 euro of rent”. A higher index means two things: advertised prices are high, as compared to rental offer; or rental offer prices are cheaper, compared to property prices. In other terms, a lower index can be interpreted as locations where the development of rental offers may be specifically profitable for real-estate investors.

Table 3d - Synthesis of harmonized indicators built from various data sources.

Code	Label	Measure unit	Based on,
PRICE_SURF	Price paid for properties	Euros per square meters	A
OFFERS_PRICESURF	Advertized price for properties,	Euros per square meters	B
RENT_PRICESURF	Advertized price for rental offers	Euros per square meters	B
PROFIT	Profitability index (Ratio between advertized price per sq. meter for property and advertized price per sq. meter for rental offers	Index	B
TRANS_SQ_METERS_LOC	Local affordability: Number of months of local income required to buy 1sq. meter (ratio between price paid per sq. meter for property and median local income/12)	Month	A-C
TRANS_SQ_METERS_NATD5	National affordability: Number of months of national income required to buy 1sq. meter (ratio between price paid per sq. meter for property and median income/12)	Month	A-C
TRANS_SQ_ METERS_DIFF	Difference between local and national affordability (transactions)	Month	D
BUY_SQ_METERS_LOC	Local affordability (real-estate transaction offers)	Month	B-C
BUY_SQ_METERS_NATD5	National affordability (real-estate transaction offers)	Month	B-C
BUY_SQ_ METERS_DIFF	Difference between local and national affordability (transaction offers)	Month	D
RENT_SQ_METERS_LOC	Local affordability (rental): Number of days of local income required to rent 1sq. meter (ratio between advertized price per sq. meter for rental and median local income/365)	Days	B-C
RENT_SQ_METERS_NATD5	National affordability (rental): Number of days of national income required to rent 1sq. meter (ratio between advertized price per sq. meter for rental and median local income/365)	Days	B-C
RENT_SQ_METERS_DIFF	Difference between local and national affordability (rental)	Days	D
PRICE_SUM_SM	Price paid for properties, sum, smoothed (cf part 3.5.2 for details)	Euros	A-B
SURFACE_SUM_SQ	Surface of properties, sum, smoothed (cf part 3.5.2 for details)	Square meters	A-B
PRICE_SQ_SM	Price paid for properties (Euros per square meters), smoothed	Euros / square meters	D

46Considering the availability of data, and combining data with the computation of statistical parameters (Q25, Q50, Q75, etc.) whenever it was deemed relevant, the database is structured with 84 variables for the FUA of Paris, 83 for Geneva-cross-border area, 108 for Avignon (real-estate transactions available for 2 time periods), 56 for Madrid and Palma de Mallorca (no real-estate transaction data), 79 for Barcelona (some real-estate transaction indicators) and 63 indicators for Poland. All variable names share a common coding system, whenever the datasets are fully comparable from a methodological point of view. The data repository includes one data file by case-study.

3.6 Quality control and data validation

47Two main issues had to be tackled with when analyzing the consistency of data produced. A first one has been the degree to which it was statistically reliable to match institutional and robust census data with scraped and maybe biased datasets. A second issue was how to address missing values, locally poor sampling of spatial data, and spatial biases. This section describes how such issues have been tackled with, and mitigation procedures that have been applied and approved by ESPON EGTC, in the process of publishing the datasets on ESPON Database Portal.

3.6.1 Quality of scraped data vs. institutional data

48The reliability of ‘big data’ when compared to institutional sources is indeed a complex question which does not have a single and clear answer. Some research focused on assessing the sampling bias of various ISP data sources, whether socio-demographics or spatial coverage quality assessment – i.e. statistical and spatial representativeness (Ruths and Pfeffer 2014; Longley, Adnan, and Lansley 2015; Shelton, Poorthui, and Zook 2015). Such studies have been conducted on a case-by-case basis, and this literature lacks references to quantitatively review, for a given spatial information task, the pros and cons of different types of big data when compared to traditional sources, by systematically comparing the picture one obtains with different sources.

49To avoid this lack of comparison between pricing references, we designed our research so as to be able to estimate the quality of scraped data, elaborating on researches that have conducted estimations and quality control on similar data sources. First, 2006 Beresewicz (2015) specifically addresses the question of comparing the representativeness of Internet Data Sources (IDS) compared to conventional data sources to analyze the real estate market in Poland. Institutional database on transactions allows to compare with harvested big data sources (real estate websites, open data sources and warehouses, Airbnb…). He also evaluates to which extent such online resources can be considered reliable to estimate quantitative information of the housing market. He compares basic variables, such as price per sq. m, floor area, number of rooms, to verify the consistency of data between institutional data and IDS. This research found the IDS data sources to be mostly consistent over time, except for larger apartments which are under-represented. He finally insists on the fact that "due to the non-sampling character of data obtained from the Internet, it is challenging to estimate standard errors for the estimated characteristics», and that transactions institutional data "do not contain any information on standard errors of estimates, which again limits the scope of comparison of distributions" (2006, p. 54).

50We obviously faced similar issues. We therefore sought to estimate the differences in datasets by comparing the statistical distribution of those constructed with institutional sources on the one hand, and those gathered by the means of real-estate website harvesting, for case-studies for which both sources were available. Elaborating on a very common indicator, price per sq. meter. Figure 3 shows that there are significant differences between institutional data and IPS data, especially in Avignon, Paris and Barcelona. Geneva, Annecy and Annemasse ISP data however seem to better fit the distribution of transaction data. We also plot on Figure 3 the distribution of price per sq. meters for case-studies for which only harvested data from ISP: it has not been possible to compare ISP data to other sources in these cases (Madrid, Palma and Warsaw, Łódź and Krakow).

52To illustrate this point, we analyzed data gathered with different sources for the Avignon FUA (Figure 4) and the Paris FUA (Figure 5). Both are functional urban areas for which both institutional data (transactions from the notaries) and scraped data (advertisements) were available for the study. Additionally, to better inform the validity of collected data, we used DVF fiscal data (the Etalab version), a dataset that was not available at the time of the study delivered to ESPON. We however used DVF in this paper because records are exhaustive, and are useful to benchmark the quality of scraped data against public records. Figure 4 and Figure 5 show the resulting distribution comparisons (box plots), a Kruskal-Wallis test and a pairwise Dunn test. Although box-plots demonstrate on average similar distribution patterns of means, median, Q1 and Q3 across the samples compared, the Kruskal-Wallis test demonstrates that distributions significantly differ one from another, and this is especially true with the number of rooms, which seems unreliable in scraped data, compared to other datasets. The pairwise comparison shows that transactions and fiscal data, although different, are also in general more similar in Avignon. In Paris FUA, all databases show significant differences in means, for every variable. Scraped data 2019 and BIEN 2012 however show similar means, and the pairwise test does not allow to reject the null hypothesis. Such a similarity in means and median is counter intuitive, given the tremendous effect of inflation in Paris. Interpretation of the result is therefore speculative, because of the lack of comparability of price references between BIEN and DVF (Casanova Enault et al., 2019; Casanova et al., 2017). A probable elaboration on the result would be that while the central city has maintained price growth in the upper segments of price/sq meters brackets, the rest of the metropolitan area has not followed up on the inflation curve: as a consequence Q3 and the upper 1.5 IQR threshold are much higher in 2019 than in 2012, and price inflation has increased heterogeneity in the statistical distribution of prices at the level of the entire FUA.

3.5.2 Spatial coverage and missing data: interpolation

54All indicators have been delivered at the LAU2 level. But some indicators have also been delivered at a finer grain. We describe the interest and methodology required to go beyond the municipal level by aggregating and interpolating the results in a 1km INSPIRE grid. This allows to go beyond the LAU2 level which, although relevant for policy making, is basically too large for observing existing inequalities for some cities defined by large territorial units, such as Paris, Barcelona or Warsaw. Indeed, larger geographical aggregates (LAU2) poorly perform when it comes to circumscribe the local structure of housing markets and living conditions, because of the local heterogeneity of local spatial patterns. Local submarkets, segments and regimes are often constructed at more local geographical levels, and grid cells have often been considered an ad hoc solution to provide a more detailed analysis when disaggregated data sources are available.

55We followed the approach that has already been applied in the case of transactions (Guérois and Le Goix 2009; Le Goix et al. 2019). It consists in computing a synthetic value based on distance and weight of the observed population, as initially proposed by Stewart (1942) for an analysis of the distribution of student population and catchment areas of American Universities, and more recently applied for socioeconomic phenomena (Grasland 2009). We infer that property markets are discrete social data, similar to Tobler’s hypothesis (Tobler and Wineburg 1971): a potential price for a specific location is a function of distance to nearby similar transactions, and also a function of the number of properties available, turnover and realized transactions. This method removes spatial bias, resolving the Modifiable Areal Unit Problem (MAUP). We apply Stewart’s potential to house price, and also to the number of sellers and buyers of each occupational category, using the SpatialPosition R package (Commenges and Giraud 2016; Commenges, Giraud, and Lambert 2016). Resulting values have been estimated negative exponential function, with a span of 5 km and a distance decay parameter of -2. An example of the resulting visualization is provided in Figure 6.

4 Use and possible reuse: some insights on comparing affordability indicators between cities

56We present maps and discuss some examples of the findings for a selection of FUAs, to highlight the different ways data from the ESPON Housing database can be used and compared to analyze some dimensions of affordability, for both the rental and property buyers’ markets.

4.1 Comparing advertised rental prices

57A first set of examples, in Poland, shows how the case cities (FUAs) reflect divergent paths of demographic and economic development in the last three decades, with dynamics regions i.e., Krakow and Warsaw, and on the other hand Łódź that became one of the fastest shrinking cities in Poland. On Figure 7, we compare the price of rental (scraped data), and the median national income. To clarify, the idea underlying the reference to national income being to what extent is it possible for a median household, living anywhere in Poland, to access affordable housing in any given neighborhood, in any city.

58The maps display the time required (number of days of median income) to rent 1 sq. m. The level of housing affordability generally reflects the geography of real estate offers, with lower availability in the municipalities with more expensive offers for sale and rental housing. Intriguingly, it is difficult to distinguish one common pattern for the three case cities/regions. Whereas in the Krakow region the lowest housing affordability (sales and rentals) characterizes the core city, suburban locations also show high rent-to-income ratio (low affordability) in Łódź and Warsaw. The lower housing affordability in some suburban municipalities is related to the relationship between the relative low incomes of the population that stay put, and the new wave of suburbanization bringing new housing further away from the core city. The patterns of housing affordability do not precisely fit the expected distance decay from the center, but is rather spatially structured within the three functional urban areas following a mosaic or fragmented structure; and some clusters of high and low values could also be distinguished.

4.2 The multifaceted aspects of affordability issues in Paris

59Another example of how we approached affordability can be covered with those three maps of the Paris FUA, a housing market that is considered very strong, tensed and highly unaffordable for local residents (Figure 9).

60The spatial structure of affordability shows very contrasted patterns in Paris, and this example offers an opportunity to highlight different ways of measuring affordability, with contrasting results.

• Affordability measured with reference to local income shows that buying an apartment downtown Paris will cost at least 3.8 months of full income by sq. meter, this is not true in the wealthiest part (16th arrondissement), where local income offset the cost of housing. This analysis of unequal affordability is strongly driven by the effects of local income: with lower income in the North-East of the FUA, the structure of housing is more and more inaccessible to local residents (Figure 8a).

• Rental affordability, when compared to local average income at the municipal level (Figure 8b) is highly detrimental to the population leaving in the Nord-East corridor of Paris, and generally speaking in the lower income areas of the East side of the inner suburbs.

• Analysis of affordability with reference to the national median income tells a different story: the entire western corridor is very unaffordable on this index (above the threshold of 2 months of income by sq. m.), while the North-East and South-East sections of the FUA correspond to areas where apartments would cost between 0.8 to 1.8 months / sq. m.

• On Figure 8d, data show what can be described as a differential accessibility, in other words the difference in purchasing power between national and local municipal incomes. This represents the difference between the financial effort of a local resident compared with the financial effort of an outsider willing to move in. While most of the Western side is rather unaffordable (purple), and out of reach for an average household based on national income, part of the region in yellow remains more or less affordable to an average household, while it is not affordable to local residents. This price-to-income spatial structure highlights how property markets pressurize the potential of ownership and decent access to housing in the inner ring of the region, in former blue-collar and now gentrifying neighborhoods especially.

4.3 Affordability in a segmented and heterogeneous rental market: Geneva and Annecy

61A last example of how we analyzed affordability in FUAs can be covered with the example of the cross-border region, with two FUAs: Geneva and Annecy. The housing market of Geneva-Annecy FUA is structured by a very segmented and heterogeneous rental market. Switzerland is a country of tenants, and ownership is significantly higher in France (59,9% in Savoie). Figure 9a shows that the centrality of Geneva does not structure the spatial distribution of rent on a center-periphery model: the average rent in the city center of Geneva is below rental prices in the surrounding LAU2. Higher income municipalities located on the south coast of the Leman lake (Cologny, Collonge-Bellerive, etc.) show in contrast higher prices: these are the locations the least affordable. Rentals are also more expensive in the French peripheral areas located on the Jura foothills (Pays de Gex: Divonne-les-Bains, Gex, Cessy, Sergy), and also on the Southern border, in the well connected, highly accessible, areas (highways) between France and Switzerland (St-Julien-en-Genevois). Such results from the dataset are counter-intuitive, as the tense situation of the rental market in Geneva city has often been commented. However, a main explanation for relatively “lower” rental prices in Switzerland is due to the institutional protection regarding tenants and rental prices evolution. This leads to another conception of inequalities and housing: where are rental markets the more profitable for landlords (Figure 9b). One common approach is to calculate how much one landlord should invest for 1 euro of rent. Peripheral areas stand as the most profitable, on average, which is an element to keep in mind when we want to address where property values are capitalized upon by owners and landlords.

4.4 Comparison between FUAs: a need to go further

62To summarize the current state of the data produced, Figure 10 shows how estimates based on national income distribution vs. local income offer a very contrasting views on affordability when comparing between FUAs.

• For lower-income households, Geneva, Warsaw, Krakow, Madrid and Barcelona are by far the least affordable FUAs for ownership, and for ownership only in Paris and Geneva FR. This is mitigated by the rental structure in Geneva, Paris and Avignon, because of regulations of the rental provision systems.

• For median national income reference, roughly defining the pivotal middle-class, the least affordable cities are Geneva, Warsaw, and Krakow, and to some extent Paris, for ownership applicants. Rental is the least affordable in Polish and Spanish cities.

• The high level of prices does not actually translate into an issue of “accessibility” or affordability for the 10% of the wealthiest households, but to some extent, Geneva, Warsaw, Krakow and Paris are the most exclusive markets for the wealthier part of the population, regarding ownership especially.

63One of the goals of the study was to provide a framework and some tools to bring together data from various sources. The goal of the data paper is to make this resource available broadly in a transferable and reproducible manner. The time-frame (8 months) and the resources of the study did not allow to expand the analysis to time-space dynamics of affordability, as it would require to collect and harmonize price and income across time, to amplify the scope of the database produced. This is seen as the most urgent step to take to expand this analysis. Second, with this methodological proposal, the data collected should allow to analyze the spatial patterns of inequalities stemming from unequal capitalization of housing wealth some areas, vs. vulnerability of households in others, to better link the analysis to the body of work on the variegated forms of financialization of housing in Europe. The collection of other case-studies, with a variety of situation, would help to compare and characterize some of the inequalities that structure access to housing in European cities. These are the future steps for investigation, and hopefully to soon complement the data provided.

5 Dataset description

• Name: «Unequal housing affordability across European cities» [Dataset] Nakala.

• Project and methodological framework published by ESPON HOUSING - Big Data for Territorial Analysis and Housing Dynamics (technical report) : URL : https://www.espon.eu/​big-data-housing.

• Repository type: Nakala [dataset].

• Dataset available:

• Permanent URL : https://nakala.fr/​10.34847/​nkl.aaea911g

• DOI : 10.34847/nkl.aaea911g

• Supplemental materials (a reproducible example, Barcelona case study): URL : https://nakala.fr/​10.34847/​nkl.aaea911g

5.1 Language

64English

5.2 Spatial coverage

1. FUAs and core cities. Data delivery includes two datasets on European cities (967 core cities and 676 Functional Urban Areas, FUAs). It covers the indicators available on Eurostat, which are relevant for analyzing housing issues and characteristics. These indicators have been used in the project to contextualize the 10 case-study cities, bench marked with other European cities.

2. Detailed data have been collected and harmonized for 10 case-studies in selected cities in France, Poland, Spain and Switzerland, defined by Functional Urban Areas delineations. All indicators are available at LAU2 level version 2011 (https://ec.europa.eu/​eurostat/​fr/​web/​nuts/​local-administrative-units). The indicators are also provided at the European grid level (EU 1KM GRID, GEOSTAT 2011), depending on data availability

   • Paris, FR

   • Avignon, FR

   • Warsaw, PL

   • Łódź,PL

   • Krakow, PL

   • Madrid, SP

   • Barcelona, SP

   • Palma de Mallorca, SP

   • Geneva transnational area CH, including Annecy and Annemasse, FR

65Datasets includes missing values. It corresponds generally to data gap, for instance in peripheral territorial units, less populated areas, and places with no transaction or available housing unit on property/rental markets.

5.3 Files, formats and content

66Data files have been released according to ESPON EGTC standards, on the ESPON database system. All datasets are provided with complete metadata. Two series of outputs are made available.

5.3.1 Selected Eurostat indicators, Main Data

67The following files contain original Eurostat indicators and metadata for cities and FUAs

• main-data/ind_1352_ho_diff_csv.zip: Housing survey: is it easy to find good housing in your city? - "Strongly disagree" (HO_DIFF)

• main-data/ind_1351_ho_easy_csv.zip: Housing survey: is it easy to find good housing in your city? - "Strongly agree" (HO_EASY)

• main-data/ind_1323_emp_est_csv.zip: Employment (share) - in real estate activities (EMP_EST)

• main-data/ind_1322_emp_hot_csv.zip: Employment (share) - in restauration, hotels and transports (EMP_HOT)

• main-data/ind_1321_emp_inds_csv.zip: Employment (share) - in Industry (EMP_INDS)

• main-data/ind_1313_pp65sp_csv.zip: Population (total) - age group 65+ (Pp65sp)

• main-data/ind_1312_pp4564_csv.zip: Population (total) - age group 45-64 (Pp4564)

• main-data/ind_1311_pp2544_csv.zip: Population (total) - age group 25-44 (Pp2544)

• main-data/ind_1304_pp0024_csv.zip: Population (total) - age group 0-24 (Pp0024)

• main-data/ind_1276_beds_csv.zip: Beds per resident population (BEDS)

• main-data/ind_1275_nights_csv.zip: Nights in tourist accomodation (NIGHTS)

• main-data/ind_1274_wf_hedu_csv.zip: Educational attainment level (share of age group 25-64 at level 5 to 8 ISCED - %) (WF_HEDU)

• main-data/ind_1273_st_hedu_csv.zip: Students in higher education (‰) (ST_HEDU)

• main-data/ind_1269_unemp_csv.zip: Unemployment rate (%) (UNEMP)

• main-data/ind_1268_ho_own_csv.zip: Households owning their own dwelling (%) (HO_OWN)

• main-data/ind_1267_ho_sing_csv.zip: Single households (%) (HO_SING)

• main-data/ind_1266_ho_size_csv.zip: Average size of households (HO_SIZE)

• main-data/ind_1265_ho_area_csv.zip: Housing size (average size of dwellings) (HO_AREA)

• main-data/ind_1260_pop_csv.zip: Population (total) (POP)

5.3.2 Aggregated and harmonized data

68The following data files contains compiled Eurostat indicators for 967 core cities and 676 FUAs.

• Housing European indicators - FUA: other-data/FUA_DB_41llvVF.xlsx: Housing_Europe_FUA

• Housing European indicators - Core cities: other-data/Core_cities_DB_bT7w9TL.xlsx: Housing_Europe_Core_Cities

69The following data files contains collected and harmonized data describing affordability and housing markets, at LAU2 and EU 1KM GRID levels.

• Housing indicators - Avignon (FR) : other-data/Indicators_AVIGNON_8V0nlBU.xls: Housing_Avignon

• Housing indicators - Geneva (CH-FR) : other-data/Indicators_GENEVA_MyrppRg.xls: Housing_Geneva

• Housing indicators - Warsaw - Łódź - Krakow (PL) : other-data/Indicators_POLAND_0gGnhNs.xls: Housing_Warsaw_Krakow_Lodz

• Housing indicators - Madrid (ES) : other-data/Indicators_MADRID_5vfODLL.xls: Housing_Madrid

• Housing indicators - Palma (ES) : other-data/Indicators_PALMA_vZQZpRB.xls: Housing_Palma

• Housing indicators - Paris (FR) : other-data/Indicators_PARIS_CukwCI7.xls: Housing_Paris

6 Creation date

70June 2020

Supplemental Materials

71A reproducible example, on the Barcelona case-study, is provided as an HTML RMarkdown document at the following URL (permanent data repository): https://nakala.fr/​10.34847/​nkl.aaea911g

Acknowledgements

72This paper was prepared under the ESPON project, "Big Data for Territorial Analysis and Housing Dynamics - Wellbeing of European citizens regarding the affordability of housing", which is gratefully acknowledged.

73The transactions BIEN proprietary database was made available by Paris Notaire Service, on the behalf of the Chamber of the Notaries, under an agreement contracted by the LabEx DynamiTe (ANR-11-LABX-0046) consortium and the Univ. Paris 1 Pantheon-Sorbonne. Data provided by the INSEE (1 km grid) contains public sector information, made available under ©INSEE, 2013.

74Mattia Mazzoli is funded by the Conselleria d’Innovació, Recerca i Turisme of the Government of the Balearic Islands and the European Social Fund with grant code FPI/2090/2018. J.J.R., M.M. and P.C. also acknowledge funding from the Spanish Ministry of Science and Innovation, the AEI and FEDER (EU) under the grant PACSS (RTI2018-093732-B-C22) and the Maria de Maeztu program for Units of Excellence in R&D (MDM-2017-0711).

75And infinite thanks to the Menthonnex 2019 #AcWri team: “We have opened the space for writing. Let’s make it hospitable”. The usual disclaimers apply.