THE ECONOMICS OF SOCIAL HOUSING:

IMPLICATIONS FOR WELFARE, CONSUMPTION, AND LABOR MARKET

COMPOSITION

M.A.C. Kattenberg
ISBN 978-94-91870-06-4
Tjalling C. Koopmans Dissertation Series
USE 019
Printed by Ridderprint, Ridderkerk

All maps of the Netherlands

c Kadaster / Centraal Bureau voor de Statistiek
(2013)
Map of European countries is constructed using Eurostat (2013)

c 2014 M.A.C. Kattenberg

All rights reserved. No part of this book may be reproduced or transited in

any form by any electronic or mechanical means (including photocopying,
recording or information storage and retrieval) without permission from the
author.

This book was typeset using LATEX.

 THE ECONOMICS OF SOCIAL HOUSING:
IMPLICATIONS FOR WELFARE, CONSUMPTION, AND LABOR MARKET
COMPOSITION

De economie van sociale woningbouw:

implicaties voor welvaart, consumptie en de opbouw van de
beroepsbevolking
(met een samenvatting in het Nederlands)

Proefschrift

ter verkrijging van de graad van doctor

aan de Universiteit Utrecht
op gezag van de rector magnificus,
prof. dr. G.J. van der Zwaan,
ingevolge het besluit van het college voor promoties
in het openbaar te verdedigen op
woensdag 14 mei 2014 des middags te 2.30 uur

door

Marcus Adrianus Cornelis Kattenberg

geboren op 4 september 1983 te Amsterdam

Promotoren: Prof. dr. W.H.J. Hassink
Prof. dr. J. Plantenga
Manuscript committee: Prof. dr. J.A. Bikker
Prof. dr. C. van Ewijk
dr. M. Goos
Prof. dr. J.N. van Ommeren
Prof. dr. F.G. van Oort
Acknowledgements

The acknowledgement section is often used to appreciate the people who con-
tributed to the book as well as those who contibuted to the life of the author.
Although both sets of people are intersecting, I have listed only the first group
for their contributions to this dissertation. Contributions to my private life
will be acknowledged privately.
First of all, I would like to express my gratitude to my two supervisors,
Wolter Hassink and Janneke Plantenga. Wolter, at the start of my PhD you
told me that “empirical research is a craft, one that takes time to learn”. Thank
you for your guidance throughout my learning process, you taught me a lot.
Janneke, thank you for making time in your bussy schedule to straighten the
lines of my narrative. I’m indebted to both of you.
The research undertaken in this dissertation was cofinanced by the Min-
istry of the Interior and Kingdom Relations. I thank Jaap Pot and Jeroen
van der Waart for initiating our collaboration: Without them this dissertation
would not have materialized. Also, I would like to thank them and Wouter
van Honstede, Francois Paes, Wouter Roorda and Jurjen van der Vlist for par-
ticipating in the sounding board and for their questions on the implications of
my work on theirs.
Next, I would like to thank the members of the manuscript committee for
their presence in it and their feedback on the manuscript. Grazie mille to Joep
Steegmans for reading and coloring the manuscript. I thank Adriaan Kalwij
for his advise on chapter three. Chapter seven uses data on sectoral produc-
tion and the labor force that has been collected by TNO. I am grateful to Walter
Manshanden from TNO for sharing this data with me.

i
ii Acknowledgement

I’d like to thank my colleagues at USE, and especially Marcel van den Berg,
Martijn Boermans, Han-Hsin Chang, Martijn Dröes, Wolter Hassink, Suzanne
Heijnen, Adriaan Kalwij, Sarah Rezaei Khavas and Joep Steegmans for the
mutual gains from exchanging ideas, advice and Stata or LATEX code.
Finally, I would like to thank my friends, my family and my family-in-
law for their interest in my work. Your questions encouraged me to explain
the results (and their usefullness) such that they can be understood by non-
economists. A special word of gratitude goes to my wife, Titia Schippers.
Titia, thank you for setting such a great example.

Mark Kattenberg
Haarlem, March 2014
Contents

Acknowledgements i

List of Tables v

List of Figures viii

1 Introduction 1
1.1 The economics of social housing . . . . . . . . . . . . . . . . . . 1
1.2 Institutional setting of the Dutch housing market . . . . . . . . . 4
1.3 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Insights for the future of social housing in the Netherlands . . . 15

Part A: A micro-economic perspective on rent control 17

2 Why do the rich live in social housing? 19

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Tenure choice and consumption . . . . . . . . . . . . . . . . . . . 22
2.3 Regulating entry into rent-controlled housing . . . . . . . . . . . 30
2.4 Implications for welfare . . . . . . . . . . . . . . . . . . . . . . . 34
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.A Application to CES-utility . . . . . . . . . . . . . . . . . . . . . . 42
2.B Graphical representation . . . . . . . . . . . . . . . . . . . . . . . 47

3 Consumption and housing market tenure 53

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

iii
iv Contents

3.4 Estimation strategy life-cycle consumption model . . . . . . . . 62

3.5 Results life-cycle consumption model . . . . . . . . . . . . . . . 65
3.6 Estimation strategy budget allocation model . . . . . . . . . . . 72
3.7 Results budget allocation model . . . . . . . . . . . . . . . . . . 75
3.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.A Propensity Score Matching . . . . . . . . . . . . . . . . . . . . . . 82

4 Rent control benefits and housing demand 87

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.2 Economic framework . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.4 Rent control and household mobility . . . . . . . . . . . . . . . . 100
4.5 Empirical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.A Results conditional fixed effect logit estimator . . . . . . . . . . 112

5 Type of housing tenure and wages: Sorting or incentives? 115

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.2 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.4 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.5 Estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.6 Type of housing and worker skills . . . . . . . . . . . . . . . . . 138
5.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
5.A The within estimator with many unobserved individual effects 147
5.B Classification into sectors . . . . . . . . . . . . . . . . . . . . . . 148
5.C Full tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

Part B: A macro-economic perspective on rent control 151

6 Housing and labor market rigidities: Consequences for skill compo-

sition 153
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
6.2 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
Contents v

6.3 Effect of an idiosyncratic productivity shock on labor relocation 162

6.4 Rent control and labor relocation . . . . . . . . . . . . . . . . . . 164
6.5 Wage rigidities and labor relocation . . . . . . . . . . . . . . . . 169
6.6 Rent control and wage rigidities combined . . . . . . . . . . . . 174
6.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
6.A Mathematical derivations . . . . . . . . . . . . . . . . . . . . . . 180

7 Rent control and local skill composition 187

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
7.2 Theoretical model . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
7.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
7.4 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
7.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
7.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
7.A Derivations number of workers in a city . . . . . . . . . . . . . . 220
7.B Housing in 1947 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222

8 Summary 223
8.1 Summary of this dissertation . . . . . . . . . . . . . . . . . . . . 224
8.2 Limitations and directions for future research . . . . . . . . . . . 232
8.3 Policy implications . . . . . . . . . . . . . . . . . . . . . . . . . . 234

References 237

Samenvatting 247

Curriculum Vitae 263

TKI dissertation series 265

List of Tables

1.1 Studied economic effects of social housing . . . . . . . . . . . . . 3

3.1 Sample adjustments . . . . . . . . . . . . . . . . . . . . . . . . . . 58

3.2 Budget shares . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.3 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.4 Homeownership rate for various definitions of agegroups . . . 62
3.5 Overview hypotheses explaining the level of consumption . . . 65
3.6 Results non-durable consumption (dummy specification) . . . . 66
3.7 Results non-durable consumption (interaction specification) . . 67
3.8 Results non-durable consumption (PSM specification) . . . . . . 70
3.9 Allocation of budget . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.10 Allocation of budget (PSM subsample) . . . . . . . . . . . . . . . 78
3.11 Allocation of budget (un)restricted tenants (PSM) . . . . . . . . 79
3.12 Propensity score selection regression . . . . . . . . . . . . . . . . 84
3.13 Comparison treatment and control group PSM sample . . . . . 86

4.1 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 95

4.2 LPM of probability to move . . . . . . . . . . . . . . . . . . . . . 106
4.3 LPM of moving within and from the social housing sector . . . 108
4.4 Fixed effect logit results . . . . . . . . . . . . . . . . . . . . . . . 113

5.1 Average commuting distance by housing tenure . . . . . . . . . 125

5.2 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.3 Estimation results . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.4 Effect of indicators . . . . . . . . . . . . . . . . . . . . . . . . . . 134

vii
viii List of Tables

5.5 Estimation results for various subsamples . . . . . . . . . . . . . 137

5.6 Estimation results instrumental variables . . . . . . . . . . . . . 137
5.7 Spatial sorting according to housing tenure type . . . . . . . . . 144
5.8 Coefficients and standard errors Table 5.3 . . . . . . . . . . . . . 149
5.9 Coefficients and standard errors Table 5.5 . . . . . . . . . . . . . 150

6.1 Allocation of rent-control housing vouchers . . . . . . . . . . . . 165

7.1 Summary statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 198

7.2 Share of high-skilled workers and social housing over time . . . 200
7.3 Correlation high-skilled workers and rent controlled housing . 204
7.4 Results spatial LSDV . . . . . . . . . . . . . . . . . . . . . . . . . 211
7.5 Results spatial LSDV (dummy specification) . . . . . . . . . . . 213
7.6 Spatial spillover effects . . . . . . . . . . . . . . . . . . . . . . . . 215
List of Figures

1.1 Social housing in Europe and The Netherlands . . . . . . . . . . 6

1.2 Distribution of owner-occupied and social housing value . . . . 7

2.1 Share of tenants of social housing with high income . . . . . . . 20

2.2 Share of tenants of social housing with high income and house
value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3 Contribution to the welfare loss as a function of income and
market rent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.4 Indirect utility as function of income . . . . . . . . . . . . . . . . 46
2.5 Division of income over housing and the composite good . . . . 46
2.6 Optimal choice of tenure if Y ≤ Y Al . . . . . . . . . . . . . . . . 48
2.7 Optimal choice of tenure if Y ≥ Y Al . . . . . . . . . . . . . . . . 50
2.8 Effects on welfare . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3.1 Homeownership and age . . . . . . . . . . . . . . . . . . . . . . 61

3.2 Marginal effect of house prices and age . . . . . . . . . . . . . . 68
3.3 Marginal effect house prices (PSM) . . . . . . . . . . . . . . . . . 71
3.4 Propensity score for the full and the matched sample . . . . . . 85

4.1 Median and observed marginal price of rent-controlled housing 98

4.2 Kernel density plot . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.3 Marginal prices of rent-controlled housing and house value . . 99
4.4 Transitions by destination and the rent-controlled marginal price
of housing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.5 Transitions by destination and income . . . . . . . . . . . . . . . 102

ix
x List of Figures

5.1 Density of municipal wages by housing tenure . . . . . . . . . . 123

5.2 Relative municipal wages of homeowners and tenants of social
housing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.3 Commuting distance (in km) by type of housing tenure . . . . . 125
5.4 Distribution skills and location premium . . . . . . . . . . . . . 139
5.5 Spatial distribution skills and location premium . . . . . . . . . 140
5.6 Average skills of homeowners and tenants of social housing . . 141
5.7 Moran’s I scatterplot of municipal skill level . . . . . . . . . . . 142

6.1 Rent control and the change in population in b . . . . . . . . . . 166

6.2 Inflow and outflow of workers for intermediate values of α . . . 169
6.3 Increasing labor relocation under wage rigidities . . . . . . . . . 173
6.4 Wage rigidities and the change in population in b . . . . . . . . 174
6.5 Housing and labor market rigidities combined (I) . . . . . . . . 176
6.6 Housing and labor market rigidities combined (II) . . . . . . . . 177
6.7 f (ucH ) for different values of ucT and δ . . . . . . . . . . . . . . . 180

7.1 Equilibrium at the housing market . . . . . . . . . . . . . . . . . 192

7.2 Equilibrium at the labor markets . . . . . . . . . . . . . . . . . . 193
7.3 Share of high-skilled workers by region, 1981 and 2006 . . . . . 201
7.4 High-skilled workers and local high-skilled wages . . . . . . . . 202
7.5 High-skilled workers and local house prices . . . . . . . . . . . . 202
7.6 Rent-controlled housing by region, 1981 and 2006 . . . . . . . . 203
7.7 Rent control and high-skilled workers . . . . . . . . . . . . . . . 205
7.8 Rent-controlled housing and labor market density . . . . . . . . 206
7.9 Spatial plot of residuals . . . . . . . . . . . . . . . . . . . . . . . 217
7.10 Spatial dependence in the residuals . . . . . . . . . . . . . . . . . 217
Chapter 1

Introduction

1.1 The economics of social housing

The aim of social housing is to provide dwellings to households that have
difficulties finding housing on their own. To keep social housing affordable,
rents in the social housing sector are often regulated below the market rent.1
However, this changes the costs and benefits of living in social housing, which
might in turn unintendedly alter the behavior of its inhabitants and other eco-
nomic agents.
Many economists believe that such a nominal rent freeze has several neg-
ative effects.2 It would cause an undersupply and lower quality of social
housing, while on the demand side of the housing market it would create
excess demand. In addition, social housing provides tenants an incentive not
to move, which might increase unemployment duration.3 The opposition of
these economists to rent control is perhaps best illustrated with a quote by
Lindbeck and Samuelson (1977): "Next to bombing, rent control seems to be
the most efficient technique so far known for destroying cities."
1 For this reason, I will use the terms ‘social housing’ and ‘rent-controlled housing’ inter-

changeably.
2 In a stratified random sample of 1990 American Economic Association members, 93.5 per-

cent of the participants agreed (strongly or with provisions) with the statement that rent control
reduces quality and quantity of housing, see Alston et al. (1992).
3 Although they focus on homeownership, this point is also made by Van Ewijk and Van

Leuvensteijn (2009).
2 Chapter 1.

However, rent control is hardly ever a mere nominal rent freeze. It often
entails a comprehensive set of regulations to determine a ‘fair rent’ and to
govern adjustments to the rent ceiling (over time and differentiated by loca-
tion) that reflect local housing demand or local housing costs. Rent control is
often accompanied by regulations on property rights, as well as regulations
on housing supply and maintenance. In the Netherlands, such an elaborate
institutional setting is in place, which might mitigate some of the economic ef-
fects of rent control.4 Consequently, rent control might be less distortive than
understood based on a nominal rent freeze.5
This dissertation aims to better understand the economic effects of rent
control on the local economy, especially on local labor markets. The first part
of this dissertation investigates the impact of rent control on the local econ-
omy, in particular the way self-selection into social housing is related to eco-
nomic and location amenities (chapters two and four) and whether this has
implications for the level and allocation of the consumption budget (chapters
two and three). A further subject of study is whether the wage differential be-
tween homeowners and tenants of social housing is the result of low-skilled
workers sorting into social housing, or whether living in social housing (or
homeownership) itself affects wages (see chapter five).
The second part of this dissertation studies the interaction of local hous-
ing markets and local labor markets. In isolation, both housing market price
rigidities due to rent control and labor market price rigidities reduce the geo-
graphical mobility of workers. Chapter six looks at the joint effects of housing
and labor market price rigidities on the labor location.
Chapter seven considers the effect of the share of rent-controlled housing
stock on local skill composition. As social housing is allocated to low-income
workers and as these workers are hedged against changes in the house price,
a negative effect of social housing on the share of high-skilled workers is ex-
4 These more elaborate regulatory schemes are often referred to as secondary generation rent

control, in contrast to the ‘first generation’ nominal rent freeze. See Arnott (1995); Lind (2001).
However, secondary generation rent control schemes are quite heterogeneous across countries.
See Fallis (1988) for several explanations of the heterogeneity of rent control schemes. Lind (2001);
Scanlon and Whitehead (2007, 2008) provide descriptions of contemporary rent control schemes
in Europe.
5 Turner and Malpezzi (2003) mention that specific regulations under specific market condi-

tions determine whether rent control is beneficial or not. In line with this, Arnott (1995) argues
that a well-designed rent-control program can be beneficial.
1.1. The economics of social housing 3

pected. This is an additional channel by which local housing market charac-

teristics influence local labor markets that has not explicitly been investigated
before. Table 1.1 gives an overview of the economic effects of social housing
that are considered in this dissertation.

Table 1.1: Studied economic effects of social housing

Economic effect Typea Chapter

Part A: microeconomic
Timing of housing consumption T 2
Allocation of consumption budget T 2
Level of consumption budget E 3
Allocation of non-durable consumption budget E 3
Timing of housing consumption E 4
Wages of workers E 5

Part B: macroeconomic
Location of workers T 6
Location of workers (local skill composition) E 7
a Type of research is either empirical (E) or theoretical (T).

Throughout this dissertation, the economic effects of social housing are an-
alyzed from three specific angles. The first angle is that the local housing mar-
ket status and local labor market status are mutually dependent. People differ
with respect to observable characteristics (such as age, gender, and employ-
ment industry) as well as with respect to ‘soft indicators’ such as preferences,
ability, or character. Ability and character might have important consequences
for workers’ labor market position, but might also be important for the self-
selection into social housing or owner-occupied housing. I seek to correct for
the mutual dependence of housing and labor market positions and the relation
with unobserved ‘soft’ worker characteristics. This is done using propensity
score matching, panel data estimators, and instrumental variables techniques.
A second angle that is explicitly taken into account concerns the hetero-
geneity of space. Economic activity such as the location of industries, firms
and jobs are unequally divided over the country. This also holds for local pub-
lic goods and consumer amenities such as the accessibility of natural parks,
recreation facilities and open space. These differences between areas are in-
ternalized into the market price of housing, but not into the controlled rent.
As a result, tenants of social housing in the economic core gain more from
4 Chapter 1.

rent control than tenants of social housing in the periphery. Consequently, the
economic effects of social housing differ across space.
A third angle that I consider is the spatial dependence of aggregate eco-
nomic variables. Spatial dependence reflects Tobler’s Law that “everything
is related to everything else, but near things are more related than distant
things” (Tobler, 1970). The many jobs in the Amsterdam area not only in-
crease housing demand in Amsterdam, but also in nearby regions. Spatial
econometric techniques are used to control for the spatial dependence of local
labor and housing market characteristics.

1.2 Institutional setting of the Dutch housing mar-

ket
The object of study in this dissertation is the economic effects of social housing
in the Netherlands. Generally speaking, the costs and benefits of rent control
are determined by the institutional design of the housing market. The insti-
tutional nature of the housing and land markets in the Netherlands is very
elaborate and exceeds mere rent control. Three important elements can be
discerned:

1. Social housing is ubiquitous and of relatively good quality

2. The controlled rent and its annual increase are uniformly regulated

3. Strict urban planning and zoning regulations apply

Social housing is ubiquitous and of relatively good quality

An important feature of the Dutch housing market is that social housing is

ubiquitous and of relatively good quality (see Figures 1.1 and 1.2). About one
third of the housing stock consists of social housing.6 Figure 1.1 shows that in
6 See Donders et al. (2010). There are no statistics on the exact numbers of rent-controlled

housing. However, there are several indicators and surveys that all indicate that the rent-
controlled housing sector is very large. Donders et al. (2010) and Romijn and Besseling (2008)
use the WoON-surveys. Romijn and Besseling (2008) find that 95 percent of rental houses have
a rent below the rent ceiling. An alternative statistic considers the ownership of rent-controlled
housing. According to Regeer and Van Daalen (2011), housing agencies were responsible for
managing 2.2 million houses in 2011, about one third of the total housing stock. On average 98
percent of their stock is rent-controlled (Van Daalen et al., 2012).
1.2. Institutional setting of the Dutch housing market 5

the majority of municipalities, the social housing sector makes up 21-30 per-
cent of the total housing stock. There is a positive relationship in the Nether-
lands between housing density and the share of housing that is owned by
housing agencies: In the two largest cities, Amsterdam and Rotterdam, more
than half of the total housing stock is owned by housing agencies. The magni-
tude of these numbers emphasizes the relevance of knowing how rent control
of social housing influences outcomes on local housing and labor markets.7
Figure 1.2 shows the distribution of total housing stock values (exclud-
ing private rental housing) and social housing stock values. Social housing is
clearly more dominant among the lower end of the value distribution. How-
ever, Figure 1.2 also indicates that the share of social housing in the middle
of the value distribution (between e175,000 and e225,000) is about as large
as that of owner-occupied housing. Among the higher house values, social
housing still makes up about ten percent of the housing stock. This shows
that social housing in the Netherlands is of relatively good quality.
Recent empirical studies show that the quality of housing provided by
housing agencies in the Netherlands can be said to be efficient, as the marginal
willingness to pay for social housing quality is close to its marginal costs.8
Housing agencies can provide social housing of the desired quality as they
are independent, non-profit organizations. There are several reasons why they
can offer such a large stock of social housing of relatively good quality.
First, the majority of the social housing stock was built when the hous-
ing corporations were effectively public organizations. Second, they benefit
from financial guarantees, which lower their risk premium. Third, housing
agencies often have a local monopoly or oligopoly in providing social hous-
ing (Donders et al., 2010). More often than not, they receive a discount from
the municipality if they acquire land to construct social housing. Fourth, the
7 However, the relevance of these insights is not limited to the Netherlands alone. The share

of rent-controlled housing is often substantial in other European countries as well (Scanlon and
Whitehead, 2007). Local housing markets where controlled housing forms a substantial share of
the housing stock can be found in for instance California, New York and New Jersey (Simmons-
Mosley and Malpezzi, 2006). Worldwide, about 40 percent of the population lives in rental hous-
ing with some form of controls on rents paid (Malpezzi, 1993). Furthermore, there is discussion
across Europe to adopt a more value based system to determine the rents of social housing as
applied in the Netherlands and England (Scanlon and Whitehead, 2007).
8 See Van Ommeren and Koopman (2011); Van Ommeren and De Graaf-Zijl (2013).
6 Chapter 1.

Figure 1.1: Social housing in Europe and The Netherlands

The share of the housing stock that consists of social housing for eight European countries
(Scanlon and Whitehead, 2007). 35 percent of the Dutch housing stock consists of social
housing. In comparison, this is 25 percent in Austria and 21 percent in Denmark. These
countries form the top three countries with largest share of social housing sector. In
contrast, the size of the social housing stock is only four percent in Hungary, six percent in
Germany and eight percent in Ireland.

The share of the municipal housing stock in 2010 that is rented out by a (social) housing
agency (in Dutch: “Huurwoningvoorraad in bezit van woningcorporatie”). In the majority
of municipalities, the social housing sector equals 21-30 percent of the housing stock. In
dense city municipalities such as Amsterdam, Rotterdam and Utrecht, the share of housing
that is social housing is between 35 and 50 percent.
1.2. Institutional setting of the Dutch housing market 7

Figure 1.2: Distribution of owner-occupied and social housing value

The y-axis yields the share of housing in each house value interval for all housing and social
housing (shaded). The difference between the two, is the share of owner-occupied housing
within each interval. Thus almost five percent of all housing in the Netherlands has a value
between 20 and 21 thousand euro (non-shaded bar). Two percent of all housing in the
Netherlands has a value between 20 and 21 thousand euro and is social housing (shaded
bar).

value of their properties has increased more than 200 percent over the period
1990-2006, which has increased the collateral of housing corporations. Nowa-
days, housing agencies have been selling parts of their property to finance
new construction.9
In the Netherlands, housing agencies are neither public organizations nor
private, profit maximizing corporations. Therefore, they are not effectively
restrained by administrative or market forces and could arguably be more
efficient.10

The controlled rent and its annual increase are uniformly regulated

Both the level of the controlled rent and the rent increases are regulated. At
each point in time, the rent that has to be paid for social housing equals the
9 Please note that this situation differs considerably from a private market in which profit-

maximizing housing agencies are unable to benefit from such arrangements. Turner and Malpezzi
(2003) observe that the majority of research is concerned with private markets, whereas in fact
many social housing markets are public.
10 See for instance Koning and Van Leuvensteijn (2010). They also point at a third problem,

namely that internal control of housing agencies is insufficient as well.

8 Chapter 1.

rent at the start of the contract, augmented with the annual rent increase. Rent
regulation prescribes that for social housing, the rent should always be below
the rent ceiling (in Dutch: maximaal toegestane huur), which is based on char-
acteristics of the house such as its size and the number of rooms that can be
individually heated.11 The annual increase of the rent (in Dutch: maximaal
toegestane huurverhoging) should not exceed a maximum that is based on the
inflation rate. It is possible for rents to rise in excess of the maximum allowed
rent increase after an investment in housing quality that has been approved
by the tenant.
The rent subsidy on controlled housing has resulted in longer waiting lists
for social housing, especially in popular cities. In general, social housing
within a local housing market is offered through an on-line platform open
to all households. Households looking for social housing need to register. The
on-line platform includes a description of any vacant housing and their rents.
Households can express interest in a limited number of houses. The available
housing is then generally assigned according to waiting time.12
Many households that are registered as ‘looking for housing’ do not ac-
tively search. Kromhout et al. (2006) mention that between 11 to 24 percent of
the households on a waiting list have responded to a house offer in previous
six months. Half of the households registered as ‘looking for housing’ indicate
they do so to build up waiting time. The allocation mechanism of social hous-
ing is consequently beneficial to those who can afford to wait and detrimental
to those who need social housing on short notice.
In addition to rent control, around 1.1 million households benefit from rent
support. Tenants are eligible for rent support if their income is below a cer-
tain threshold and the rent they pay for their (independent) house is below
a ceiling equal to the maximum allowed rent.13 Demand for owner-occupied
housing is subsidized as well using fiscal treatment. The rationale for this is
11 To be more precise each house characteristic is rewarded with a number of points. As of July

1st , 2008, one point is awarded for every square meter of floor space. Two points are assigned to
each room that can be heated individually by central heating. And a maximum of 25 points could
be assigned based on characteristics that are not tied to the house, such as the proximity of stores
and public parks. See Huurcommissie (2008).
12 If the house is labeled to be used by a particular group of households, it is often assigned

according to waiting time within that group of households.

13 Total rent regulations and rent support benefits accrue to e14.5 billion (Donders et al., 2010).
1.2. Institutional setting of the Dutch housing market 9

that homeownership might create positive externalities.14 As the price elas-

ticity of housing supply in the Netherlands is very low, demand subsidy for
the owner-occupied market increases house prices.15 The combination of rent
control of social housing and fiscal treatment of owner-occupied housing am-
plifies the wedge between house prices on the rental (social housing) and
owner-occupied housing markets.

Strict urban planning and zoning regulations apply

The third particular feature of the Dutch housing market is the presence of
strict urban planning and zoning regulations. These are imposed to preserve
local public goods such as open space and to regulate externalities on the
housing market. Urban planning and zoning regulations are binding restric-
tions.16 Formally, they have to be updated every ten years, but it is possible
to change them earlier by approved municipal amendments. As a results of
these regulations, the elasticity of housing supply in the Netherlands reflects
the political response to increased house prices. This might explain the low
level of price elasticity of housing supply in the Netherlands.
The planning system also has its effect on the growth of cities. The Dutch
housing policy has two conflicting goals. Since the housing shortage in the
aftermath of the Second World War, political forces have been very much con-
cerned with house construction and the house construction sector. While con-
struction ‘in city areas’ (in Dutch: binnenstedelijk gebied) is encouraged, the
construction of housing still requires the provision of residential land. At the
same time, however, the Dutch housing policy is geared towards the conser-
vation of open space. As a result, population growth did not always occur in
large cities (which could amplify agglomeration externalities), but occurred in
smaller ‘growth-cities’ (in Dutch: groeikernen).17
14 DiPasquale and Glaeser (1999) present evidence that suggest the existence of such exter-

nalities. These externalities are mentioned as a justification for the mortgage tax deduction (see
(MinVROM, 2000, p 296); MinBZK (2011)). In 2005, homeowners in the Netherlands received a
net fiscal subsidy of e14.25 billion (Donders et al., 2010).
15 See Donders et al. (2010); CSED (2010); Vermeulen (2008). Vermeulen (2008) presents evi-

dence that the price elasticity of the supply of owner-occupied housing is .04 percent, whereas
the hypothesis that the price elasticity of total housing supply is zero could not be rejected.
16 Evidence is provided by the fact that land increases in value once it is declared to be residen-

tial land, see Vermeulen (2008).

17 This might be why Zipf’s Law is found not to hold in the Netherlands, see chapter three in

Groot et al. (2010).

10 Chapter 1.

1.3 Outline
This dissertation has been divided into two parts. The first part studies how
heterogeneous workers self-select into social housing and how this affects con-
sumption expenditure, the allocation of the consumption budget to housing
and other goods, and the income of workers. It does so by presenting a the-
oretical model in chapter two and three empirical studies in the chapters that
follow. The second part considers the macroeconomic effects of social hous-
ing on local housing and labor markets. Chapter six then investigates how la-
bor market and housing market price rigidities affect the composition of cities
following changes in local workforce productivity. In closing, chapter seven
considers the effect of the social housing stock on the share of high-skilled
workers.

Chapter two

Social housing is not allocated randomly to workers. Tenants of social hous-

ing with high incomes benefit more from rent control than tenants with low
incomes (Gyourko and Linneman, 1989). Also, the prevalence of high-income
workers in social housing increases with local house prices (Van Daalen et al.,
2012). This is peculiar as tenants of social housing do not pay these market
prices. In chapter two I show that both observations can be explained by local
amenities influencing the self-selection into social housing.
It is shown that rent control changes the allocation of the budget such that
some tenants of social housing are unrestricted in their consumption of hous-
ing quality and other are restricted. ‘Unrestricted tenants of social housing’
consume on their demand curve, provided they can adjust the quality of so-
cial housing after a change in income.18
The income level at which workers decide to leave the social housing sec-
tor is shown to be determined by the level of location amenities. This refines
the argument by Glaeser (2003) that rent control leads to social mixing only
if local housing demand if high. I expand on this by arguing that the compo-
18 As the majority of moves by rent-controlled tenants is within the rent-controlled housing

sector, it is natural to assume that rent-controlled tenants can adjust the amount of rent-controlled
housing services.
1.3. Outline 11

sition of the social housing stock is biased against social mixing: In growing
cities with many location amenities, rent-controlled tenants will have higher
incomes, thereby reducing social mixing.
Finally, the welfare analysis in this chapter shows that the contribution to
the welfare loss of social housing increases with the income of the tenant and
the market rent. Therefore the welfare loss created by rent control can be re-
duced by decreasing the social housing stock or by reallocating social housing
from high-income workers to low-income workers (especially in areas with
high market rents).

Chapter three

The literature on the relationship between house prices and consumption has
been inconclusive whether the correlation is the result of wealth effects or that
consumption and house prices are driven by a common factor.19 The analy-
sis in this chapter follows the strategy of Attanasio et al. (2009) and considers
whether the national house price index over the period 1980-1999 has influ-
enced the consumption of homeowners differently than that of tenants.20
In line with the common factor hypothesis I conclude that the marginal
effect of house prices on consumption is positive and decreases with age.
Results from the propensity score matching procedure confirm the positive
marginal effect of house prices on consumption that decreases with age for
both homeowners and tenants. Based on the findings in chapter three, one
can conclude that the current crisis on the Dutch housing market is insuffi-
cient to explain the low propensity to consume in the Netherlands.
In the second part of chapter three it is tested whether housing market
tenure type leads to changes in the allocation of the budget over basic goods,
19 Campbell and Cocco (2007) find that old households increase consumption after a positive

change in house prices, whereas both young and old tenants reduce consumption, which they
interpret as evidence in favor of the wealth hypothesis. Bostic et al. (2009) estimate that the price
elasticity of consumption to house prices is about 6 percent. Using aggregate data Chen et al.
(2010) find evidence for a positive effect of house prices on consumption as well.
On the other hand, studies by Attanasio and Weber (1994); Attanasio et al. (2009) show that
young households increase consumption more in response to changing house prices than old
households, which they interpret as evidence in favor of the common factor hypothesis. They
do, however, employ different empirical strategies and use different survey years. Cristini and
Almudena (2013) consider the specifications used by Campbell and Cocco (2007) and Attanasio
et al. (2009) and conclude that evidence in favor of the common factor hypothesis is more robust.
20 The data does not allow for the separate consideration of house prices across regions.
12 Chapter 1.

housing (in general), recreation and alcohol and tobacco. Again, propensity
score matching is used to select similar household pairs. This allows for the
computing of the average treatment effect (in each year) of living in rent-
controlled housing on consumption, conditional on the consumption budget
and household characteristics.
The results suggest that tenants of social housing spend less on housing
and more on recreation. The average treatment effect has been falling over
time, which is in line with a falling difference between imputed market rents
and observed controlled rents. In line with the model of chapter two I find
evidence that tenants of social housing spend more on recreation if they are
similar in characteristics to homeowners.

Chapter four

Chapter four tests whether the income level at which workers leave the so-
cial housing sector increases with the level of local amenities. This hypothesis
follows directly from the model presented in chapter two. Previous empirical
research concludes that rent control reduces the overall propensity to move
house.21 Chapter four hypothesizes that this effect might be different for tran-
sitions from the social housing sector and transition within the social housing
sector. Making this distinction between the direction of the transitions is infor-
mative on the relationship between house prices and the allocation of social
housing, on the misallocation costs of rent control and on the rate social hous-
ing becomes available for new entrants in the short run.
Estimation results are based on a unique household panel dataset con-
structed from administrative records provided by Statistics Netherlands. The
results suggest that the effect of rent control benefits on household mobility is
large for transitions within the social housing sector, whereas it is limited for
transitions from the social housing sector. In line with the model presented in
chapter two the probability to leave the social housing sector decreases with
rent control benefits for high-income workers only. In contrast, all tenants of
21 See,
for instance, Gyourko and Linneman (1989); Ault et al. (1994); Nagy (1995); Munch and
Svarer (2002); Simmons-Mosley and Malpezzi (2006) for empirical results using various estima-
tion methods including OLS, logistic regression, and duration models.
1.3. Outline 13

social housing reduce transitions within the social housing sector if rent con-
trol benefits increase.

Chapter five

In the Netherlands workers living in rent-controlled housing earn on aver-

age 25 percent lower wages compared to homeowners. Chapter five studies
whether this wage differential is the result of the sorting of low-skilled work-
ers into social housing, as implied by chapter two. Alternatively, the wage
differential could reflect a causal relationship between housing market tenure
type. Both social housing and homeownership might influence wage rates as
they reduce labor market mobility by increasing the transaction costs of mov-
ing (De Graaf et al., 2009).
As of today no study has empirically investigated the effect of rent control
on wages. Two studies have investigated the effect of homeownership on
wages with diverging results: Munch et al. (2008) find that homeowners earn
about 5 percent higher wages than tenants, whereas Coulson and Fisher (2009)
conclude that homeowners in the United States earn on average 30 percent
lower wages.
The panel nature of the data allows to control for the sorting effect of low-
skilled workers into social housing, which is shown to be an important driver
of the wage differentials. There is no evidence that either homeownership or
living in social housing influences the wage rate.

Chapter six

Local housing markets and local labor markets are not independent. Chapter
six studies the workings of both the labor and housing market when char-
acterized by price rigidities. This is done by combining the spatial model of
Moretti (2011) with the labor market matching framework developed by Pis-
sarides (2000).
Price rigidities in the labor market enter the model in the form of nomi-
nal wage rigidities that are the result of a supra-regional bargaining process.
They reduce the mobility of labor if there are limited spatial differentials in
14 Chapter 1.

unemployment rates or if wages are not adjusted in response to local changes

in productivity.
Price rigidities in the housing market enter in the form of housing vouch-
ers that enable the holder to rent housing at the rent ceiling. Their effect on
the housing market crucially depends on the distribution of rent control hous-
ing vouchers among workers of different skill levels: If rent control housing
vouchers are allocated mainly to low-skilled workers, both the inflow of high-
skilled workers and outflow of low-skilled workers after a productivity shock
to high skilled labor will be lower. The reverse occurs if rent-control housing
vouchers are allocated mainly to high-skilled workers.
In general, housing market price rigidities are found to amplify labor mar-
ket price rigidities after a productivity shock to high-skilled labor. Vice versa,
the effect of housing market price rigidities on the relocation of high- and low-
skilled workers can be seen to be generally lower if labor market price rigidi-
ties are present. In contrast, housing market price rigidities mitigate labor
market price rigidities after a productivity shock to low-skilled labor.

Chapter seven

Moretti (2011, 2013) shows how local skill composition is driven by the net
local wage rate, adjusted for local housing costs. As rent control impacts the
housing costs in the social housing sector, it could affect the growth of cities
and their skill composition (see chapter six). This might reduce agglomeration
externalities that stem from the clustering of high-skilled workers.22 If so, this
provides a new channel by which rent control distorts outcomes on the local
labor market, in addition to its effect on unemployment.
This hypothesis is tested using longitudinal panel data on forty NUTS 3
regions in the Netherlands (in Dutch: COROPs). This data is augmented with
information on productivity and employment for eight sectors in these years
at the NUTS 3 level. I condition on the local high-skilled wage premium and
local house prices, which might be correlated with changes in (unobserved)
location amenities. This creates an endogeneity problem. Therefore these
22 Moretti (2004) presents evidence that clustering of high-skilled workers increases nominal

wages of all skill types. For the Netherlands, De Groot et al. (2013) find that the wage elasticity to
the percentage of high-skilled workers equals 2.9-5 percent (depending on skill-type).
1.4. Insights for the future of social housing in the Netherlands 15

variables are instrumented using the data on sectoral productivity and em-
ployment, as well as data on historic housing density.
Rent control is found to reduce the percentage of high-skilled workers in a
region. Ceteris paribus a ten percentage point increase of the rent-controlled
housing stock is found to reduce the percentage of high-skilled workers in a
region with 1.5 percentage points. This effect seems to be limited: If the so-
cial housing stock of Utrecht would be halved from forty to twenty percent
of the housing stock, the percentage of high-skilled workers in Utrecht would
increase with three percentage points only. The low magnitude of this effect is
presumably due to the self-selection of high-income tenants into social hous-
ing in places with many local amenities, as described in chapter two.

1.4 Insights for the future of social housing in the

Netherlands
The institutional reform of the Dutch housing sector is on the political agenda.
As I have no doubt that the social housing sector will be transformed in the
near future, I end this chapter by setting out some insights provided by this
dissertation that may contribute to the discussion on the future of social hous-
ing in the Netherlands.
It is shown in chapter three that tenants of social housing spend more on
basic goods, recreation and alcohol and tobacco. However, chapter two shows
these gains of social housing are smaller than the costs of rent control. There-
fore, the private benefits of rent control do not give an economic rationale to
increase the social housing sector, quite the contrary. Chapter four concludes
that social housing reduces the propensity to move, especially for transitions
within the social housing sector. I do not find evidence that this influences the
wage rate of workers living in social housing.
I show in chapter two that high-skilled workers self-select into social hous-
ing if house prices are high, which cannot be prevented by entry regulations
based on income. Public interest in this sorting mechanism has increased and
the pricing mechanism of social housing has been adjusted in order to prevent
16 Chapter 1.

this.23 Chapter two presents an argument against high-skilled workers living

in social housing that is based on efficiency: High-income workers contribute
more to the welfare loss than low-income workers. Therefore the welfare loss
created by rent control can be reduced by decreasing the social housing stock
or by reallocating social housing from high-income workers to low-income
workers (especially if market rents are high).24
Reducing the share of social housing that is inhabited by high-income
households is often considered to be ‘fair’. Also, it improves the efficiency on
the housing market. It should be realized, however, that this will increase lo-
cal labor market distortions. It is shown in chapter seven that while a negative
relationship between the size of the social housing stock and the percentage of
high-skilled workers exists, it is but modest in size. If one would impose leg-
islation that ensures social housing is inhabited by low-income workers only,
the effect of social housing on local skill composition is expected to be larger.
Therefore, improving the allocation of social housing to low-skilled workers
provides an ‘efficiency vs. efficiency trade-off’: It will increase efficiency on
the housing market and decrease it on the labor market.

23 As of July 2011, the rent of new contracts is allowed to increase more in attractive locations

in the country (the so-called schaarstegebieden). As of July 2013 rents in the social housing sector
were made dependent on income: Rents were allowed to increase with the inflation rate plus
2.5, 3.5 and 4 percent maximums for low-, middle-, and high-income households. Here, middle-
income households are defined as earning an annual income between e33,614 and e43,000. The
government aims to introduce a new pricing mechanism on 1 July 2014 in which rents of social
housing are based on house values, see BZK (2013).
24 The welfare loss can be eliminated by decontrolling all social housing and letting the winners

compensate the losers. However, such a policy is likely to be unfeasible.

Part A:
A micro-economic
perspective on rent control of
social housing

17
Chapter 2

Why do the rich live in social

housing?

2.1 Introduction
In many municipalities in the Netherlands a considerable share of the social
housing segment is inhabited by rich households.1 Figure 2.1 illustrates that
the proportion of social housing that is occupied by high-income households
is spatially clustered. In the north-east of the country, the proportion of the
social housing stock inhabited by households with income exceeding e33,000
is less than 25 percent, whereas in the center of the country this ratio is over
40 percent.
Also, there exists a positive correlation between house prices and the share
of social housing inhabited by rich households. This is shown in Figure 2.2
where the size of each circle reflects the size of the municipal housing stock.
The positive relationship between house value and the proportion of the social
housing stock inhabited by high-income households seems not to be driven
the size of the municipal housing stock.
1I refer to households earning 33 thousand euros annually or more as ‘high-income house-
holds’, although this equals the 2011 average affordable income (in Dutch: besteedbaar inkomen).
Statistics Netherlands (CBS) refers to them as scheefwoners.
20 Chapter 2.

Figure 2.1: Share of tenants of social housing with high income

Source data: Van Daalen et al. (2012).

The figure shows the municipal share of tenants of social housing with high
income. High-income households are defined to earn annual income equal to
e33,000 or more.

Figure 2.2: Share of tenants of social housing with high income and house
value

Source: own calculations based on Van Daalen et al. (2012); CBS (2013).
Size of the circle represents size of municipal housing stock.
2.1. Introduction 21

Entry regulations do not differ across the country, and therefore cannot ex-
plain this positive correlation. Affordability cannot explain it either, as tenants
of social housing do not pay market rents and as the rent ceiling is uniformly
regulated across the country. Instead, I argue that self-selection of heteroge-
neous workers into social housing based on location amenities generates the
positive correlation between house prices and the share of rich households in
social housing. It is shown that this self-selection mechanism increases the
welfare costs of social housing.
A key observation is that location amenities are internalized in the market
rent of housing, but not in the controlled rent of social housing. This allows
tenants of social housing to consume location amenities for free. Therefore, the
more location amenities, the higher will be the income at which workers move
to uncontrolled housing. This creates a positive correlation between income
of rent-controlled tenants and local house prices, even if entry into the rent-
controlled housing stock is uniformly regulated. It also explains why high-
income households benefit more from rent control than low-income house-
holds (Gyourko and Linneman, 1989).
This chapter presents a stylized theoretical model in which rational work-
ers maximize utility by consuming housing quality, a composite good and lo-
cal amenities. workers can consume social housing of different quality as long
as it does not exceed a predefined quality threshold.2,3 Workers self-select into
either social housing or uncontrolled housing, which allows to distinguish
three groups: Some workers will rent uncontrolled housing. The remaining
workers, living in social housing, can be subdivided into those who will con-
sume on their expansion path (‘unrestricted tenants of social housing’) and
those who will consume off it (‘restricted tenants of social housing’). This
implies that the costs associated with rent control differ from those based on
2 As the majority of moves by rent-controlled tenants is within the rent-controlled housing

sector, the assumption that rent-controlled tenants can adjust quality of rent-controlled housing
is natural. A model where tenants on rent support can choose the number of housing services
consumed is presented in Koning and Ridder (1997); Le Blanc and Laferrère (2001); Kim et al.
(2004).
3 The economic literature uses housing services instead of housing quality as the market rent

of a dwelling depends on characteristics of the house and the location amenities it provides access
to. However, as I focus on the fact that location amenities (part of housing services, but not of
house quality) are not priced in the controlled rent, using housing services is not desirable.
22 Chapter 2.

models in which all tenants of social housing consume off their demand curve
by construction.4
Amenities have been shown to be important for the location decision of
workers (see among others Brueckner et al. (1999); Glaeser et al. (2001); Van
Duijn (2013)). I show that location amenities not only influence location de-
cisions of high-income and low-income workers in the uncontrolled housing
market (Brueckner et al., 1999), but also in the rent-controlled housing market.
Also, I show that the welfare costs of rent control are increasing in location
amenities.
Finally, I show that high-income workers contribute more to the welfare
loss of rent controlled social housing than low-income workers. This implies
that the self-selection of high-income tenants of social housing increases the
welfare costs of it. Stated differently, policy that reallocates social housing
from high-income workers to low-income workers will reduce the welfare loss
of rent control.5
The remainder of this chapter is organized as follows. Section 2.2 presents
the theoretical model. Section 2.3 shows that the correlation between income
of social tenants and house prices does not disappear if entry regulations are
imposed. Welfare implications of the self-selection into social housing are dis-
cussed in section 2.4. Finally, section 2.5 concludes.

2.2 Tenure choice and consumption

I model the choice of type of housing tenure in a partial equilibrium frame-
work. As in Lyytikäinen (2006), the model is based on a standard model of
consumption of two goods with a budget restriction: workers maximize util-
ity by consuming housing quality q, a composite good x and location ameni-
ties A given their budget Y. Utility is increasing and strictly concave in the
4 Many partial market equilibrium analyses of housing demand and rent control stress that

rent control provides benefits and costs to rent-controlled tenants, see Arnott (1995); Turner and
Malpezzi (2003); Simmons-Mosley and Malpezzi (2006). In these models workers consume on
their demand curve at uncontrolled prices and suddenly are subject to rent control. As a result
they necessarily consume off their demand curve: given the rent-controlled price, they would be
willing to consume rent-controlled housing of better quality than they currently do.
5 Of course, this policy is second-best. The welfare loss of social housing can be eliminated

by decontrolling social housing. If winners compensate the losers no agent will be worse off.
However, this first best solution is likely to be politically infeasible.
2.2. Tenure choice and consumption 23

consumption of housing quality, the composite good and location amenities.6

Housing and the composite good are normal goods.
There are two types of housing with different prices: rent-controlled hous-
ing that can be rented at r̄ per unit of quality, and uncontrolled housing with
market rent rl . The subscript l refers to location. Local market rents are in-
creasing in the level of amenities and the controlled rent is independent of
local amenities.7
Workers are assigned a local housing market and are presented the option
to rent controlled housing. However, rent-controlled housing is available up
to a quality threshold q = q̄. Thus, supply of controlled housing quality is
perfectly elastic on the domain q = [0, q̄]. Supply of uncontrolled housing
quality is perfectly elastic. Workers can consume uncontrolled housing in all
regions and moving costs are zero.
Denote the type of housing tenure with τ = o for uncontrolled tenants
and home-owners, and τ = c for controlled tenants. The budget constraint of
workers equals x = Y − r τ q, where x is the numeraire good. Given the local
housing market for which the housing voucher is valid, the decision of the
worker is to maximize utility by choosing the optimal type of tenure.

max U τ ( x, q, Al ) s.t. x = Y − rτ q
x,q,τ

rl if τ = o

with rτ =
r̄ if τ = c


2.2.1 Utility of uncontrolled workers

Let U o ( x, q, Al ) denote utility if workers chooses uncontrolled housing. Un-

controlled workers maximize utility if the ratio of marginal benefits equals the
6 I do not impose further restrictions on the shape of the utility function. Section 2.A shows

results hold for a general CES-utility function.

7 Scanlon and Whitehead (2007, 2008) illustrate that across Europe a large part of the rental

housing sector is regulated based on either construction costs, income of tenants or house charac-
teristics, but not on location amenities. Furthermore, to keep the model simple, it is assumed that
the presence of rent-controlled housing does not increase house prices of uncontrolled housing.
This assumption does not drive the outcomes of the model.
24 Chapter 2.

ratio of marginal costs as in equation (2.1).

∂U o ∂U o
= rl (2.1)
∂q ∂x

Indirect utility V o is a function of income and the price of uncontrolled hous-

ing as in equation (2.2).

U o ( x, q, Al ) = V o (Y, rl ) (2.2)

Indirect utility of uncontrolled workers does not depend on the local level of
amenities, because any increase in location amenities is offset by a correspond-
ing increase in house prices, to which workers respond by changing consump-
tion such that utility does not change. This occurs whenever equation (2.3) is
satisfied.

∂rl ∂U o /∂Al 1
= (2.3)
∂Al ∂U o /∂x q

Proposition 1. No spatial arbitrage: if tenants of uncontrolled housing maximize

utility, and local house price change with the local amenity stock according to (2.3) ,
then utility of uncontrolled workers U o ( x, r, Al ) is independent of the local amenity
level.

Proof. No spatial arbitrage implies that workers in uncontrolled housing can-

not increase utility by moving to another location l. As moving is costless, no
spatial arbitrage requires that

U o ( x, q, Al ) = U o ( x, q, Al 0 ) ∀ l, l 0

This condition can only hold if the full derivative of U o ( x, q, Al ) towards Al

is zero. To solve substitute the budget constraint x = Y − rl q into the utility
function and take the full derivative of utility to local amenities:

dU o ( x, q, Al ) ∂U o ∂U o ∂q ∂U o
 
∂rl ∂q
=− q+ rl + +
dAl ∂x ∂Al ∂Al ∂q ∂Al ∂Al
2.2. Tenure choice and consumption 25

dU o ( x,q,Al )
equate dAl to zero and rearrange to get

∂U o ∂U o ∂q ∂U o
 
∂rl ∂q
q+ rl = +
∂x ∂Al ∂Al ∂q ∂Al ∂Al

Utility maximizing uncontrolled workers consume on their demand curve,

such that equation (2.1) holds. Use (2.1) to replace the first term at the right
∂U ∂q
hand side with ∂x ∂Al rl , which cancels out after opening up the brackets. Solv-
∂rl
ing for ∂Al yields equation (2.3).
Equation (2.3) implies any increase in utility from consuming more loca-
tion amenities, is offset by having to pay higher (uncontrolled) house prices.
Therefore, tenants of uncontrolled housing cannot increase utility by changing
the consumption of location amenities.

2.2.2 Utility of rent-controlled workers

Let U c ( x, q, Al ) denote utility if the worker rents controlled housing in l. The

indirect utility function is defined as V c . Note that V c depends on the local
level of amenities in location l.8

U c ( x, q, Al ) = V c (Y, r̄, Al ) (2.4)

Uncontrolled workers maximize utility if they spend their budget on housing

and the composite good such that the ratio of marginal benefits equals the
ratio of marginal costs. If they spend their budget according to this decision
rule, they could not be better off and are ‘on their demand curve’.

∂U c ∂U c
= r̄ (2.5)
∂q ∂x

As controlled housing can only be provided up to the quality level q̄, U c ( x, q, Al )

is continuous in q on the domain q = (0, q̄]. Define Y s as the income level such
that the worker is on the expansion path and the worker consumes housing of
quality q̄.

8 AsV c depends on more than prices and income, it is not an indirect utility function in the
true meaning of the word.
26 Chapter 2.

Definition 1. Y s is the income level at which workers in controlled housing consume

on their demand curve and consume housing of quality equal to q̄.

If income is below Y s , each marginal increase in income is spend on both

controlled housing quality and the composite good according to (2.5). How-
ever, if income equals at least Y s and the worker remains in controlled hous-
ing, each marginal increase in income can only be spend on the composite
good. This increases utility as well, but less than if the worker would be on
the demand curve.

Proposition 2. Indirect utility of utility maximizing workers in rent-controlled hous-

ing V c (Y, r̄, Al ) increases with income with a kink at Y s . Derivatives of utility to-
wards income are given by

∂V c (Y,r̄,Al ) ∂U c ∂x ∂U c ∂q ∂U c ∂x ∂V c (Y,r̄,Al )
∂Y |Y <Y s
= ∂x ∂Y + ∂q ∂Y > ∂x ∂Y = ∂Y |Y s <Y

Proof. workers maximize utility if equation (2.5) holds. As ∂U c /∂ql > 0 and
∂U c /∂xl > 0 each marginal increase in income is divided over both goods xl
and ql . However, as the maximum level of ql equals q̄, it follows from def-
inition 1 that V c (Y, r̄, Al ) is increasing and continuous in Y on the domain
Y = [0, Y s ). If income exceeds Y s , workers can spend each marginal increase
in income on the composite good only. As ∂U c /∂xl > 0, V c (Y, r̄, Al ) is increas-
ing and continuous in Y on the domain Y = (Y s , ∞). Therefore V c (Y, r̄, Al )
has a kink at Y s .

Proposition 3. Indirect utility of workers in controlled housing increases with the

local amenity level.

Proof. Substitute x c = Y − r̄qc into the utility function for controlled tenants
and take the full derivative to local amenities:

dU c ( x, q, Al ) ∂U c ∂U c ∂q ∂U c
 
∂r̄ ∂q
=− q+ r̄ + +
dAl ∂x ∂Al ∂Al ∂q ∂Al ∂Al

∂r̄
Note that ∂Al = 0 as the rent ceiling does not depend on local amenities.
Workers in rent-controlled housing maximize utility if equation (2.5) holds.
∂U c ∂q c ∂q
Use (2.5) to substitute ∂q ∂Al = r̄ ∂U
∂x ∂Al , which then cancels out. Therefore,
2.2. Tenure choice and consumption 27

the full derivative of utility of controlled workers to the local amenity stock
equals the partial derivative (which is positive). As a result, the indirect utility
function of rent-controlled workers depend on income, the rent ceiling and the
local level of amenities as in equation (2.4).

2.2.3 Choice of tenure type and consumption quantities

Utility of uncontrolled workers with income Y is equal across the country as

location amenities are perfectly priced into house prices (proposition 1). Yet,
utility of rent-controlled workers with income Y varies across the country with
the level of location amenities (proposition 3). In this subsection, I will show
that workers choose to live in controlled or uncontrolled housing in location l
depending on their income and the amenity level.

Proposition 4. Workers with income equal to or below Y s rent controlled housing.

Proof. Workers in uncontrolled and controlled housing maximize utility if equa-

tions (2.1) and (2.5) hold. As controlled housing can be rented up to the quality
threshold q̄, equation (2.5) can only hold if Y ≤ Y s . Divide equation (2.1) by
(2.5) and solve for rl /r̄ to derive equation (2.6).

rl (∂U o /∂q) / (∂U o /∂x )

= (2.6)
r̄ (∂U c /∂q) / (∂U c /∂x )

As rl /r̄ > 1 for all l and utility is concave and increasing in the composite
good x and housing quality q, equation (2.6) implies that controlled workers
can consume more of the composite good than uncontrolled workers given
equal consumption of housing quality q. As they can consume more, utility of
rent-controlled tenants exceeds that of uncontrolled workers if Y ≤ Y s .

Next, define Y Al as the income level such that the worker is indifferent be-
tween living in controlled and non-controlled housing. It follows from propo-
sition 3 that Y Al increases with the local amenity level Al given rl .

Definition 2. Y Al is the income level such that the worker is indifferent between
living in controlled and non-controlled housing.
28 Chapter 2.

Proposition 5 states that workers with income below Y Al will rent con-
trolled housing, whereas workers with income exceeding Y Al will rent uncon-
trolled housing.

Proposition 5. Utility maximizing workers with income exceeding Y Al will con-

sume uncontrolled housing of quality exceeding q̄ at the market rent rl .

Proof. Workers with income Y Al are indifferent between renting controlled or

uncontrolled housing (Definition 2). Workers in uncontrolled housing with
income Y Al consume more housing quality (q > q̄) and fewer units of the
composite good than workers with equal income in controlled housing.
Utility is increasing and concave in consumption of the composite good.
Therefore, if both types spend any marginal increase in income solely on the
composite good, utility of uncontrolled workers with income Y Al increases
more than utility of controlled workers with this income level. Thus workers
with income exceeding Y Al are better off renting uncontrolled housing.

Proposition 6. Utility maximizing workers with income in between Y s and Y Al will

consume rent-controlled housing of quality q̄.

Proof. Based on Definition 2 and Proposition 5 the following equations holds:

V o (Y Al , rl ) =V c (Y Al , r̄, Al ) (Definition 2) (2.7)

V o (Y, rl ) >V c (Y, r̄, Al ) if Y > Y Al (Proposition 5) (2.8)

As V o is monotonically increasing in Y and V c is a monotonically increas-

ing function in Y on the domain Y = (Y s , ∞), it follows that V o (Y, rl ) <
V c (Y, r̄, Al ) for Y s < Y < Y Al .

Propositions 4, 5 and 6 state that the optimal tenure type is a function of

income and location amenities. Workers with income equal or smaller than
Y Al will rent controlled housing, whereas workers with income exceeding will
rent uncontrolled housing.
Note that the group of rent-controlled workers consists of two subgroups.
The first group with income Y ≤ Y s consumes housing of quality q ≤ q̄ at
the controlled rent. Their consumption bundle is optimal as they consume
2.2. Tenure choice and consumption 29

according to equation (2.5). Hence they are unrestricted tenants that consume
on their demand curve.
The second group of rent-controlled workers with income Y s < Y ≤ Y Ac
would be willing to consume housing of higher quality than q̄ at the controlled
rent r̄. Thus they consume off their demand curve. As their desired bundle is
infeasible they are referred to as restricted tenants of rent-controlled housing.

Definition 3. Unrestricted tenants of rent-controlled housing are workers who rent

controlled housing and consume on their demand curve. By proposition 4, all workers
with income Y ≤ Y s are unrestricted tenants of rent-controlled housing.

Definition 4. Restricted tenants of social housing are workers who rent social hous-
ing and consume off their demand curve. By proposition 5, they are defined as workers
in l with income Y s < Y < Y Al .

For unrestricted tenants of rent-controlled housing and tenants of uncon-

trolled housing, consumption of the composite good and housing quality is
defined by the expansion path and budget restriction. Restricted tenants of
controlled housing spend r̄ q̄ on housing and Y − r̄ q̄ on the composite good.
Thus consumption of housing quality and the composite good is given by
equation (2.9), where the superscript ∗ indicates the value is in accordance
with equation (2.1) or equation (2.5).

Y − r τ q ∗

if Y ≤ Y s or Y ≥ Y Al
x= (2.9)
Y − r̄ q̄ if Y s < Y < Y Al


A graphical representation of the model is presented in section 2.B.

2.2.4 Variation in income across local housing markets

This subsection discusses how spatial variation in location amenities creates

spatial variation in the local income level of tenants of social housing as shown
in Figure 2.2. For simplicity, consider a region with workforce of mass equal to
one. Let f (Y ) denote the probability density function of income. Recall that all
tenants with income equal or below Y Al will live in rent-controlled housing.
Thus the share of workers in rent-controlled housing sl is given by equation
30 Chapter 2.

2.10.

Z Y Al
sl = f (Y )dY (2.10)
0

o l =1 − s l

Average income of tenants in rent-controlled housing av(Y )sl is given by equa-

tion (2.11) and increasing in Y Al .

Z Y Al
av(Y )sl = f (Y )YdY (2.11)
0

Consider an arbitrary income threshold Ỹ that is lower than Y Al . Let s+

l denote
the share of workers in rent-controlled housing for whom income exceeds this
arbitrary threshold.

R Y Al
f (Y )dY
s+
l = RỸ A (2.12)
Y l
0 f (Y )dY

It follows from equation (2.12) that the share of all rent-controlled tenants
with income exceeding Ỹ increases with Y Al . As both uncontrolled house
prices and Y Al are increasing in the level of local amenities, equation (2.12)
implies there is a positive relationship between the share of rent-controlled
workers with income exceeding a certain threshold value and the level of lo-
cation amenities in the housing market.

2.3 Regulating entry into rent-controlled housing

This section shows that entry regulations of the social housing stock does not
alter the outcomes of the model. As a result entry regulations are unable to
prevent that social housing is rented by high-income workers.
The model is adapted to contain two periods. Workers perfectly smooth
income over both periods, such that expected income in both periods is equal.
At the start of each period, workers are allowed to change housing tenure. Al-
location of rent-controlled housing is regulated based on income: Only work-
ers with income equal or below Ȳ are allowed to enter rent-controlled hous-
2.3. Regulating entry into rent-controlled housing 31

ing. Ensure the constraint on income is binding for restricted tenants of rent-
controlled housing by assuming Y s < Ȳ < Y r,A in all l.9
To study how location amenities influence the income at which households
change tenure type, I allow for a positive income shock γ that occurs to all
workers in between period one and two.10 Let subscripts denote time peri-
ods, such that Y1 is income in period one and Y2 = Y1 + γ equals income in
period two. A crucial assumption, in line with regulations on rent-controlled
housing in the Netherlands, is that the income ceiling no longer applies once
workers live in rent-controlled housing. Hence the exit out of rent-controlled
housing is not regulated.

The equations on page 32 summarize the outcomes of the model. As the in-
come ceiling is binding, all workers with income below the income ceiling will
rent social housing in period one and those above it will rent uncontrolled
housing. Consumption of housing increases with income for unrestricted ten-
ants of rent-controlled housing. This is also true for tenants of uncontrolled
housing, who are those workers with income exceeding the income threshold
Ȳ.
In the second period the (binding) income ceiling for rent-controlled hous-
ing no longer determines who rents controlled or uncontrolled housing. Rent-
controlled workers will move to the uncontrolled housing sector in the second
period, only if the income shock raises income above Y Al . As the maximum
income of rent-controlled workers in period one equals Ȳ and all workers re-
ceive the unexpected income shock γ, workers with income in period two
equal to or smaller than min(Ȳ + γ, Y Al ) will be living in rent-controlled hous-
ing. All other workers will rent uncontrolled housing in the second period.
In the first period, all workers with income below the income threshold
will be living in rent-controlled housing, see equation (2.13).

Z Ȳ
0
s1 = f (Y1 )dY1 (2.13)
0

< Y s yields similar results. The income constraint is not binding if Y Al < Ȳ.
9 Ȳ
10 It
occurs to all workers such that the share of workers in rent-controlled housing exceeding
a certain threshold is tractable. Results do not depend on this assumption.
 32
Tenure choice and consumption of the composite good and housing quality in each period are given by the following equations:

Choice of tenure

First period Second period

 
c if Y1 ≤ Ȳ c if Y2 ≤ min(Ȳ + γ, Y Al )
 
τ1 = τ2 =
o

otherwise o

otherwise

Consumption

First period Second period

 
∗ if Y1 ≤ Y s ∗ s
Y1 − r̄qc Y2 − r̄qc , if Y2 ≤ Y

 


 

x1∗ = Y1 − r̄ q̄ if Y s ≤ Y1 ≤ Ȳ x2∗ = Y2 − r̄ q̄ if Y s ≤ Y2 ≤ min(Ȳ + γ, Y Al )

 

 
Y − r o q ∗

if Ȳ < Y1 Y − r o q∗ , if min(Ȳ + γ, Y Al ) < Y

1 c 2 c 2
Note that Y r,A < Ȳ + γ indicates that some workers who were rent-controlled in period one switch to uncontrolled housing in period

Chapter 2.
two. In contrast, if Ȳ + γ < Y r,A those workers who rented controlled housing in period one, will still do so in period two.
2.3. Regulating entry into rent-controlled housing 33

The share of workers in rent-controlled housing in period 2 is given by equa-

tion (2.14). The upper limit of the second integral contains Y S + γ as all work-
ers with income equal or below Y s in period one rented controlled housing.

Z min(Y Al ,Ȳ +γ)

0
s2 = f (Y2 )dY2 (2.14)
0

Therefore, the average income of workers in rent-controlled housing in peri-

ods 1 and 2 equals:

Z Ȳ
av(Y )s0 = f (Y1 )Y1 dY1 (2.15)
1 0
Z min(Y Al ,Ȳ +γ)
av(Y )s0 = f (Y2 )Y2 dY2 (2.16)
2 0

Consider an income threshold Ỹ that is lower than the minimum of Y Al and

0
Ȳ + γ. Let s2+ denote the share of workers in rent-controlled housing in period
two for whom income exceeds this threshold (relative to the number of rent-
controlled workers). Equation (2.17) increases with Y Al as long as conditions
are such that some workers change tenure type because of the income shock
(provided Y Al < Ȳ + γ).

R min(Y Al ,Ȳ +γ)

+0
f (Y2 )dY2
s2 = RỸ , (2.17)
min(Y Al ,Ȳ +γ)
0 f (Y2 )dY2

As both uncontrolled house prices and Y Al are increasing in the level of local
amenities, equation (2.17) implies there is still a positive relationship between
the share of rent-controlled workers with income exceeding a certain thresh-
old value and the level of location amenities in the housing market.
Compare equation (2.10) to (2.14) and equation (2.12) to (2.17) under the
assumption that Y = Y2 . They indicate that size of the rent-controlled hous-
ing stock (in period two) and the distribution of income among workers in
rent-controlled housing is not affected by the income threshold if the income
at which workers will move from the social housing sector is lower than the
income ceiling plus the positive income shock. Thus regulating entrance into
rent-controlled housing on income does not change the allocation of rent-
34 Chapter 2.

controlled housing if regulations on entry are loosely binding (Ȳ is high) or

in locations with few location amenities (Y Al is small).

2.4 Implications for welfare

The previous part of this chapter has shown that the income of workers and
location amenities determine the self-selection of high-income workers into
social housing. Entry regulation based on income could not prevent this self-
selection. In this section, I will discuss the welfare implications of this sorting
mechanism on the housing market.
The costs of rent control equal the rent the lessor of social housing forgoes.
Equation (2.18) defines these costs, where demand for housing as a function
of income is given by q∗ = D (Y ). Given the price difference between social
housing and uncontrolled housing, the costs of rent control are increasing in
income as long as Y < Y s and remain constant as long as Y = [Y s , Y Al ]. If
Y Al < Y the costs of rent control become zero as workers will rent uncon-
trolled housing.

(rl − r̄ ) D (Y ) if Y ≤ Y Al

C (Y, rl ) = (2.18)
0 if Y Al < Y


Next, define the benefits of rent control B(Y, rl ) (in equivalent variation) as the
increase in income a worker in uncontrolled housing would require in order
to be indifferent between renting social or uncontrolled housing.

Definition 5. Equivalent variation B(Y, rl ) is the income increase required by work-

ers in uncontrolled housing to be indifferent between renting social housing or uncon-
trolled housing: V o (Y + B(Y, rl ), rl ) = V c (Y, r̄, Al ).

It follows from Definition 2 that B(Y Al , rl ) = 0. 11 As housing is a normal

good, workers substitute away from consuming housing quality if the rent
11 Workers with income exceeding Y A L do not live in social housing, and therefore their bene-

fits from rent control are set to zero. Would they be living in social housing, Proposition 5 implies
that workers with income exceeding Y A L would be willing to give up income (B(Y, rl ) < 0) in
order to leave the social housing sector.
2.4. Implications for welfare 35

increases. This implies that the costs of rent control of social housing are at
least as large as the benefits of social housing and rent control is inefficient.
Let x ∗ (Y + C (Y, rl ), rl ) be the utility maximizing consumption of the com-
posite good if the worker would receive the costs of rent control C (Y, rl ) in
addition to his income Y while renting uncontrolled housing. Let x ∗ (Y, r̄ ) de-
note the optimal consumption of the composite good if the worker would rent
social housing at the controlled rent r̄. As x is the numeraire, these quantities
can be used to compute the welfare loss if income is at most Y s . Therefore, the
welfare loss of rent control is given by equation (2.19).

L(Y, rl ) =C (Y, rl ) − B(Y, rl )


 x ∗ (Y + C, rl ) − x ∗ (Y, r̄ )

 if Y ≤ Y s


L(Y, rl ) = q̄(rl − r̄ ) − B(Y, rl )
 if Y s < Y ≤ Y Al (2.19)


if Y ≤ Y Al

0

Proposition 7. Rent control of social housing creates a welfare loss if Y ≤ Y s .

Proof. The proof follows from the fact that for any discrete price change, Slut-
sky compensated demand will be larger than Hicksian compensated demand
(Mas-Colell et al., 1995, p. 72). The welfare loss for someone with income
Y ≤ Y s who lives in a city with market rent rl equals L(Y, rl ) = x ∗ (Y +
C (Y, rl ), rl ) − x ∗ (Y, r̄ ). L(Y, rl ) is positive if x ∗ (Y + C (Y, rl ), rl ) > x ∗ (Y, r̄ )
which occurs if Y < Y s .
Definition 5 implies that the points [ x ∗ (Y + C (Y, rl ), rl ), q∗ (Y + C (Y, rl ), rl )]
and [ x ∗ (Y, r̄ ), q∗ (Y, r̄ )] are on the same indifference curve. Also, as x ∗ (Y +
C (Y, rl ), rl ) and x ∗ (Y, r̄ ) are interior optimal values, they are described by
equations (2.1) and (2.5). As r̄ < rl and the utility function is strictly con-
cave, the two bundles can only be described by equations (2.1) and (2.5) if
x ∗ (Y + C (Y, rl ), rl ) > x ∗ (Y, r̄ ) and q∗ (Y + C (Y, rl ), rl ) < q∗ (Y, r̄ ). Therefore,
the costs of rent control always exceed the benefits if Y < Y s .

Proposition 8. Rent control of social housing creates a welfare loss that increases
with income Y and the market rent rl if Y < Y s .
36 Chapter 2.

Proof. Define Y < Y 0 such that Y, Y 0 < Y s . Use equation 2.19 to write:

L(Y 0 , rl ) − L(Y, rl ) > 0

 ∗ 0
x (Y + C (Y 0 , rl ), rl ) − x ∗ (Y + C (Y, rl ), rl ) > x ∗ (Y 0 , r̄ ) − x ∗ (Y, r̄ ) (2.20)
  

Thus the welfare loss increases with income if the change in optimal consump-
tion of the composite good if the worker were compensated in equivalent in-
come for renting uncontrolled housing increases more than the change in opti-
mal consumption of the composite good if the worker’s income increases and
he would rent social housing.
Again, the quantities in equation (2.20) are interior optimal solutions that
are described by equations (2.1) and (2.5). As r̄ < rl and housing and the com-
posite good are normal goods, workers will consume more of the composite
good if (compensated variation of) income increases. This implies equation
2.20 holds and therefore the welfare loss from living in rent-controlled social
housing increases with income.
Next, define rl < rl 0 to prove that the welfare loss increases with the mar-
ket rent. Workers in the uncontrolled market consume on their expension path
given by equation (2.1) and the composite good is a normal good. The defi-
nition of a normal good states that its consumption is increasing in income.
Therefore, x ∗ (Y + C (Y, rl 0 ), rl 0 ) > x ∗ (Y + C (Y, rl ), rl ) if C (Y, rl 0 ) > C (Y, rl ).
Equation 2.19 shows this condition is true.

It follows from equation 2.18 that the costs of providing rent control do not
increase with income once income exceeds Y s , simply because consumption
of housing quality cannot increase. However, as the benefits from rent control
decrease if income goes to Y Al from below, the welfare loss of providing rent
controlled social housing remains positive.

Proposition 9. Rent control of social housing creates a welfare loss if Y s < Y ≤ Y Al ,

that is increasing in income.

Proof. The welfare loss of rent control equals the costs of rent control minus
the benefits. It follows from equation 2.18 that the welfare costs of rent control
are constant on the domain Y = (Y s , Y Al ]. Thus for the welfare loss to increase
2.4. Implications for welfare 37

with income, it is sufficient that the benefits of rent control decrease with in-
come on this domain. From Definition 5 it follows that Y + B(Y, rl ) ≤ Y Al for
any value of Y = [Y s , Y Al ]. This can only be true if B(Y, rl ) is decreasing in
income.
Assume that benefits of rent control B0 (rl ) would not decrease in income
but remain constant. Then there would exists a value of income Y A L − B0 (rl ) <
Y 0 < Y Al such that compensating variation of income exceeds Y Al . This im-
plies that indirect utility V o (Y 0 + B0 (rl ), rl , Al ) > V o (Y Al , rl , Al ). This can-
not be true as Definitions 2 and 5 and Proposition 2 require that V o (Y 0 +
B0 (rl ), rl ) < V o (Y Al , rl ). Thus benefits of rent control cannot be constant
(nor increase), and therefore the benefits of rent control should decrease in
income if Y = [Y s , Y Al ]. As on this domain the costs of rent control are con-
stant, this implies that the welfare loss of rent control is increasing in income
if Y = [Y s , Y Al ].

Proposition 10. Rent control of social housing creates a welfare loss if Y s < Y ≤
Y Al , that is increasing in the market rent.

Proof. This follows from the proof of Proposition 8.

Propositions 7 to 10 are summarized in Figure 2.3. Figure 2.3 illustrates

that rent control is inefficient as it creates a positive welfare loss if income ex-
ceeds zero. The Hicks-Kaldor principle that ‘winners can compensate losers’
holds as all workers in social housing would be better off if they would be
compensated for renting uncontrolled housing by receiving the welfare costs
in addition to their income. However, few schemes in which winners com-
pensate losers are observed, indicating that designing such a scheme might be
unfeasible.
Figure 2.3 also illustrates that the contribution to the welfare loss increases
with income. Conditional on the market rent high-income workers in social
housing contribute more to the welfare loss than low-income workers. There-
fore, the misallocation costs of rent control are not only larger if housing is
allocated randomly among tenants (Glaeser and Luttmer, 2003), but that self-
selection of high-income tenants into social housing increases the misalloca-
tion costs of rent control even more. This implies that the welfare loss of rent
38 Chapter 2.

control can be reduces by reallocating social housing to low-income workers.

Regulation that limits the income level of tenants of social housing will reduce
the welfare loss of social housing. This might provide a politically feasible
mechanism by which the welfare loss of rent control might be reduced.
Finally, Figure 2.3 shows that social housing is more distortive if the mar-
ket rent increases. This has two reasons. First, this increases the slope of the
function that describes the contribution to the welfare as a function on in-
come. Second, it increases the income level at which workers remain in the
0
social housing sector from Y Al to Y Al .
2.4. Implications for welfare 39

Figure 2.3: Contribution to the welfare loss as a function of income and market
rent

6
c q̄(rl 0 − r̄ )
Contribution to welfare loss

c q̄(rl − r̄ )




t
en
tr
ke
ar
m

t
ren
h

et
ig

k
H

r
ma
o w
L

  c c -
0
Ys Y Al Y Al Income

Contribution of a worker with income Y to the welfare loss from social housing
for different levels of market rent. The contribution to the welfare loss is always
positive (Propositions 7 and 9). The contribution to the welfare loss is larger if
market rents are high (Propositions 8 and 10) and increasing in income (Propo-
sitions 8 and 9).
The contributions to the welfare loss are defined up to the point Y Al as workers
with higher income rent uncontrolled housing. Y Al is increasing in the market
rent, see Proposition 3 and Definition 2. In the point Y Al the contribution to the
welfare loss equals the costs of rent control as the benefits from rent control are
zero: B(Y Al , rl ) = 0 (Definitions 2 and 5). The costs of rent control are given
by q̄(rl − r̄ if Y = [Y s , Y Al ], see equations (2.9) and (2.18).
40 Chapter 2.

2.5 Conclusion
In attractive cities the rich live in social housing as this enables them to con-
sume the location amenities for free. This mechanism has been explained us-
ing a stylized theoretical model in which rational workers maximize utility
by consuming housing quality, a composite good and local amenities. The
more location amenities are present, the higher will be the income level at
which rent-controlled tenants start renting uncontrolled housing. This ex-
plains the positive relation between uncontrolled house prices and income of
rent-controlled tenants and why tenants with high-income benefit more from
rent control.
A contribution of the model is that it allows for the self-selection of high-
skilled workers into social housing if the market rent is high. As a result, three
types of workers can be considered.
The first group consists of workers with income below Y s live in social
housing and consume on their demand curve, as the optimal level of social
housing quality does not exceed the quality threshold. The second group of
workers has an income between Y s and Y Al . workers in this group maximize
utility if they consume the maximum quality of rent-controlled housing and
spend the remainder on the composite good. These workers are restricted in
their consumption and consume off their demand curve. The third group con-
sists of workers with income exceeding Y Al . This group lives in uncontrolled
housing and consumes on the demand curve again.
Another contribution of the model is that it illustrates how the self-selection
of high-skilled workers into social housing influences the welfare loss of social
housing. Previous research has indicated that the welfare loss of rent control
might be larger if social housing is allocated randomly (Glaeser and Luttmer,
2003). I show that self-selection into social housing increases the welfare loss
of social housing.
I find that social housing is always inefficient: The contribution of a worker
to the welfare loss is positive regardless of his income. However, the contri-
2.5. Conclusion 41

bution to the welfare loss is increasing in both income and the market rent.
Therefore, policy that reduces the self-selection of high-income workers into
social housing if market rents are high will reduce the welfare loss associated
with rent control. The same holds for policy that reallocates social housing
from high-income workers to low-income workers.
The result that the income level at which workers move from the rent-
controlled housing segment depends on local amenities, has important impli-
cations for the effect of rent control on local inequality. Glaeser (2003) men-
tions that rent-control reduces segregation only in growing cities as home-
owners in declining cities will have lower income (are more similar to rent-
controlled tenants). In addition to this, the model in this chapter suggests that
rent-controlled tenants are more similar to uncontrolled tenants in growing
cities, because of the self-selection of high-income into controlled housing in
these places.
In the remaining chapters I test the model’s implications for consumption
and tenure choice. Equation (2.9) indicates that consumption patterns dif-
fer between tenants who rent controlled and uncontrolled housing, because
the housing rent influences consumption of the composite good as well. In
chapter three I study whether the budget shares devoted to basic goods, hous-
ing, recreation and alcohol and tobacco are different for home-owners (uncon-
trolled workers) and workers in rent-controlled housing. Equation (2.9) also
shows how consumption patterns of restricted rent-controlled workers differ
from those of unrestricted tenants of rent-controlled housing. This is tested in
chapter three as well.
In chapter four I test the hypothesis that the income at which workers con-
sume uncontrolled housing is increasing in the amenity level. I do so by con-
sidering the relationship between the mobility of workers and the difference
between the controlled rent and market rent of their house. The panel di-
mension of the data allows to control for the sorting of heterogeneous work-
ers (with different ability or preferences) into rent-controlled housing across
the country. In addition, I distinct between moving house within the rent-
controlled housing sector and between moving house out of the rent-controlled
housing sector into the uncontrolled housing sector.
42 Chapter 2.

2.A Application to CES-utility

I follow Brueckner et al. (1999) and show that our results apply to a general
three input CES-utility function. Consider the following application in which
workers of type τ = o, c maximize a CES-utility function (with constant re-
turns to scale) subject to a budget constraint.

max U s.t. Y = px + r τ q, r τ = rl , r̄
1
ρ
ρ
U ( x, q, Al ) = ax ρ + bqρ + (1 − a − b) Al

Note that only the composite good x and housing quality q are priced and
enter the budget constraint. The level of location amenities is internalized into
the local price for uncontrolled housing r o = rl , but not in that of controlled
housing r c = r̄. As before, I have rl > r̄ for all l.
If an interior solution exists utility maximizing workers will spend their
budget Y over housing and the composite good such that the ratio of marginal
benefits equals the ratio of marginal costs. The expansion path is given by:

 −σ
b p 1
x= q, σ= 1− ρ
a rτ

and the (potentially infeasible) consumption of housing quality and the com-
posite good that maximizes utility equals
 
∗τ aσ Y
x = σ
p a σ p 1− σ + b σ ( r τ )1− σ
σ   (2.21)
∗τ b Y
q = τ σ
(r ) a σ p 1− σ + b σ ( r τ )1− σ

x ∗ τ and q∗ τ in (2.21) only maximize utility if they are feasible and if they are an
interior solution. The model assumes that any level of uncontrolled housing
quality can be supplied, and therefore the combination of x ∗ o and q∗ o in equa-
tion (2.21) maximize utility for tenants in uncontrolled housing. For controlled
tenants this is slightly different as controlled housing can only be supplied up
to q̄. The definition of Y s states that for tenants in controlled housing an inte-
rior solution according to (2.21) only exists as long as income is not larger than
2.A. Application to CES-utility 43

Y s . This allows to solve for Y s as

r̄ σ  σ 1−σ σ 1− σ

Y s = q̄ a p + b r̄
bσ

Consequently, indirect utility of home-owners is given by equation (2.22).

 1− ρ 1
σ 1− σ ρ ρ
o
V (Y, rl , Al ) = a p + bσ rl1−σ Y + (1 −
ρ
a − b) Al (2.22)

The indirect utility function of rent-controlled workers is given by equa-

tion (2.23).

i1

h
1− ρ ρ
a σ p 1− σ + bσ r̄1−σ
ρ ρ
if Y ≤ Y s


 Y + (1 − a − b ) A l
V c (Y, r̄, Al ) = h  ρ i1
 a Y −r̄q̄ + bq̄ρ + (1 −
 ρ ρ
a − b) Al if Y ≥ Y s

p

(2.23)

Equations (2.22) and (2.23) show that workers with income Y ≤ Y s are al-
ways better off consuming rent-controlled housing. Note that they maximize
utility by consuming on their demand curve.
Y Ac is the income level at which a tenant is indifferent between consum-
ing controlled housing of quality q > q̄ at the market rent rc or consuming
uncontrolled housing quality q = q̄ at the controlled rent r̄. As tenants of rent-
controlled housing with income exceeding Y s are in a corner solution, solving
for Y Ac is tedious. However, proving its existence is less cumbersome.

Proposition 11. The income level at which tenants are indifferent between renting
controlled or uncontrolled housing exists.

Proof. Note that if Y = Y s workers prefer to rent q̄ housing quality at the

controlled rent over renting uncontrolled housing: V o (Y s , r l ) < V ( Y s , r̄, Al ).
Next, given their income level, controlled tenants who consume controlled
housing for free, are always better off than tenants who pay a positive con-
trolled rent r̄. Suppose there is an income level Ỹ Ac such that a worker is in-
different between consuming uncontrolled housing at the market rent, or con-
suming rent-controlled housing q̄ for free. If Ỹ Ac exists workers with income
Ỹ Ac prefer to consume uncontrolled housing at the rent rl over consuming un-
44 Chapter 2.

controlled housing at a positive controlled rent: V o (Ỹ Al , r l ) > V ( Ỹ Al , r̄, Al ) if

0 < r̄ < rl .
As V c and V o are monotonic increasing functions in Y the intermediate
value theorem states that Y Al exists, provided that Ỹ Al exists.
First, define Ṽ c as the indirect utility function if Y > Y s and r̄ = 0.

h  ρ i1
Ṽ c = a Yp + bq̄ρ + (1 − a − b) Al
ρ ρ
(2.24)

Ỹ Al is the income level for which Ṽ c (Ỹ Al , r̄ = 0, Al ) − V o (Ỹ Al , rl , Al ) = 0.

Define ∆ as the difference of both indirect utility functions taken to the power
γ and plug in equations (2.22) and (2.24):

 ρ  ρ
∆ ≡ V O − Ṽ C = γY ρ − bq̄ρ (2.25)
  1− ρ 
σ 1− σ σ 1− σ a
γ= a p + b rl − pρ > 0

As bσ rl1−σ > 0 it follows that γ ≥ 0. Therefore, the difference between both

indirect utility functions taken to the power γ is a linear upward sloping func-
tion in income. Thus the difference in utility terms is concave and increasing
in Y. This completes the proof that Ỹ Al exists. From this it follows that Y Al
exists as well.

Figure 2.4 graphs the indirect utility functions in equations (2.22) and (2.23).
If income is below Y Al indirect utility is maximized if the worker consumes
rent-controlled housing. If income exceeds Y Al the worker consumes uncon-
trolled housing. Figure 2.4 also shows that if income is between Y s and Y Ac
utility maximizing workers (restricted controlled tenants) end up in a corner
solution: they consume controlled housing of quality q̄ and spend the remain-
der of their income on the composite good.
Finally, I consider expenditures. Consumption of the housing quality is
given by equation (2.26) and consumption of the composite good by equation
2.A. Application to CES-utility 45

(2.27).
  
bσ Y
if Y ≤ Y s



 r̄ σ
a p
σ 1 − σ +b r̄
σ 1 − σ


q∗ = q̄ if Y s < Y < Y Al (2.26)

  
 σ
 brσ Y
if Y ≥ Y Al


c aσ p1−σ +bσ rc1−σ
  
aσ Y
if Y ≤ Y s



 p σ
a p
σ 1 − σ +b r̄
σ 1 − σ


x ∗ = Y −pr̄q̄ if Y s < Y < Y Al (2.27)

  
 σ
 pa σ Y
if Y ≥ Y Al


σ 1− σ
a p +b rc σ 1− σ

Figure 2.5 summarizes this by plotting how the budget share going to
housing q/Y and the composite good x/Y changes with income. Under the
shown parameterization of the CES-utility function the decrease in rent causes
unrestricted controlled workers to spend a larger share of their budget on
housing. For restricted controlled tenants the share of budget spend on hous-
ing is decreasing in income. This occurs as they cannot increase consumption
of housing, and therefore the budget share declines with income.
46 Chapter 2.

Figure 2.4: Indirect utility as function of income

Indirect utility of unrestricted tenants of social housing, restricted tenants of

social housing and tenants of uncontrolled housing as a function of income. Pa-
rameters are fixed at a = b = 1/3, ρ = .25, p = 1, r̄ = h̄ = 2, rc = 2.2.

Figure 2.5: Division of income over housing and the composite good

Budget allocated to housing quality and the composite good for unrestricted ten-
ants of social housing, restricted tenants of social housing and tenants of uncon-
trolled housing as a function of income. Parameters are fixed at a = b = 1/3,
ρ = .25, p = 1, r̄ = h̄ = 2, rc = 2.2.
2.B. Graphical representation 47

2.B Graphical representation

Th left panel of Figure 2.6 shows the optimal consumption of workers with
income not exceeding Y Al if they would be consuming uncontrolled housing.
Similarly, the right panel does so if they would consume social housing. Both
panels have units of housing quality on the x-axis and units of the composite
good on the y-axis.
The downward sloping straight lines in the left panel are the budget lines
on the uncontrolled housing market with slope −rl . The lower line is the
budget line for workers with income Y s . The upper line is the budget line
for workers with income Y Al . All points below the budget line are feasible
consumption bundles given the income of the worker. The curved lines U1 ,
U2 and U3 are indifference curves, where utility curves further from the origin
indicate higher utility levels.
In the left panel of Figure 2.6 workers with budget Y s maximize utility at
U1 if they consume q∗ units of uncontrolled housing and x ∗ units of the com-
posite good. For workers with budget Y Al all points below the budget line
and on the curve U2 are feasible. However, workers maximize utility if they
0 0
consume on the intersection with the budget line and U3 in the point (q∗ , x ∗ ).
The upward sloping vector going through the welfare maximizing consump-
tion bundles and the origin is the expansion path (“E.P”). All consumption
bundles on it are combinations of uncontrolled housing quality and the com-
posite good that maximize utility.
The right panel of Figure 2.6 shows optimal consumption if workers would
consume social housing. The downward sloping straight lines in the left panel
are still the budget lines for workers with income Y s and Y Al , however, the
slope is less steep at −r̄. They become vertical at q̄ as social housing is not
available at higher quality levels.
For workers with income Y s all points on U1 that are on or below the bud-
get line are feasible. These workers maximize utility if they consume exactly q̄
units of social housing and x̃ ∗ units of the composite good such that they are
on the utility curve U2 . As U1 < U2 workers with income below Y s are better
off renting social housing (Proposition 4).
48 Chapter 2.

Figure 2.6: Optimal choice of tenure if Y ≤ Y Al

Uncontrolled housing Social housing

composite good x

composite good x
6 6

p x = Y Al p x = Y Al
A @
A @
A @
A @
x = YAs + B @
0
x̃ ∗
A A @
A A
A A
A A
p x = YAs A p x = Ys
A
A
A
A E.P.
@ ˜
Unfeasible E.P.
A A >
 @
A A A  @ 

x∗
0
A A A.  x = Y s −@B̃
A A A U3 @ @ 


U3

A A U2 A x̃ ∗ @ @ U2
x∗ A @  3


A A
 A U1 A ˜ @
E.P. U1
A
p p
 AA  @
A 0
A - @ -
O q∗ q∗ housing quality q O q̄ housing quality q

Consumption of the composite good and housing quality given income. The left panel shows (op-
timal) consumption if uncontrolled housing quality would be consumed, the right panel shows
(optimal) consumption if social housing quality would be consumed.

Note that the kink in the budget line does not affect workers with income
Y s or lower, as they can consume the utility maximizing units of social housing
(which will not exceed q̄) and the composite good (which will not exceed x̃ ∗ )
˜ Therefore, workers with income not exceeding Y s
on the expansion path E.P.
are unrestricted tenants of social housing (see Definition 3).
For workers with income exceeding Y s this is different: As they cannot
consume more than q̄ units of housing quality at the social housing sector,
consuming on the expansion path is unfeasible. Therefore, these workers con-
sume exactly q̄ units of social housing and spend the remainder on the com-
posite good. Thus if income rises over Y s , only consumption of the composite
good increases. Their consumption bundle is not on the expansion path (Def-
inition 4).
Workers with income Y Al maximize utility if they consume q̄ units of hous-
0
ing quality and x̃ ∗ units of the composite good, which yields utility equal to
2.B. Graphical representation 49

U3 . If they consume uncontrolled housing, their optimal consumption bundle

0 0
equals (q∗ , x ∗ ) which yields utility equal to U3 as well. Workers with income
Y Al are therefore indifferent between living in uncontrolled or social housing
(Definition 2). Also, this implies that workers with income Y s ≤ Y < Y Al will
consume social housing of quality q̄ (Proposition 6).

The left panel of Figure 2.7 shows the optimal choice of consumption if in-
come is at least Y Al and workers rent uncontrolled housing. Workers with
income Y + can attain a maximum utility level of U4 if they consume the bun-
dle q∗” , x ∗” which is on their expansion path E.P.
The right panel of Figure 2.7 shows this situation if the worker were to
rent social housing. In this case the utility level U4 cannot be reached, as U4
is located above the budget line or to the right of q̄. Thus tenants with income
exceeding Y Al will prefer to rent uncontrolled housing (Proposition 5).
Figures 2.6 and 2.7 show that workers with income below Y Al will choose
to live in social housing, whereas those with income exceeding Y Al will rent
uncontrolled housing. Moreover, the group of workers in social housing can
be subdivided into workers with income not exceeding Y s whose consump-
tion bundle is on their expansion path (‘unrestricted tenants of social housing’)
and those whose consumption bundle is not on the expansion path (‘restricted
tenants of social housing’).
Figure 2.8 shows the welfare effects of rent control of social housing for a
worker with income Y s . This worker would have utility U1 if he would rent
uncontrolled housing, rent control of social housing enables him to rent q̄ units
of social housing and utility increases to U2 . The costs of rent control born by
the lessor of social housing equal C = q(rl − r̄ ). This can be measured in units
of the composite good on the vertical axis. The benefits of rent control of social
housing can be expressed in equivalent variation of income as the increase in
income B he would require in order to be indifferent between living in social
housing or uncontrolled housing. Rent control leads to a loss in welfare equal
to C − B.12
12 Alternatively, one could use compensating variation in income to express benefits from rent

control as done in the right panel of Figure 2.6. Compensating variation is defined as the income
50 Chapter 2.

Figure 2.7: Optimal choice of tenure if Y ≥ Y Al

Uncontrolled housing Social housing

composite good x

composite good x
6 6
x= Y+ x = Y+
A @
A @
p x =AY Al p x = Y Al@
A @
A A @ @
A A @ @
A A @
A A @
A A @
A @
A
A A
A A
A A
A A
A A E.P. ˜
Unfeasible E.P.
A A >

x ∗” A A 

A U  U4
A UA3 4 
U3
 A 

 A A 
3

 A A
 ˜ 
E.P.
A
p p
 A 
A A - -
O q̄ q ∗” O q̄ housing quality q
housing quality q

Consumption of the composite good and housing quality given income. The left panel shows (op-
timal) consumption if uncontrolled housing quality would be consumed, the right panel shows
(optimal) consumption if social housing quality would be consumed.
2.B. Graphical representation 51

Figure 2.8: Effects on welfare

6

composite good x
x = Ys + C
A x = Ys + B
AA
AA
AA
AA
AA
AA
AA
AA
AA
AA
rx = Y s AA
A@ AA
A@ AA
A @ AA E.P.
A @ AA >

A @ AA 
  
A @ AA  
A @ A 
A @ AA  
  Unfeasible E.P. ˜
A @ AA 
A @AA   
A 
@ A
x̃ ∗
A
@A
AA U2
3
A  
A  AA
A 
  AA
 A
A U1
 AA

 ˜
E.P. AA
 A AAA
r


 AA -
O
q̄ housing quality q

The figure shows the contribution to the welfare loss for a worker with income
Y s . Equivalent variation of income for a worker with income Y s at market rent
rl equals B. The costs of rent control equals C = q̄(rl − r̄ ). Costs of rent control
always exceed the benefits.

B̃ tenants of social housing would be willing to give up in order to remain in social housing. In
general B 6= B̃. A special case occurs if income equals Y Al as then B = B̃ = 0.
Chapter 3

Consumption and housing

market tenure

3.1 Introduction
Rising house prices have been used to explain the increase in consumption
in the 1990s and early 2000s. If correctly so, housing tenure might have im-
portant implications for consumer spending and the business cycle (Leamer,
2007). This chapter provides an all round perspective on the relationship be-
tween housing market tenure and consumption by investigating the effect it
has on the level of consumption and the allocation of the consumption budget.
The first part studies whether housing market tenure influences the level of
consumption. I consider three hypotheses that might explain the co-movement
of house prices and expenditures. An unexpected increase in house prices
might result in (unexpected) equity gains (wealth hypothesis) or reduce the
credit constraint (credit constrained hypothesis). Both might be an incentive
for home-owners to raise consumption (Campbell and Cocco, 2007). On the
other hand, the co-movement of both consumption and house prices might
be the result of both variables being influenced by common factors (common
factor hypothesis, see Attanasio et al. 2009, 2011).
In the second part I investigate whether housing market tenure is im-
portant for the allocation of the budget over consumer goods. According to
54 Chapter 3.

the model of chapter two some tenants of rent-controlled housing are better
off consuming the maximum number of rent-controlled housing services at
the controlled rent instead of consuming more housing services at the mar-
ket rent. These households underconsume housing and overconsume other
goods. Thus rent control changes the allocation towards the composite good.

3.2 Literature
Homeowners who plan to live in their house for a long time are perfectly
hedged against price changes, and therefore consumption is not affected by
housing prices (Sinai and Souleles, 2005; Campbell and Cocco, 2007). How-
ever, this does not hold for households planning to move down. According
to the wealth hypothesis an increase in house prices increases expected future
wealth, which enables them to raise consumption (Attanasio et al., 2011). For
homeowners who plan to trade up or households entering the owner-occu-
pied housing market the effect of house prices on consumption is ambiguous
(Attanasio et al., 2009): On the one hand, their future wealth increases the po-
tential to consume. On the other hand, since both the prices of the current and
the future house go up, they might need to save more in order to finance their
move.
According to the wealth hypothesis the effect of house prices on consump-
tion by tenants is ambiguous as well. A negative effect of housing prices
on consumption is expected as tenants planning to enter the owner-occupied
housing market need to save more to finance the house. However, if an in-
crease in housing prices discourages tenants from entering the owner-occupied
housing market altogether, they might respond by increasing consumption,
see Engelhardt (1994).
The second explanation for the effect of house prices on consumption fo-
cuses on the collateral provided by housing. An increase in house prices or
innovations to the financial sector reduce the credit constraint of homeown-
ers as their net-equity increases (see among others Muellbauer and Murphy
(1990); Aoki et al. (2004); Iacoviello (2004); Chen et al. (2010)). According to
3.2. Literature 55

the collateral hypothesis housing prices influence the consumption by home-

owners only, but should not influence consumption by tenants.
Finally, according to the common factor hypothesis there is no causal rela-
tionship between house prices and consumption. Instead, their co-movement
is explained by both variables being related to a common factor. For instance,
an unexpected increase in productivity might increase consumption. It might
also increase the demand for housing. Now, if the price-elasticity of housing
is low, the productivity shock will result in rising house prices (King, 1990;
Pagano, 1990; Attanasio et al., 2009). If indeed productivity is the driver of
both processes, young households are expected to increase consumption more
compared to old households, as their change in lifetime income from labor is
larger (Attanasio et al., 2009, 2011).
As mentioned before, the evidence on which mechanism explains the co-
movement of house prices and consumption is mixed. Using micro-data,
Campbell and Cocco (2007) find that old households increase consumption
after a positive change in house prices, whereas young and old tenants re-
duce consumption, which they interpret as evidence in favor of the wealth
hypothesis. Bostic et al. (2009) estimate that the price-elasticity of consump-
tion to housing prices is about six percent. Using aggregate data Chen et al.
(2010) find evidence for a positive effect of house prices on consumption as
well. However, Attanasio and Weber (1994) find that young households in-
crease consumption more in response to house prices than old households,
which they interpret as evidence in favor of the common factor hypothesis. A
similar conclusion using more data points is found by Attanasio et al. (2009).
Cristini and Almudena (2013) consider the specifications used by Campbell
and Cocco (2007) and Attanasio et al. (2009) and conclude that evidence in fa-
vor of the common factor hypothesis is more robust.

The economic literature on consumption allocation studies how household

characteristics and labor market characteristics influence the consumption of
households. Kalwij et al. (1998) study how household demographics, the bud-
get and female labor supply influence the allocation of budget over consumer
goods. Kalwij and Salverda (2004) investigate how the allocation of budget is
56 Chapter 3.

influenced by employment status, the budget and household characteristics,

where they especially focus on the determinants of spending on consumer ser-
vices. Ferrer-Carbonell and van den Bergh (2004) use a similar framework to
study the determinants of unsustainable consumption. Kalwij (2005) studies
how the arrival of children affect households income, savings and expendi-
ture.
Studies on rent control mainly consider the effect on the consumption of
housing services (see overviews by Arnott (1995); Turner and Malpezzi (2003)).
Evidence on the effect of rent control on non-housing consumption is limited.
Le Blanc and Laferrère (2001) find that rent-controlled tenants consume ten
percent more housing services and eleven percent more on other goods. Wang
and Kinsey (1994) find that strictly rationed housing has a positive spillover
effect on savings, but not on consumption.

3.3 Data
It is investigated whether housing tenure changes the level and allocation of
the consumption budget by estimating life cycle consumption model and En-
gel curves. I use data from the Dutch budget survey (in Dutch: Budget On-
derzoek). Data has been collected by Statistics Netherlands over the years
1978-2000 and in 2003 and 2004. It has previously been used by Statistics
Netherlands to construct the price index and socio-economic statistics. The
data has been used for quantitative scientific research before.1 The data are
collected over the period 1980-1999. The first two waves are omitted in order
to avoid measurement errors related to the start-up of the project as suggested
in Kalwij et al. (1998). The waves in 2003 and 2004 are not used either as
household income and rents are differently defined in these years.
The data is collected by Statistics Netherlands using a stratified sampling
scheme, where households are stratified according to social-economic posi-
tion (self-employed, employed, not active on labor market), household size,
gender of single-person household and annual household income (Gebruiker-
shandboek Budgetonderzoek 1999). As the response rate is generally low (for
1 See Kalwij et al. (1998); Ferrer-Carbonell and van den Bergh (2004); Kalwij (2005)
3.3. Data 57

instance 23 percent in 1999), sample weights are used in all tables, statistics
and regressions to provide output that is representative for the Netherlands.
All households keep a ’regular’ daily record of all expenses per item if
expenses exceed the threshold of around 16 euro (35 Dutch guilders), except
when being on holiday. Holiday expenses are recorded in a separate holiday
record. Expenses less than the threshold are recorded in a separate intensive
record for a short period of time. The period in which the intensive record has
been kept has been reduced to about a month in 1979 to seven or eight days
in 1998.
Total expenditures on non-durables are defined as the sum of expenditures
on the categories basic goods, housing, recreation and alcohol and tobacco.
These categories contain 80 percent of total household expenditures on aver-
age. The part from total expenditures that is spend on non-durables (or that is
saved) is omitted to prevent the creation of measurement error due to expen-
ditures that carry on into the (unobserved) future. Total spending is grouped
into the following categories:

1. Basic goods: Expenditures on food, clothing and personal care (exclud-

ing expenditures on alcoholic beverages, smoking, and health insur-
ance).

2. Housing: Expenditures on rent (tenants) and imputed rent (homeown-

ers), heating and light. This group does not contain expenditures on
durable goods like heaters, lamps, etc.

3. Recreation: Expenditures on holidays (in the country and abroad), recre-

ation and transport.

4. Alcohol and tobacco: Expenditures on alcoholic beverages and smok-

ing.

Some selections are made on the total sample of 47,021 household observa-
tions. Households with unknown type of tenure, with a head of household
aged below eighteen or living in uncontrolled rental housing are removed
from the sample. Tenants in the liberalized sector are removed from the es-
58 Chapter 3.

timation sample as the number of observations is insufficient.2 Self-employed

are not included in the sample as one cannot distinguish between firm and
household expenditures. See Table 3.1 for an overview.

Table 3.1: Sample adjustments

Observations
Description left
Starting number of observations 47,021
Remove households with unknown tenure type 46,716
Keep households with head of household 18 or older 46,714
Remove tenants in liberalized housing sector 46,277
Remove self-employed (including farmers) 41,430
Remove missing observations 35,795
Remove households with zero income or expenditure 35,783

As can be seen from Table 3.2 expenditure and consumption patterns of

households vary considerably. On average, households in rental housing have
lower total expenditures. They spend relatively more on alcohol and tobacco
and basic goods, whereas they spend less on housing and recreation.

Table 3.2: Budget shares

Tenants Owner-occupied Difference

(1) (2) (1-2)
mean sd mean sd difference p-value
Expenditure (e)a 35,907 14,147 50,572 18,958 -14,664.9 <0.0001
Budget shares(%)a
Basic goods 37.4 10.8 36.0 11.2 1.4 <0.0001
Housing 32.1 11.4 32.3 10.4 -0.2 0.0617
Recreation 27.0 13.5 28.8 14.7 -1.8 <0.0001
Alcohol and tobacco 3.5 3.7 2.9 3.1 0.6 <0.0001
a Based on 35,783 observations, 20,674 on homeowners and 15,109 on tenants of social housing.

The surveys include measures of net annual income, household composi-

tion. Job and house characteristics are included as well. These variables are
included as household composition might influence the demand for certain
type of goods. The same holds for labor market position, which might influ-
ence the amount of trust the household has in the stability of his earnings.
2 In most of the years there are 25 observations (or less), which is considered a too low number

to base conclusions on. Including tenants in the liberalized sector does not change results.
3.3. Data 59

Finally, house characteristics are used to compare the preferences for house-
holds for house characteristics.
Table 3.3 lists the variables that are included in the specification. It also
shows the difference in means between tenants and owner-occupied house-
holds. In general the average realization for tenants is different from that of
owner-occupied households. On average, owner-occupied households have
larger families and are better educated. Interestingly, the probability that a
household is an owner-occupied household is greater among households with
head aged below 50. They also have better earning jobs and are less likely to
be unemployed. Finally, owner-occupied households live in larger houses on
average. Because of space, cohort dummies have not been tabulated. How-
ever, they are strongly correlated with age.
60 Chapter 3.

Table 3.3: Descriptive statistics

Tenants Homeowners Difference

1 2 1-2
mean st. dev mean st. dev. mean p-value
Household characteristicsa
Ln of household size 0.77 0.54 1.01 0.48 -0.24 <.0001
(Ln of households size) squared 0.89 0.84 1.25 0.89 -0.36 <.0001
Household composition indicators: One if. . .
Single person household 0.25 0.43 0.1 0.31 0.15 <0.0001
Single family household 0.73 0.45 0.88 0.33 -0.15 <0.0001
Multiple family household 0.01 0.09 0.01 0.09 <0.01 0.915
One child in household 0.13 0.34 0.15 0.36 -0.02 <0.0001
Two children in household 0.16 0.36 0.24 0.43 -0.08 <0.0001
Three children in household 0.05 0.21 0.09 0.29 -0.04 <0.0001
Indicator age of household head: One if. . .
18 ≤ age < 25 0.04 0.20 0.01 0.11 0.03 <0.0001
25 ≤ age < 30 0.14 0.35 0.09 0.28 0.05 <0.0001
30 ≤ age < 35 0.13 0.34 0.17 0.37 -0.04 <0.0001
35 ≤ age < 40 0.12 0.32 0.18 0.38 -0.06 <0.0001
40 ≤ age < 45 0.09 0.28 0.13 0.34 -0.04 <0.0001
45 ≤ age < 50 0.07 0.26 0.10 0.30 -0.03 <0.0001
50 ≤ age < 55 0.06 0.24 0.07 0.26 -0.01 0.0033
55 ≤ age < 60 0.07 0.25 0.06 0.24 0.01 0.2658
60 ≤ age < 65 0.07 0.26 0.07 0.25 <0.01 0.0165
65 ≤ age < 70 0.08 0.28 0.06 0.24 0.02 <0.0001
70 ≤ age < 75 0.06 0.25 0.04 0.18 0.02 <0.0001
75 ≤ age < 80 0.04 0.19 0.02 0.13 0.02 <0.0001
80 ≤ age < 85 0.01 0.11 0.01 0.08 0 <0.0001
85 ≤ age < 90 <0.01 0.06 <0.01 0.04 <0.01 <0.0001
90 ≤ age < 95 <0.01 0.02 <0.01 0.01 <0.01 0.4183
Indicator highest educational level obtained by head of household or spouse: One if. . .
Primary level 0.23 0.42 0.12 0.32 0.11 <0.0001
Secundary level, first step 0.25 0.44 0.18 0.39 0.07 <0.0001
Secundary level, second step 0.32 0.47 0.34 0.47 -0.02 0.0092
Tertiary level, first step 0.15 0.35 0.27 0.44 -0.12 <0.0001
Tertiary level, second step 0.05 0.21 0.09 0.29 -0.04 <0.0001

Job characteristicsa
Employment indicator: one if the main wage earner is. . .
Employee 0.56 0.50 0.75 0.43 -0.19 <0.0001
Not employed 0.44 0.50 0.25 0.43 0.19 <0.0001
Type of job indicator: one if the job is. . .
Scientific 0.11 0.31 0.21 0.41 -0.10 <0.0001
Higher managerial 0.04 0.19 0.09 0.29 -0.05 <0.0001
Other 0.42 0.49 0.44 0.50 -0.02 <0.0001

House characteristicsa
Type of house indicator: one if the house is/has. . .
A detached single family house 0.01 0.12 0.21 0.41 -0.20 <0.0001
An appartment 0.43 0.49 0.09 0.28 0.34 <0.0001
A single family row house 0.54 0.50 0.69 0.46 -0.15 <0.0001
Constructed prior to 1945 0.33 0.47 0.42 0.49 -0.09 <0.0001
Constructed within 1945 and 1979 0.55 0.50 0.45 0.50 0.10 <0.0001
Constructed in 1980 or later 0.12 0.32 0.13 0.33 -0.01 0.0003
One or two rooms 0.07 0.26 0.02 0.13 0.05 <0.0001
Three to four rooms 0.47 0.50 0.24 0.43 0.23 <0.0001
Five or more rooms 0.46 0.50 0.74 0.44 -0.28 <0.0001

Income and expenditurea

Annual net income 49,774.41 20,392.07 70,257.29 27,564.35 -20,482.88 <.0001
Annual expenditures 35,894.45 14,172.15 50,567.8 18,964.12 -14,673.35 <.0001
Annual expenditures on. . .
Basic goods 13,589.2 6,851.03 18,142.36 8,240.53 -4,553.16 <0.0001
Housing 10,321.56 2,311.18 15,177.76 4,435.88 -4,856.2 <0.0001
Recreation 10,678.53 8,583.56 15,766.48 11,580.08 -5,087.95 <0.0001
Alcohol and tobacco 1,305.15 1,496.84 1,481.19 1,696.38 -176.04 <0.0001
a
Based on 35,783 observations, 20,674 on homeowners and 15,109 on tenants of social housing.
3.3. Data 61

Homeownership rate among age groups

Next, I turn to the definition of young, middle aged and old households and
the homeownership rate among these groups. Figure 3.1 illustrates that the
homeownership rate in the Netherlands follows an inverse-u distribution in
age.

Figure 3.1: Homeownership and age

Dots represent homeownership rates for different five year intervals of age. Dots
are placed at the middle of the interval. ‘ABHL cutoff’ indicate the cutoff values
for age used by Attanasio et al. (2009). ‘cutoff’ is the cutoff presented in this
chapter.

Attanasio et al. (2009), hereafter denoted as ABHL, define young and old
households as households with a head younger than 35 or older than 60.
Their classification results in the homeownership rate being over 50 percent
for young households and below 50 for old households (see Table 3.4). There-
fore, I define young households to be those with a head of household younger
than 30, whereas old households are those with a head older than 70. Table 3.4
indicates this results in the homeownership rate being around 40 percent for
both young and old households. The cutoffs used in this chapter better reflect
the distribution of the homeownership rate over the various age categories.
62 Chapter 3.

Table 3.4: Homeownership rate for various definitions of agegroups

homeownership rate
Definition This chaptera ABHLb
Young 42.8 53.7
Middle aged 62.2 64.6
Old 41.1 48.6
a Middle aged households aged between 30 and 70
b Middle aged households aged between 35 and 60

3.4 Estimation strategy life-cycle consumption model

It is investigated whether housing tenure influences the level of consumption
by specifying life cycle consumption along the lines of Attanasio et al. (2009).
Denote consumption for household h in period t as Xth . Assume this can be
approximated as a function of a lifetime wealth W h and the place in the life-
cycle. The effect of the place in the life cycle is a function of age, which reflects
differences in needs, household composition and discount rate over the course
of life. ε is an error term that reflects unanticipated shocks to lifetime wealth
and current income, see equation (3.1).

Xth = k ( ageh )W h exp(εht ) (3.1)

Take logs and separate log k( ageh ) into a part that depends on age only f ( ageh )
and a part that depends on household characteristics γ0 zth . This yields equa-
tion (3.2).

log( Xth ) = f ( ageh ) + γ0 zth + log(W h ) + εht (3.2)

Taking the average across cohorts yields the pseudo-panel equivalent, where
the household specific lifetime wealth W h is represented by the cohort-specific
average αc . This equation can be estimated on individual data for household
h in cohort c (indicated with superscript ch) in equation (3.3). Here νtch reflects
the deviation in consumption from the cohort average, with expectation as-
sumed to equal zero. Both methods assume that the expectation of εct is zero
3.4. Estimation strategy life-cycle consumption model 63

over the sample period and the average of lifetime wealth in each cohort can
be represented by a cohort fixed effect. Also, the age profile is assumed to be
fixed across cohorts.

log( Xtch ) =αc + f ( agec ) + γ0 zch c ch

t + ε t + νt (3.3)

Equation (3.3) specifies the basic life-cycle consumption path for households.
The level of consumption changes with the cohort, because the expected life-
time wealth changes with the productivity level. Equation (3.3) does not in-
clude income as expected lifetime income is incorporated in the deterministic
and constant terms in zch ch
t , whereas unexpected shocks to income enter via νt .
Adding income as a regressor would not be in line with the theoretical life
cycle framework that is applied. Also, if current income is correlated with
future income, current income will be correlated with the error term εct (At-
tanasio et al., 2009).
Adding house prices hpt to this baseline model helps to identify whether
shocks to house prices can be considered an unexpected innovation within
the life cycle model. In order to distinguish the wealth and collateral hypoth-
esis from the common factor hypothesis Attanasio et al. (2009) suggest to add
a function of real house prices hpt and analyze its effect for young, middle
aged and old households. Thus, they estimate equation (3.4). The sign and
magnitude of the δ coefficients indicate whether the effect of house prices on
consumption is larger for young households or old households, which identi-
fies whether house prices influence consumption via the wealth/collateral or
via the common factor mechanism.

log( Xtch ) =αc + f ( agec ) + γ0 zch

t

+ δ1 hpt Y + δ2 hpt M + δ3 hpt O + µch

t (3.4)

µch c ch
t = ε t + νt

Note that equation (3.4) assumes that Ŷ = M̂ = Ô = 0 and therefore these

covariates do not need to be included. Otherwise, as Y is correlated with
g(hpch
t )Y omitted variable bias might occur (Brambor et al., 2006). Therefore
64 Chapter 3.

equation (3.4) is extended by including these constitutive terms in equation

(3.5). Also, as the division into young, middle aged and old households de-
pends on arbitrary cutoff-points, an interaction model that interacts house
prices with age (and age squared) is specified in equation (3.6).

log( Xtch ) =αc + f ( agech ) + γ0 zch

t +Y+ M+O

+ δ1 hpt Y + δ2 hpt M + δ3 hpt O + µch

t (3.5)

log( Xtch ) =αc + f ( agech ) + γ0 zch

t

+ δage agech hpt + δage2 ( agech )2 hpt + µch

t (3.6)

µch c ch
t = ε t + νt

To summarize, a life-cycle consumption model is estimated where house prices

are interacted with age to distinguish the wealth/collateral hypothesis from
the common factor hypothesis. Various specifications are considered. Table
3.5 summarizes how the sign on the interaction effects can be used to distin-
guish the different hypotheses. Note that without making an explicit distinc-
tion between home-owners and tenants (as in the row overall) it is not possible
to distinguish the wealth hypothesis from the collateral hypothesis.
To overcome this limitation, one would want to interact house prices with
tenure status. However, unobserved preferences regarding consumption of
housing and other goods might make a comparison between home-owners
and tenants suffer from selectivity bias. Propensity Score Matching (PSM) is
used to generate a subsample of homeowners and tenants with similar ob-
served characteristics. Section 3.A explains the PSM-procedure and shows the
first stage results.
3.5. Results life-cycle consumption model 65

Table 3.5: Overview hypotheses explaining the level of consumption

Hypothesis ME effect house prices Interaction with age

overall positivea follows distribution homeownership rate
wealth homeowners positive follows distribution homeownership rate
tenants negative or
positive
overall positive follows distribution homeownership rate
collateral homeowners positive follows distribution homeownership rate
tenants none
overall positive decreasing in age
common factors homeowners positive decreasing in age
tenants positive decreasing in age
a
Under the wealth hypothesis it is unlikely that any negative effect for tenants dominates
the positive effect for homeowners.

3.5 Results life-cycle consumption model

Table 3.6 shows regression results in which age dummies for young, mid-
dle aged and old households are interacted with house prices.3 Column one
shows results for the dummy specification by Attanasio et al. (2009) with cut-
offs at 35 and 60 years. Evidence is provided for a positive relationship be-
tween consumption and aggregate house prices that increases with age. This
cannot be interpreted as evidence in favor of the wealth-hypothesis, as the
relationship between homeownership rate and age is inverse u-shaped.
However, using other cutoff values in column two, I find that the con-
sumption of middle-aged households reacts strongest to house prices, fol-
lowed by young households and old households. This order reflects the av-
erage share of homeownership among those age groups (see Table 3.4) and
therefore is evidence in favor of the wealth-hypothesis.
In column three, the specification is extended using the constitutive dummy
variables for middle-aged and old households. If significant, these variables
should be included to obtain consistent estimates. Including the constitutive
terms changes the interpretation of the results. Evidence is provided that the
positive relationship between consumption and house prices decreases with
age, which is in line with the common factor hypothesis.
3 House prices are in 10,000 euro. Including the log of real house prices yields similar results

for all specifications.

66 Chapter 3.

In column four the log of net income is included in the specification. The
variable is likely to be endogenous, so results should be interpreted with care.
Most interaction terms become insignificant upon including income, which
indicates that the interaction terms in the specification in column three are
influenced by income. No clear-cut pattern arises as only middle-aged house-
holds are found to adjust consumption with respect to house prices. These
results are in line with the explanation that income is a common factor that
drives both consumption and house prices.

Table 3.6: Results non-durable consumption (dummy specification)

(1) (2) (3) (4)

Interaction real house prices with. . .
Young households ABHL (age<35) 0.0181∗∗∗
(0.0014)
Middle-aged households ABHL (35<=age<60) 0.0195∗∗∗
(0.0012)
Old households ABHL (age>=60) 0.0259∗∗∗
(0.0015)
Young households (age<30) 0.0178∗∗∗ 0.0258∗∗∗ 0.0039
(0.0015) (0.0034) (0.0030)
Middle-aged households (30<=age<70) 0.0222∗∗∗ 0.0214∗∗∗ 0.0090∗∗∗
(0.0012) (0.0013) (0.0011)
Old households (age>=70) 0.0132∗∗∗ 0.0129∗∗∗ 0.0019
(0.0016) (0.0035) (0.0029)

Control variables
Middle-aged households (age<30) 0.0734∗∗ -0.0087
(0.0282) (0.0245)
Old households (age>=70) 0.0670 0.0056
(0.0425) (0.0363)
Age 0.0183∗∗∗ 0.0074∗∗∗ 0.0066∗∗∗ 0.0007
(0.0016) (0.0017) (0.0017) (0.0014)
Age squared -0.0003∗∗∗ -0.0001∗∗∗ -0.0001∗∗∗ -<0.0001∗∗
(<0.0001) (<0.0001) (<0.0001) (<0.0001)
log net annual income 0.5590∗∗∗
(0.0085)
Constant 10.5071∗∗∗ 10.5523∗∗∗ 10.4915∗∗∗ 4.3503∗∗∗
(0.1560) (0.1434) (0.1445) (0.1318)
adj. R2 0.462 0.463 0.463 0.622
F-statistic interation terms house prices 112.7 123.4 105.5 22.2
F-statistic p-value <0.0001 <0.0001 <0.0001 <0.0001
Degrees model 38 38 40 41
N 35,783 35,783 35,783 35,783
Standard errors in round parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Variables not shown: 17 cohort dummies; log household size; log household size squared; dummies for SPH, SFH MFH; one dummy for not employed;
three dummies for type of job, and four for educational level.
Dependent variable: log non-durable expenditure. Real house price measure in 10,000 euro.

These results might be driven by the (ad hoc) cutoff points that define
young, middle aged and old households. Therefore house prices are inter-
acted with a continuous measure of age and with age in quadratic form. House
prices are interacted with the quadratic form of age, as this allows for a non-
linear interaction effect as is expected under the wealth hypothesis.
3.5. Results life-cycle consumption model 67

Results are given in Table 3.7. Column one shows that consumption is
positively related to house prices. The relationship between consumption and
age is concave with the optimum at about age 40.4 In column two house prices
are interacted with age and the squared value of age. Evidence is found in
favor of the common factor analysis, as the marginal effect of house prices on
consumption decreases with age. A change in real house prices with e10,000
increases consumption with about 2.3 percent for households around age 30,
but with 1.8 percent for households around age 65, see the upper left panel in
Figure 3.2.
After including the log of net income, evidence is found that the marginal
effect of house prices on consumption is inverse u-shaped, see column three or
the bottom left panel in Figure 3.2. For heads of households with age between
35 and 70 a positive effect of house prices on consumption is found. Around
age 50 the response in consumption is maximized: an increase in house prices
with 10,000 euro increases consumption with about one percent.
The right panels in Figure 3.2 shows the marginal effect of age on con-
sumption for a range of house prices. Evidence in favor of a marginal effect of
age on consumption is not found (conditional on cohort dummies).

Table 3.7: Results non-durable consumption (interaction specification)

(1) (2) (3)

Real house prices (in 10,000 euro) 0.0208∗∗∗ 0.0275∗ -0.0182
(0.0012) (0.0113) (0.0097)
Interaction age and real house prices -<0.0001 0.0011∗∗
(0.0005) (0.0004)
Interaction age squared and house prices -<0.0001 -<0.0001∗∗
(0.0000) (0.0000)
Age 0.0161∗∗∗ 0.0141∗∗∗ -0.0022
(0.0013) (0.0041) (0.0035)
Age squared -0.0002∗∗∗ -0.0002∗∗∗ -<0.0001
(<0.0001) (<0.0001) (<0.0001)
Log net annual income 0.5600∗∗∗
(0.0086)
Constant 10.4626∗∗∗ 10.4122∗∗∗ 4.4681∗∗∗
(0.1502) (0.1748) (0.1497)
adj. R2 0.462 0.462 0.622
Degrees model 36 38 39
N 35,783 35,783 35,783
Standard errors in round parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Variables not shown: 17 cohort dummies, ln household size, ln household size squared, dummies for SPH, SFH MFH
one dummy for not employed, 3 for type of job, and four for educational level
Dependent variable: log non-durable expenditure

4 ∂log(c) = 0.0161 − 2(0.0002)age which is zero when age is about 40.

∂age
68 Chapter 3.

Figure 3.2: Marginal effect of house prices and age

Marginal effect house prices (l) and age (r)

(Table 3.7, column 2)

Marginal effect house prices (l) and age (r), controlling for annual net income
(Table 3.7, column 3)

Next, the effect of house prices on consumption is considered for home-

owners and tenants separately. The endogeneity of homeownership status
is corrected for using propensity score matching (see section 3.A for the first
stage results of the PSM-procedure). Column one of Table 3.8 indicates that
in general consumption increases with house prices. An increase in house
prices with 10,000 euro raises consumption with 1.7 percent. Also, the rela-
tionship between age and consumption is found to be concave, with the opti-
mum around the age of 50.
Column two shows that the change in consumption with respect to house
prices is lower for tenants than for homeowners. A 10,000 euro increase in
house prices raises consumption with about 1.7 percent for homeowners. This
is about one percent for tenants.5
5 The coefficient on house prices equals 0.0166. The coefficient on the interaction term "House

prices and tenants" equals -0.0067. The difference is about one percent.
3.5. Results life-cycle consumption model 69

Column three of Table 3.8 shows results for the specification in which local
house prices are interacted with age for homeowners and tenants separately.6
Results indicate that homeowners and tenants of different age respond dif-
ferently to house prices. For both types of tenure status, the marginal effect
of house prices decreases with age, see Figure 3.3: For homeowners aged be-
tween 18 and 50 the marginal effect of an increase in house prices with 10,000
euro ranges between two and one percent. For tenants the effect is around
one percent for households aged between 18 and about 35 only. As a positive
effect of house prices on consumption is found to exist for (a subsample of)
tenants, the collateral hypothesis is ruled out.
Column four shows results if income is included in the specification. Now,
no real evidence is found that house prices influence the consumption of home-
owners or tenants. The lower panels of Figure 3.3 show that no significant
marginal effect is found except for homeowners aged younger than 23. How-
ever, this group is very small in size (as is shown in the histogram) and most
likely this positive marginal effect is the result of the linear specification of the
interaction.
Overall, most of the evidence points in the direction of the common factor
hypothesis as the marginal effect of house prices on consumption decreases
with age and becomes insignificant if income (a potential common factor) is
included. The propensity score matching reveals that young homeowners and
young tenants consume more if house prices increase. For tenants, the effect
of house prices on consumption becomes insignificant if they are about 35. For
homeowners this occurs if they are about 60. Thus for home-owners the cor-
relation between house prices and consumption (caused by common factors)
is stronger than for tenants.

6 Including interaction of house prices with age squared for homeowners and tenants sepa-

rately yield insignificant coefficients and insignificant marginal effects for both house prices and
age.
70 Chapter 3.

Table 3.8: Results non-durable consumption (PSM specification)

(1) (2) (3) (4)

Tenant -0.0305 -0.0105 0.0252 0.1966∗∗
(0.0200) (0.0526) (0.0756) (0.0656)
Real house prices 0.0168∗∗∗ 0.0166∗∗∗
(0.0021) (0.0021)
House prices and tenants -0.0072∗∗ -0.0067∗∗ 0.0188∗∗ 0.0032
(0.0024) (0.0024) (0.0066) (0.0058)
House prices and homeowners 0.0301∗∗∗ 0.0159∗
(0.0069) (0.0062)
Interaction of house prices with age . . .
for homeowners -0.0003∗ -0.0002
(0.0001) (0.0001)
for tenants -0.0002 <0.0001
(0.0001) (0.0001)

Age 0.0108∗∗∗ 0.0110∗∗∗

(0.0018) (0.0021)
Age squared -0.0002∗∗∗ -0.0002∗∗∗
(<0.0001) (<0.0001)
Age and tenants -0.0004 0.0098∗∗∗ 0.0018
(0.0021) (0.0021) (0.0018)
Age squared and tenants -<0.0001 -0.0002∗∗∗ -0.0001∗∗
(<0.0001) (<0.0001) (<0.0001)
Age and homeowners 0.0109∗∗∗ 0.0095∗∗∗
(0.0022) (0.0020)
Age squared and homeowners -0.0001∗∗∗ -0.0001∗∗∗
(<0.0001) (<0.0001)
Log net annual income 0.5222∗∗∗
(0.0085)
Constant 10.7714∗∗∗ 10.7648∗∗∗ 10.6539∗∗∗ 4.5727∗∗∗
(0.2242) (0.2192) (0.2141) (0.1744)
adj. R2 0.362 0.363 0.363 0.519
Degrees model 38 40 42 43
N 16,386 16,386 16,386 16,386
Standard errors in round parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Variables not shown: 17 cohort dummies, log household size, log household size squared, dummies for SPH, SFH MFH
one dummy for not employed, 3 for type of job, and four for educational level
Dependent variable: Log non-durable expenditure
3.5. Results life-cycle consumption model 71

Figure 3.3: Marginal effect house prices (PSM)

Marginal effect house prices for homeowners (l) and tenants (r)
(Table 3.8, column 3)

Marginal effect house prices for homeowners (l) and tenants (r), controling for annual net income
(Table 3.8, column 4)
72 Chapter 3.

3.6 Estimation strategy budget allocation model

To investigate whether housing tenure alters the allocation of the consumption
budget an Almost Ideal Demand System is estimated.7 Deaton and Muell-
bauer (1980a) start from the log of the cost function, which they define as

log(C (U, p)) = A( p) + uB( p), (3.7)

with
0 0
A( p) = a0 + ∑ a g log( p g ) + 21 γ∗g0 g ∑ ∑ log( p g )log( p g )
g g g

B ( p ) = β 0 ∏ ( p g )bg
g

Where all α, β, b and γ are unknown weights that have to be estimated and
0
p g , p g are prices of the goods g and g0 6= g. Let h∗ g (U, p) be the Hicksian de-
mand for good g given utility level U and price vector p. As the cost function
is given by C = ∑ g p g h∗ g (U, p) the derivative of the log of the cost function to
the log of price for good g yields the budget share of good g. Therefore, it can
be shown that the budget share w g is a function of a constant for good g (α g ),
0
the prices of all other goods p g and the price index B( p):8

0
w g = a g + ∑ γg0 g log( p g ) + (bg )uB( p), (3.8)
g0

where γg0 g = (γ∗gg0 + γ∗g0 g )/2 = γgg0

To simplify this equation replace total costs C (U, p) with total expenditure
X under the assumption that they are equivalent. Solve equation (3.7) for u,
where the price index A( p) is indicated by log( P). Substitute this into (3.8) to
7 This part is based on Deaton and Muellbauer (1980a) and chapters two and three in Deaton

and Muellbauer (1980b).

8 To see why, start with w g = ∂A( p) + ∂uB( p) ∂p g ∂A( p) ∂uB( p) g
∂log( p g ) ∂p g ∂log( p g ) = ∂log( p g ) + ∂p g p . The partial deriva-
∂A( p) 0
tive of A( p) to log( p g ) equals ∂log( p g )
= a g + ∑ g0 γg0 g log( p g ). Define β 0 ∏ g ( p g )bk = B( p) such
∂uB( p)
that the partial derivative of uB( p) to p g equals ∂p g = bg uβ 0 ∏ g ( p g )bg = bg uB( p)( p g )−1 . This
yields equation (3.8).
3.6. Estimation strategy budget allocation model 73

derive equation (3.9).

0
w g = a g + ∑ γg0 g log( p g ) + bg log( X ) − bg log( P) (3.9)
h

Thus the budget share devoted to consumption of good g by household i is

a function of a constant α g , the log of total expenditure Xi , and prices of all
0
goods p g , and the price index.9
g
The budget share of each of the consumption categories (wi ) is regressed
on total consumption (Xi ), household characteristics (Hi ), a vector including
eighteen fixed year effects (Dt ) to control for price differences and a vector in-
cluding nineteen fixed year effects for tenants only (RDt ) to control for prices
g
for rent-controlled tenants. ε i is an i.i.d. error term, see equation (3.10). Su-
perscripts g indicate whether the consumption is on basic goods, housing,
recreation or alcohol and tobacco. The β g ,ζ g and ζ gR are parameters that have
to be estimated.

1999 1999
∑ ∑
g g g g g gR g
wi = β 0 + β 1 Xi + β 2 Hi + ζ t Dt + ζ t0 RDt0 + ε i (3.10)
t=1982 0
t =1981

with g = 1, .., 4; i = 1, .., N

On the right hand side of equation (3.10) Xi measures the log of total expendi-
ture by household i. As the budget share has total expenditure in the denomi-
nator, Xi is instrumented using annual income Yi . Hi is a matrix of household
characteristics that might affect preferences for certain goods. The variables in
Hi are age of the household head (in groups of five years), size of the house-
hold, squared size of the household, dummies for household type (single,
family, family plus others), dummies for the number of children present in
the household (one, two or three, or more) and the highest educational degree
in the household (attained by either the head of the household or the spouse,
if present).
9 It satisfies the first property of demand (adding up), whenever ∑ g α g = 1 and ∑ g γg0 g =
∑ g β g = 0, which holds by definition. Furthermore, the conditions of homogeneity and symmetry
can be tested. Homogeneity holds if and only if ∑ g γg0 g = 0, ∀ g0 . Symmetry holds if and only
if γg0 g = γgg0 . However the negativity property can not easily be reproduced by the system and
cannot be controlled by restricting the values of the estimated parameters. Although w g = [0, 1]
(by definition) estimating equation (3.9) might lead to predictions outside this interval, as X =
[0, ∞). Both drawbacks make the system ‘almost ideal’ (see Barten (1993) for both statements).
74 Chapter 3.

Dt is a vector that indicates whether the observation is in year t. It reflects

the effect of year specific factors such as prices and the economic cycle. Addi-
tionally, RDt is defined as a ‘rental year’ dummy. RDt is one if the survey year
equals t and the household rents controlled housing, and is zero otherwise.
As a result RDt captures the year effect that is specific to tenants. Under the
assumption that homeowners and tenants have equal preferences, this RDt
captures the effect of lower rents on the consumption budget shares.
A test whether consumption of good g differs between homeowners and
gR gR
tenants because of rents is testing the joint hypothesis whether ζb1981 = ζb1982 =
gR
. . . = ζb1999 = 0. Under the null hypothesis the effect of prices is equal for both
gT
homeowners and tenants. If the null hypothesis is rejected, ζ t contains the
difference in consumption of good g in year t by tenants because of lower
rents. It exactly identifies this effect if there are no other variables that are
constant within a year that are different for homeowners and tenants.
Additionally, it is investigated whether consumption patterns differ within
the group of rent-controlled tenants. This hypothesis can be tested under
the assumption that rent-controlled tenants with a high propensity to live in
owner-occupied housing are on average more restricted in their consumption
of housing services than controlled tenants with a low propensity to live in
owner-occupied housing. Then, equation (3.11) can be estimated. Here, the
dependent variable is an indicator equal to one for being a homeowner (hi ).
Explanatory variables are annual net income Yi , its squared value Yi2 , a ma-
trix of explanatory variables Zi and a set of year indicators Dt . νi is an i.i.d.
error term. The gammas and deltas are parameters that have to be estimated.
Equation (3.11) is estimated as a logistic regression to ensures that 0 < b
hi < 1.

hi = γ0 + γ1 Yi + γ2 Yi2 + γ3 Zi + ∑ δt Dt + νi (3.11)
t

with i = 1, .., N; t = 1981, .., 1999

Explanatory variables included in Zi are indicators of household, job and

house characteristics. Household characteristics are included in the selection
regression as household composition might affect the demand for housing. As
rent-controlled housing is of different quality (generally smaller), this might
3.7. Results budget allocation model 75

affect the selection into a particular type of housing. Similarly, job character-
istics are included to measure the effect of income on tenure choice. Also, the
type of job is a proxy reflecting ability of the household. Finally, house charac-
teristics are included such that households in equal housing are matched. As a
result, the selected households have similar revealed preferences for housing.
Each household’s propensity to be a home-owner 0 < ĥi < 1 is predicted.
I include Pi = ĥi and its interaction with rent-controlled tenant status PiR in
the specification. Then, a positive and significant coefficient on PiR indicates
that on average tenants who are more similar to home-owners overconsume
the good.

1999 1999
∑ ∑
g g g g g gR
w i = β 0 + β 1 Xi + β 2 H i + ζ t Dt + ζ t0 RDt0
t=1982 t0 =1981
g gR g
+ Pi + Pi + εi (3.12)

with g = 1, .., 4; i = 1, .., N; t = 1981, .., 1999

3.7 Results budget allocation model

Table 3.9 shows the base result using the full sample. Homeowners do not
alter the allocation of their budget over time, as most of the year dummies
are insignificant. In contrast, almost all year dummies are significant for ten-
ants. For all consumption categories it holds that the difference between ten-
ants and homeowners declines over time. This is in line with the fact that
the difference between imputed rents and rents decreased over time. On av-
erage, tenants spend six percent of their budget less on housing, which they
spend on basic goods (plus three percent), recreation (plus three percent) and
alcohol and tobacco (plus 0.6 percent).10 This indicates that, ceteris paribus,
homeowners and tenants differ in the allocation of budget over consumption
categories.
10 The average coefficients do not add up to zero, because of rounding.
76 Chapter 3.

Table 3.9: Allocation of budget

(1) (2) (3) (4)

Basic goods Housing Recreation Alcohol and tobacco
coef. s.e. coef. s.e. coef. s.e. coef. s.e.
Ln of expenditure -0.0020 (0.0039) -0.1600∗∗∗ (0.0029) 0.1600∗∗∗ (0.0048) 0.0025∗ (0.0012)
Tenant in year 1980 0.0300∗∗∗ (0.0046) -0.0710∗∗∗ (0.0031) 0.0380∗∗∗ (0.0050) 0.0032 (0.0017)
Tenant in year 1981 0.0270∗∗∗ (0.0045) -0.0680∗∗∗ (0.0036) 0.0360∗∗∗ (0.0049) 0.0058∗∗∗ (0.0017)
Tenant in year 1982 0.0340∗∗∗ (0.0050) -0.0830∗∗∗ (0.0039) 0.0390∗∗∗ (0.0052) 0.0093∗∗∗ (0.0018)
Tenant in year 1983 0.0390∗∗∗ (0.0039) -0.0710∗∗∗ (0.0031) 0.0280∗∗∗ (0.0043) 0.0042∗∗ (0.0014)
Tenant in year 1984 0.0360∗∗∗ (0.0038) -0.0690∗∗∗ (0.0030) 0.0240∗∗∗ (0.0042) 0.0084∗∗∗ (0.0014)
Tenant in year 1985 0.0350∗∗∗ (0.0051) -0.0730∗∗∗ (0.0031) 0.0320∗∗∗ (0.0047) 0.0058∗∗∗ (0.0014)
Tenant in year 1986 0.0340∗∗∗ (0.0038) -0.0780∗∗∗ (0.0028) 0.0360∗∗∗ (0.0043) 0.0075∗∗∗ (0.0014)
Tenant in year 1987 0.0360∗∗∗ (0.0043) -0.0830∗∗∗ (0.0030) 0.0400∗∗∗ (0.0048) 0.0075∗∗∗ (0.0015)
Tenant in year 1988 0.0330∗∗∗ (0.0060) -0.0660∗∗∗ (0.0046) 0.0240∗∗ (0.0088) 0.0085∗∗∗ (0.0018)
Tenant in year 1989 0.0320∗∗∗ (0.0052) -0.0670∗∗∗ (0.0034) 0.0300∗∗∗ (0.0057) 0.0055∗∗ (0.0017)
Tenant in year 1990 0.0270∗∗∗ (0.0061) -0.0710∗∗∗ (0.0039) 0.0390∗∗∗ (0.0087) 0.0054∗∗∗ (0.0014)
Tenant in year 1991 0.0290∗∗∗ (0.0066) -0.0730∗∗∗ (0.0052) 0.0370∗∗∗ (0.0072) 0.0067∗∗∗ (0.0020)
Tenant in year 1992 0.0240∗∗∗ (0.0051) -0.0680∗∗∗ (0.0037) 0.0370∗∗∗ (0.0055) 0.0066∗∗∗ (0.0017)
Tenant in year 1993 0.0220∗∗∗ (0.0049) -0.0620∗∗∗ (0.0036) 0.0350∗∗∗ (0.0053) 0.0040∗ (0.0018)
Tenant in year 1994 0.0260∗∗∗ (0.0049) -0.0650∗∗∗ (0.0037) 0.0350∗∗∗ (0.0052) 0.0042∗ (0.0017)
Tenant in year 1995 0.0130∗∗ (0.0049) -0.0440∗∗∗ (0.0036) 0.0260∗∗∗ (0.0051) 0.0056∗∗ (0.0018)
Tenant in year 1996 0.0160∗∗∗ (0.0049) -0.0390∗∗∗ (0.0035) 0.0170∗∗ (0.0051) 0.0064∗∗∗ (0.0018)
Tenant in year 1997 0.0190∗∗∗ (0.0051) -0.0460∗∗∗ (0.0035) 0.0190∗∗∗ (0.0051) 0.0083∗∗∗ (0.0018)
Tenant in year 1998 0.0200∗∗∗ (0.0051) -0.0430∗∗∗ (0.0037) 0.0160∗∗ (0.0052) 0.0068∗∗∗ (0.0019)
Tenant in year 1999 0.0180∗∗ (0.0056) -0.0330∗∗∗ (0.0038) 0.0096 (0.0057) 0.0058∗∗ (0.0019)
Year 1981 -0.0073 (0.0061) 0.0120∗ (0.0049) -0.0027 (0.0069) -0.0022 (0.0020)
Year 1982 -0.0100 (0.0100) 0.0340∗∗∗ (0.0079) -0.0190 (0.0110) -0.0039 (0.0032)
Year 1983 -0.0180 (0.0140) 0.0280∗∗ (0.0110) -0.0081 (0.0160) -0.0014 (0.0042)
Year 1984 -0.0120 (0.0180) 0.0330∗ (0.0140) -0.0160 (0.0200) -0.0039 (0.0055)
Year 1985 -0.0190 (0.0220) 0.0380∗ (0.0170) -0.0180 (0.0260) -0.0020 (0.0068)
Year 1986 -0.0077 (0.0270) 0.0380 (0.0210) -0.0270 (0.0310) -0.0030 (0.0082)
Year 1987 0.0002 (0.0310) 0.0350 (0.0240) -0.0320 (0.0360) -0.0030 (0.0095)
Year 1988 0.0190 (0.0360) 0.0170 (0.0270) -0.0330 (0.0410) -0.0030 (0.0110)
Year 1989 0.0230 (0.0400) 0.0190 (0.0310) -0.0400 (0.0460) -0.0023 (0.0120)
Year 1990 0.0280 (0.0430) 0.0180 (0.0340) -0.0440 (0.0500) -0.0020 (0.0130)
Year 1991 0.0210 (0.0480) 0.0300 (0.0370) -0.0470 (0.0560) -0.0038 (0.0150)
Year 1992 0.0210 (0.0530) 0.0290 (0.0410) -0.0470 (0.0610) -0.0034 (0.0160)
Year 1993 0.0240 (0.0580) 0.0250 (0.0440) -0.0490 (0.0670) -0.0003 (0.0170)
Year 1994 0.0190 (0.0620) 0.0320 (0.0480) -0.0520 (0.0720) 0.0009 (0.0190)
Year 1995 0.0250 (0.0660) 0.0180 (0.0510) -0.0460 (0.0760) 0.0023 (0.0200)
Year 1996 0.0210 (0.0710) 0.0260 (0.0550) -0.0470 (0.0820) -<0.0001 (0.0210)
Year 1997 0.0240 (0.0750) 0.0250 (0.0580) -0.0490 (0.0870) -0.0002 (0.0230)
Year 1998 0.0250 (0.0800) 0.0270 (0.0610) -0.0540 (0.0920) 0.0020 (0.0240)
Year 1999 0.0330 (0.0840) 0.0120 (0.0650) -0.0470 (0.0970) 0.0019 (0.0250)
Age -0.0040 (0.0045) -0.0015 (0.0035) 0.0054 (0.0052) -<0.0001 (0.0014)
Age squared <0.0001 (<0.0001) <0.0001∗∗∗ (<0.0001) -<0.0001∗∗∗ (<0.0001) -<0.0001∗∗∗ (<0.0001)
Secundary, first step -0.0074∗∗∗ (0.0018) -0.0002 (0.0014) 0.0080∗∗∗ (0.0019) -0.0004 (0.0007)
Secundary, second step -0.0098∗∗∗ (0.0020) -0.0015 (0.0016) 0.014∗∗∗ (0.0021) -0.0024∗∗ (0.0007)
Tertiary, first step -0.0075∗∗ (0.0023) 0.0013 (0.0017) 0.0100∗∗∗ (0.0025) -0.0042∗∗∗ (0.0008)
Tertiary, second step -0.0037 (0.0031) 0.0084∗∗∗ (0.0023) 0.0004 (0.0034) -0.0050∗∗∗ (0.0010)
Employee 0.0049 (0.0090) -0.0086 (0.0069) 0.0087 (0.011) -0.0050 (0.0029)
Academic job -0.0005 (0.0091) -0.0016 (0.0069) 0.0022 (0.011) -<0.0001 (0.0029)
Managerial job 0.0065 (0.0091) 0.0023 (0.0070) -0.0088 (0.011) 0.0001 (0.0029)
Other 0.0057 (0.0090) -0.0087 (0.0069) -0.0011 (0.010) 0.0041 (0.0029)
adj. R2 0.128 0.545 0.299 0.052
Chi2 tenant year 582.8 5370.3 573.7 223.8
P-value <0.0001 <0.0001 <0.0001 <0.0001
ATE tenant-year 0.03 -0.06 0.03 0.006
Standard error ATE 0.001 0.0010 0.001 0.0005
T-value ATE 20.8 -65.3 20.6 13.7
Degrees model 75 75 75 75
N 35783 35783 35783 35783
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Dependent variable: share of budget spend on each category; log expenditure instrumented using ln net income
Variables not shown: Constant, 17 cohort dummies, log household size, log household size squared, dummies for SPH, SFH MFH
one dummy for not employed, three for type of job, and four for educational level
3.7. Results budget allocation model 77

Next, the budget share equations are re-estimated using the PSM subsam-
ple to control for potential endogeneity. Estimation results based on 2SLS are
given in Table 3.10. Results are similar to those for the complete sample, al-
though the tenant-year dummies for basic goods and alcohol and tobacco are
no longer significant at the end of the sample period. However, on average,
tenants still spend six percent of their budget less on housing, which they
spend on basic goods (plus two percent), recreation (plus three percent) and
alcohol and tobacco (plus 0.6 percent). Thus the propensity score matching
procedure does not significantly alter our findings with respect to the average
treatment effect of living in rent-controlled or owner-occupied housing.11
Finally, I investigate whether the consumption pattern for restricted ten-
ants is different compared to that of unrestricted tenants. The propensity score
for being a homeowner is used as a measure of whether the household is re-
stricted. Here I assume that rent-controlled tenants who are very likely to be
a home-owner differ in their consumption patterns from households who are
very unlikely to be a home-owner. Table 3.11 shows that consumption pat-
terns of a ‘more typical home-owner’ differs from that of a ‘typical tenant’.
A head of household that is likely to be a home-owner spends more of his
budget on housing, and less on recreation. However, for tenants, this effect
is reduced. This indicates that especially restricted tenants spend less of their
budget on housing and more on recreation.

11 The estimates on other variables differ significantly between the PSM-sample and the full

sample.
78 Chapter 3.

Table 3.10: Allocation of budget (PSM subsample)

(1) (2) (3) (4)

Basic goods Housing Recreation Alcohol and tobacco
coef. s.e. coef. s.e. coef. s.e. coef. s.e.
Log of expenditure 0.0050 (0.0080) -0.2000∗∗∗ (0.0050) 0.2000∗∗∗ (0.0100) 0.0020 (0.0020)
Tenant in year 1980 0.0100∗ (0.0070) -0.0600∗∗∗ (0.0050) 0.0400∗∗∗ (0.0080) 0.0040 (0.0020)
Tenant in year 1981 0.0200∗∗ (0.0070) -0.0500∗∗∗ (0.0050) 0.0300∗∗∗ (0.0070) 0.0050∗ (0.0020)
Tenant in year 1982 0.0200∗∗ (0.0070) -0.0700∗∗∗ (0.0050) 0.0400∗∗∗ (0.0070) 0.0100∗∗∗ (0.0020)
Tenant in year 1983 0.0300∗∗∗ (0.0060) -0.0700∗∗∗ (0.0040) 0.0300∗∗∗ (0.0060) 0.0050∗ (0.0020)
Tenant in year 1984 0.0300∗∗∗ (0.0050) -0.0700∗∗∗ (0.0040) 0.0300∗∗∗ (0.0060) 0.0100∗∗∗ (0.0020)
Tenant in year 1985 0.0300∗∗∗ (0.0050) -0.0700∗∗∗ (0.0040) 0.0300∗∗∗ (0.0050) 0.0060∗∗ (0.0020)
Tenant in year 1986 0.0300∗∗∗ (0.0050) -0.0700∗∗∗ (0.0040) 0.0400∗∗∗ (0.0060) 0.0060∗∗ (0.0020)
Tenant in year 1987 0.0400∗∗∗ (0.0060) -0.0800∗∗∗ (0.0040) 0.0400∗∗∗ (0.0070) 0.0040 (0.0020)
Tenant in year 1988 0.0300∗∗∗ (0.0070) -0.0600∗∗∗ (0.0050) 0.0200∗∗ (0.0080) 0.0090∗∗∗ (0.0030)
Tenant in year 1989 0.0400∗∗∗ (0.0080) -0.0700∗∗∗ (0.0050) 0.0300∗∗∗ (0.0080) 0.0060∗ (0.0030)
Tenant in year 1990 0.0200 (0.0100) -0.0700∗∗∗ (0.0070) 0.0500∗∗ (0.0200) 0.0060∗∗ (0.0020)
Tenant in year 1991 0.0200∗ (0.0100) -0.0600∗∗∗ (0.0070) 0.0300∗∗ (0.0100) 0.0060 (0.0030)
Tenant in year 1992 0.0200∗∗ (0.0070) -0.0600∗∗∗ (0.0050) 0.0300∗∗∗ (0.0080) 0.0090∗∗ (0.0030)
Tenant in year 1993 0.0200∗∗∗ (0.0070) -0.0600∗∗∗ (0.0050) 0.0300∗∗∗ (0.0080) 0.0020 (0.0030)
Tenant in year 1994 0.0300∗∗∗ (0.0070) -0.0600∗∗∗ (0.0050) 0.0300∗∗∗ (0.0070) 0.0030 (0.0030)
Tenant in year 1995 0.0100 (0.0070) -0.0500∗∗∗ (0.0050) 0.0300∗∗∗ (0.0080) 0.0030 (0.0030)
Tenant in year 1996 0.0200∗∗ (0.0070) -0.0400∗∗∗ (0.0050) 0.0200∗ (0.0070) 0.0040 (0.0020)
Tenant in year 1997 0.0100 (0.0070) -0.0400∗∗∗ (0.0050) 0.0200∗∗ (0.0080) 0.0090∗∗ (0.0030)
Tenant in year 1998 0.0200∗ (0.0070) -0.0400∗∗∗ (0.0050) 0.0200∗ (0.0080) 0.0080∗∗ (0.0030)
Tenant in year 1999 0.0060 (0.0080) -0.0200∗∗∗ (0.0050) 0.0100 (0.008) 0.0020 (0.0030)
Year 1981 -0.0100 (0.0090) 0.0060 (0.0070) 0.0040 (0.0100) -0.0005 (0.0030)
Year 1982 -0.0040 (0.0200) 0.0300∗ (0.0100) -0.0200 (0.0200) -0.0050 (0.0050)
Year 1983 -0.0300 (0.0200) 0.0300∗ (0.0200) -0.0050 (0.0200) -0.0007 (0.0060)
Year 1984 -0.0200 (0.0300) 0.0400 (0.0200) -0.0100 (0.0300) -0.0050 (0.0080)
Year 1985 -0.0300 (0.0300) 0.0400 (0.0200) -0.0100 (0.0400) -0.0010 (0.0100)
Year 1986 -0.0200 (0.0400) 0.0500 (0.0300) -0.0200 (0.0500) 0.0003 (0.0100)
Year 1987 -0.0200 (0.0500) 0.0500 (0.0300) -0.0300 (0.0500) 0.0006 (0.0100)
Year 1988 0.0020 (0.0500) 0.0200 (0.0400) -0.0300 (0.0600) 0.0001 (0.0200)
Year 1989 0.0010 (0.0600) 0.0400 (0.0400) -0.0400 (0.0700) 0.0007 (0.0200)
Year 1990 0.0200 (0.0600) 0.0400 (0.0500) -0.0500 (0.0700) 0.0005 (0.0200)
Year 1991 -0.0040 (0.0700) 0.0400 (0.0500) -0.0400 (0.0800) -0.0010 (0.0200)
Year 1992 -0.0050 (0.0800) 0.0500 (0.0600) -0.0400 (0.0900) -0.0020 (0.0200)
Year 1993 0.0010 (0.0900) 0.0500 (0.0600) -0.0500 (0.1000) 0.0030 (0.0300)
Year 1994 -0.0090 (0.0900) 0.0500 (0.0700) -0.0500 (0.1000) 0.0040 (0.0300)
Year 1995 0.0040 (0.1000) 0.0500 (0.0700) -0.0600 (0.1000) 0.0060 (0.0300)
Year 1996 -0.0100 (0.1000) 0.0600 (0.0800) -0.0500 (0.1000) 0.0020 (0.0300)
Year 1997 -0.0020 (0.1000) 0.0600 (0.0800) -0.0600 (0.1000) 0.0020 (0.0300)
Year 1998 -0.0040 (0.1000) 0.0600 (0.0900) -0.0600 (0.1000) 0.0020 (0.0400)
Year 1999 0.0090 (0.1000) 0.0400 (0.0900) -0.0600 (0.1000) 0.0060 (0.0400)
Age -0.0030 (0.0070) -0.0030 (0.0050) 0.0060 (0.0080) -<0.0001 (0.0020)
Age squared <0.0001 (<0.0001) <0.0001∗ (<0.0001) -<0.0001∗ (<0.0001) -<0.0001∗∗∗ (<0.0001)
Secundary, first step -0.0080∗∗ (0.0030) 0.0040 (0.0020) 0.0050 (0.0030) -0.0010 (0.0009)
Secundary, second step -0.0100∗∗∗ (0.0030) 0.0050∗ (0.0020) 0.0100∗∗ (0.0030) -0.0040∗∗∗ (0.0010)
Tertiary, first step -0.0100∗∗ (0.0030) 0.0100∗∗∗ (0.0030) 0.0050 (0.0040) -0.0050∗∗∗ (0.0010)
Tertiary, second step -0.0080 (0.0050) 0.0200∗∗∗ (0.0040) -0.0020 (0.0060) -0.0070∗∗∗ (0.0020)
Employee 0.0050 (0.0100) 0.0030 (0.0100) -0.0070 (0.0200) -0.0010 (0.0050)
Academic job 0.0060 (0.0100) -0.0100 (0.0100) 0.0090 (0.0200) -0.0050 (0.0050)
Managerial job 0.0100 (0.0100) -0.0100 (0.0100) 0.0040 (0.0200) -0.0040 (0.0050)
Other 0.0070 (0.0100) -0.0200 (0.0100) 0.0090 (0.0200) 0.0004 (0.0050)
Constant 0.5000 (0.6000) 2.4000∗∗∗ (0.4000) -2.0000∗∗ (0.7000) 0.1000 (0.2000)
adj. R2 0.129 0.562 0.277 0.046
Chi2 tenant year 284.9 3091.7 344.7 146.6
P-value <0.0001 <0.0001 <0.0001 <0.0001
ATE tenant-year 0.02 -0.06 0.03 0.006
Standard error ATE 0.002 0.001 0.002 0.0006
T-value ATE 14.3 -51.3 17.1 10.3
Degrees model 75 75 75 75
N 16,386 16,386 16,386 16,386
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Dependent variable: share of budget spend on each category; log expenditure instrumented using ln net income
Variables not shown: 17 cohort dummies, log household size, log household size squared, dummies for SPH, SFH MFH
one dummy for not employed, three for type of job, and four for educational level
3.7. Results budget allocation model 79

Table 3.11: Allocation of budget (un)restricted tenants (PSM)

(1) (2) (3) (4)

Basic goods Housing Recreation Alcohol and tobacco
coef. s.e. coef. s.e. coef. s.e. coef. s.e.
Log of expenditure 0.0170∗ (0.0079) -0.2100∗∗∗ (0.0055) 0.1900∗∗∗ (0.0100) 0.0027 (0.0023)
Tenant in year 1980 0.0220∗ (0.0087) -0.0470∗∗∗ (0.0057) 0.0220∗ (0.0100) 0.0028 (0.0027)
Tenant in year 1981 0.0250∗∗ (0.0081) -0.0400∗∗∗ (0.0060) 0.0110 (0.0094) 0.0041 (0.0028)
Tenant in year 1982 0.0280∗∗ (0.0089) -0.0570∗∗∗ (0.0064) 0.0190 (0.0100) 0.0100∗∗∗ (0.0027)
Tenant in year 1983 0.0410∗∗∗ (0.0074) -0.0550∗∗∗ (0.0054) 0.0110 (0.0088) 0.0035 (0.0024)
Tenant in year 1984 0.0380∗∗∗ (0.0075) -0.0540∗∗∗ (0.0055) 0.0061 (0.0091) 0.0094∗∗∗ (0.0023)
Tenant in year 1985 0.0380∗∗∗ (0.0074) -0.0530∗∗∗ (0.0053) 0.0110 (0.0091) 0.0046∗ (0.0023)
Tenant in year 1986 0.0370∗∗∗ (0.0077) -0.0570∗∗∗ (0.0055) 0.0160 (0.0097) 0.0043 (0.0024)
Tenant in year 1987 0.0460∗∗∗ (0.0081) -0.0630∗∗∗ (0.0057) 0.0140 (0.0099) 0.0026 (0.0025)
Tenant in year 1988 0.0370∗∗∗ (0.0086) -0.0450∗∗∗ (0.0059) 0.0003 (0.0100) 0.0082∗∗ (0.0030)
Tenant in year 1989 0.0430∗∗∗ (0.0093) -0.0590∗∗∗ (0.0059) 0.0120 (0.0110) 0.0047 (0.0029)
Tenant in year 1990 0.0250∗ (0.0097) -0.0540∗∗∗ (0.0060) 0.0250 (0.0130) 0.0043 (0.0024)
Tenant in year 1991 0.0300∗∗ (0.0110) -0.0490∗∗∗ (0.0078) 0.0140 (0.0130) 0.0044 (0.0034)
Tenant in year 1992 0.0290∗∗ (0.0091) -0.0470∗∗∗ (0.0062) 0.0110 (0.0100) 0.0073∗ (0.0031)
Tenant in year 1993 0.0320∗∗∗ (0.0089) -0.0460∗∗∗ (0.0062) 0.0130 (0.0100) 0.0009 (0.0030)
Tenant in year 1994 0.0340∗∗∗ (0.0088) -0.0470∗∗∗ (0.0060) 0.0120 (0.0100) 0.0021 (0.0029)
Tenant in year 1995 0.0210∗ (0.0091) -0.0360∗∗∗ (0.0062) 0.0130 (0.0100) 0.0021 (0.0031)
Tenant in year 1996 0.0310∗∗∗ (0.0090) -0.0310∗∗∗ (0.0058) -0.0033 (0.0100) 0.0033 (0.0028)
Tenant in year 1997 0.0200∗ (0.0090) -0.0300∗∗∗ (0.0060) 0.0027 (0.0100) 0.0073∗ (0.0029)
Tenant in year 1998 0.0260∗∗ (0.0092) -0.0280∗∗∗ (0.0062) -0.0044 (0.0110) 0.0065∗ (0.0029)
Tenant in year 1999 0.0150 (0.0098) -0.0069 (0.0062) -0.0085 (0.0110) 0.0009 (0.0031)
Year 1981 -0.0094 (0.0094) 0.0054 (0.0072) 0.0045 (0.0100) -0.0005 (0.0032)
Year 1982 -0.0039 (0.0150) 0.0250∗ (0.0110) -0.0170 (0.0170) -0.0050 (0.0048)
Year 1983 -0.0260 (0.0200) 0.0310∗ (0.0150) -0.0046 (0.0230) -0.0007 (0.0064)
Year 1984 -0.0200 (0.0270) 0.0360 (0.0200) -0.0110 (0.0300) -0.0052 (0.0083)
Year 1985 -0.0270 (0.0330) 0.0400 (0.0240) -0.0120 (0.0380) -0.0011 (0.0100)
Year 1986 -0.0230 (0.0400) 0.0420 (0.0290) -0.0200 (0.0460) 0.0005 (0.0120)
Year 1987 -0.0170 (0.0470) 0.0430 (0.0340) -0.0280 (0.0540) 0.0008 (0.0140)
Year 1988 0.0013 (0.0530) 0.0200 (0.0390) -0.0220 (0.0610) 0.0003 (0.0160)
Year 1989 0.0007 (0.0600) 0.0340 (0.0440) -0.0360 (0.0680) 0.0010 (0.0180)
Year 1990 0.0160 (0.0640) 0.0310 (0.0470) -0.0480 (0.0710) 0.0009 (0.0200)
Year 1991 -0.0037 (0.0730) 0.0380 (0.0530) -0.0330 (0.0830) -0.0010 (0.0220)
Year 1992 -0.0053 (0.0790) 0.0400 (0.0580) -0.0330 (0.0910) -0.0013 (0.0240)
Year 1993 0.0012 (0.0860) 0.0400 (0.0630) -0.0440 (0.0990) 0.0033 (0.0260)
Year 1994 -0.0085 (0.0930) 0.0460 (0.0680) -0.0420 (0.1100) 0.0044 (0.0290)
Year 1995 0.0045 (0.0990) 0.0430 (0.0720) -0.0550 (0.1100) 0.0067 (0.0310)
Year 1996 -0.0100 (0.1100) 0.0520 (0.0770) -0.0440 (0.1200) 0.0029 (0.0330)
Year 1997 -0.0019 (0.1100) 0.0480 (0.0820) -0.0480 (0.1300) 0.0025 (0.0350)
Year 1998 -0.0037 (0.1200) 0.0520 (0.0870) -0.0510 (0.1400) 0.0030 (0.0370)
Year 1999 0.0094 (0.1300) 0.0310 (0.0920) -0.0470 (0.1400) 0.0067 (0.0390)
Age -0.0028 (0.0068) -0.0022 (0.0049) 0.0050 (0.0078) -<0.0001 (0.0021)
Age squared <0.0001 (<0.0001) <0.0001∗ (<0.0001) -<0.0001∗ (<0.0001) -<0.0001∗∗∗ (<0.0001)
Pr.score -0.0130 (0.0072) 0.0400∗∗∗ (0.0053) -0.0260∗∗ (0.0093) -0.0017 (0.0020)
Pr.score and tenant -0.0130 (0.0093) -0.0270∗∗∗ (0.0062) 0.0370∗∗ (0.0120) 0.0023 (0.0025)
Constant 0.4200 (0.6100) 2.4900∗∗∗ (0.4400) -2.0200∗∗ (0.7200) 0.1100 (0.1900)
adj. R2 0.122 0.567 0.279 0.046
Chi2 tenant year 78.3 376.6 31.1 34.5
P-value <0.0001 <0.0001 0.05 0.02
ATE tenant-year 0.03 -0.05 0.010 0.005
Standard error ATE 0.005 0.004 0.007 0.001
T-value ATE 5.9 -12.6 1.5 3.1
Degrees model 77 77 77 77
N 16,386 16,386 16,386 16,386
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
Dependent variable: share of budget spend on each category; log expenditure instrumented using ln net income
Variables not shown: 17 cohort dummies, log household size, log household size squared, dummies for SPH, SFH MFH
one dummy for not employed, three for type of job, and four for educational level
80 Chapter 3.

3.8 Conclusion
In this chapter the effect of housing tenure on the level and allocation of the
consumption budget has been investigated. Significant differences in con-
sumption behavior between homeowners and (rent-controlled) tenants are
found to exist. First, the relationship between housing market prices and
consumption is most likely to be driven by the common factor hypothesis.
I conclude this as the marginal effect of housing prices on consumption de-
creases with age, whereas a parabolic relationship would be expected under
the wealth hypothesis.
Using a propensity score matching procedure to control for the endogene-
ity of housing market tenure, I find that the effect of house prices on con-
sumption is positive for both home-owners and tenants. For both types, it is
concluded that the marginal effect decreases with age, which is in line with
the common factor hypothesis too. The effect of house prices is found to be
smaller for tenants than for homeowners.
Second, an Almost Ideal Demand System has been estimated to see whether
consumption patterns of controlled and uncontrolled households differ. It is
concluded that consumption patterns of homeowners and controlled tenants
are very different. Controlled tenants spend less on housing and more on ba-
sic goods, recreation and alcohol and tobacco. Evidence based on propensity
score matching suggests that this difference in consumption patterns is indeed
due to rent-control and not to heterogeneous preferences.
Additionally, the consumption patterns of homeowners and tenants have
been converging over time. This is line with the observation that rents in the
controlled housing sector have been increasing more than the imputed rents in
the owner-occupied sector. Results suggest that consumption differences be-
tween restricted and unrestricted tenants exist, but they are limited. Tenants
that most resemble homeowners spend less on housing and more on recre-
ation.
3.8. Conclusion 81

It would be interesting to augment the analysis presented in this chapter

in various directions. For instance, future research for the Netherlands could
include local measures of housing prices to capture spatial variation. This
would also allow to compute local measures of rent control benefits which
could be used to test whether consumption of tenants is related to rent control.
Also, it would be interesting to study the effect of housing market tenure on
the consumption of durables or saving behavior.
82 Chapter 3.

3.A Propensity Score Matching

If owner-occupied and renting households have equal preferences for the con-
sumption goods, the RDt dummy variables in equations (3.10) and (3.12) cap-
ture the effect of lower rents on the consumption budget shares. However,
if preferences for consumption goods differ with tenure type, part of RD
d t re-

flects the effect of individual preferences and parameter estimates are biased
and inconsistent. This mechanism is not unlikely: Table 3.3 shows that home-
owner have higher income. Research by Christensen (2007) finds that tastes
g
are correlated with income. Let αi denote individual preferences, and assume
its expectation is different for households living in owner-occupied and rent-
controlled housing. Now, equation (3.10) is given by

wi = β 0 + β 1 Xi + β 2 Hi + ∑ ζ t Dt + ∑ ζ t RDt + µi
g g g g g gR g
(3.13)
t t
g g g
µi = αi + ε i

with g = 1, .., 4; i = 1, .., N PM ; t = 1981, .., 1999

In an ideal setting, panel data estimation techniques would be used to con-

g
trol for the presence of unobserved heterogeneity αi . As I do not observe the
observations in a panel, nearest neighbour propensity score matching (PSM)
is used to control for unobserved heterogeneity. PSM constructs a N pm < N
subsample of homeowners and tenants with similar propensity to be a home-
hi and
owner using the (selection) equation (3.11). Under the assumption that b
g
αi are strongly correlated, households with equal propensity to be a home-
g
owner no longer differ structurally in terms of preferences αi .
3.A. Propensity Score Matching 83

If this is true, the parameter estimates for the PSM subsample are unbi-
ased.12, 13 Equation (3.12) is re-estimated as

1999 1999
∑ ∑
g g g g g gR
w i = β 0 + β 1 Xi + β 2 H i + ζ t Dt + ζ t0 RDt0
t=1982 t0 =1981
5 5
∑ Pci + ∑
g
+ PcR0 i + ε i (3.14)
c =2 0 c =1

with g = 1, .., 4; i = 1, .., N; t = 1981, .., 1999

Matching results

First, I present estimation results of the propensity score selection regression

(Table 3.12). It follows that the propensity to be living in owner-occupied
housing increases with income (and concave on the domain in the sample).
Households with a head younger than 40 or older than 60 are more likely
to be living in owner-occupied housing. The propensity to be home-owner
increases with education. Conditional on education, the type of job is not sig-
nificant at the five percent level. Finally, homeowners inhabited larger houses.
A caliper of two percent of the standard deviation of the propensity score
is imposed, which results in 7,543 matched pairs. Using this caliper, the null
hypothesis that the difference in group means of the variables in Zi is zero
cannot be rejected at the five percent level, see Table 3.13.14
Figure 3.4 shows that the propensity score matching procedure succeeds
in selecting a homogeneous sample of both homeowners and tenants out of
two very different subsamples. First, the solid and long-dashed line indicates
that the distribution of propensity to be a home-owner is very different for
12 The propensity score matching procedure is to estimate the propensity score b hi . Next it
matches to each owner-occupied household a household that rents housing such that the differ-
ence in propensity scores is minimized. Matching is done without replacement, i.e. each renting
household can be assigned to only one owner-occupied household. In order to control the quality
of the match a maximum distance threshold (caliper) is defined. Matched pairs with difference
in propensity score exceeding this procedure are not considered. Here, the caliper is based on the
standard error of the propensity score.
13 However, results based on PSM are less efficient than those provided by panel fixed effects.
g
Naturally, if αi is only weakly correlated with b hi , propensity score matching does not solve the
estimation bias.
14 The standard deviation of the propensity score (SDPS) equals 29.95 percent. A caliper of 2.5

percent of the SDPS (or higher) results in a statistically significant difference in group mean of (at
least) the squared value of income. Contrary, an extreme caliper of .02 percent of the SDPS yields
5,879 matched pairs and gives similar results compared to those reported.
84 Chapter 3.

homeowners and tenants. It is skewed to the right with a peak around 0.95
for owner-occupied households and skewed to the left with peak around 0.1
for tenants. However, for the subsample of matched observations the distri-
bution of propensity scores for the selected homeowners overlaps with that of
matched tenants on almost the complete domain.15

Table 3.12: Propensity score selection regression

(1)
Owner-occupied
coef. s.e.
Income
Annual net income 0.0648∗∗∗ (0.0023)
Annual net income squared -0.0000∗∗∗ (0.0000)
Household characteristics
Age -0.3191∗∗ (0.1094)
Age squared 0.0004∗∗∗ (0.0001)
log(household size) -0.4093 (0.4048)
log(household size) squared -0.3179 (0.1798)
Single person household 0.7797∗∗ (0.2417)
One family household 0.5276∗∗∗ (0.1301)
Multiple family household 0.2924 (0.1985)
One child present 0.4833∗∗∗ (0.0566)
Two children present 0.9254∗∗∗ (0.0669)
Three children present 1.1741∗∗∗ (0.1010)
Secundary, first step 0.2363∗∗∗ (0.0453)
Secundary, second step 0.5616∗∗∗ (0.0500)
Tertiary, first step 0.6545∗∗∗ (0.0548)
Tertiary, second step 0.4225∗∗∗ (0.0777)
Job characteristics
Employee 0.0778 (0.2371)
Academic job 0.3023 (0.2386)
Managerial job 0.2582 (0.2423)
House characteristics
Other 0.2245 (0.2358)
Detached SFH 2.4752∗∗∗ (0.0738)
Apartment -1.7544∗∗∗ (0.0386)
Other -0.8753∗∗∗ (0.1148)
Prior to 1945 0.8514∗∗∗ (0.0315)
After 1980 0.1250∗∗ (0.0473)
Three or four rooms -0.0607 (0.0817)
Five or more rooms 0.2358∗∗ (0.0846)
Cohort dummies YES
Year dummies YES
Constant 22.0356∗ (9.5325)
N 35,690
pseudo R2 0.293
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001

15 As should be the case, the smaller the caliper, the more the distributions of matched home-

owners and tenants overlap.

3.A. Propensity Score Matching 85

Figure 3.4: Propensity score for the full and the matched sample

Table 3.13 shows the mean and standard deviation of several variables for
tenants and owner-occupied households in the matched sample. It also shows
the difference in means and the p-value that corresponds to an equality of
means test. The table shows that for all households, job and house characteris-
tics the null hypothesis of equal means for tenants and homeowners cannot be
rejected. This also holds for income and the squared value of income. How-
ever, their expenditure and allocation over basic goods, housing, recreation
and alcohol and tobacco differs. Table 3.13 shows that rent-control tenants
spend on average 3,480 euro less than homeowners. Also, the budget share
spend on housing is on average four percent lower for tenants, which they
spend on basic goods (two percent more), recreation (one percent more) and
alcohol and tobacco (one percent more). This suggests that housing tenure
status might increase consumption and the allocation of the budget.
86 Chapter 3.

Table 3.13: Comparison treatment and control group PSM sample

Tenants Owner-occupied Difference

(1) (2) (1-2)
mean std. dev. mean std. dev. mean p-value
Age 45.590 15.120 45.440 15.210 0.150 0.514
Age squared 2,307.360 1,516.390 2,295.950 1,520.730 11.410 0.631
Log of household size 0.920 0.520 0.920 0.520 0.000 0.846
Single person household 0.160 0.370 0.160 0.370 0.000 0.798
One family household 0.820 0.380 0.820 0.380 0.000 0.855
Multiple family household 0.010 0.090 0.010 0.090 0.000 0.656
One child present 0.150 0.350 0.150 0.350 0.000 0.860
Two children present 0.210 0.410 0.210 0.410 0.000 0.789
Three or more children present 0.070 0.250 0.070 0.260 0.000 0.830
Primary education 0.170 0.370 0.170 0.370 0.000 0.707
Secondary education first step 0.250 0.430 0.250 0.430 -0.010 0.312
Secondary education second step 0.350 0.480 0.350 0.480 0.000 0.844
Tertiary education first step 0.180 0.390 0.180 0.380 0.000 0.492
Tertiary education second step 0.050 0.220 0.050 0.220 0.000 0.778
Cohort head of household
1895− <1900 0.000 0.020 0.000 0.020 0.000 0.739
1900− < 1905 0.010 0.070 0.010 0.080 0.000 0.672
1905− <1910 0.010 0.110 0.010 0.100 0.000 0.607
1910 − <1915 0.030 0.170 0.030 0.170 0.000 0.851
1915 − <1920 0.040 0.200 0.040 0.200 0.000 0.938
1920 − <1925 0.070 0.250 0.070 0.250 0.000 0.712
1925 − <1930 0.070 0.260 0.070 0.260 0.000 0.739
1930 − <1935 0.070 0.250 0.070 0.250 0.000 0.877
1935 − <1940 0.070 0.250 0.070 0.260 0.000 0.539
1940 − <1945 0.080 0.280 0.080 0.270 0.000 0.648
1945 − <1950 0.130 0.330 0.130 0.330 0.000 0.869
1950 − <1955 0.150 0.350 0.140 0.350 0.000 0.690
1955 − <1960 0.140 0.340 0.140 0.350 0.000 0.734
1960 − <1965 0.080 0.280 0.080 0.270 0.000 0.630
1965 − <1970 0.040 0.200 0.040 0.200 0.000 0.416
1970 − <1975 0.020 0.120 0.020 0.120 0.000 0.949
1975 − <1980 0.000 0.050 0.000 0.040 0.000 0.479

Job characteristicsa
Employee 0.670 0.470 0.670 0.470 0.000 0.740
Not employed 0.330 0.470 0.330 0.470 0.000 0.740
Academic job 0.140 0.350 0.140 0.350 0.010 0.358
Managerial job 0.050 0.220 0.050 0.210 0.000 0.384
Other 0.480 0.500 0.480 0.500 -0.010 0.492
No job 0.330 0.470 0.330 0.470 0.000 0.727

House characteristicsa
Detached SFH 0.030 0.160 0.030 0.180 0.000 0.072
Apartment 0.200 0.400 0.210 0.400 -0.010 0.361
Row house SFH 0.760 0.430 0.750 0.430 0.010 0.123
Constructed prior to 1945 0.370 0.480 0.380 0.480 0.000 0.735
Constructed between 1945 1979 0.520 0.500 0.520 0.500 0.000 0.815
Constructed in 1980 or later 0.110 0.310 0.100 0.300 0.000 0.356
1 or 2 rooms 0.030 0.170 0.030 0.170 0.000 0.784
3 or 4 rooms 0.350 0.480 0.370 0.480 -0.010 0.084
5 or 6 rooms 0.620 0.490 0.600 0.490 0.010 0.073

Income and expenditurea

Annual net income 57,460 21,000 57,040 20,510 420 0.192
Annual net income squared 3.74e+09 3.00e+09 3.67e+09 2.93e+09 6.87e+07 0.138
Expenditure 40,108.61 14,669.42 43,590.70 1,5501.77 -3482.09 <0.0001
Basic goods 0.380 0.100 0.360 0.120 0.020 <0.0001
Housing 0.300 0.100 0.340 0.110 -0.040 <0.0001
Recreation 0.290 0.130 0.270 0.170 0.010 <0.0001
Alcohol and tobacco 0.040 0.040 0.030 0.030 0.010 <0.0001
Household characteristicsa
a
Based on 8,193 renting and 8,193 owner-occupied consumers
Chapter 4

Rent control benefits and

housing demand

4.1 Introduction
In many countries price schedules control the social housing market in or-
der to guarantee access to housing for low-income households or to prevent
extreme price rises.1 Price controls set the price of housing below the market
price, which leads to many different distortions in the housing market.2 Previ-
ous studies found a negative relation between the amount of rent control and
the probability of moving house.3 These estimates are not fully informative
about the size of the distortion for the demand for houses in the non-controlled
1 In Europe various pricing schemes exist. The price ceiling might depend on construction

costs (Austria, Denmark, France), income of the tenant (Germany, Netherlands, UK (England,
Ireland)), negotiations between landlords and tenants (Sweden), local or central authorities deci-
sion or a combination of them (Scanlon and Whitehead, 2007, p. 16).
2 Price controls lead to an inefficient allocation of housing (Glaeser and Luttmer, 2003) and

it reduces the consumer surplus (Bulow and Klemperer, 2012). Furthermore, intertemporal con-
sumption of housing services is affected as higher rent control leads to a lower probability of
moving houses (Munch and Svarer, 2002; Simmons-Mosley and Malpezzi, 2006). Finally, rent
control also affects the labor market through its influences on matching between employers and
employees. Rent-controlled tenants are less likely to move out of their region for a job and more
likely to accept jobs within their own region (Svarer et al., 2005).
3 Empirical evidence from New York indicates that rent-controlled households exhibit lower

overall mobility rates. The durations of tenancy is longer in the rent-controlled sector and mobil-
ity rates are lower (Gyourko and Linneman, 1989; Ault et al., 1994; Nagy, 1995; Simmons-Mosley
and Malpezzi, 2006). A similar conclusion for Denmark is reached by Munch and Svarer (2002).
They find that the durations of tenancy for a typical household is six years longer if the benefits of
rent control are within the highest ten percent of rent-control benefits compared to the ten percent
lowest.
88 Chapter 4.

sector, because tenants do not lose their entitlements to rent-control benefits

if they move within the rent-controlled housing sector. Hence, it is likely that
rent control is less distortive for transitions within the rent-controlled sector,
whereas the distortion may be even more severe for transitions between both
sectors. So far, no study of rent control has distinguished between the con-
trolled and non-controlled segments of the housing market as a destination of
the household’s transition.
This chapter analyzes the effect of rent control on the transition by house-
holds from the controlled social housing sector to either of both destinations.
Both transitions within the rent-controlled housing sector (the within tran-
sition) and transitions from the rent-controlled housing sector to the non-
controlled sector (the between transition) are considered. For both types of
transitions the probability that a household moves house is inversely related
to the amount of rent control.
Two different forms of rent seeking are introduced that give rise to both
transitions. First, tenants in rent-controlled accommodation might be less
likely to move to non-controlled housing, because they want to retain the
financial benefits of rent control. I will refer to this as ’rent-keeping behav-
ior’. Second, households may increase the benefits of rent control accruing
to them by transitioning within the rent-controlled housing sector. This will
be denoted by ’rent seeking behavior’. For both the within transition and the
between transition there is a negative effect of rent control on the probability
to move house.
The empirical analysis is based on administrative information collected
from Dutch households over the period 2006-2008. About one third of the
housing stock in the Netherlands consists of rent-controlled social housing.
Rents are regulated uniformly, which ensures that the controlled rent level
is exogenous from local housing market characteristics. The measure of rent
control is the controlled rent divided by the house value. The rich nature of the
dataset enables us to control for confounding factors such as household com-
position, labor market position and local housing market characteristics. The
panel dimension enables us to control for unobserved heterogeneity, which
4.2. Economic framework 89

is important as households with innate low mobility may self-select into con-
trolled social housing.
The estimates indicate that the effect of rent control on the probability of
transition is different for both destinations. With respect to the between tran-
sition only the high-income households exhibit a decreased transition rate to
the non-controlled housing sector in a reaction to tighter rent controls. For
the within transition all households respond to higher rent-control stringency
through a decreased probability of transition. These estimates indicate that
rent control has a substantial effect on demand for controlled housing. By
contrast, the spillover effect on uncontrolled housing is relatively modest,
which can be explained by a reduced financial attainability for the low-income
households in the non-controlled sector. In other words, the negative relation-
ship between rent control and the probability of moving is driven mainly by
the decreased transition within the rent-controlled housing sector. From this it
is concluded that rent control induces rent seeking behavior of Dutch house-
holds, but that rent keeping behavior is limited.
The remainder of this chapter is organized as follows. This section is fol-
lowed by an explanation in section 4.2 of how rent control influences transi-
tions from and within the rent-controlled housing sector. Section 4.3 describes
the data. Section 4.4 presents evidence on the relationship between rent con-
trol and household mobility. Section 4.5 explains the estimation strategy. Esti-
mation results are presented in section 4.6 and section 4.7 concludes.

4.2 Economic framework

Based on the economic framework below four hypotheses are formulated. The
intuition behind this framework is that rent control changes demand for hous-
ing in both the controlled and non-controlled segments. However, if market
rents are too high, low-income households cannot afford to purchase a house
in the uncontrolled sector. Consequently, changes in rent-control benefits will
not influence their transition behavior from the rent-controlled housing sector.
Housing is assigned a controlled rent r based on the physical characteris-
tics of the house (h). The market rent p is assumed to be a function of h and the
90 Chapter 4.

access it provides to the local amenity level A(·). The level of local amenities
is defined by its location in the housing market, L, and it is measured in utility
terms. Thus, the rent ceiling and market rent are defined respectively as:

r = f (h)

p = g(h) + A( L)

The physical characteristics of the house and local amenity level are defined
such that they are always positive, h, A(·) > 0. Furthermore, it is assumed that
the controlled rent and market rent are increasing in housing quality, although
∂r ∂p
the shape of this positive relationship is not defined: ∂h > 0, ∂h > 0. Finally,
market rents are increasing in the level of local amenities provided
∂p
∂A > 0.4
Household i earns income yi . I assume that landlords of uncontrolled
housing impose financial constraints. In order to rent uncontrolled housing
the tenant’s income must exceed a certain threshold γ, that is based on the
yi
market rent: γ = g(h f ,L f )
.5 Next, define rent-control benefits b as

b(h, L) = p(h, L) − r (h)

Households derive utility from housing services, amenities provided by the

local housing market and consumption. For simplicity it is assumed that
households are risk neutral, and that utility is additive separable. Let the
utility derived from inhabiting a dwelling of quality h be H (h), and define
the utility derived from amenities in local labor market L as A( L). Super-
script u and s indicate uncontrolled and controlled social housing. House-
holds that live in uncontrolled housing have utility U u (h, L) and households
in controlled social housing will have utility U c (h, L):

U u (h, L) = H (h) + A( L) + y − p

U c (h, L) = H (h) + A( L) + y − r
4 Note that this relation implies perfect pricing of local amenities in the uncontrolled market

as for any house of quality h̃ the difference in market price equals the difference in utility derived
from the local amenities: p(h̃, Li ) − p(h̃, L j ) = A( Li ) − A(l j ).
5 For owner-occupied housing, these financial constraints reflect income requirements set by

mortgage providers.
4.2. Economic framework 91

Note that, ceteris paribus, controlled households have higher utility levels
than households living in uncontrolled housing.
Households receive a costless house offer. The house offer may be for
an uncontrolled house λu (drawn from a bivariate Poisson-distribution λu =
G u (h, L)) or for a rent-controlled house λc (drawn from a Poisson-distribution
λc = G c (h, L)). The house that is offered is exogenous to the characteristics of
the household. Because households observe the characteristics h and location
L of the house they are offered, they are able to determine the market rent
of the house offer and therefore the value of the offer (in terms of their own
utility).
If households in rent-controlled housing accept a house offer for an uncon-
trolled house λu , they will incur moving costs m. After moving their utility is
U (λu ). Similarly, if they decide to move from one controlled house to another
controlled house, their utility equals U (λc ).

U (λu ) = H ( hu ) + A( Lu ) + y − g( hu , Lu ) − m

U (λc ) = H ( hc ) + A( Lc ) + y − f ( hc ) − m

Let a transition within the rent controlled sector be denoted as tw = 1 and

a transition from the rent-controlled housing sector as tb = 1, where super-
scripts w and b refer to within transfers and between transfers respectively.
Households in rent-controlled housing who receive an offer for uncontrolled
housing will move if the utility from the house offer minus moving costs ex-
ceeds the utility of the current dwelling. Thus, the value of tb is given by
equation (4.1), which shows that a household will move if two conditions are
met. First, the change in utility from house characteristics minus the change in
housing costs (measured in market rents) should exceed the benefits from rent
control of the current house plus moving costs. Second, the financial threshold
attached to the uncontrolled house offers should not be binding.




 1 if H (hu ) − H (hi ) + A( Lu ) − A( Li ) − g(hu , Lu )


tib = ≥ b(hi , Li ) + m and yi /g(hu , Lu ) > γ (4.1)




0 otherwise
92 Chapter 4.

Second, households in controlled social housing, who receive an offer for con-
trolled social housing, will move if the utility derived from this offer minus
the moving costs U (λw ) − m exceeds the utility of the current dwelling U c .
Rearranging and using b = p − r gives




 1 if H (hc ) − H (hi ) + A( Lu ) − A( Li ) − g(hu , Lu ) + g(hi , Li )+


tiw = b ( h c , L c ) ≥ b ( hi , Li ) + m (4.2)




0 otherwise

A household will move within the rent-controlled housing sector if the change
in utility attributed to house characteristics minus the change in housing costs
(in terms of market rents) exceeds the change in benefits from rent control
of the current house plus moving costs. This condition is similar to the one
that describes the transition from the rent-controlled to uncontrolled housing,
although there are important differences.
The first difference is that financial constraints only affect transitions out of
the rent-controlled housing sector, but they do not restrict transitions within
the rent-controlled housing sector. Second, even in the absence of financial
constraints (yi /g(hu , Lu ) > γ), households are more likely to move within the
rent-controlled housing sector than from the rent-controlled housing sector.
f
To see this, note that if hc = hu and Lc = Lu , then P(tiw = 1) > P(ti = 1).
Equations (4.1) and (4.2) are used to derive hypotheses about transition
behavior from and within the rent-controlled housing sector and the size of
rent control benefits. Equation (4.1) implies that, ceteris paribus, an increase
in rent-control benefits b(hi , Li ) reduces the number of transitions from the
rent-controlled housing sector.

Hypothesis 1: Transitions from the rent-controlled housing sector are de-

creasing in rent-control benefits
Provided that Yi /g(h f , L f ) > γ, an increase in rent-control benefits b(hi , Li ) de-
creases the probability that the first inequality in equation (4.1) holds. Ceteris paribus,
this reduces the probability that the utility of the (uncontrolled) house offer net of mov-
ing costs exceeds the utility from remaining in the current house.
4.2. Economic framework 93

The same line of reasoning holds for transitions within the rent-controlled
housing sector.

Hypothesis 2: Transitions within the rent-controlled housing sector are de-

creasing in rent-control benefits
An increase in rent-control benefits b(hi , Li ) decreases the probability that the in-
equality in equation (4.2) holds. Ceteris paribus, this reduces the probability that the
utility of the (rent-controlled) house offer net of moving costs exceeds the utility from
living in the current house.

The key difference between equations (4.1) and (4.2) is that an additional in-
come restriction is imposed. In order to move from the rent-controlled hous-
ing sector, the uncontrolled house offer should be affordable (yi /g(hu , Lu ) >
γ). Suppose that for household i this restriction does not hold. In this case, a
change in rent-control benefits b(hi , Li ) does not affect the probability of tran-
sition to the uncontrolled housing sector, as the household cannot move to the
offered house. This leads to the following hypotheses:

Hypothesis 3: Transitions out of the rent-controlled housing sector of finan-

cially constrained households are not affected by rent-control benefits
Consider a household that receives a house offer λu for uncontrolled housing. The
household is financially constrained, yi /g(hu , Lu ) ≤ γ. Ceteris paribus, a change in
rent-control benefits does not change their probability of accepting the uncontrolled
house offer, as the second condition in equation (4.1) is not met and does not depend
on rent control.

Hypothesis 4: Transitions within the rent-controlled housing sector of fi-

nancially constrained households are affected by rent-control benefits
Consider a household that receives a house offer λc for controlled social housing. In
this case, an increase in rent-control benefits decreases the probability they will accept
the controlled house offer, as it increases the RHS of equation (4.2).
94 Chapter 4.

4.3 Data
Empirical evidence is based on data from various administrative datasets that
were provided by Statistics Netherlands over the years 2006, 2007 and 2008.
The starting point is a database recording all employees (SSB Banen), self-
employed (SSB Zelfstandigen) and households on rent support (Raamwerk hu-
urtoeslag, provided by the Ministry of Internal Affairs) in 2006, 2007 and 2008.
These data sets contain information on labor market status, and they were
matched with their administrative records of municipalities (Gemeentelijke Ba-
sis Administratie), which allowed us to observe the address of the household
as well as additional administrative information on household composition.
I tracked transitions of individual households across different addresses over
the three years. As employees may not be the only income source for the
household, I obtained household income from the dataset Integraal Inkomens-
bestand.
To obtain a measure of the type of transition, the data sets were matched
with the WRG woonruimteregister verrijkt which contains information on the
dwelling type and the value of all houses in the Netherlands. Additional lo-
cal housing market characteristics were derived from the Volkshuisvesting In-
formatieSysteem (VoIS) that is used for policy-making and monitoring by the
Dutch Ministry of Internal Affairs. These additional variables provided the
size of the housing stock and the share of social rental housing within the mu-
nicipality.
Because of computational limitations, a random sample of five percent
of the population of (self-) employed households and households on rent-
support has been selected. Observations were only included in the analy-
sis if the head of household was older than 30, and had complete records on
household composition and address. For each household only the head of the
household was selected. The sample contains 91,703 households that were
observed over a maximum of three years. It gives us an unbalanced panel of
234,020 observations, that will be used for the empirical analyses.6
6 Tenants who rent from private agents were omitted from the analysis, as they are more likely

to rent uncontrolled housing at uncontrolled rents. A household is considered rent controlled if it

rents from a housing agency and the value of the house is less than 300 thousand euro. Around
4.3. Data 95

Table 4.1: Summary statistics

mean sd av(<=p5)a av(>=p95)b

General statisticsc
Transition dummy (1 if within transition or between transition) 5.25
Within transitions (1 if within transition) 3.61
Between transitions (1 if between transition) 1.80

Household summary statisticsc

Marginal price of rent-controlled housingd 270.17 83.93 137.73 494.20
Market marginal price of (rent-controlled) housinge 433.93 56.66 349.96 587.38
House value (in 100,000 euro) 1.56 0.48 0.77 2.79
Income of households in first income decile Y117 (in 10,000 euro) 0.95 0.39 -0.16 1.16
Income of households in second income decile Y117−129 (in 10,000 euro) 1.23 0.03 1.17 1.28
Income of households in third income decile Y129−140 (in 10,000 euro) 1.34 0.03 1.29 1.39
Income of households in fourth income decile Y140−150 (in 10,000 euro) 1.44 0.03 1.40 1.49
Income of households in fifth income decile Y150−161 (in 10,000 euro) 1.55 0.03 1.50 1.60
Income of households in sixth income decile Y161−175 (in 10,000 euro) 1.67 0.04 1.61 1.74
Income of households in seventh income decile Y175−194 (in 10,000 euro) 1.83 0.05 1.75 1.93
Income of households in eight income decile Y194−222 (in 10,000 euro) 2.07 0.08 1.94 2.21
Income of households in ninth income decile Y222−267 (in 10,000 euro) 2.42 0.13 2.22 2.65
Income of households in tenth income decile Y267 (in 10,000 euro) 3.42 1.52 2.68 7.23
Age of head of household is between 30 up to 40 dummy (1 if true) 0.21
Age of head of household is between 40 up to 50 dummy (1 if true) 0.27
Age of head of household is between 50 up to 65 dummy (1 if true) 0.30
Age of head of household is between 65 up to 75 dummy (1 if true) 0.10
Age of head of household is 75 or higher (1 if true) 0.12
Marital status dummy (1 if married) 0.33
Multiple adults dummy (1 if there are multiple adults in the household) 0.38
Children dummy (1 if there are children in the household) 0.37
Nojob dummy (1 if without job) 0.36
Self-employed dummy (1 if self-employed) 0.06
Fixed contract dummy (1 if fixed contract type) 0.54
Fulltime contract dummy (1 if fulltime contract) 0.40
Fulltime and fixed contract dummy (1 if fulltime and fixed contract) 0.36

Municipality summary statisticsf

Size of municipalities in sample (number of observations within municipality) 1,338.18 2,312.26 164.00 8,437.55
Size of housing stock (in units) 15,888.55 28,804.71 2,151.82 100,690.00
Size of rental housing stock (in units) 5,228.67 13,332.34 400.35 41,906.45
Percentage of housing stock that is rent-controlled (in percentages) 27.05 7.59 13.05 44.16
Median value owner-occupied housing stock (in 100,000 euro) 2.42 0.57 1.47 3.88

District summary statisticsg

Size of district in sample (number of observations within district) 321.29 418.24 11.37 1,707.91
Median value owner-occupied housing stock (in 100,000 euro) 1.99 0.69 0.98 3.92
Median value rent-controlled housing stock (in 100,000 euro) 1.51 0.38 0.84 2.38
Percentage of housing stock that is rent-controlled (in percentages) 0.36 0.18 0.09 0.79
Median income of employed households (in 1000 euro) 22.43 2.94 17.28 30.20

Neighborhood summary statisticsh

Size of neighborhood in sample (number of observations within neighborhood) 96.97 105.98 3.39 406.85
Median value owner-occupied housing stock (in 100,000 euro) 1.89 0.80 0.92 3.93
Median value rent-controlled housing stock (in 100,000 euro) 1.55 0.52 0.81 2.92
Percentage of housing stock that is rent-controlled (in percentages) 0.45 0.23 0.07 0.95
Median income of employed households (in 1000 euro) 21.95 3.66 16.15 30.47

a
Average value of the 0-5th percentile
b
Average value of the 95-100th percentile
c
Based on 234,020 observations for 91,703 households
d Monthly controlled rent
Measure of rent-control stringency defined as House value/100,000 . Lower values imply larger rent-control benefits.
e monthly market rent
Calibrated measure of market marginal price of housing defined as House value/100,000 . monthly market rent is defined as
(4000 + 0.031(House value))/12.
f
433 unique municipality observations
g
665 unique district observations
h
2381 unique neighborhood observations
96 Chapter 4.

Summary statistics for all variables used in the empirical analysis are pro-
vided in Table 4.1. The measures of within transitions and between transitions
are similarly defined. The measure of within transitions equals 1 if the first
transition in a particular year is to rent-controlled housing and zero elsewhere.
I focus on the first transition within the calendar year, as only a small number
of households move more than once in the same calender year. Similarly, the
measure of between transitions equals one if the first transition in a particular
year is to non-controlled housing and zero elsewhere. Using these measures
for moves and their directions, the overall probability that a household de-
cides to move equals 5.25 percent. The transition rate to non-controlled hous-
ing is lower than the transition rate to rent-controlled housing. 3.61 percent
of transitions are within the rent-controlled housing sector and 1.8 percent are
between transitions.
The average house value of rent-controlled tenants is 156,000 euro. Aver-
age standardized income equals 18.000 euro, which is far below the median
income of 33,000 euro. This result is not surprising, given that I have selected
households living in rent-controlled housing. Around one-third of the sam-
ple include a married couple, and around 37 percent of the households have
children. In around 38 percent of the households multiple adults are living,
indicating that the majority of households have only one adult. 53 percent of
households are on rent support. 36 percent of the households do not have a
wage earner residing, 54 percent of households are on fixed contracts. 40 per-
cent of the households sampled have a fulltime contract, for 36 percent this is
fixed and fulltime. Almost six percent of the population are self-employed.
The set of local housing market characteristics consists of the median house
value for every municipality, district and neighborhood, where districts and
neighborhoods are classified by Statistics Netherlands. In total I observe all
433 municipalities, 2099 districts and 6074 neighborhoods in 2008. However,
data was not drawn from all districts and neighborhoods, because not all dis-
tricts and neighborhoods contain (controlled) rental housing. All 433 munic-
ipalities are sampled. The average number of observations in a municipality

three percent of the total housing stock managed by housing agencies are non-controlled hous-
ing(CSED, 2010).
4.3. Data 97

equals 1338. Furthermore, the sample contains 665 districts (339 observations
on average) and 2381 neighborhoods (98 observations on average). All hous-
ing market statistics at the municipality, district and neighborhood level have
been computed using the local population for each spatial region.

Measuring rent-control benefits

Next, I detail the construction of the measure of rent-control benefits. The

data set WRG woonruimteregister verrijkt contains a variable that measures the
value of the house if it were to be sold freely at the market (in Dutch: WOZ-
waarde). The value of the house is assessed annually for both owner-occupied
and rental housing. The assessment is based on the characteristics of the house
and the selling price of comparable housing in the neighborhood that has been
sold in the months prior to the month of assessment.7 The data set contains
only rents for tenants who are on rent support. The individual rent is proxied
by the median rent of households on rent support within the neighborhood.
This is a valid approach, because the rent support program is very generous,
and therefore households on rent support are located across the entire spec-
trum of regulated housing. The stock of controlled social housing is almost
homogeneous within neighborhoods and regulated rents vary little within
neighborhoods.8
The individual benefit of living in rent-controlled housing is computed as
the median rent in a neighborhood j (r j ) divided by the value of the house
of the tenant (Vi ), which is divided by e100,000. This measure of regulation
benefits equals the marginal price of rent-controlled housing in the case that a
house with value of e100,000 provides one unit of housing services. This ap-
proach is similar to the one undertaken by Munch and Svarer (2002) although
these authors divide by the market rent which is assumed to be a constant
fraction of the house value.9
7 Owners of housing pay taxes over the value of the house, and can comply if, according to

them, the value of the house is wrongly assessed.

8 On average about 75 percent of the observed controlled rents within a neighborhood are

within 20 percent of the median value.

9 In Denmark the yearly market rent was equal to eight percent of the house value (Munch and

Svarer, 2002). If the market rent is a fraction of the house value, both methods give statistically
equal results.
98 Chapter 4.

Figure 4.1 shows the marginal price of rent-controlled housing based on

the median rent against the marginal price of rent-controlled housing based
on the individual rent for 5000 tenants who receive rent support. It illus-
trates that both measures give similar results. A bivariate regression of the
real marginal price of rent-controlled housing on the marginal price of rent-
controlled housing based on the median rent yields a coefficient equal to 0.82
and is significant at the one percent level (the robust standard error equals
0.0189 based on 141.861 observations). The R2 equals 60.7 percent.

Figure 4.1: Median and observed marginal price of rent-controlled housing

The graph is based on a random subsample of 5,000 observations as Statistics

Netherlands does not allow the exportation of individual observations. Therefore
the house value is adjusted with a random jitter between -2,000 and 2,000 euro.
As the adjustment is small compared to the house value, it does not affect Figure
4.1.

Figure 4.2 shows kernel density estimates of the observed measure of rent
control benefit (marginal price of rent-controlled housing) and the market rent
of rent-controlled housing (market marginal price of housing). The market
marginal price of housing has been calculated by dividing the market rent by
the house value measured in 100,000 euro. Here, I take as market rent the
user-costs of owner-occupied housing, which has been established by Koning
et al. (2006) to equal to 3,000 euro plus 3.1 percent of the value of the house
annually. The left panel shows that the distribution of the marginal price of
4.3. Data 99

Figure 4.2: Kernel density plot

The graph is based on a random subsample of 5,000 observations as Statistics

Netherlands does not allow the exportation of individual observations. Therefore
the house value is adjusted with a random jitter between -2,000 and 2,000 euro.
As the adjustment is small compared to the house value, it does not affect Figure
4.2. The graph is representative for the full, non-jittered sample which is used in
the regression analysis.

Figure 4.3: Marginal prices of rent-controlled housing and house value

The graph is based on a random subsample of 5,000 observations as Statistics

Netherlands does not allow the exportation of individual observations. Therefore
the house value is adjusted with a random jitter between -2,000 and 2,000 euro.
As the adjustment is small compared to the house value, it does not affect the
pattern in Figure 4.3. The graph is representative for the full, non-jittered sample
which is used in the regression analysis.
100 Chapter 4.

rent-controlled housing is wider and to the left of the market marginal price
of housing. Nearly all observations of the marginal price of rent-controlled
housing are within 75 and 500 euro per unit of housing services, with a peak
at around 200. By contrast, nearly all observations of the market marginal
price of housing, are within 300 and 600 euro per unit of housing services,
with a peak at 400.
Figure 4.3 shows the difference between the observed marginal price of
rent-controlled housing and the calibrated market marginal price of housing.
Each observed marginal price of rent-controlled housing is presented as a dot,
while the market price of rent-controlled housing is indicated by the black line.
The figure shows that both the marginal prices of housing are convex to the
origin in house value. Moreover, the difference between the market marginal
price of rent-controlled housing and the rent-controlled marginal price of rent-
controlled housing is increasing in house value. Tenants of high-valued hous-
ing receive more rent-control benefits than tenants of low-valued housing. For
a very small minority of observations the marginal price of rent-controlled
housing exceeds the market price of housing.

4.4 Rent control and household mobility

Next, I focus on household transitions and relate it to the amount of rent con-
trol. Figure 4.4 shows that the probability of moving from the rent-controlled
housing sector is smaller than the probability of moving within this housing
sector for all levels of rent-control benefits. It seems that transitions from the
rent-controlled housing sector are barely related to the marginal price of rent-
controlled housing. By contrast, households that pay a high marginal price
for rent-controlled housing (receiving low rent-control benefits) move more
often within the rent-controlled housing sector than households that pay a
low marginal price of rent-controlled housing (receiving a higher level of rent-
control benefits). This is important preliminary evidence that both types of
transition are fundamentally different.
Figure 4.5 indicates that both types of transition are affected differently
by income. First, the probability of moving from the rent-controlled housing
4.4. Rent control and household mobility 101

sector is increasing in income, but there is no evident relationship between in-

come and transitions within the rent-controlled housing sector. This suggests
that financial constraints play a minor role in explaining transitions within the
rent-controlled housing sector, but that they are important for transitions out
of the rent-controlled housing sector. This is exacerbated by the finding that
households with annual standardized income below 25,000 euro move more
often within the rent-controlled housing sector, whereas households with a
higher income are more likely to move to uncontrolled housing.
In sum, Figures 4.4 and 4.5 provide crude evidence in favor of some of
the listed hypotheses. Figure 4.4 shows that the relationship between transi-
tions from the rent-controlled housing sector and the marginal price of rent-
controlled housing is unclear. At first sight, rent control benefits do not ap-
pear to influence transitions out of the rent-controlled housing sector. The
data provides no evidence in favor of Hypothesis 1 that transitions from the
rent-controlled sector are decreasing in rent-control benefits (increasing in the
marginal price of rent-controlled housing). Figure 4.4 also shows that the
probability of transition within the rent-controlled housing sector is increas-
ing in the marginal price of rent-controlled housing, providing some evidence
in support of Hypothesis 2.
Finally, Figure 4.5 shows that the probability of a transition out of the rent-
controlled housing is strongly increasing in income. This indicates that afford-
ability of housing is important for transitions from the rent-controlled housing
sector, which provides evidence supporting Hypothesis 3. By contrast, it also
shows that the effect of income on the probability of moving to a house within
the rent-controlled sector is weak. At best, it is slightly decreasing, which can
be explained by special arrangements for particular low-income groups. This
evidence is conflicting with Hypothesis 4. To further investigate the hypothe-
ses, I turn to regression analysis.
102 Chapter 4.

Figure 4.4: Transitions by destination and the rent-controlled marginal price

of housing

Figure 4.5: Transitions by destination and income

4.5. Empirical model 103

4.5 Empirical model

To investigate the three hypotheses more carefully three Linear Probability
Models are specified. First, in order to compare the results for the Netherlands
with those for the US and Denmark, I estimate how rent-control stringency
influences the probability of moving. At this stage the destination of the move
is neglected. The specification is

10 9
mit = β 0 + ∑ βk r ∗ Ykit + ∑ γl Ylit + Xit η + δt + ai + ε it (4.3)
k =1 l =1

i = 1, . . . , N; t = 1, . . . , Ti

where the dependent variable m indicates household moves; subscripts i and

t refer to the i-th household and t-th year, respectively. m is explained by rent-
control benefits r, where I allow the effect of rent-control on mobility to differ
across ten income groups (r ∗ Yk ) as in equation (4.3). Additional controls for
household, job and local housing market effects are included in X. The vector
X consists of variables identifying features of households, their labor mar-
ket positions and the local housing market to control for confounding factors
influencing housing demand, affordability and the offer rate respectively.10
Time dummies δt are also included. Finally, αi is an individual unobserved
fixed effect and ε it is an i.i.d. error term.
Next, I control for the destination of the move and estimate two additional
empirical specifications, see equation (4.4). Both equations have different de-
pendent variables, but their right-hand-side (RHS) variables are identical to
those in equation (4.3). The first equation is for z = b, where b represents the
between transition of the household from the controlled to the non-controlled
sector. In the second equation z = w, where w refers to the transition of house-
10 A list of included variables is found in Table 4.1. Many of these characteristics vary very

little over the sample period. As a result, coefficients on these regressors are identified by a small
group who switch status. For those that do not switch status, the effect ends up in the individual
fixed effect αiz . Therefore, the fact that the almost time-invariant variables are identified for a small
group of households does not influence the parameter estimates. In particular, as the measure of
rent-control benefits is continuous, its coefficient is identified for all households in the estimation
sample.
104 Chapter 4.

holds within the controlled sector. The specifications are

10 9
mitz = βz0 + ∑ βzk r ∗ Ykit + ∑ γlz Ylit + Xit η z + δtz + aiz + εzit (4.4)
k =1 l =1

z = b, w; i = 1, . . . , N; t = 1, . . . , Ti

For the between transition, the dependent variable mitb is an indicator that be-
comes one if the household moves from controlled to non-controlled housing.
Similarly, for the within transition mitw is one if the household moves from
controlled social housing to controlled social housing. Estimation of equation
(4.4) allows us to test the hypotheses. Hypothesis 1 and Hypothesis 2 are con-
firmed if the estimated coefficients on rent-control stringency (βzk ) are positive.
In order to test whether rent-control benefits decrease transitions between and
within the rent-controlled housing sector, I also compute the average marginal
effect of a change in rent-control stringency.11
Hypothesis 3 states that transitions out of the rent-controlled housing sec-
tor of financially constrained households are not affected by rent-control ben-
efits. As low-income households are more likely to be financially constrained,
I would expect βbk to be small and insignificant if k is a low number.
Finally, Hypothesis 4 states that transitions within the rent-controlled hous-
ing sector low-income households are affected by rent-control benefits as fi-
nancial constraints are absent. According to this hypothesis, βw
k is positive and
significant for all income deciles, also for those including low-income house-
holds.
Equation (4.4) is estimated as a Linear Probability Model (LPM). Some
studies have made use of duration models in which competing risks have
been estimated. Also, conditional fixed effects logit models have been used.
Duration models can be used to control for left and right censoring of the data
and to estimate competing risks models. In this chapter duration models are
not used as the period of observation is very small. Estimators have been de-
veloped to estimate the contribution of individual specific effects to duration
time in multiple spell data.12 Although the LPM estimation procedure does
there are ten parameters I test whether ζ = ∑k βzk /10 = 0.
11 As
12 However,
this estimator is not (yet) available in the remote access environment of Statistics
Netherlands.
4.6. Results 105

not control for duration explicitly, it does control for ‘innate immobility’ of the
worker (by estimating the unobserved fixed effects) which are expected to be
correlated with duration in the house.
Conditional fixed effects models have the advantage that the prediction of
the dependent variable is between zero and one, whereas an LPM might lead
to under- or overshooting. A drawback of logit models is that the parameter
estimate on an interaction term does not equal the marginal effect of that vari-
able and might even be of opposite sign (Ai and Norton, 2003, p.123). In ad-
dition, the conditional fixed effects model is identified only for workers who
move. This might result in an estimation sample that is reduced in size consid-
erably. In LPM models the parameter estimate equals the marginal effect and
it can be estimated on a larger sample as only those who move to competing
housing drop out the estimation sample. Results using the conditional fixed
effects logit estimator are presented in section 4.A.

4.6 Results
This section discusses the estimated coefficients and their relationship with
Hypotheses 1-4. First, we present estimates of equation (4.3) is estimated to
determine whether estimation results for the overall transition rate are sim-
ilar to those found in previous research. The estimated coefficients in Table
4.2 imply that the higher the rent-control benefits, the less likely households
are to move. These results are consistent with previous empirical work. The
parameter of the average marginal effect equals 0.0291 and is significant at
the one percent level. Thus, a one standard deviation increase in rent-control
stringency decreases the probability of moving with 2.9 percentage points. As
the average transition rate equals 5.25 percent (see Table 4.1), this effect is also
economically significant. In addition, rent control has a stronger effect on mo-
bility for high-income groups compared to low-income groups.
Furthermore, households in high-value housing move less often. For each
100,000 euro increase in house value, the probability of moving decreases by
4.8 percentage points. I also find that households in high-income districts
and neighborhoods move less often. Each increase of the median income in
106 Chapter 4.

Table 4.2: LPM of probability to move

Dependent variable:
Transition=1
Interaction terms (normalized) rent control stringency r and income group dummies a,b
r ∗ Y117 0.0196***
(0.0067)
r ∗ Y117−129 0.0185***
(0.0066)
r ∗ Y129−140 0.0159**
(0.0067)
r ∗ Y140−150 0.0153**
(0.0067)
r ∗ Y150−161 0.0175***
(0.0067)
r ∗ Y161−175 0.0212***
(0.0068)
r ∗ Y175−194 0.0306***
(0.0071)
r ∗ Y194−222 0.0350***
(0.0074)
r ∗ Y222−267 0.0499***
(0.0077)
r ∗ Y267 0.0676***
(0.0086)

Income dummies. Base is household with annual income below 11,700 euro
Y117−129 0.0042
(0.0108)
Y129−140 0.0103
(0.0121)
Y140−150 0.0163
(0.0131)
Y150−161 0.0077
(0.0137)
Y161−175 0.0032
(0.0144)
Y175−194 -0.0139
(0.0156)
Y194−222 -0.0063
(0.0169)
Y222−267 -0.0131
(0.0183)
Y267 -0.0109
(0.0216)

Household, house and local housing market characteristicsc

House value (in 100,000 euro) -0.0482***
(0.0150)
Median income in district (in 1,000 euro) -0.0089*
(0.0046)
Median income in neighborhood (in 1,000 euro) -0.0179***
(0.0031)
Observations 234,020
Number of households 91,703
R2 (within) 0.024
F-test income dummies F(9, 91702) = 0.72
Prob > F = 0.6871
F-test interaction terms of r with income dummies F(10, 91702) = 8.67***
Prob > F = <0.0001
Average marginal effect of the interaction terms of r with the income dummies 0.0291 ***
0.0063
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
a
Interaction term where r is interacted with an income dummy YA− B . r measures rent control stringency and is defined
monthly controlled rent
as House value/100,000 . The mean of r equals 270 with standard deviation 84. Lower values of r imply larger rent-control
benefits. YA− B is a dummy equal to one if income is greater than or equal to A and smaller than B, and is zero otherwise.
b
The average marginal effect of rent control stringency -computed using the Delta method- equals 0.0291 *** and is
significant at the 1 percent significance level with an standard error equal to 0.0063.
c
Explanatory variables included but not shown: dummies for being on rent support, having multiple adults in the
household, being married, having children, age group and indicating contract type (fixed, fulltime, fulltime and fixed or
having no job), being self-employed and two year dummies. Median value of owner-occupied housing in the municipality,
median house value and median house value of rent-controlled housing in the district and neighborhood. Percentage of
rent-controlled housing in the district and neighborhood. Housing stock, rent-controlled housing stock and percentage of
rent-controlled housing in the municipality. The inclusion of municipality dummies does not change estimated
parameters.
4.6. Results 107

district by 1,000 euro decreases the probability of moving by 0.9 percentage

points. This effect is even stronger at the neighborhood level. Here, each
additional 1,000 euro increase in median income decreases the probability of
moving by 1.8 percentage points. All three estimates indicate that households
in attractive places move less often.
Next, columns one and two of Table 4.3 provide estimation results for tran-
sitions out of and within the rent-controlled housing sector separately. First,
evidence is provided that rent-control benefits have a stronger effect on within
transitions than on between transitions. The average marginal effect of the
rent-controlled marginal price of housing on within transitions equals 0.0291
and is significant at the 1 percent level. A one standard deviation increase in
the rent-controlled marginal price of housing increases the within transition
rate by 2.9 percentage points. However, for between transitions this coeffi-
cient is 0.0009 and insignificant at the 10 percent level. These results provide
evidence for Hypothesis 2, but not for Hypothesis 1.
Mixed evidence is found for Hypothesis 3. In line with the hypothesis, an
one standard deviation decrease in the rent-controlled price of housing de-
creases the probability of a between transition by 2.7 percentage points for the
tenth income decile.
However, a positive coefficient is found for households in the first four in-
come deciles. The coefficients are statistically significant at the 5 percent level.
Their economic significance is low: For the four lowest income groups, a one
standard deviation increase in the rent-controlled marginal price of housing
decreases the probability of a between transition by 0.5 percentage point. This
is around one quarter of the effect on within transitions for the same income
group.
I find that the probability of a within transition is increasing in the rent-
controlled marginal price of housing. The lower are rent-control benefits, the
more likely a household is to move house within the rent-controlled sector.
moreover, this relationship is increasing in income. A one standard deviation
increase in the rent-controlled marginal price of housing increases the proba-
bility of moving house within the controlled sector by 2.3 percentage points
108 Chapter 4.

Table 4.3: LPM of moving within and from the social housing sector

Dependent variables: transition =1

Within Between
Interaction terms (normalized) rent control stringency r and income group dummies a,b
r ∗ Y117 0.0228*** -0.0050**
(0.0066) (0.0023)
r ∗ Y117−129 0.0217*** -0.0050**
(0.0065) (0.0021)
r ∗ Y129−140 0.0194*** -0.0049**
(0.0066) (0.0021)
r ∗ Y140−150 0.0190*** -0.0052**
(0.0661) (0.0022)
r ∗ Y150−161 0.0197*** -0.0035
(0.0066) (0.0023)
r ∗ Y161−175 0.0232*** -0.0029
(0.0067) (0.0024)
r ∗ Y175−194 0.0306*** 0.0005
(0.0069) (0.0028)
r ∗ Y194−222 0.0368*** 0.0001
(0.0071) (0.0033)
r ∗ Y222−267 0.0465*** 0.0076*
(0.0074) (0.0039)
r ∗ Y267 0.0515*** 0.0269***
(0.0080) (0.0054)

Income dummies. Base is household with annual income below 11,700 euro
Y117−129 0.0061 -0.0017
(0.0104) (0.0036)
Y129−140 0.0130 -0.0035
(0.0116) (0.0045)
Y140−150 0.0192 -0.0031
(0.0125) (0.0052)
Y150−161 0.0152 -0.0080
(0.0130) (0.0057)
Y161−175 0.0086 -0.0062
(0.0136) (0.0063)
Y175−194 -0.0066 -0.0111
(0.0146) (0.0077)
Y194−222 -0.0135 0.0019
(0.0157) (0.0094)
Y222−267 -0.0223 0.0012
(0.0167) (0.0112)
Y267 -0.0192 -0.0120
(0.0190) (0.0158)

Household, house and local housing market characteristicsc

House value (in 100,000 euro) -0.0478*** 0.0006
(0.0148) (0.0051)
Median income in district (in 1,000 euro) -0.0097** 0.0001
(0.0046) (0.0013)
Median income in neighborhood (in 1,000 euro) -0.0179*** -0.0003
(0.0030) (0.0009)
Observations 228,939 222,354
Number of households 89,310 89,267
R2 (within) 0.017 0.032
F-test income dummies F( 9; 89,309) = 1.12 F( 9; 89,266) = 0.84
Prob > F = 0.3423 Prob > F = 0.5836
F-test interaction terms of r with income dummies F( 10; 89,309) = 5.99*** F( 10; 89,266) = 4.68***
Prob > F = <0.0001 Prob > F = <0.0001
Average marginal effect of the interaction terms of r with the income dummies 0.0291 *** 0.0009
0.0063 0.0023
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

a
Interaction term where r is interacted with an income dummy YA− B . r measures rent control stringency and is defined
monthly controlled rent
as House value/100,000 . The mean of r equals 270 with standard deviation 84. Lower values of r imply larger rent-control
benefits. YA− B is a dummy equal to one if income is greater than or equal to A and smaller than B, and is zero otherwise.
b
The average marginal effect of rent control stringency -computed using the Delta method- equals 0.0291 *** for
transitions within the rent-controlled housing sector. It is significant at the 1 percent significance level with an standard
error equal to 0.0063. The average marginal effect of rent control stringency on between transitions equals 0.0009 and is
insignificant at the 10 percent significance level as its standard error equals 0.0023.
c
Explanatory variables included but not shown: dummies for being on rent support, having multiple adults in the
household, being married, having children, age group and indicating contract type (fixed, fulltime, fulltime and fixed or
having no job), being self-employed and two year dummies. Median value of owner-occupied housing in the municipality,
median house value and median house value of rent-controlled housing in the district and neighborhood. Percentage of
rent-controlled housing in the district and neighborhood. Housing stock, rent-controlled housing stock and percentage of
rent-controlled housing in the municipality. The inclusion of municipality dummies does not change estimated
parameters.
4.7. Conclusion 109

for the first income decile. However, for the tenth income decile this is 5.2
percentage points. This provides evidence in favor of Hypothesis 4
In additional, evidence is found that within transitions are affected by
house value and local housing market characteristics. There is no evidence
that these variables influence between transitions. If house value increases by
100,000 euro, the probability of making a within transition decreases by 4.8
percentage points. The estimated coefficient that measure the effect of house
value on between transitions is insignificant.
Similarly, it is concluded that median income level in the district and neigh-
borhood influence within transitions, but not between transitions. This effect
is stronger at the neighborhood level. The coefficients on house value, me-
dian income in the district and median income in the neighborhood show that
households are less likely to make within transitions if they consume housing
of better quality, where quality refers to the house and the location. I found
no evidence that between transitions are influenced by house value, median
income in the district or median income in the neighborhood.

4.7 Conclusion
Price ceilings change demand for the controlled good as well as for substitute
goods. The economic literature has established that rent control reduces hous-
ing demand, but did not distinguish between the effect on demand for con-
trolled and uncontrolled housing. Two separate mechanisms why rent control
might affect housing demand for controlled and uncontrolled housing differ-
ently have been presented in this chapter, exposing a deeper relationship be-
tween rent control and housing demand. Transitions out of the rent-controlled
sector are explained by rent keeping behavior: the lower the benefits house-
holds receive from rent-control, the more likely they are to move to uncon-
trolled housing, as they give up fewer benefits from rent control. Transitions
within the rent-controlled housing sector are explained by rent seeking be-
havior: the lower the benefits households receive from rent-control, the more
likely they are to move to controlled social housing, as this might increase
their rent-control benefits. The conclusions are threefold.
110 Chapter 4.

The first conclusion is that the negative relationship between rent con-
trol and housing demand is explained entirely by a fall in demand for rent-
controlled housing. In line with the aforementioned previous empirical lit-
erature on rent control and housing demand rent control benefits are found
to lower the overall transition rate. On average, a one-standard deviation in-
crease in rent-control benefits decreases the overall transition rate with 2.9
percentage points. No evidence is found that, on average, rent control bene-
fits influence demand for uncontrolled housing. Thus, I find evidence in favor
of rent seeking, but no evidence in favor of rent keeping.
The second conclusion is that income moderates the effect of rent control
on demand for uncontrolled housing. Although I do not find evidence for rent
keeping behavior on average, the highest income decile engages in rent keep-
ing behavior and postpones transitions to the uncontrolled housing sector in
response to rent control. The economic model attributes this difference to the
financial unattainability of uncontrolled housing for low-income households.
The third conclusion is that income moderates the effect of rent control on
demand for controlled social housing. Rent-seeking behavior is increasing in
income (for all ten income deciles). Thus high-income households are more
likely to move house within the rent-controlled housing sector if rent-control
benefits are low compared to low-income households. This empirical finding
is in line with the economic framework as well.
Previous research stressed that the allocation of rent-controlled housing is
often imperfect (Glaeser and Luttmer, 2003), which leads to opportunities for
rent seeking (Bulow and Klemperer, 2012). By specifically considering transi-
tions within the rent-controlled housing sector, it can be investigated empiri-
cally how rent control influences the allocation of controlled social housing.
Future research could work along this path to further investigate the mech-
anisms that lead to rent seeking and rent keeping behavior. A puzzling para-
dox arises. As households do not lose all their rent-control benefits if they
move within the rent-controlled housing sector, one would expect the effect
of rent control on between-transitions to be larger than the effect on within-
transitions. However, as I find no effect of rent control on transitions from the
rent-controlled housing sector, the evidence points in the opposite direction.
4.7. Conclusion 111

The results presented in this chapter have important consequences for the
allocation of rent-controlled housing to new and existing tenants. Given a
fixed supply of rent-controlled housing, new tenants can only enter the rent-
controlled housing sector if existing tenants leave. It is shown that rent control
only reduces the exit rate out of rent-controlled housing for the top income
decile. This would imply that rent-control has a smaller negative effect on the
offer rate of controlled social housing to new entrants than was previously
understood.
112 Chapter 4.

4.A Results conditional fixed effect logit estimator

Table 4.4 shows the parameter estimates if equations (4.3) and (4.4) are esti-
mated using a conditional fixed effect logit estimator. Results should be inter-
preted with care as the marginal effect of that variable and might even be of
opposite sign (Ai and Norton, 2003, p.123). Nonetheless, the pattern that arises
is similar to that of the LPM.
Again, the estimates on transitions within the social housing sector in col-
umn two are similar to the estimates on the overall transition rate in column
one. Thus the conclusion that transitions within the social housing sector
dominate the effect of rent-control benefits on household mobility is not spe-
cific to the LPM estimation method as it is found as well using a conditional
fixed effects logit model.
Also, transitions from the social housing sector (column three) seem to be
less responsive to rent-control benefits than transitions within the social hous-
ing sector (column two). Only transitions from the social housing sector by
high-income households seem to be affected by rent-control benefits. In con-
trast, transition within the social housing sector are affected by rent-control
benefits by low-income, middle and high-income workers (column two). A
similar pattern has been presented in section 7.5 using the LPM model.
4.A. Results conditional fixed effect logit estimator 113

Table 4.4: Fixed effect logit results

Dependent variable: Transition

Overal Within Between
Interaction terms rent control stringency r and income group dummies a
r ∗ Y117 0.0024*** 0.0020*** 0.0115*
(0.0007) (0.0008) (0.0066)
r ∗ Y117−129 0.0018** 0.0017** -0.0012
(0.0008) (0.0008) (0.0076)
r ∗ Y129−140 0.0010 0.0011 0.0031
(0.0008) (0.0008) (0.0073)
r ∗ Y140−150 0.0009 0.0009 0.0003
(0.0008) (0.0008) (0.0082)
r ∗ Y150−161 0.0017** 0.0013 0.0140**
(0.0008) (0.0008) (0.0070)
r ∗ Y161−175 0.0018** 0.0016** 0.0089
(0.0008) (0.0008) (0.0068)
r ∗ Y175−194 0.0033*** 0.0031*** 0.0089
(0.0008) (0.0009) (0.0061)
r ∗ Y194−222 0.0032*** 0.0036*** 0.0097
(0.0008) (0.0009) (0.0060)
r ∗ Y222−267 0.0044*** 0.0051*** 0.0084
(0.0009) (0.0010) (0.0058)
r ∗ Y267 0.0044*** 0.0047*** 0.0102*
(0.0010) (0.0012) (0.0062)
Income dummies. Base is household with annual income below 11,700 euro
Y117−129 -0.0151 0.0647 -0.3720
(0.0724) (0.0754) (0.5430)
Y129−140 -0.1260 0.0235 -0.8160
(0.0775) (0.0821) (0.5110)
Y140−150 -0.0641 0.1370 -0.7580
(0.0835) (0.0893) (0.5310)
Y150−161 -0.1790** 0.0517 -0.8590*
(0.0853) (0.0924) (0.5020)
Y151−175 -0.0940 0.1340 -0.5410
(0.0861) (0.0951) (0.4760)
Y175−194 0.0622 0.2540** -0.2630
(0.0875) (0.0996) (0.4430)
Y194−222 0.3330*** 0.5040*** -0.1060
(0.0890) (0.1040) (0.4290)
Y222−267 0.8500*** 0.9930*** 0.6530
(0.0931) (0.1100) (0.4350)
Y267 1.4760*** 1.4230*** 1.4270***
(0.1030) (0.1280) (0.4540)
Household, house and local housing market characteristicsb
House value (in 100,000 euro)1 -0.6190*** -0.4890*** 4.7390***
(0.1330) (0.1340) (1.4050)
Median income in district (in 1,000 euro) -0.0488 -0.0434 -0.3870
(0.0333) (0.0334) (0.3810)
Median income in neighborhood (in 1,000 euro) -0.1460*** -0.1230*** 0.0244
(0.0239) (0.0243) (0.2430)
Observations 24,069 19,038 4,947
Number of households 8,965 6,842 2,116

a
Interaction term where r is interacted with an income dummy YA− B . r measures rent control stringency and is defined
monthly controlled rent
as House value/100,000 . The mean of r equals 270 with standard deviation 84. Lower values of r imply larger rent-control
benefits. YA− B is a dummy equal to one if income is greater than or equal to A and smaller than B, and is zero otherwise.
b
Explanatory variables included but not shown: dummies for being on rent support, having multiple adults in the
household, being married, having children, age group and indicating contract type (fixed, fulltime, fulltime and fixed or
having no job), being self-employed and two year dummies. Median value of owner-occupied housing in the municipality,
median house value and median house value of rent-controlled housing in the district and neighborhood. Percentage of
rent-controlled housing in the district and neighborhood. Housing stock, rent-controlled housing stock and percentage of
rent-controlled housing in the municipality.
Chapter 5

Type of housing tenure and

wages: Sorting or incentives?

5.1 Introduction
In the Netherlands workers living in social housing earn on average 25 per-
cent lower wages compared to homeowners.1 This wage differential might be
the result of various sorting mechanisms as low-income households are more
likely to rent social housing. Alternatively, this wage differential might be
explained by a causal effect of housing tenure on wages, that occurs if hous-
ing tenure affects search behavior on the labor market. In this chapter it is
studied whether housing market tenure influences wages, or whether the ob-
served wage differentials between homeowners and tenants are the results of
sorting.
The wage differential between homeowners and workers in social hous-
ing might be the result of various sorting mechanisms. Low-income workers
are more likely to live in rent-controlled housing as rent-controlled housing is
mostly distributed towards them. Also, workers with low wages do not want
to consume better quality housing at the market rent (see chapter 2). In ad-
dition, if housing of a certain tenure is spatially concentrated, its inhabitants
1 In contrast, tenants of private rental housing on average earn 2.5 percent lower wages.
116 Chapter 5.

might predominantly be working in nearby regions, and the industries, firms

and jobs located therein. This might influence wages.
However, wage differentials might also be influenced by the type of hous-
ing tenure as both homeownership and rent control reduce the geographical
mobility of workers (De Graaf et al., 2009).2 On the one hand, a lower geo-
graphical mobility will weaken the labor market position, as the area in which
unemployed workers look for jobs becomes smaller and the number of job
offers decreases. This will negatively affect wages if workers respond by low-
ering their reservation wage. On the other hand, the lower geographical mo-
bility might have a positive effect on wages: As workers are more likely to
stay with the firm, they might receive more training that increases their hu-
man capital. Also, workers with lower geographical mobility might provide
more effort to prevent becoming unemployed.
The literature has shown that rent-controlled households are more likely to
become unemployed and to work within their local labor market.3 Homeown-
ers are found to have a lower probability to become unemployed, but they ex-
hibit longer unemployment duration conditional on becoming unemployed.4
Two studies have investigated the effect of homeownership on wages. Munch
et al. (2008) use a mixed proportional hazard model to estimate the effect of
homeownership on wages in Denmark. They find that that homeowners earn
about 5 percent higher wages than tenants. In contrast, Coulson and Fisher
(2009) use an instrumental variables on cross-sectional data for the United
States. They find that homeowners in the United States earn on average 30
percent lower wages compared to tenants.
In this chapter, I consider the effect of homeownership and living in rent-
controlled housing on wages.5 Detailed administrative data for over one mil-
2 Homeownership does so by increasing the transaction costs of moving. Living in rent-

controlled housing does so as moving to rent-controlled housing generates high (often non-
pecuniary) transaction costs such as queing. In addition, rent control creates transaction costs
in the form of in kind benefits that are lost if households move to uncontrolled housing.
3 See Hughes and McCormick (1987); Minford et al. (1988); McCormick (1997) for the UK,

Flatau et al. (2003) for Australia, Munch and Svarer (2002); Svarer et al. (2005) for Denmark, and
De Graaf et al. (2009) for several European countries including the Netherlands.
4 The so-called Oswald-hypothesis, see Oswald (2009). Overviews on the relation between

homeownership and unemployment are Van Ewijk and Van Leuvensteijn (2009); Havet and
Penot (2010).
5 Thus, the baseline type of housing market tenure is living in private market rental housing

instead of living in (private or social) rental housing as in Munch et al. (2008) and Coulson and
Fisher (2009).
5.2. Literature 117

lion workers is used to distinguish the sorting effect from the incentive effect
of housing tenure. The panel nature of the data allows to control for the sort-
ing effect of low-skilled workers into rent-controlled housing. Several controls
for the region, industry, firm and job in which the worker works are provided,
and type of housing tenure is instrumented to infer the causal effect of housing
tenure on wages.
It is concluded that conditioning on worker skill is very important if one is
to study the effect of housing tenure on local wage rate. In the end, no signif-
icant effect of living in rent-controlled housing (compared to being a private
tenant) is found. Without instrumentation, a small statistical significant pos-
itive effect for being a homeowner is found: homeowners have a 0.3 percent
higher hourly wage. Thus the coefficient is not very significant economically.
After instrumenting homeownership and living in social housing there is no
longer evidence that the type of housing tenure influences the wage rate.
This paper continuous as follows: In section 5.2 the literature on housing
tenure and the sorting and incentives effects is discussed. Section 5.3 explains
the data that are used. The empirical strategy is explained in section 5.4. Sec-
tion 5.5 lists results on the effect of housing market tenure on wages. In section
5.6 evidence on the relationship between worker skill and sorting into social
housing is presented. Conclusions are presented in section 5.7.

5.2 Literature

5.2.1 Framework

The sorting effect of housing tenure type arises as workers with high income
are more likely to buy owner-occupied housing and workers with low-income
are more likely to rent social housing. However, the incentive effect of housing
tenure type entails that housing tenure reduces worker mobility, which the
worker anticipates by accepting jobs that pay less or by providing more effort.
Panel data is used to distinguish the sorting effect from the incentive effect
of housing tenure. Denote the log wage of worker i at time t as wit , which
is a function of homeownership hit or living in rent-controlled social housing
cit . Furthermore, consider the wage rate to be influence by time-invariant (un-
118 Chapter 5.

observed) ability of the worker αi and a matrix of observable worker and job
characteristics Xit . In this specification, the housing market tenure types cap-
ture how housing market tenure type influences the wage rate (by changing
incentives), whereas observed and unobserved worker characteristics take up
the effect of sorting. See equation (5.1), where ε it is an i.i.d. error term with
zero expectation and constant variance.

wit = β 1 hit + β 2 cit + αi + β 3 Xit +ε it (5.1)

| {z } | {z }
Incentives Sorting

The initial specification equals a multivariate regression of wages on home-

ownership dummy h and a dummy for living in rent-controlled housing c
only. This specification is biased if ability or variables in Xit are correlated with
homeownership or living in social housing.6 Then, controls are provided for
sorting by considering the coefficients on homeownership and living in rent-
controlled housing if several indicators for the region, the industry, the firm
and the job in which the worker works are included in Xit . Finally, αi is es-
timated to control for unobserved ability of the worker. The sorting effect of
housing tenure type is measured by the effect of homeownership and the ef-
fect of living in rent-controlled housing. The incentive effect of housing tenure
type is given by E[hit |αi , Xit ] for homeownership and E[cit |αi , Xit for living in
rent-controlled housing.

5.2.2 Sorting of workers

I control for various sorting mechanisms by estimating unobserved worker

characteristics that do not vary over time and by controlling for region, indus-
try, firm and the job in which the worker is employed. The literature on ag-
glomeration economies has shown that heterogeneous workers sort into het-
erogeneous places. High-skilled workers have been sorting into cities that
provide more amenities (Glaeser et al., 2001), that provide higher returns to
6 Workers with a high ability earn higher wages on average. If the correlation between home-

ownership and ability or variables in Xit is positive (as hypothesized), omitting them from equa-
tion (5.1) would bias the estimated coefficient of β 1 upwards. Similarly, if the correlation between
living in rent-controlled housing and ability or variables in Xit is negative, omitting them would
bias the coefficient of β 2 downwards.
5.2. Literature 119

education (Moretti, 2004) and that increase productivity from more effective
face-to-face exchange of information (Glaeser and Mare, 2001). As workers
sort across skill and types of housing tenures are not evenly distributed, this
might create a relationship between wages and type of housing tenure that is
due to sorting. This implies that a specification of the effect of housing tenure
on wages should control for the worker’s skill, location and characteristics of
industry, firm and job in which the worker works.7

5.2.3 Type of housing market tenure and wages

Living in rent-controlled or owner-occupied housing increases the transaction

costs of moving house. Rent control provides tenants with in kind benefits
that are lost if the household moves to uncontrolled housing. Also, controlled
housing is often allocated based on non-pecuniary costs such as waiting time.
Homeowners face similar non-pecuniary transaction costs as they have to wait
to sell their house. In addition, they often face substantial pecuniary moving
costs in the form of a tax on transferring residence (as shown by Belot and
Ederveen (2005) for 15 European countries).
The increased transaction cost of moving house reduce the geographical
mobility of both rent-controlled and owner-occupied workers. Therefore they
are less likely to accept jobs outside the local labor market. If immobile work-
7 A large body of literature points at including these control variables. Already in the nine-

teenth century, the idea was put forward that asset sharing among firms could increase local pro-
ductivity. Marshall (1890) argued that firms in dense areas could benefit from lower transportation
costs, industry specific inputs, or benefit from local labor pools suitable to their industry. How-
ever, even in todays world in which falling transportation and communication costs have evoked
the ‘death of distance’ (Cairncross, 2001; Friedman, 2006), regional wage differentials have been
rising and spatial sorting of households has increased (Moretti, 2012). Todays explanations for
agglomeration economies are therefore largely based on the way cities enable face-to-face contact
to exchange ideas (Glaeser, 2011).
In the view of Marshall (1890), Arrow (1962) and Romer (1986) clustering allows firms to benefit
from inter-knowledge spillovers. Local specialisation in an industry enhances the flow of ideas
within industries. Following Glaeser et al. (1992), these externalities due to specialization are also
referred to as MAR-externalities. On the other hand, in the view of Jacobs (1969), knowledge spills
over between complementary industries as practice in the one industry lead to innovations in the
other. The interaction of ideas across different industries stirs innovation and productivity. These
externalities are referred to as Jacobian-externalities, and are the result of diversification. Finally,
Porter (1998) argues that clustering of firms in a location increases information spillover, but also
increases competition. This spurs the adaptation of new ideas as all firms have to perform better
to beat their competitors.
Besides spatial heterogeneity in the productivity level of places, the local wage rate might be
influenced by the spatial sorting of industries, firms and the jobs they provide. Firms in a partic-
ular industry sort into cities as dense labor markets might improve the match between job tasks
and workers skills. Also, in line of the models by Melitz (2003), increased competition in dense
areas might lead to the survival of high-productivity firms in cities.
120 Chapter 5.

ers become unemployed, their hesitation to move house to accept a job might
negatively affect unemployment duration. Various studies found that immo-
bile workers, such as homeowners and tenants of rent-controlled housing, are
more likely to work within their local labor market and have increased unem-
ployment spells.8
However, workers with a prolonged expected unemployment duration
(conditional on becoming unemployed) might lower their reservation wage
in response. Thus housing market transaction costs not only influence unem-
ployment and job mobility, but also whether a wage offer is accepted. Munch
et al. (2006) develop a theoretical search model in which unemployed work-
ers distinct between jobs within and outside their local labor market. They
show that workers with higher transaction costs of moving (homeowners)
lower their reservation wage for jobs within their local labor market com-
pared to workers with lower transaction costs of housing (tenants of private
housing). Similarly, the reservation wage for jobs outside their local labor
market is higher. As workers with increased transaction costs are overrep-
resented in jobs within their local labor market, homeowners and tenants of
rent-controlled housing might earn lower wages compared to tenants of pri-
vate housing, ceteris paribus.
Increased transaction costs of moving house reduces mobility and increases
the expected duration of unemployment. As a result, these workers might
provide more effort to prevent becoming unemployed. Individual homeown-
ers have been found to become unemployed less often.9 Also, their lower
geographical mobility makes it more likely that homeowners stay with the
firm for a long time period. This might be an incentive to both the employer
and employee to increase human capital of the worker. Thus workers who are
8 Hughes and McCormick (1987); Minford et al. (1988); Flatau et al. (2003); Munch and Svarer

(2002); Svarer et al. (2005) find that rent-controlled workers are more likely to work local jobs and
to be unemployed. De Graaf et al. (2009) study both homeownership and rent controlled tenants
and their risk to become unemployed for fourteen European countries. On average, job-to-job
mobility of rent-controlled tenants is three percentage points higher than that of private tenants,
and their entry rate into unemployment is eleven percentage point higher. However, there is large
varation across the sampled countries in sign of the effects.
9 However, they have also been found to move out of unemployment faster. See among oth-

ers Oswald (2009); Partridge and Rickman (1997); Green and Hendershott (2001); Flatau et al.
(2003); Munch et al. (2006); Van Vuuren (2007); Battu and Phimister (2008); Munch et al. (2008).
Overviews are given by Van Ewijk and Van Leuvensteijn (2009); Havet and Penot (2010).
5.2. Literature 121

less likely to move house, have an incentive to provide more effort and are
more likely to receive training. This might increase wages.10
The empirical literature on the effect of housing market transaction costs
on wages is small and ambiguous. So far, two papers have investigated how
wages are affected by increased housing transaction costs in the form of home-
ownership. I did not find empirical research on the effect of rent control on
wages. Munch et al. (2008) model job duration, the wage rate and the selec-
tion into homeownership simultaneously using a mixed proportional hazard
model. They estimate the effect of homeownership on wages while control-
ling for unobserved random worker characteristics. They identify the effect
of homeownership on wages using the panel dimension of their data. Also,
they instrument homeownership using the local homeownership rate in the
local labor market. Their results indicate that Danish homeowners receive a
positive wage premium of about five percent compared to tenants of private
housing.
Coulson and Fisher (2009) use cross-sectional data from the US census to
test the effect of homeownership on income (not wages) using various OLS
and IV specifications. They develop a theoretical search model in which the
wage firms offer depends on the local homeownership rate in the local labor
market. From this they conclude that the local homeownership rate, a com-
monly used instrument of individual homeownership, is not valid. Therefore
they instrument homeownership with a dummy variable that equals one if
the first two children in the households are of equal sex. Given preference of
households for a boy and a girl, these households are more likely to have a
third child. Consequently, these households have larger families and live in
owner-occupied housing more often.11 They find that that homeowners in the
United States earn 42 percent lower wages compared to renters.
10 Also, if unemployment benefits are means tested, the cost of unemployment might be larger

for homeowners (Van Vuuren, 2007). If homeowners receive unemployment benefits for a smaller
time period, whereas tenants receive unemployment benefits as long as they are unemployed, the
reservation wage for homeowners might be lower compared to that of private tenants for jobs
within and outside the local labor market.
11 Also, they suggest to follow DiPasquale and Glaeser (1999) and use the local homeownership

rate stratified to race or income. However, this variable is higly collinear to their homeownership
variable.
122 Chapter 5.

5.3 Data
Several administrative data sets provided by Statistics Netherlands are com-
bined to test whether the wage differentials between homeowners and tenants
in the Netherlands is due to sorting or is the result of housing market tenure.
For each worker the highest paying job that lasted for more than 28 days is
selected (obtained from SSBBanen). The dataset GEMSTPLBUS contains for
each worker the municipality in which his job is located. The highest paying
job of each worker is matched to the municipality in which it is located. This is
done for the population of workers with one job or with multiple jobs within
the same municipality.
Industry, firm and job characteristics are based on SSBBanen as well. It
is used to construct twelve sectoral indicators that correspond to groups of
ISIC indicators (see section 5.B). Additionally, ten dummies for firm size are
computed based on the average number of employees employed by the firm
in each year. Finally, job characteristics like tenure, the type of contract, and
hours worked are provided as well.
I proxy observed worker skills and effort using age-dummies, the presence
of children in the household and indicators for marital status that are mea-
sured in the municipal administration GBA. Time-invariant indicators such as
gender or belonging to a minority group cancel out in the within transforma-
tion. Also, the GBA data set provides a house indicator, which is merged with
the file WRG verrijkt. This allows to distinguish homeowners from tenants
living in private or rent-controlled housing.
In total there are 2,884,112 worker-year observations of which 73.3 percent
is on homeowners, 22.7 percent is on rent-controlled tenants and 3.9 percent
is on tenants of private housing.
Figure 5.1 shows the density of the municipal wage rate for home-owners,
tenants of private rental housing and tenants of social housing. The figures in-
dicate that the municipal wage rate of tenants of social housing is below that of
tenants of private housings, which in turn is (somewhat) below that of home-
owners. Figure 5.2 shows the relative municipal wage for homeowners (com-
pared to that of tenants of private rental housing) against the relative wage
of tenants of social housing. Circle size reflects municipality size. The major-
5.3. Data 123

ity of observations is in the upper left quadrant, which indicates that in most
municipalities homeowners earn more and tenants of rent-controlled housing
earn less than tenants of private rental housing. The upper right quadrant
indicates there are some (small) municipalities where both tenants of social
housing and homeowners earn more than tenants of social housing. Finally,
there are quite some municipalities where both homeowners and tenants of
social housing earn less than tenants of private housing. These municipalities
are in the bottom left quadrant.

Figure 5.1: Density of municipal wages by housing tenure

Density plot of the average municipal wages earned between 2006 and 2008 for
workers living in social housing, workers living in private rental housing and
workers who live in owner-occupied housing.

Both tenants in rent-controlled housing and homeowners are expected to

have local jobs more often. Table 5.1 shows the mean distance, measured as
the crow flies, between the (centroid of) the municipality where the worker
lives and the (centroid of) the municipality in which he works. Although this
is a rough approximation for commuting time which defines the boundary
of local labor markets, it indicates that on average tenants of private housing
have longest commutes, followed by homeowners. Tenants of social housing
124 Chapter 5.

Figure 5.2: Relative municipal wages of homeowners and tenants of social

housing

Scatter plot of the relative municipal wages earned between 2006 and 2008
by workers living in owner-occupied housing against relative municipal wages
earned by workers living in social housing. Both measures are relative to the
municipal wage earned by workers living in private rental housing.
5.3. Data 125

Table 5.1: Average commuting distance by housing tenure

Commuting distancea
Homeowners 15.91
Tenants of social housing 12.64
Tenants of private housing 17.02
a in kilometres as the crow flies

have the lowest commuting distance. This result is striking as it is often as-
sumed that transaction costs are lowest for tenants of private rental housing
and therefore they can lower commuting time by moving.
Figure 5.3 shows the percentage of workers within a commuting distance
interval according to type of housing tenure. Thus 50 percent of all workers
living in rent-controlled housing have a commuting distance that is smaller
than 5 kilometer. Homeowners commute more, as about 40 percent of home-
owners have a commuting distance smaller than five kilometer. Summary
statistics for the variables used in the analysis are shown in Table 5.2.

Figure 5.3: Commuting distance (in km) by type of housing tenure

Commuting distance (in kilometres) for workers living in owner-occupied hous-

ing (homeowners), workers living in private rental housing and workers living
in social housing. The numbers above bars indicate cumulative share of the work-
force
126 Chapter 5.

Table 5.2: Summary statistics

Variablesa mean st.dev. av(p5)b av(p95)c

Log daily wage 4.5655 0.4576 3.8321 5.7399
Age 42.3632 10.2750 24.6784 60.3727

Indicators housing tenure: one if. . .

Tenant of social housing 0.2273 0.4191
Homeowner 0.7332 0.4423
Tenant of private housing 0.0394 0.1946

Indicators worker characteristics: one if. . .

Not married 0.3345 0.4718
Widowed 0.0075 0.0861
Divorced 0.0792 0.2700
Female 0.3728 0.4836

Indicators for job characteristics: one if. . .

Has temporary contract 0.1113 0.3145
Works [2-4) hours 0.0111 0.1050
Works [4-6) hours 0.0724 0.2591
Works [6-8) hours 0.1391 0.3461
Works [8-10) hours 0.7760 0.4169
Tenure equals [4-39) weeks 0.0154 0.1232
Tenure equals [39-74) weeks 0.0748 0.2630
Tenure equals [74-122) weeks 0.0907 0.2871
Tenure equals [122-198) weeks 0.0952 0.2934
Tenure equals [198-287) weeks 0.1040 0.3053
Tenure equals [287-379) weeks 0.1113 0.3145
Tenure equals [379-491) weeks 0.1195 0.3243
Tenure equals [491-689) weeks 0.1227 0.3281
Tenure equals [689-1002) weeks 0.1283 0.3344

Indicators for firm characteristics: one if. . .

Firm size equals [9-22) workers 0.0910 0.2876
Firm size equals [23-53) workers 0.0968 0.2957
Firm size equals [54-133) workers 0.1057 0.3074
Firm size equals [134-311) workers 0.1097 0.3125
Firm size equals [312-725) workers 0.1101 0.3129
Firm size equals [726-1564) workers 0.1075 0.3098
Firm size equals [1567-3238) workers 0.1056 0.3073
Firm size equals [3248-10664) workers 0.1024 0.3032
Firm size equals [10679-86840) workers 0.0913 0.2880
a
Based on 2,888,441 observations
b
Average value for the 5th percentile and lower
b
Average value for the 95th percentile and lower
5.4. Empirical strategy 127

5.4 Empirical strategy

The log of the hourly wage (wit ) is a linear function of worker skills (X, αi ), the
region where the worker works (R), the type of industry the worker works
in (S), firm productivity (F) and job characteristics (J). Dummy variables for
type of tenure are included as c is equal to one if the worker is living in social
housing and the dummy h is one if he is a homeowner. The default type of
tenure is therefore being a tenant of uncontrolled housing. Additionally, fixed
year effects δ are included. Thus equation (5.2) is estimated, where ε is an i.i.d.
error term and the betas are vectors of parameters that have to be estimated.

wit =αi + cit + hit + Xit β X + Rrt β R + Sst β S + Ff t β F

+ Jit β J + δt + ε it (5.2)

i =1, 2, . . . n; t = 2006, 2007, 2008; r = 1, 2, . . . 443; s = 1, 2, . . . 7

Our measure of worker skills includes unobserved worker fixed effects such
as gender and ability. It also includes observable worker characteristics for
worker age, tenure at the firm and marital status.12 I control for the munici-
pality R in which the worker’s job is located by including 430 fixed effect at
the municipal level. In addition, eleven industry fixed effects S are included,
together with nine dummies for firm size in F to proxy for firm productivity.
Equation (5.2) is estimated using OLS after the equation is ‘within trans-
formed’ (see section 5.A). To identify the fixed industry effect Ss and the fixed
part of the region effect µc it is required that there is enough worker mobility
across sectors and municipalities such that all industries and municipalities
are ‘connected’ by worker geographical and labor market mobility. Because
of the large sample size this condition is met. The coefficient for the industry
business activities and the coefficient for the municipality Amsterdam are set
to zero.
12 Age is decomposed into groups of five years (baseline: workers aged at least 35 and younger

than 40). Tenure at the firm is measured in deciles (baseline being the fifth decile). Marital status
is indicated using dummy variables for ‘being divorced’, ‘being widowed’ and ‘never has been
married’ (baseline: ‘being married’).
128 Chapter 5.

Models along the lines of Roback (1982) and Moretti (2011) illustrate how
wages might be set with respect to unobserved amenities. This might bias
results if unobserved amenities influence the sorting of workers into regions
and influences the wage rate. However, if the unobserved local amenities
are invariant over the sample period, the fixed part of the regional effects R
will capture both variation in exogenous productivity (say: due to location)
and variation in the wage bargaining process due to local amenities. As the
panel consists of three consecutive years, the assumption that local consumer
amenities are fixed seems reasonable.13
The estimates for αi , cit , hit , δt and the β-parameters are used to decompose
predicted wages into contributions of time, skills, regions, industry, firms, jobs
and housing tenure to predicted wages. I follow Combes et al. (2008) who sug-
gest to measure the conditional variance and covariance terms in one measure
called “the effect”. The effect of a variable equals its sample correlation coeffi-
cient with the dependent variable, multiplied by its own standard deviation.
This measure can be shown to be equal to the sum of the variance of the vari-
able divided by the variance of the dependent variable and all of the variable’s
covariance terms (Gibbons et al., 2013).

Identification

Equation (5.2) contains dummy variables for homeownership and renting con-
trolled housing. They equal the average wage difference between homeown-
ers (or tenants living in social housing) compared to tenants of private rental
housing conditional on other variables included. The desired specification
in equation (5.2) controls for unobserved ability of the worker using worker
specific fixed effects αi . This control variable is important as it is used to distin-
guish sorting in housing tenure type from the incentives these housing types
provide. If αi is excluded from the specification in equation (5.2), the coeffi-
cients on homeownership measures both the (average) skill of homeowners
and the causal effect of homeownership on wages. Thus to measure the effect
of skill based sorting into housing market tenure, estimates of homeowner-
13 Combes et al. (2008) mention another endogeneity bias may occur if regions specialize into

sectors because they pay high wages. However, this seems not to be problematic if the covered
time period is small and changes in specialization rates are limited.
5.4. Empirical strategy 129

ship in equation (5.2) with controls for worker fixed effects are compared to
those without. The same line of reasoning holds for the dummy for living in
rent-controlled housing.
One might be concerned that the effect of housing tenure on the wage rate
is driven by reversed causality. This would occur if shocks to the worker’s
wage induce him to change housing tenure. The setup of the data implies that
a change in housing market tenure is measured before wages are measured.
This timing of events prevents some protection against reversed causality.14
Furthermore, the rich nature of the data set allows to include controls for
changes in earning potential that are determined by (changes in) age, mar-
ital status and the presence of children, together with indicators for region,
industry, firm productivity, and -potentially very important- job contract char-
acteristics and hours worked.
As a robustness check the variables that measure housing market tenure
type are instrumented. Coulson and Fisher (2009) conclude that the local
homeownership rate, a commonly used instrument of individual homeown-
ership, is not valid under wage bargaining if firms take the homeownership
rate in the municipality into account in determining their wage offer. Yet, their
suggested instrument would lead to a considerable reduction in sample size
(to about 186 thousand observations).
Therefore, the approach suggested by DiPasquale and Glaeser (1999) is
followed and the local homeownership rate stratified to income is used as an
instrumental variable. To be precise, being a homeowner and living in so-
cial housing are instrumented with the homeownership rate for the first in-
come decile in the neighborhood, and the homeownership rate among work-
ers within the 10th and 25th percentile in the neighborhood.
Consider homeownership. The homeownership rate among these income
groups at the neighborhood level is positively correlated with the probability
to be a homeowner in the neighborhood because of a supply effect. Workers
who move to neighborhoods with a large stock of owner-occupied housing are
more likely to become homeowners. Similarly, they are less likely to rent social
14 Housing market tenure is registered at the first of January in each year. The daily wage in

each year is based on labor income over the course of the year. Thus if a worker gets a raise in year
t and subsequently decides to move within the same year, the raise will be incorporated partly in
the wage rate for year t.
130 Chapter 5.

housing. Thus the instrument is supposed to be most relevant in explaining

housing tenure for workers who change tenure type. This is not problematic
as the housing tenure dummies are only identified for those workers who
switch tenure type.
The suggested instruments are exogenous if employers determine their
wage offer based on the homeownership rate as a whole within local labor
markets and do not consider homeownership among low-income households.
Also, (changes in) the homeownership rate among low-income households at
the neighborhood level should be ‘different enough’ from (changes in) the
homeownership rate within the local labor market. I find some evidence sug-
gesting that individual wage outcomes do not depend on the homeownership
rate among low-income households.

5.5 Estimation results

Table 5.3 presents the main estimation results for eight nested specifications.15
The smallest specification in column one regresses the log hourly wage on a
dummy for being a homeowner and a dummy for living in social housing.
One by one, control variables are included. The final specification in column
eight includes the full specification as in equation (5.2).
Without additional controls, evidence is found that wages vary according
to housing tenure type. The hourly wage of tenants of social housing is about
18.2 percent below that of tenants in private rental housing. Homeowners earn
7.5 percent higher wages compared to workers in private rental housing. The
combined effect indicates that homeowners earn wages that are 25.7 percent
higher than tenants of social housing.
In columns two to seven controls are added for the year of observation,
for observed worker skills, for the location of the job, for the industry, and for
firm and job characteristics. This changes parameter estimates for homeown-
ership slightly. In general, the estimated coefficient of living in social housing
remains negative around a value of minus 19 percent. The coefficient of home-
15 Table 5.8 shows results for the same specification, including the estimates on control vari-

ables. Results are similar to those where standard errors have been clustered on the neighbor-
hood level. Especially the conclusions with respect to the housing market tenure variables do not
change.
5.5. Estimation results 131

ownership is positive at a value of about two percent. Together, the wage

differential between homeowners and tenants of social housing is about 21
percent.
In the final column unobserved worker skills are estimated to control for
the self-selection into housing tenure based on worker skill. The absolute
value of the estimated coefficient on living in social housing drops consid-
erably and becomes insignificant at the five percent level. From this it is con-
cluded that the negative coefficient on the dummy for living in social housing
in specifications one to seven is biased by unobserved worker skills and that
low-skilled workers have been sorting into social housing. The value of the
coefficient on homeownership drops as well, but it remains significant. Its
value suggests that homeowners earn wages that are 0.5 percent higher than
wages of tenants in private housing. This is evidence that skilled workers sort
into owner-occupied housing.
The estimates in column eight suggest that the positive effects of home-
ownership slightly dominate its negative effects. For tenants of social hous-
ing, the two effects cancel out (or are both insignificant) and no evidence in
favour of an effect of living in social housing on the wage rate is found. It
is also concluded that homeowners earn wages that are 0.8 percent higher
than tenants living in social housing, conditional on worker unobserved char-
acteristics. Thus homeowners earn wages that are 0.5-0.8 percent higher than
tenants, depending on type of housing the worker rents. These estimates have
the same sign as that by Munch et al. (2008), however, they are considerably
smaller.16
Table 5.4 shows the contribution of worker observed skills, time and lo-
cation, industry firm and job characteristics to the explained variation in the
log of hourly wages. Specification seven indicates that, without conditioning
on unobserved worker skills, the largest contribution comes from “observed
worker skills” (0.08), “Type of industry” (0.05) and “Living in social hous-
ing” (0.04). They have the largest effect as they correlate strongly with the
log hourly wage and they contain considerable variation. In contrast, being a
16 The baseline group of Munch et al. (2008) equals tenants. They do not distinguish between

tenants of private sector housing or social housing.

132 Chapter 5.

homeowner is also correlated with log hourly wages, but its variation is very
small.
However, column eight shows that by far the largest contribution towards
the log hourly wage comes from “unobserved worker skills” (0.85), followed
by job characteristics (0.01) and observed worker skills (0.01). Note that the
effects of type of industry and living in social housing drop, which suggests
that in estimation seven the effect of unobserved worker skills was captured
by the indicators for living in social housing and the industry dummies.
 5.5. Estimation results
Table 5.3: Estimation results

Dependent variable: log hourly wage

1 2 3 4 5 6 7 8
coef s.e. coef s.e. coef s.e. coef s.e. coef s.e. coef s.e. coef s.e. coef s.e.
Living in social housing -0.1824 0.0014 -0.1815 0.0014 -0.2053 0.0014 -0.2052 0.0013 -0.1894 0.0013 -0.1897 0.0013 -0.1885 0.0013 -0.0029* 0.0017
Being a homeowner 0.0750 0.0014 0.0756 0.0014 0.0100 0.0013 0.0255 0.0013 0.0223 0.0013 0.0218 0.0013 0.0201 0.0012 0.0050 0.0014
Time controls YES YES YES YES YES YES YES
Worker observed skills YES YES YES YES YES YES
Regional dummies YES YES YES YES YES
Industry dummies YES YES YES YES
Firm characteristics YES YES YES
Job characteristics YES YES
Worker fixed effects YES
Worker groups 1141683
Degrees model 3 5 19 461 472 481 491 1142173
R2 0.0551 0.0564 0.1300 0.1625 0.2028 0.2054 0.2114 0.8848
Combined effect (hit − cit ) 0.2574 0.0006 0.2571 0.0006 0.2153 0.0006 0.2307 0.0006 0.2117 0.0006 0.2115 0.0006 0.2086 0.0006 0.0080 0.0011
All coefficients except * significant at the 0.1 promille significance level. * significant at the 10 percent significance level. regressions based on 2,884,112 observations.

133
 134
Table 5.4: Effect of indicators

1 3 4 5 6 7 8
corr. s.d. effect corr. s.d. effect corr. s.d. effect corr. s.d. effect corr. s.d. effect corr. s.d. effect corr. s.d. effect
Log hourly wage 1 0.4567 1 1 0.4567 1 1 0.4567 1 1 0.4567 1 1 0.4567 1 1 0.4567 1 1 0.4567 1
Being a homeowner 0.2231 0.0332 0.0162 0.2231 0.0044 0.0022 0.2231 0.0113 0.0055 0.2231 0.0098 0.0048 0.2231 0.0096 0.0047 0.2231 0.0089 0.0044 0.2231 0.0022 0.0011
Living in social housing 0.2327 0.0764 0.0389 0.2327 0.0860 0.0438 0.2327 0.0860 0.0438 0.2327 0.0794 0.0404 0.2327 0.0795 0.0405 0.2327 .0790 0.0402 0.2327 0.0012 0.0006
Time 0.0376 0.0153 0.0013 0.0376 0.0154 0.0013 0.0376 0.0152 0.0012 0.0375 0.0152 0.0012 0.0376 0.0157 0.0013 0.0375 0.0270 0.0022
Observed worker skills 0.3010 0.1255 0.0827 0.3019 0.1251 0.0827 0.3020 0.1205 0.0797 0.3020 0.1197 0.0792 0.3022 0.1164 0.0770 0.1725 0.0286 0.0108
Location workplace 0.1610 0.0829 0.0292 0.1581 0.0716 0.0248 0.1578 0.0710 0.0245 0.1578 0.0705 0.0244 0.0165 0.0164 0.0006
Type of industry 0.2517 0.0939 0.0518 0.2511 0.0946 0.0520 0.2512 0.0941 0.0517 0.062 0.0262 0.0036
Firm characteristics 0.0597 0.0246 0.0032 0.0623 0.0245 0.0033 0.0821 0.0156 0.0028
Job characteristics 0.1158 0.0357 0.0091 0.1382 0.0394 0.0119
Worker fied effects 0.9285 0.4187 0.8512
Specification numbers correspond to the specifications in Table 5.3. To improve readability results for specification 2 are not shown.
“corr” equals the correlation coefficient with the log of hourly wages. “s.d.” equals the standard deviation.
“effect” equals (corr )(s.d.) divided by the standard deviation of the log of the hourly wage. Hence, the effect in the first row is one by construction

Chapter 5.
5.5. Estimation results 135

Results for various subsamples

Table 5.5 presents the results after re-estimating equation (5.2) for different
subsamples.17 This is done to consider whether control variables in the spec-
ification can correct for subgroups of workers sorting into social or owner-
occupied housing.
In column one the estimation sample contains all workers who are work-
ing 24 hours a week or more, to correct for any bias caused by the self-selection
of workers with parttime jobs into social housing. A very small, statistically
significant effect is found that suggest living in social housing reduces the
wage earned with 0.3 percent. However, as the effect is only significant at
the five percent level and the residual degrees of freedom exceed 1.4 million,
the evidence is considered weak. Homeowners are found to earn 0.3 percent
higher wages compared to tenants of social housing.
Column two limits the estimation sample to male workers (who work 24
hours a week or more). This might be important, as the labor market behavior
of females might depend on that of their partner. However, the estimation
results in this subsample are alike those in the full specification of Table 5.3.
In the third column, equation (5.2) is re-estimated for male workers who
are at least thirty and who work more than 24 hours a week. This to control for
all workers with parttime jobs, labor market behavior of females and the self-
selection of young households into social housing. Results are similar to those
found in column one, only the evidence for both tenants of social housing and
owner-occupied housing is weak.
Finally, it might be that the coefficient on control variables are different
for homeowners and tenants of social housing. To control for any bias that
comes from omitting these covariates, equation (5.2) is estimated separately
on a subsample of tenants in column four and a subsample of homeowners
and tenants of private rental housing in column five. Both specifications yield
parameter estimates on the type of homeownership variable that are equal to
those found in column eight of Table 5.3.
Overall, the results found on the homeownership indicators in column
eight of Table 5.3 are in line with the results found for various subsamples.
17 Table 5.9 presents the results including coefficients and standard errors for control variables.
136 Chapter 5.

This provides evidence that these results are not biased by subgroups of work-
ers sorting into a particular kind of housing.

Results using instrumental variables

In this section instrumental variables are used to control for the potential en-
dogeneity of homeownership or living in social housing. The type of hous-
ing indicators are instrumented using the homeownership rate among the
first decile and among workers between the first decile and the first quar-
tile. Columns one and two of Table 5.6 show the first stage regressions. The
instrumental variables are relevant as they are individually and jointly signif-
icant. The coefficients for the homeownership first stage regression are posi-
tive, whereas those for the regression for living in social housing are negative.
This is conform expectations.
In column three equation (5.2) is estimated for the subsample for which
instruments are available (without instrumenting the type of housing tenure
variables). Results are similar to the previous estimates, with a negative albeit
insignificant coefficient for living in social housing. The coefficient for being a
homeowners is positive and significant with a value of 0.53 percent.
However, the coefficients become insignificant once homeownership and
living in social housing are instrumented. This is in line with previous results
for living in social housing. The insignificant coefficient on homeownership
suggests that not too much weight should be attached to the small, positive
coefficient for homeownership that was found in column eight of Table 5.3.
5.5. Estimation results 137

Table 5.5: Estimation results for various subsamples

Dependent variable: log hourly wage

1 2 3 4 5
Living in social housing -0.0026 ** -0.0021 -0.0030 * -0.0008
(0.0012) (0.0015) (0.0018) (0.0033)
Being a homeowner 0.0033 **** 0.0048 **** 0.0030 ** 0.0043 ***
(0.0010) (0.0013) (0.0015) (0.0016)
Time controls YES YES YES YES YES
Worker observed skills YES YES YES YES YES
Regional dummies YES YES YES YES YES
Industry dummies YES YES YES YES YES
Firm characteristics YES YES YES YES YES
Job characteristics YES YES YES YES YES
Worker fixed effects YES YES YES YES YES
Observations 2,338,343 1,669,985 1,441,815 674,583 2,155,097
Worker groups 921,631 644,847 540,518 288,211 841,248
Degrees model 922,121 645,337 541,006 288,700 841,737
R2 0.9428 0.9413 0.9404 0.8683 0.8829
Combined (hit − cit ) 0.0059 ***** 0.0069 ***** 0.0059 ***** 0.0008 0.0043 ***
(0.0008) (0.0010) (0.0013) (0.0033) (0.0016)
Standard errors within parentheses.
*, **, ***, ****, ***** significant at the 10, 5, 1, 0.1 and 0.001 percent significance level.
Specification 1): All workers working 24 hours a week or more
Specification 2): All male workers working 24 hours a week or more
Specification 3): All male workers, aged 30 or older, working 24 hours a week or more
Specification 4): All tenants of social and private rental housing
Specification 5): All homeowners and tenants of private rental housing

Table 5.6: Estimation results instrumental variables

Dependent variable 1): Being a homeowner

Dependent variable 2): Living in social housing
Dependent variable 3), 4): log hourly wage
1 2 3 4
Instrumental variables: local homeownership rate among:
First decile 0.1322 ***** -0.1192 *****
(0.0010) (0.0008)
th
First decile and 25 percentile 0.2262 ***** -0.1998 *****
(0.0012) (0.0010)
Instrumented variables:
Living in social housing -0.0027 1.0602
(0.0017) (0.7337)
Being a homeowner 0.0053 **** 0.9477
(0.0015) (0.6530)
Control variables:
Time controls YES YES YES YES
Worker observed skills YES YES YES YES
Regional dummies YES YES YES YES
Industry dummies YES YES YES YES
Firm characteristics YES YES YES YES
Job characteristics YES YES YES YES
Worker fixed effects YES YES YES YES
Observations 2,799,366 2,799,366 2,799,366 2,799,366
Worker groups 1,113,369 1,113,369 1,113,369 1,113,369
Degrees model 1,113,860 1,113,860 1,113,860 1,113,860
R2 0.885 0.855
F 33,169.470 36,276.580
Combined 0.1125
(0.0809)
Stage 1 1 2
Standard errors within parentheses.
*, **, ***, ****, ***** significant at the 10, 5, 1, 0.1 and 0.001 percent significance level.
138 Chapter 5.

5.6 Type of housing and worker skills

Based on the estimates in column eight of Table 5.3 the predicted wage of each
worker is divided into a part that depends on worker skills and a part that
depends on the location of the industry and a remainder term. Worker skills
are measured using the predicted value of the coefficients on observed worker
skills and the unobserved worker fixed effects. Location characteristics are
those indicators which vary with location in the short term, such as the fixed
location effect, fixed industry effects and fixed firm effects. Also, differences
in job characteristics Jit0 are considered, which equal the job characteristics in
J minus the indicators on length of tenure at the firm (which is worker-firm
specific).

ŵit = α̂i + Xit β̂ X + Rrt β̂ R + Sst β̂ S + Ff t β̂ F + Jit0 β̂ J +ζ it

| {z } | {z }
skills location

ζ it =ĉit + ĥit + δ̂t + f tit β̂ f t

Next, the municipal average of both predicted worker skills and predicted
location characteristics is computed. Figure 5.4 shows the relationship be-
tween the normalized local wage premium and the local skill level. It shows
no correlation between both variables. This negative relationship between lo-
cal worker skills and local productivity has been found by the literature on
agglomeration economies and can be explained by high productivity regions
being able to set lower wages because they offer lower probabilities to be-
come unemployed.18 In Figure 5.5 the spatial distribution of skills and local
wage premium are plotted. The left panel shows the spatial distribution of
skills, which is indicative of spatial clustering according to skill. Skill levels
are higher in the economic core. The spatial distribution of location premia in
the right panel shows a less clearcut spatial pattern.
Figure 5.6 shows the average skill level of homeowners and tenants of so-
cial housing as a function of the number of workers in a municipality that
18 A similar relationship is found for productivity of firms and worker skills, see Eeckhout and

Kircher (2011).
5.6. Type of housing and worker skills 139

Figure 5.4: Distribution skills and location premium

Relationship between the normalized local wage premium in a municipality and

the normalized average skill level in the municipality. Size of the circle indicates
the relative size of the workforce in the municipality.

live in social housing. First, the average skill level of homeowners exceeds
that of tenants of social housing. Second, it shows a clear positive relationship
between average skill level and the share of the workforce in social housing.
Despite the fact that the average skill of homeowners and tenants of social
housing is increasing in the share of the workforce that rents social housing,
the sorting mechanism of homeowners and tenants seems to be different. Fig-
ure 5.7 shows a positive relationship between the average skills of homeown-
ers and the average skills of homeowners in neighbouring regions. However,
for workers living in social housing, this pattern is less clear-cut and seems to
be reversed: the higher the average skill level of workers in social housing in
a municipality, the lower is the average skill level of workers in social hous-
ing in neighbouring regions. This pattern is confirmed if the average skill of
workers according to housing tenure type is regressed on the local wage pre-
mium, the number of jobs per squared kilometer, the share of the workers in
social housing and a spatially lagged independent variable and error term as
140 Chapter 5.

Figure 5.5: Spatial distribution skills and location premium

5.6. Type of housing and worker skills 141

Figure 5.6: Average skills of homeowners and tenants of social housing

Relationship between the share of the workforce that lives in social housing (at the municipal
level) against the average normalized skill level of homeowners and tenants of social housing
at the municipal level). Size of the indicator reflects the size of the municipal workforce.
142 Chapter 5.

Figure 5.7: Moran’s I scatterplot of municipal skill level

Moran’s I scatterplot of the spatial lag of the normalized average skill in neighbouring re-
gions against the normalized average skill of workers for workers living in owner-occupied
housing (homeowners) and workers living in social housing. The average skill level in neigh-
bouring regions is based on an inverse distance spatial weight matrix with cutoff value at 15
km. Larger cutoff-distances increase the (absolute value of) the slopes. Size of the indicator
reflects the size of the municipal workforce.
5.6. Type of housing and worker skills 143

in Table 5.7. For homeowners and tenants of private social housing a positive
coefficient on the spatial lag of the average skill level is found. The smaller
the cutoff distance in the spatial weight matrix, the stronger the parameter
estimates becomes and the stronger is the sorting mechanism implied. A neg-
ative parameter estimate on the average skill level in neighbouring regions is
found. Again, the absolute value of the coefficient is decreasing in the cutoff
distance of the spatial weight matrix.
Differences in sorting in the social housing sector and other housing sec-
tors are also indicated by the non-spatial coefficients. The null hypothesis that
there is no relationship between the average skill of workers and the share of
the workforce that lives in social housing is not rejected (for homeowners and
tenants). However, the skill level of tenants (in private and social housing)
and of homeowners respond differently to the local wage premium and the
number of jobs in the municipality. This suggest sorting patterns differ be-
tween tenants of social housing, tenants of private housing and homeowners.
144 Chapter 5.

Table 5.7: Spatial sorting according to housing tenure type

Dependent variable: Average skill level of workers

Cutoff distance 15 km
(1A) (2A) (3A)
Homeowners Social housing Private housing
Spatial parameters: spatial lag of
Average skill level 0.6279∗∗∗ -0.4316∗∗∗ 0.4058∗
(0.0706) (0.0402) (0.1671)
Error term 0.1021 0.8185∗∗∗ -0.1057
(0.0743) (0.0699) (0.2151)
Main:
Local wage premium -0.0396∗ -0.0814∗∗∗ -0.0391
(0.0196) (0.0098) (0.0199)
Jobs per km2 0.0321∗∗∗ -0.0036 0.0505∗∗∗
(0.0077) (0.0052) (0.0090)
Share of workers in social housing -0.0602 -0.2999 0.7726∗
(0.1775) (0.1955) (0.3410)
Constant -0.0324 -0.6472∗∗∗ -0.2732∗∗∗
(0.0340) (0.0395) (0.0800)
Observations 443 443 443

Cutoff distance 25 km
(1B) (2B) (3B)
Homeowners Social housing Private housing
Spatial parameters: spatial lag of
Average skill level 0.4632∗∗∗ -0.3536∗∗∗ 0.4262∗∗∗
(0.0481) (0.0301) (0.1002)
Error term 0.1552∗∗ 0.6359∗∗∗ -0.1978
(0.0514) (0.0362) (0.1212)
Main:
Local wage premium -0.0443∗ -0.0859∗∗∗ -0.0360
(0.0201) (0.0107) (0.0203)
Jobs per km2 0.0287∗∗∗ -0.0017 0.0470∗∗∗
(0.0076) (0.0052) (0.0091)
Share of workers in social housing -0.2333 -0.6024∗∗ 0.4461
(0.1656) (0.2145) (0.3350)
Constant -0.0070 -0.6383∗∗∗ -0.1774∗
(0.0324) (0.0433) (0.0807)
Observations 443 443 443

Cutoff distance 35 km
(1C) (2C) (3C)
Homeowners Social housing Private housing
Spatial parameters: spatial lag of
Average skill level 0.3469∗∗∗ -0.2881∗∗∗ 0.3518∗∗∗
(0.0423) (0.0260) (0.0786)
Error term 0.1610∗∗∗ 0.5011∗∗∗ 0.0001
(0.0478) (0.0482) (0.0858)
Main
Local wage premium -0.0476∗ -0.0862∗∗∗ -0.0420∗
(0.0205) (0.0107) (0.0207)
Jobs per km2 0.0300∗∗∗ -0.0017 0.0452∗∗∗
(0.0074) (0.0054) (0.0092)
Share of workers in social housing -0.1905 -0.5605∗∗ 0.3156
(0.1628) (0.2074) (0.3177)
Constant -0.0201 -0.6870∗∗∗ -0.1506∗
(0.0320) (0.0440) (0.0762)
Observations 443 443 443
5.7. Conclusion 145

5.7 Conclusion
In this chapter it is studied whether the observed wage differential between
homeowners and tenants is the result of sorting into housing tenure type or
whether type of housing tenure influences wages. It is concluded that the
entire difference in wages between homeowners and tenants of private rental
housing and between tenants of social and private housing can be explained
by sorting.
Administrative data for the Netherlands is used to construct a panel of
more than one million workers who are followed during the years 2006 to
2008. The dataset allows to regress hourly wages on worker characteristics
and the region, industry, firm and job in which each worker is employed. The
panel nature of the data is used to control for unobserved worker heterogene-
ity. These ‘unobserved worker fixed effects’ might reflect ability or expecta-
tions on worker ability. The results suggest that these unobserved worker
characteristics are very important to control for the sorting of heterogeneous
workers into housing type. Without controls on worker unobserved hetero-
geneity, one would conclude that wages of tenants of social housing are 20
percent lower than those of tenants of private housing. Similarly, one would
conclude that wages of homeowners are two percent higher than those of
tenants of private rental housing. After controlling for unobserved worker
characteristics, wage differentials caused by housing tenure almost entirely
disappear. Tenants of social housing do not earn lower wages than tenants
of private housing, ceteris paribus. Homeowners are found to earn slightly
higher wages compared to tenants of private rental housing, however, the pa-
rameter estimate is very small in economic terms: for various specifications,
homeowners are found to earn 0.3-0.48 percent higher wages. The parame-
ters on housing tenure become insignificant if they are instrumented using
the homeownership rate for low-income workers in the neighborhood.
Thus it is concluded that housing tenure hardly influences wages, if con-
trols for unobserved worker heterogeneity are included in the specification.
Stated otherwise, the observed wage differentials between homeowners and
tenants of social and private rental housing are entirely due to sorting of het-
erogeneous workers into different kinds of housing tenure.
146 Chapter 5.

It is concluded that spatial sorting patterns between homeowners and work-

ers in social housing are structurally different: the average skill level of home-
owners is characterized by positive spatial sorting, whereas the skill level of
tenants of social housing exhibits negative spatial sorting.
5.A. The within estimator with many unobserved individual effects 147

5.A The within estimator with many unobserved

individual effects
Because of the many individual effects αi that have to be estimated, the within
transformation of the dependent and independent variables is computed us-
ing OLS on the within transformed data. The following equation is estimated:

ln˜(wit ) = X̃it β X + R̃ct β R + S̃st β S + F̃f t β F + J̃it β J + rc

˜ it + δ̃t + ε̃ it

Where ˜ indicates the within transformation ln˜(wit ) = ln(wit ) − ∑t ln(wit )/Ti ,

where Ti ≥ 2 indicates the number of times the individual is observed. The
‘within estimates’ of the β-parameters equal the LSDV-parameters.
Next, one can obtain the value of the within transformed error term, which
is consists of the constant, individual worker effect and idosyncratic error term
using

β̂ 0 + α̂i + ε̂ it = ln(wit ) − Xit β̂ X + Rct β̂ R + Sst β̂ S + Ff t β̂ F + Jit β̂ J + rc

ˆ it + δ̂t

β 0 and αi cannot be identified without an identifying restriction on the distri-

bution of αi . Use the identifying restriction that the mean of αi is zero. Use
this and the assumption that E[ε] = 0 to identify β̂ 0 :

N
N −1 ∑ β̂ 0 + α̂i + ε̂ it = β̂ 0

Define the error term minus the constant as v̂it :

N
v̂it = β̂ 0 + α̂i + ε̂ it − N −1 ∑ β̂ 0 + α̂i + ε̂ it = α̂i + ε̂ it

Finally, αi can be obtained as the average within individual of vit and ε̂ it as the
time-demeaned variation in v̂it .

Ti
α̂i = Ti−1 ∑ v̂it
Ti
ε̂ it = v̂it − Ti−1 ∑ v̂it
148 Chapter 5.

As said, the ‘within estimates’ of the β-parameters equal the LSDV-parameters.

However, the standard errors are based on insufficient degrees of freedom and
need to be corrected. The standard errors are corrected for having g fewer
degrees of freedom. Let NT be the total number of observations for N indi-
viduals, and let k be the number of explanatory variables. LSDV computes
standard errors se LSDV based on NT − (k − 1) degrees of freedom. The fixed-
effects or within estimator computes standard errors se FE based on NT − N −
q
−(k−1)
(k − 1) degrees of freedom. Thus se FE = 2 NTNT se
− N −(k−1) LSDV
.

5.B Classification into sectors

All firms are grouped into twelve sectors based on their isic-code.

Industrya Description (CBS) Letter (CBS) ISIC codes

Agriculture, Hunting and Forestry A 01, 02
1
Fishing B 05
2 Manufacturing D 15–37
3 Construction F 45
Wholesale and retail trade;
4 G 50, 51, 52
Repair of motor verhicles, personal and household goods
5 Hotels and restaurants H 55
6 Transport, storage and communication I 60–64
7 Financial intermediation J 65, 66, 67
8 Real estate, renting and business activities K 70–74
Public administration and defence;
9 L 75
Compulsory social security
10 Education M 80
11 Health and social work N 85
Other community, social and personal service activities;
12 O 90–93
Culture, recreation and other service activities
. Mining and Quarrying; C 10, 11, 14
. Electricity, gas and water supply E 40, 41
. Private households with employed persons; P 95
. Extra-territorial organizations and bodies Q 99
a "." not included (mostly because of non-occurence in some of the municipal population).
5.C. Full tables 149

5.C Full tables

Table 5.8: Coefficients and standard errors Table 5.3

Dependent variable: log hourly wage

1 2 3 4 5 6 7 8
coef s.e. coef s.e. coef s.e. coef s.e. coef s.e. coef s.e. coef s.e. coef s.e.

Housing tenure
Living in social housing -0.1824 0.0014 -0.1815 0.0014 -0.2053 0.0014 -0.2052 0.0013 -0.1894 0.0013 -0.1897 0.0013 -0.1885 0.0013 -0.0029* 0.0017
Being a homeowner 0.0750 0.0014 0.0756 0.0014 0.0100 0.0013 0.0255 0.0013 0.0223 0.0013 0.0218 0.0013 0.0201 0.0012 0.0050 0.0014
Fixed year effects
Year 2006 -0.0386 0.0006 -0.0349 0.0006 -0.0354 0.0006 -0.0346 0.0006 -0.0345 0.0006 -0.0359 0.0006 -0.0646 0.0004
Year 2007 -0.0297 0.0006 -0.0288 0.0006 -0.0289 0.0006 -0.0289 0.0006 -0.0291 0.0006 -0.0294 0.0006 -0.0439 0.0003
Observed worker characteristics: one if children aged
6 or less 0.0087 0.0007 0.0062 0.0007 0.0054 0.0006 0.0053 0.0006 0.0046 0.0006 0.0028 0.0007
between 6 and (inc.) 12 0.0172 0.0008 0.0158 0.0008 0.0162 0.0008 0.0162 0.0008 0.0161 0.0007 0.0039 0.0008
between 12 and (incl.) 16 0.0287 0.0006 0.0303 0.0006 0.0280 0.0006 0.0281 0.0006 0.0283 0.0006 0.0035 0.0007
Observed worker characteristics: one if
Not married -0.0146 0.0007 -0.0250 0.0007 -0.0248 0.0007 -0.0242 0.0007 -0.0241 0.0007 -0.0157 0.0015
Widowed -0.0854 0.0029 -0.0852 0.0029 -0.0823 0.0028 -0.0822 0.0028 -0.0818 0.0028 -0.0046 0.0056
Divorced -0.0557 0.0010 -0.0644 0.0010 -0.0629 0.0009 -0.0624 0.0009 -0.0615 0.0009 0.0006 0.0020
Aged between 20 and 25 -0.3402 0.0017 -0.3229 0.0016 -0.3078 0.0016 -0.3044 0.0016 -0.2918 0.0016 -0.0767 0.0025
Aged between 25 and 30 -0.1984 0.0011 -0.1900 0.0011 -0.1887 0.0010 -0.1872 0.0010 -0.1810 0.0010 -0.0461 0.0018
Aged between 30 and 35 -0.0690 0.0010 -0.0681 0.0010 -0.0695 0.0009 -0.0690 0.0009 -0.0674 0.0009 -0.0151 0.0012
Aged between 40 and 45 0.0368 0.0009 0.0362 0.0009 0.0370 0.0009 0.0367 0.0009 0.0356 0.0009 -0.0004 0.0011
Aged between 45 and 50 0.0739 0.0010 0.0731 0.0010 0.0696 0.0010 0.0695 0.0009 0.0671 0.0010 -0.0109 0.0015
Aged between 50 and 55 0.0999 0.0010 0.0997 0.0010 0.0903 0.0010 0.0903 0.0010 0.0870 0.0010 -0.0269 0.0019
Aged between 55 and 60 0.1153 0.0011 0.1153 0.0011 0.1052 0.0011 0.1048 0.0011 0.1007 0.0011 -0.0488 0.0023
Aged between 60 and 65 0.1597 0.0015 0.1558 0.0015 0.1427 0.0015 0.1427 0.0015 0.1393 0.0015 -0.0733 0.0028
Fixed municipality effects
Municipality of workplace YES YES YES YES YES
Fixed industry effects
Industry 1 -0.2296 0.0025 -0.2190 0.0025 -0.2203 0.0025 -0.0074 0.0051
Industry 2 -0.1154 0.0009 -0.1227 0.0009 -0.1244 0.0009 0.0084 0.0018
Industry 3 -0.1042 0.0011 -0.1001 0.0011 -0.1037 0.0011 -0.0007 0.0026
Industry 4 -0.1643 0.0009 -0.1608 0.0009 -0.1612 0.0009 -0.0099 0.0018
Industry 5 -0.3372 0.0021 -0.3331 0.0021 -0.3286 0.0021 -0.0355 0.0053
Industry 6 -0.0303 0.0011 -0.0264 0.0011 -0.0284 0.0011 0.0654 0.0023
Industry 7 0.1626 0.0012 0.1772 0.0013 0.1722 0.0013 0.0273 0.0021
Industry 9 -0.0207 0.0010 -0.0160 0.0011 -0.0172 0.0011 -0.0053 0.0026
Industry 10 0.0779 0.0011 0.0622 0.0011 0.0634 0.0011 -0.0326 0.0032
Industry 11 -0.1121 0.0009 -0.1179 0.0009 -0.1182 0.0009 -0.0411 0.0027
Industry 12 -0.1013 0.0015 -0.1041 0.0015 -0.1028 0.0015 -0.0027 0.0027
Fixed firm size effects: one if firm size equals
9≤ size ≤ 22 -0.0132 0.0012 -0.0112 0.0012 0.0108 0.0014
23≤ size ≤53 0.0172 0.0012 0.0197 0.0012 0.0244 0.0016
54≤ size ≤133 0.0418 0.0011 0.0446 0.0011 0.0379 0.0017
134≤ size ≤311 0.0589 0.0011 0.0615 0.0011 0.0362 0.0018
312≤ size ≤725 0.0497 0.0012 0.0519 0.0012 0.0384 0.0019
726≤ size ≤1564 0.0254 0.0012 0.0278 0.0012 0.0437 0.0019
1567≤ size ≤3238 0.0137 0.0012 0.0161 0.0012 0.0459 0.0020
3248≤ size ≤10664 0.0489 0.0012 0.0511 0.0012 0.0376 0.0021
10679≤ size ≤86840 -0.0117 0.0013 -0.0064 0.0013 0.0603 0.0022
Job characteristics: one if worker holds
Temporary contract -0.0667 0.0009 -0.0148 0.0009
Job characteristics: one if tenure at firm in weeks equals
4≤ tenure <39 0.2452 0.0021 0.0346 0.0022
39≤ tenure <74 -0.0175 0.0012 -0.1156 0.0018
74≤ tenure <122 -0.0028 0.0011 -0.0936 0.0018
122≤ tenure <198 0.0029 0.0011 -0.0757 0.0018
198≤ tenure <287 -0.0058 0.0011 -0.0520 0.0018
287≤ tenure <379 -0.0206 0.0010 -0.0354 0.0017
379≤ tenure <491 -0.0094 0.0010 -0.0209 0.0016
491≤ tenure <689 0.0058 0.0010 -0.0155 0.0015
689≤ tenure <1002 -0.0119 0.0010 -0.0050 0.0011
Worker fixed effects YES

Worker groups 1141683

Degrees model 3 5 19 461 472 481 491 1,142,173
R2 0.0551 0.0564 0.1300 0.1625 0.2028 0.2054 0.2114 0.8848
Combined effect (hit − cit ) 0.2574 0.0006 0.2571 0.0006 0.2153 0.0006 0.2307 0.0006 0.2117 0.0006 0.2115 0.0006 0.2086 0.0006 0.0080 0.0011
All coefficients except * significant at the 0.1 promille significance level. * significant at the 10 percent significance level.
regressions based on 2,884,112 observations.
150 Chapter 5.

Table 5.9: Coefficients and standard errors Table 5.5

Dependent variable: log hourly wage

1 2 3 4 5
coef s.e. coef s.e. coef s.e. coef s.e. coef s.e.

Housing Tenure
Living in social housing -0.0026** 0.0012 -0.0021 0.0015 -0.0030* 0.0018 -0.0008 0.0033 0.0000***** 0.0000
Being a homeowner 0.0033**** 0.0010 0.0048**** 0.0013 0.0030** 0.0015 0.0000***** 0.0000 0.0043*** 0.0016
Fixed year effects
Year 2006 -0.0753***** 0.0003 -0.0735***** 0.0003 -0.0662***** 0.0003 -0.0636***** 0.0008 -0.0640***** 0.0004
Year 2007 -0.0493***** 0.0002 -0.0485***** 0.0003 -0.0442***** 0.0003 -0.0423***** 0.0007 -0.0441***** 0.0004
Observed worker characteristics: one if children aged
6 or less -0.0005 0.0006 -0.0005 0.0006 -0.0020**** 0.0006 0.0055*** 0.0019 0.0016** 0.0008
Between 6 and (incl.) 12 0.0030***** 0.0006 0.0033***** 0.0007 0.0031***** 0.0007 -0.0014 0.0019 0.0057***** 0.0009
Between 12 and (incl.) 16 0.0032***** 0.0005 0.0038***** 0.0006 0.0050***** 0.0006 0.0013 0.0016 0.0056***** 0.0008
Observed worker characteristics: one if
Not married -0.0154***** 0.0011 -0.0202***** 0.0014 -0.0206***** 0.0018 -0.0100*** 0.0037 -0.0172***** 0.0018
Widowed -0.0095** 0.0041 -0.0122** 0.0052 -0.0098* 0.0052 0.0061 0.0112 -0.0054 0.0065
Divorced 0.0019 0.0014 0.0007 0.0018 0.0020 0.0018 -0.0035 0.0041 0.0000 0.0025
Aged between 20 and 25 -0.0869***** 0.0017 -0.1055***** 0.0023 -0.0677***** 0.0048 -0.0807***** 0.0030
Aged between 25 and 30 -0.0477***** 0.0013 -0.0562***** 0.0016 -0.0429***** 0.0036 -0.0475***** 0.0021
Aged between 30 and 35 -0.0162***** 0.0008 -0.0188***** 0.0010 -0.0239***** 0.0010 -0.0121***** 0.0025 -0.0155***** 0.0013
Aged between 40 and 45 0.0038***** 0.0008 0.0061***** 0.0009 0.0106***** 0.0009 -0.0070*** 0.0024 0.0019 0.0012
Aged between 45 and 50 -0.0029*** 0.0011 0.0004 0.0013 0.0096***** 0.0014 -0.0257***** 0.0033 -0.0054*** 0.0018
Aged between 50 and 55 -0.0170***** 0.0014 -0.0134***** 0.0017 0.0007 0.0017 -0.0462***** 0.0041 -0.0189***** 0.0022
Aged between 55 and 60 -0.0395***** 0.0017 -0.0360***** 0.0020 -0.0172***** 0.0020 -0.0677***** 0.0048 -0.0398***** 0.0027
Aged between 60 and 65 -0.0630***** 0.0020 -0.0596***** 0.0024 -0.0365***** 0.0024 -0.0959***** 0.0055 -0.0620***** 0.0032
Fixed municipality effects
Municipality of workplace YES YES YES YES YES YES YES
Fixed industry effects
Industry 1 0.0073** 0.0035 0.0074* 0.0040 0.0154**** 0.0045 -0.0085 0.0105 -0.0040 0.0060
Industry 2 0.0142***** 0.0012 0.0148***** 0.0014 0.0191***** 0.0015 0.0140**** 0.0036 0.0073**** 0.0021
Industry 3 0.0041** 0.0018 0.0067**** 0.0019 0.0068*** 0.0022 0.0053 0.0053 -0.0034 0.0031
Industry 4 -0.0042*** 0.0013 -0.0032** 0.0015 0.0055**** 0.0016 -0.0165***** 0.0037 -0.0081**** 0.0022
Industry 5 -0.0338***** 0.0037 -0.0333***** 0.0048 -0.0192**** 0.0057 -0.0447***** 0.0095 -0.0307***** 0.0067
Industry 6 0.0431***** 0.0016 0.0493***** 0.0019 0.0545***** 0.0021 0.0573***** 0.0043 0.0685***** 0.0028
Industry 7 0.0289***** 0.0015 0.0314***** 0.0019 0.0305***** 0.0020 0.0466***** 0.0048 0.0227***** 0.0023
Industry 9 0.0035* 0.0019 0.0025 0.0024 -0.0047* 0.0026 0.0210***** 0.0051 -0.0152***** 0.0031
Industry 10 -0.0270***** 0.0025 -0.0346***** 0.0032 -0.0370***** 0.0036 -0.0097 0.0064 -0.0454***** 0.0038
Industry 11 -0.0250***** 0.0021 -0.0240***** 0.0031 -0.0262***** 0.0033 -0.0200**** 0.0053 -0.0508***** 0.0032
Industry 12 -0.0060*** 0.0020 0.0019 0.0025 0.0021 0.0027 0.0152*** 0.0049 -0.0102*** 0.0033
Fixed firm size effects: one if firm size equals
9≤ size ≤ 22 0.0095***** 0.0010 0.0116***** 0.0012 0.0117***** 0.0013 0.0114**** 0.0030 0.0099***** 0.0017
23≤ size ≤ 53 0.0197***** 0.0011 0.0218***** 0.0013 0.0208***** 0.0015 0.0256***** 0.0034 0.0239***** 0.0019
54≤ size ≤ 133 0.0295***** 0.0012 0.0325***** 0.0014 0.0305***** 0.0016 0.0340***** 0.0036 0.0389***** 0.0020
134≤ size ≤ 311 0.0322***** 0.0013 0.0346***** 0.0015 0.0327***** 0.0017 0.0326***** 0.0038 0.0373***** 0.0021
312≤ size ≤ 725 0.0422***** 0.0013 0.0450***** 0.0016 0.0437***** 0.0017 0.0316***** 0.0039 0.0404***** 0.0022
726≤ size ≤ 1,564 0.0380***** 0.0014 0.0406***** 0.0017 0.0383***** 0.0018 0.0392***** 0.0040 0.0451***** 0.0023
1,567≤ size ≤ 3,238 0.0401***** 0.0014 0.0441***** 0.0018 0.0423***** 0.0019 0.0446***** 0.0041 0.0465***** 0.0024
3,248≤ size ≤ 10,664 0.0348***** 0.0015 0.0306***** 0.0018 0.0258***** 0.0020 0.0410***** 0.0042 0.0364***** 0.0024
10,6749≤ size ≤ 86,840 0.0575***** 0.0016 0.0585***** 0.0019 0.0526***** 0.0021 0.0560***** 0.0045 0.0610***** 0.0026
Job characteristics: one if worker holds
Temporary contract -0.0236***** 0.0006 -0.0245***** 0.0008 -0.0269***** 0.0009 -0.0136***** 0.0017 -0.0148***** 0.0011
Job characteristics: one if tenure at the firm in weeks equals
4 ≤ tenure < 39 0.0197***** 0.0015 0.0134***** 0.0018 0.0057*** 0.0019 0.0416***** 0.0046 0.0314***** 0.0025
39 ≤ tenure < 74 -0.1100***** 0.0013 -0.1145***** 0.0015 -0.1218***** 0.0015 -0.1122***** 0.0039 -0.1160***** 0.0020
74 ≤ tenure < 122 -0.0817***** 0.0013 -0.0880***** 0.0015 -0.0983***** 0.0015 -0.0818***** 0.0039 -0.0974***** 0.0020
122 ≤ tenure < 198 -0.0639***** 0.0013 -0.0702***** 0.0015 -0.0804***** 0.0015 -0.0668***** 0.0039 -0.0787***** 0.0020
198 ≤ tenure < 287 -0.0424***** 0.0013 -0.0488***** 0.0015 -0.0581***** 0.0015 -0.0470***** 0.0039 -0.0530***** 0.0020
287 ≤ tenure < 379 -0.0273***** 0.0012 -0.0329***** 0.0015 -0.0407***** 0.0015 -0.0329***** 0.0038 -0.0355***** 0.0020
379 ≤ tenure < 491 -0.0156***** 0.0012 -0.0205***** 0.0014 -0.0263***** 0.0014 -0.0221***** 0.0036 -0.0196***** 0.0019
491 ≤ tenure < 689 -0.0095***** 0.0011 -0.0138***** 0.0013 -0.0174***** 0.0013 -0.0153***** 0.0033 -0.0150***** 0.0017
689 ≤ tenure < 1,002 0.0005 0.0008 -0.0012 0.0009 -0.0040***** 0.0009 -0.0040* 0.0024 -0.0056***** 0.0012
Worker fixed effects YES YES YES YES YES

Observations 2,338,343 1,669,985 1,441,815 674,583 2,155,097

Worker groups 921,631 644,847 540,518 288,211 841,248
Degrees model 922,121 645,337 541,006 288,700 841,737
R2 0.9428 0.9413 0.9404 0.8683 0.8829
Combined (hit − cit ) 0.0059***** 0.0008 0.0069***** 0.0010 0.0059***** 0.0013 0.0008 0.0033 0.0043*** 0.0016
*, **, ***, ****, ***** significant at the 10, 5, 1, 0.1 and 0.01 percent significance level.
Specification 1): All workers working 24 hours a week or more
Specification 2): All male workers working 24 hours a week or more
Specification 3): All male workers, aged 30 or older, working 24 hours a week or more
Specification 4): All tenants of social and private rental housing
Specification 5): All homeowners and tenants of private rental housing
Part B:
A macro-economic
perspective on rent control

151
Chapter 6

Housing and labor market

rigidities: Consequences for
skill composition

6.1 Introduction
Many countries have witnessed increased sorting of high-skilled workers in
cities with high productivity and house prices. Moretti (2012) calls this ‘The
great divide’. Urban economists explaining this sorting mechanism have em-
phasized the interaction of local labor and housing markets. As housing pro-
vides access to local jobs "housing supply elasticity will determine whether
urban success reveals itself in the form of more people or higher incomes"
(Glaeser and Gottlieb, 2009, p. 983).1
Consequently, both housing market and labor market price rigidities influ-
ence the equilibrium in the economy as they alter market signals. On the labor
market, wage rigidities reduce the incentive to move into cities with higher
1 Local productivity shocks stimulate the inflow of labor into a region. If the elasticity of

housing supply is high, the housing stock can adapt smoothly to changes in local demand. As a
result, the city grows, and many workers are hired at a rate close to the pre-existing wage rate.
In contrast, if the elasticity of housing supply is low, workers cannot enter the city. This increases
both wages and housing costs.
154 Chapter 6.

productivity. On the housing market, rent control benefits and the illiquidity
of housing reduce labor mobility, as they are impediments to moving.
It has been acknowledged that outcomes on the housing and labor market
are dependent. Local wage adjustments become more important to clear the
labor market if labor is geographically immobile (say: due to housing mar-
ket restrictions, see Hughes and McCormick 1987). In this chapter I study the
combined effect of housing and labor market rigidities on local skill composi-
tion and derive conditions under which they enhance another and conditions
under which they cancel out. Two price rigidities are considered: On the hous-
ing market, I allow for a share of the housing stock to be subject to rent control,
which freezes the rent across the country. As a result, demand for location is
no longer internalized into the controlled house prices.
On the labor market, I consider the role of equalizing wages across the
country in wage negotiations between employer and employee representa-
tives. If regions differ in productivity, this will generate local unemployment
differentials. I adjust the matching framework developed by Mortensen and
Pissarides (see Pissarides 2000) to model how labor demand is revealed in the
local unemployment rate when wages are rigid.
In an economy with perfect labor and housing markets, a productivity
shock to high-skilled labor in a region will lead to an inflow of high-skilled
workers and an outflow of low-skilled workers. In this chapter I show how
labor and housing market price rigidities might mitigate or amplify this relo-
cation process.
First, the model shows that the effect of price rigidities on the housing
market depends on the distribution of rent-control housing vouchers among
workers of different skill. If rent-control housing vouchers are allocated mainly
among low-skilled workers, the inflow of high-skilled workers and outflow of
low-skilled workers after a productivity shock to high skilled labor will both
be lower. The reason is that low-skilled workers with a rent-control housing
voucher do not leave the core after the inflow of high-skilled workers increases
market rents. The reverse occurs if rent-control housing vouchers are allo-
cated mainly among high-skilled workers. Then, the inflow of high-skilled
workers will be larger, as increasing house prices do not deter high-skilled
6.1. Introduction 155

workers with a housing voucher from entering the region. However, they in-
crease house prices for low-skilled workers without a housing voucher, which
creates more outflow of low-skilled workers. Also, there exists an intermedi-
ate distribution such that rent-control housing vouchers increase the inflow
of high-skilled workers and reduce the outflow of low-skilled workers after a
productivity shock to high-skilled labor.
Second, the model indicates that labor market price rigidities mitigate the
effect of a productivity shock on the relocation process. This occurs if, initially,
the regions are equally productive or if the nominal wage rate is not adjusted
in response to the productivity shock. The reason is that labor market price
rigidities reduce the spatial expected wage rate differential that occurs due to
the productivity shock.
However, if the nominal wage rate of high-skilled workers is altered in
accordance to the productivity shock or if both regions differ in productiv-
ity, labor market price rigidities might amplify the relocation process. This is
more likely to occur if unemployment levels are very high, the labor market
matching process is vacancy intensive and wages are much larger than unem-
ployment benefits.
A third finding is that, in general, housing market price rigidities amplify
labor market price rigidities after a productivity shock to high-skilled labor,
whereas they mitigate labor market price rigidities after a productivity shock
to low-skilled labor. Vice versa, it is concluded that, in general, the effect
of housing market price rigidities on the relocation of high- and low-skilled
workers is lower if labor market price rigidities are present.2
This research is related to several branches of the economic literature. It
augments studies of the effect of labor market rigidities on labor supply. In
many European countries wages at the industry level are not set via bargain-
ing by individual firms and workers, but in a bargaining process between
(firm or) industry representatives and worker representatives. These negoti-
ations often take place at a supra-regional (or national) level, see for instance
Hartog et al. (2002). In this chapter I model wage negotiations at the suprare-
gional level, and allow them to influence the housing market.
2 ‘In general’ refers to the situation in which rent-control housing vouchers are mainly allo-

cated among low-skilled workers and both regions are equally productive.
156 Chapter 6.

Wage setting regimes have been found to affect regional unemployment

when wages are set according to one ‘leading’ labour market (Manacorda
and Petrongolo, 2006); or centralized wage setting with mobile workers and
government preference for immobility of workers (Caponi, 2008). These la-
bor market outcomes might influence the housing market. For instance, Ver-
meulen and Van Ommeren (2009) present empirical evidence that housing
markets provide compensation for regional unemployment. This chapter adds
to this literature on the relationship between unemployment, wages and the
housing market, as it considers the joint effect of housing and labor market
rigidities on the location decision of workers.3
The remainder of this chapter is organized as follows: Section 6.2 describes
the equilibrium conditions on the housing and labor market that constitutes
equilibrium in the economy. Section 6.3 describes how skill and place specific
productivity shock influence the location decision of workers and affect local
skill composition. Next, section 6.4 shows how the equilibrium is affected by
housing market price rigidities in the form of rent control. The effect of nom-
inal wage rigidities on the equilibrium is described in section 6.5. The com-
bined effect of housing and labor market price rigidities is derived in section
6.6. Finally, section 6.7 concludes.

6.2 The model

In this section a general equilibrium model is presented that has two regions
and workers that are high-skilled or low-skilled. Utility of workers is a func-
tion of wage income (+), local amenities (+), house prices (-) and an idiosyn-
cratic preference for either region (+). Unemployed workers move to the re-
gion that maximizes expected utility. It is assumed unemployed workers face
zero moving costs, but employed workers are tied to their work and their
moving costs are infinite. As a result, only unemployed workers react to labor
demand shocks.
Labor markets are imperfect as skill-specific matches depend on the num-
ber of vacancies and unemployed workers in the region. Each filled job pro-
3 See Hughes and McCormick (1987); Minford et al. (1988); McCormick (1997) for the UK,

Flatau et al. (2003) for Australia, Munch and Svarer (2002); Svarer et al. (2005) for Denmark.
6.2. The model 157

vides a surplus that has to be shared between the employer and the employee.
I introduce price rigidities in the labor market by assuming that employer and
employee-representatives bargain over a national wage rate. On the one hand,
the national wage rate reduces the incentive to move as employed workers
of type T earn equal wages in both regions. On the other hand, it sets the
local unemployment rate higher or lower than the locally optimal unemploy-
ment rate.4 Therefore the probability to be employed (at the national wage rate)
differs between both regions.
I also introduce housing market price rigidities. Housing prices are a func-
tion of the number of residents in a city such that attractive cities have higher
house prices. Part of the workforce is provided rent-control housing vouchers
such that they can rent housing at the rent ceiling regardless of city size. This
influences the outcomes of the model, depending on the distribution of the
vouchers among high-skilled and low-skilled workers. Workers with a rent-
control voucher are not subject to changing housing costs. Therefore, after
a skill-specific productivity shock, rent-control vouchers affect the local skill
composition in two ways: they increase the inflow of workers of that skill
type, and they prevent the outflow of workers of the other skill type.

6.2.1 Labor market matching

Workers can be high-skilled or low-skilled. Let the type of labor be denoted

with the superscript T = H, L, where H denotes high-skilled workers and
L low-skilled workers. Workers live and work in the same region c = a, b.
For simplicity, assume that the two types of labor work in firms that produce
different products using only high-skilled and low-skilled labor. They sell
their products at a fixed market price.
Firms post skill-specific vacancies depending on their required number of
matches. Assume that the number of matches LcT in any period dt is a function
of the number of skill-specific vacancies VcT , and the number of unemployed
workers UcT . Furthermore, I assume the matching process to be Cobb-Douglas
and of constant returns to skill, with general matching technology productiv-
ity µ, and vacancy intensity d. Local labor market tightness θcT is defined as
4 Except in the special case in which the national wage rate equals the local optimal wage rate.
158 Chapter 6.

the number of vacancies per unemployed worker of type T in city c. Use labor
market tightness to define the local vacancy filling rate m(θcT ) and the local
exit rate out of unemployment θcT m(θcT ).

−(1−d) d
VcT VcT
 
LcT = µ(VcT )d (UcT )1−d , m(θcT ) = µ UcT
, θcT m(θcT ) = µ UcT

6.2.2 Behavior of firms

All firms have to incur costs h > 0 each time period dt to attract labor. These
costs can be thought of as resembling search costs, hiring costs, and/or inter-
mediate costs. The return of a job of type T in c is assumed to be constant at
ycT , whereas wage paid equals wcT . With the risk free interest rate r and with
the probability that a filled jobs becomes vacant being equal to q, the expected
profit of a filled and vacant job of type T in c in each period can be expressed
as:5

rπcTe = (ycT − wcT ) + q(πcTv − πcTe )

rπcTv = −h + m(θcT )(πcTe − πcTv )

Here, superscript e indicates a filled (employed) job and v a vacant job.

The free entry condition states that firms enter the market and post jobs
until the expected profit of a vacant job is zero (πcTe = 0). Under this condition
and using the equilibrium value for the vacancy filling rate, the unemploy-
ment rate uc of workers of type T can be expressed as

1 
µ ycT −wcT
δ
ucT = with αcT = µ h r +q , δ= d
1− d
1 + αcT

This equation gives the unemployment rate as determined by the matching

technology, fundamental parameters h, q, r and the local productivity and wage
rate for a filled job. It is a decreasing function of the surplus of a filled job
ycT − wcT and matching technology µ. The unemployment rate is increasing in
the costs of filling a vacant job h as well as in the interest rate r and the job dis-
5 Imposed restrictions are that r > 0, 0 < q < 1 and h > 0. Thus firms care about future pe-

riods, there is uncertainty whether the household becomes unemployment and keeping a vacant
job is costly.
6.2. The model 159

solvement rate q. Finally, the unemployment rate is decreasing in the vacancy

intensity of the matching process d. Note that population size or number of
workers of skill-type T does not influence the local unemployment rate. This
is a common feature of these types of matching models.

6.2.3 Determination of wages

Firms open up jobs in both regions such that the expected value of opening a
vacancy is zero. Yet, each firm makes a nonzero profit on each job that is filled,
as ycT − wcT ≥ 0. This positive surplus has to be divided between employers
and employees.
Employer and employee representatives enter a Nash bargaining process
over the surplus. Simplify the bargaining process by assuming that employer
representatives bargain with respect to a reference productivity level y Tr and
employee representatives with respect to replacement benefits z.6 With bar-
gaining power of employee representatives being given by β, the Nash bar-
gaining surplus for a filled job of type T becomes:

 β   1− β
S T = y Tr − w T wT − z , 0<β<1

Taking logs and maximizing with respect to w T shows that the negotiated
optimal wage equals:

max S T : w T = y Tr − β(y Tr − z)
wT

Therefore, the derivative of the wage rate with respect to reference productiv-
ity equals (1 − β). Note that the wage rate is independent of the local produc-
tivity level and depends solely on reference productivity and unemployment
insurance. Use the expression for wcT to write the unemployment rate in loca-
tion c as:
 δ h
1 µ iδ
ucT = , where acT = µ ycT − (1 − β)y Tr − βz
1 + acT h (r + q )
6 Thus, bargaining does not take place between individual firms and workers, but between

their representatives. I explicitly assume that regional differences with respect to living conditions
(amenities, house prices) are not taken into account by employee representatives.
160 Chapter 6.

Unemployment differentials between city a and b can only come from differ-
entials in productivity, as all other variables are assumed to be equal across
the country. For reasons that become apparent later, write the first derivatives
of the local unemployment rate with respect to the local productivity level and
the reference productivity level as functions of the unemployment level:7

∂ucT ∂ucT
T
= − f (ucT )ζδ; = −(1 − β) f (ucT )ζδ (6.1)
∂yc ∂y Tr
 1−1/δ  1+1/δ
f (ucT ) = 1 − ucT ucT , ζ = µ1/δ h(r+q)
µ

The shape of f ucT depends on the local unemployment rate ucT and vacancy-

intensity of the matching process d. Figure 6.7 in section 6.A shows how
f ucT changes with ucT for different values of δ. If matching is vacancy neu-

tral or vacancy intensive (d ≥ 0.5) the function will be smaller or equal to one.
f ucT might exceed one if matching is vacancy extensive (or: unemployed

worker intensive, d < 0.5), but this only occurs if the unemployment rate is 60
percent or more.

6.2.4 Behavior of workers

Assume workers live infinite lives and are risk neutral. Employed workers of
type T earn w T , whereas unemployed workers receive replacement benefits
z < w T . Housing costs are given by pc . The value of local amenities is de-
noted Ac , whereas individual preference for location c is eic . Housing costs,
local amenities and individual preferences for a location do not depend on
employment status. I assume z, w > pc , such that unemployed workers can
locate in either region. Unemployed workers are always in search of jobs. The
expected utility of being unemployed equals:

r +q
rVicTu = (w T − pc ) − r +q+αcT
(w T − z) + Ac + eic (6.2)

Tu > U Tu .
Unemployed workers of type T decide to look for jobs in b if Uib ia
Equating (6.2) for location a and b and rearranging gives (6.3). The marginal
7 Where acT T T and


I used 2 = uc 1 − uc
(1+acT )

−1
−1/δ T
1/δ 1/δ µ
ycT − (1 − β)y Tr − βz = 1 − ucT uc µ h (r + q ) .
6.2. The model 161

unemployed worker of type T is indifferent between moving or not and for

him the equality holds.
 
r +q r +q
(eia − eib ) ≤ r +q+α Ta
− r +q+αbT
(w T − z) + ( p a − pb ) + ( Ab − A a ) (6.3)

Preferences for location a over b are i.i.d. uniformly distributed on the domain
[−s, s], independent from skill type.

eia − eib ∼ [−s, s]

The size of s determines the number of workers in b. Assume that the value of
s is large enough to restrict the decision to relocate for at least one unemployed
household.8 Then in the long run the share of workers of type T in location b
equals
 
r +q r +q
NbT N2sT = r +q+α Ta
− r +q+αbT
(w T − z) + ( p a − pb ) + ( Ab − A a ) + s (6.4)

Equation (6.4) indicates how population size of workers of type T is deter-

mined by employment conditions and the wage rate for its skill, local house
prices and the local level of amenities. Note that equation (6.4) also equals
housing demand in b by workers of skill type T. As the labor market is in
equilibrium if equation (6.4) holds, replacing pc by its value determined by
housing market supply will solve the model.

6.2.5 The housing market

Assume all housing is owned by profit maximizing agents abroad with deep
pockets. They rent out housing at the market price to uncontrolled tenants.
The price of uncontrolled housing is an increasing function of demand. Thus
 
r +q r +q
8 This restriction is met if s ≥
r +q+α T
− r +q+αbT
(w T − z) + ( p a − pb ) + ( Ab − A a ). It has a
a
similar role as the ‘no black hole condition’ in the New Economic Geography literature. The ‘no
black hole condition’ ensures that the entire economy does not collapse into one point, see Fujita
et al. (1999). Similarly, the restriction that s is large enough, ensures that unemployed workers of
type T do not locate in only one region.
162 Chapter 6.

both types of workers compete for the same uncontrolled housing.

pc = ρ + k c Nc

However, some workers are in the possession of a voucher that enables them
to rent housing at ρ regardless of location. γN with 0 ≤ γ ≤ 1 equals the stock
of housing vouchers in the economy. α L γN and α H γN are the number of low-
skilled and high-skilled workers with a housing voucher, where α L = 1 − α H .
It is assumed that the rent-control housing vouchers are randomly allocated
within the two skill groups.
Consider the location decision of rent-controlled workers. They locate ac-
cording to equation (6.4) with p a = pb = ρ. Therefore, their location is driven
by the difference in real wage and amenities in a and b and localization prefer-
ence only. City size or housing demand by other households do not influence
their decision to live in a or b. However, by living in a city, they influence the
price of uncontrolled housing. Therefore, they affect the location decision of
workers without a housing voucher.
It follows that the number of high and low-skilled workers in b is given by
equation (6.5). For notational convenience, I used the simplifying assumption
r + q = 1. Section 6.A.4 shows the derivations and also shows the number of
controlled and market tenants by skill type.
 0 0

σ + N T −α T γN
  
NT NT
NbT = 2 + σ +(1−γ) N 2s u Ta − ubT wT − z
0
N T −α T γN N T 0 0
  0 
− σ +(1−γ) N 2s
u Ta − ubT wT − z
N T −α T γN
h i
NT N
+ N − σ +(1−γ) N 2s ( Ab − Aa )
N T −α T γN N
+ (k
σ +(1−γ) N 2(k a +k b ) a
− kb ) (6.5)

6.3 Effect of an idiosyncratic productivity shock on

labor relocation
This section describes the comparative statics of our model with two types of
labor, nominal wage rigidities and rent-controlled housing if an idiosyncratic
6.3. Effect of an idiosyncratic productivity shock on labor relocation 163

productivity shocks occurs. Without loss of generality, I consider the case

when the productivity of high-skilled workers in b increases. It is assumed
both regions are identical in the price elasticity of housing supply (k a = k b ),
local amenities (A a = Ab ) and, initially, in local productivity levels as well
(y Ta = ybT ). Assume that the productivity level in b is the reference productiv-
ity level for both types of labor.
If a productivity shock to high-skilled labor occurs in b only, such that
H
yb2 H = ∆ > 0, this will influence the location of labor via its effect on
− yb1
relative employment conditions. Denote relative employment conditions for
workers of skill type T in location b as EbT . Relative employment conditions
are a function of the national skill-specific wage rate and local unemployment
rates. Define them as EbT = (1 − ubT )(w T − z) − (1 − u Ta )(w T − z), which equals
the first line of equation 6.5. Then, the effect of the idiosyncratic productiv-
ity shock might manifest itself via an effect on the wage of all high-skilled
workers w H and an effect on the local unemployment rates. More specifically,
the unemployment level in b falls as the region becomes more productive,
whereas the unemployment rate in a rises.
Starting from equation (6.5), write the change in the number of high and
low-skilled workers in b because of the change in productivity as:

∂NbH σ +( N − N H )−(1−α H )γN N H H

∆= 2s ∇ Eb ∆ >0 (6.6)
∂y Hb σ +(1−γ) N
∂NbL ( N − N H )−(1−α H )γN N H H
∆=− 2s ∇ Eb ∆ ≤0 (6.7)
∂y Hb σ +(1−γ) N

Here, I used ∇ EbH to define the change in relative employment conditions of

high-skilled workers due to the productivity shock in b under rigid wages.9

dEbH ∂EbH ∂EbH ∂Y Hr

∇ EbH = dybH
= ∂ybH
+ ∂y Hr ∂YbH
  
Hr f (u aH )
 
= 1 + (1 − β) ∂y
∂y H
1− f (ubH )
ζδ f (ubH ) w H − z +
b

∂y Hr
(1 − β) ∂y H (u aH − ubH ) (6.8)
b

9 Note that the assumption that the initial unemployment levels in a and b are equal ensures

that any change in the reference productivity level does not affect ∇ EbH . Under this condition
∇ EbH simplifies to ∇ EbH = ζδ f (ubH ) w H − z .


164 Chapter 6.

Equations (6.6) shows that an increase in productivity of high-skilled workers

in b makes city b more attractive for high-skilled workers compared to city
a. As a result, high-skilled workers move to b, which increases house prices.
As productivity of low-skilled workers in b has not changed, whereas house
prices increase, low-skilled workers are made worse off. Therefore some low-
skilled workers move to a. There is one exception to this rule: If all low-skilled
workers are on rent-control such that (1 − α H )γN = N − N H , then housing
prices will not increase for low-skilled workers. As a result, a shock to the
productivity of high-skilled labor does not lead to an outflow of low-skilled
workers.

6.4 Rent control and labor relocation

Here I highlight the role of rent-control on the allocation of high-skilled and
low-skilled workers. Equations (6.6) and (6.7) show that the change in the
population of workers depends on the size of the rent-controlled housing
stock and its distribution among high and low-skilled workers.
Basically, there are four types of allocations, see Table 6.1. In the baseline
situation, there are no rent-controlled housing vouchers (γ = 0). As there
are no vouchers to allocate, the value of α H is irrelevant. In the other type of
situations, the stock of rent-control housing vouchers γ is nonzero. In the one
extreme, all housing vouchers are in the possession of low-skilled workers
and α H = 0. This is the top right cell in Table 6.1. Another extreme allocation
would be the one in which all housing vouchers are the property of high-
skilled workers (α H = 1, the bottom left cell). Finally, the cell in the middle of
Table 6.1 refers to the situation in which the stock of rent-controlled housing
is nonzero and are distributed among both high and low-skilled workers.
6.4. Rent control and labor relocation 165

Table 6.1: Allocation of rent-control housing vouchers

Among low-skilled
none some all
none 0≤ ≤1 αH αH
=0
γ=0 γ>0
Among high-skilled some 0 < αH < 1
γ>0
all αH = 1
γ>0
α and γ are such that αγN ≤ N H and (1 − α)γN ≤ N − N H

Figure 6.1 shows the change in number of high-skilled and low-skilled

workers after an increase in productivity of high-skilled workers in b. The up-
ward sloping lines represent the change in the number of high-skilled work-
ers, the downward sloping lines are the change in the number of low-skilled
workers in b. This is done for the baseline situation and the situations in which
all rent-control housing vouchers are in the possession of high-skilled workers
or low skilled workers.
The baseline situation in which there are no rent-control housing vouch-
ers is illustrated with the solid black lines. The upward sloping solid black
line shows the inflow of high-skilled workers in b after a productivity shock
(to high-skilled workers in b). It is linear and becomes horizontal when all
the high-skilled workers in the economy live in b. As high-skilled workers
enter city b, house prices go up, and some low-skilled workers leave b. This
is represented by the downward sloping solid line, which becomes horizon-
tal when all low-skilled workers have left b. Note that city b is growing, as
the slope parameter for high-skilled workers exceeds the absolute value of the
slope parameter for low-skilled workers.
Next, consider the situation in which there are some rent-control housing
vouchers in the economy, and they are all held by low-skilled workers (α H =
0, γ > 0 in the top right cell). Then the inflow of high-skilled workers is lower
compared to the baseline situation.10 At the same time, equation (6.7) shows
that the flow of low-skilled workers out of b is lower compared to the baseline
10 This σ+( N − N H )−γN σ+( N − N H )
is because σ+ N −γN < σ+ N .
166 Chapter 6.

Figure 6.1: Rent control and the change in population in b

The graph shows the situation outlined above where the local amenities, the un-
employment rates of low-skilled labor and housing market elasticities are equal
across both cities. Furthermore, WbH /2s = 4, σ = 500/20 = 25, population size
N = 100 and number of high-skilled workers N H = 50. Positive values on the y-
axis represent the inflow of high-skilled workers into b after a shock to high-skilled
productivity in b (on the x-axis). Negative values on the y-axis indicate the outflow
of low-skilled workers out of b.
6.4. Rent control and labor relocation 167

situation.11 This situation is shown in Figure 6.1 using the dashed lines (for
γ = 0.25). The increase in productivity of high-skilled workers increases the
number of high-skilled workers in b and the number of low-skilled workers
decreases. However, the increase in high-skilled workers and the decrease in
low-skilled workers are lower compared to the baseline situation. The higher
the stock of rent-control housing vouchers, the more the lines rotate towards
the x-axis. In the extreme case in which all low-skilled workers receive rent-
control housing vouchers, the productivity increase of high-skilled workers
slightly increases the number of high-skilled workers, without leading to an
outflow of low-skilled workers. This implies that city b is still growing. Also,
the skill composition changes in favor of high-skilled workers, however, the
change is less profound.
Consider the effect of rent-control housing vouchers on the presence of
high and low-skilled workers in b if all housing vouchers are in the posses-
sion of high-skilled workers (α H = 1). Now, the shock in the productivity of
high-skilled workers raises the inflow of high-skilled workers compared to the
baseline model in which there are no rent-controlled housing vouchers. Simi-
larly, the outflow of low-skilled workers is larger than the outflow under the
baseline model. The upward sloping dotted line in Figure 6.1 shows the inflow
of high-skilled workers into b in this situation. As housing prices do not hold
back their entrance of high-skilled workers into b, the line is steeper compared
to the baseline situation without rent-control housing vouchers. Similarly, the
outflow of low-skilled workers in this scenario is larger.
Finally, consider the effect of rent-control on the inflow and outflow of
workers after the productivity shock if rent-control housing vouchers are al-
located to both low and high-skilled workers (0 < α H < 1). First, note there
exists a special value for α H for which the relocation of high-skilled workers
is not influenced by rent-control. This value equals α H = N H /(σ + N ).12 For
values of α H that are smaller, an increase in the stock of rent-control hous-
ing vouchers γ will decrease the inflow of high-skilled workers into b after
11 This ( N − N H )−γN (N−N H )
is because σ+ N −γN < σ+ N .
12 Write σ = sN and N − N H = lN. Then equate the multiplier for high-skilled workers if
1− l NH
0 ≤ α H ≤ 1 and γ > 0 to the multiplier if γ = 0. Cross-multiplication gives α H = ( s +1)
= σ+ N .
168 Chapter 6.

the described productivity shock. Likewise, rent-control housing vouchers

do not influence the outflow of low-skilled workers in the special case that
α H = (σ + N H )/(σ + N ).13 For values of α H that are smaller, an increase in
the stock of rent-control housing vouchers γ will decrease the flow of low-
skilled workers out of b after the described productivity shock.
The left panel of Figure 6.2 shows how parameter values for α H and γ
change the inflow of high-skilled workers. For all values of α H and γ the in-
flow of high-skilled workers is positive. For the chosen parameterization in
Figure 6.2, the number of high-skilled workers increases more than the base-
line model if more than 40 percent of the rent-control housing vouchers is in
the possession of high-skilled workers (α H > 0.4). The upward sloping arrow
indicates the difference with the baseline situation is increasing in the number
of rent-control housing vouchers in the economy.
If less than 40 percent of the rent-control housing vouchers is in the pos-
session of high-skilled workers (α H < 0.4) the inflow of high-skilled workers
is smaller compared to the baseline model. In this situation, the gap between
the inflow of high-skilled workers under a situation with rent-control and un-
der the baseline situation is increasing in the number of rent-control housing
vouchers in the economy.
The right panel of Figure 6.2 shows the outflow of low-skilled workers
after the productivity shock in b for different parameter values of α H and γ.
For all values of α H and γ the outflow is larger than or equal to zero.14 The
outflow of low-skilled workers is larger than the baseline model if more than
60 percent of the housing vouchers are allocated to high-skilled workers (α H >
0.6). This excess outflow is increasing in the number of rent-control housing
vouchers in the economy, as indicated by the upward sloping arrow in the
right panel.
In the case that less than 60 percent of the housing vouchers is in the pos-
session of high-skilled workers (α H < 0.6), the outflow of low-skilled workers
is smaller compared to the baseline situation. Again, the difference with the
13 Write σ = sN and N − N H = lN. Then equate the multiplier for low-skilled workers if
H
0 ≤ α H ≤ 1 and γ > 0 to the multiplier if γ = 0. Cross-multiplication gives: α H = s+s+1− l σ+ N
1 = σ+ N
14 The outflow equals zero if all low-skilled workers in b possess a rent-control housing

voucher.
6.5. Wage rigidities and labor relocation 169

baseline situation in which there are no rent-control housing vouchers in the

economy is increasing in the number of rent-control housing vouchers, as the
downward sloping arrow indicates.

Figure 6.2: Inflow and outflow of workers for intermediate values of α

Upward sloping arrow: alpha>0.4 (left panel) or alpha>0.6 (right panel): Inflow high-skilled larger than base-
line
Downward sloping arrow, α H <0.4 (left panel) or α H <0.6 (right panel): Inflow high-skilled smaller than baseline
The direction of the arrows indicate rising absolute values for the gap with the baseline model. The graph
shows the situation where I have set σ = 500/20 = 25, population size N = 100 and number of high-skilled
workers N H = 50.

Figure 6.2 illustrates that the effect of increasing the rent-controlled hous-
ing stock γ on local skill composition depends on the allocation of rent-control
housing vouchers among high and low-skilled workers. It shows that if rent-
control housing vouchers are mainly distributed among low-skilled workers
(such that α H is smaller than the value of the special cases), increasing the
rent-control housing stock will reduce the inflow of high-skilled workers into
b after a positive productivity shock to high-skilled labor. Similarly, it will re-
duce the outflow of low-skilled workers out of b after the positive productivity
shock to high-skilled labor in b.

6.5 Wage rigidities and labor relocation

Wage rigidities influence the relocation of high-skilled and low-skilled work-
ers as they constitute the term ∇ EbH in equations (6.6) and (6.7). Under rigid
wages, ∇ EbH is given by equation (6.8). However, under flexible wages ∇ EbH
is determined differently.
170 Chapter 6.

If wages can vary locally, the surplus-maximizing wage rate in c is equal

to

w̃cT = ycT − β(ycT − z)

Use this to write the surplus maximizing unemployment rate for workers of
type T in c as

1  δ h iδ
ũcT = ãcT = µ β(ycT − z)
µ
with h (r + q )
1 + ãcT

Under flexible wages the productivity shock to high-skilled workers in b only

affects the local wage rate of high-skilled workers in b. The unemployment
rate of high-skilled workers in b decreases, but the unemployment rate of
high-skilled workers in a and the unemployment rates of low-skilled work-
ers do not change. The change in the local unemployment rate in b after a
productivity shock to high-skilled workers in b equals:

∂ũcT − f (ũcT )ζδ if c = c0 and T = T0

0 = (6.9)
∂ycT0 0

otherwise
 1−1/δ  1+1/δ
f (ũcT ) = 1 − ũcT ucT ζ = µ1/δ h(r+q)
µ
,

Define ẼBH as the relative employment conditions of high-skilled workers in b.

Next, denote as ∇ ẼbH the change in relative employment conditions of high-
skilled workers due to the productivity shock of high-skilled workers in b.15

ẼbH = (1 − ũbH )(w̃bH − z) − (1 − ũ aH )(w̃aH − z)

d ẼbH ∂(1−ũbH ) ∂(w̃bH −z)
∇ Ẽ H = dybH
= ∂ybH
(w̃bH − z) + ∂ybH
(1 − ubH )
   
= f (ũbH )ζδ w̃bH − z + (1 − β) 1 − ubH (6.10)

Let the number of workers of skill type T under flexible wage determination
be given by ÑbT . Then it follows that the effect of an increase of the productiv-
ity of high-skilled workers in b equal to ∆ changes the number of high-skilled
acT
15 Where I used (1 − ucT ) = and w̃cT − z = (1 − β)(ycT − z).
1+ acT
6.5. Wage rigidities and labor relocation 171

and low-skilled workers with

∂ ÑbH σ + N − N H −(1−α H )γN N H H

∂ybH
∆ = σ +(1−γ) N 2s ∇ Ẽb ∆ (6.11)
∂ ÑbL H H
= − N − Nσ+(−( 1−α )γN N HH
∂ybH
∆ 1− γ ) N 2s ∇ Ẽb ∆ (6.12)

These are equal to the expression found under rigid wages, except for the
replacement of ∇ EbH with ∇ ẼbH . Nc1
T equals Ñ T if it is assumed that the initial
c1
local productivity levels in a and b equal the reference productivity level. This
allows to write the difference in the number of workers of skill type T in city
b under flexible and rigid wages as:
 
∂ ÑbH ∂NbH σ + N − N H −(1−α H )γN N H
 
∂ybH
− ∂ybH
∆= σ +(1−γ) N 2s ∆ ∇ ẼbH − ∇ EbH
 
∂ ÑbL ∂NbL N − N H −(1−α H )γN N H
 
∂ybH
− ∂ybH
∆=− σ +(1−γ) N 2s ∆ ∇ ẼbH − ∇ EbH

Thus, whether wage rigidities reduce or increase the adaptation in the num-
ber of high-skilled workers depends on the sign of ∇ ẼbH − ∇ EbH . Equation
(6.13) shows that wage rigidities always reduce the response of city size and the
number of high-skilled workers in the city if the reference rate is not adopted
in response to the productivity shock or if the initial unemployment rates in a
and b are equal.
  
Hr ∂y Hr
∇ ẼbH − ∇ EbH = (1 − β) 1 − ∂y ∂ybH a
u H
+ ( 1 − ∂ybH
) u H
b
 
Hr f (u aH )
 
H H
− (1 − β) ∂y
∂y H
1 − f (u H )
ζδ f ( u b ) w − z (6.13)
b b

∂y Hr
 
∇ ẼbH − ∇ EbH = (1 − β) 1 − ubH ≥ 0 if u aH = ubH or ∂ybH
=0

Consider the situation in which the reference productivity level is not altered
∂y Hr
such that ∂ybH
= 0. Under rigid wages, the wage rate for high-skilled workers
does not change, and therefore the unemployment level in a is not affected
by the productivity shock in b. However, the unemployment rate of high-
skilled workers in b falls as demand for high-skilled workers goes up with
their productivity. Therefore region b becomes more attractive for high-skilled
172 Chapter 6.

workers and workers move to b. However, the inflow of high-skilled workers

is smaller compared to the situation with flexible wages as the wage rate is
determined by the reference productivity level (the initial productivity level
in b) and does not reflect the increase in productivity of high-skilled workers in
b. This shows the increase in employment conditions for high-skilled workers
is smaller under rigid wages. Also, wage rigidities reduce the response in the
number of workers if the initial unemployment values in a and b are equal.
This holds even if the reference productivity level is adjusted.
On the one hand, wage rigidities reduce mobility of high-skilled workers
as they do not have to move to b to benefit from a wage increase. On the
other hand, the wages in a are too high with respect to its productivity level,
which increases the unemployment level in a over the optimal unemployment
level. Note that the unemployment level in a depends on the productivity in
a and the reference productivity level, whereas the wage rate depends only
on the reference productivity level. Therefore, if the reference productivity
level changes with the productivity in b and the productivity levels in a and
b are equal, the increase in u aH is smaller than the increase in wages and wage
rigidities lead to less mobility.
However, if the local unemployment levels are not equal and δ is large, the
change in adjustment of labor might be larger under rigid wages. Rigid wages
increase the change in city size and skill-composition if equation (6.14) holds.16
The colored areas in Figure 6.3 show for which values of the unemployment
rate in a and b this condition is met. They are drawn for several values of
the vacancy intensity of the matching process (recall that δ > 1 if the vacancy
intensity parameter of matching d > .5) and several values for the rescaled
wages ζ w H − z under the assumption that ∂y Hr /∂ybH = 1.

   
∂y Hr H ∂y Hr
1− u
∂ybH a
+ 1− ∂ybH
ubH  
< ζδ w H − z (6.14)
∂y Hr
f (ubH ) − f (u aH )


∂ybH

16 Note that equation (6.14) is not defined if the local unemployment levels for high-skilled

workers are equal or if the reference productivity level for high-skilled workers is not changed
after the productivity shock.
6.5. Wage rigidities and labor relocation 173

Figure 6.3: Increasing labor relocation under wage rigidities

δ = 1/3 δ=1 δ=3

ζ (w H − z) = 1

ζ (w H − z) = 5

ζ (w H − z) = 10

Colored: range of values for u aH and ubH for which wage rigidities increase labor relocation after a positive
∂y Hr
shock to productivity of high-skilled workers in b (∇ũ H − ∇u H < 0). Outcomes based on = 1 and
∂ybH
on different values of δ = d/(1 − d) and ζ (w H − z ).

From Figure 6.3 I conclude that in general ∇ Ẽ H − ∇ E H > 0. Thus, in

general, local nominal wage rigidities reduce the relocation of workers in the
economy. This implies that after a positive shock to labor productivity of high-
skilled workers in b, less high-skilled workers will move into city b compared
to a situation where wages are flexible. Likewise, the outflow of low-skilled
workers out of b is lower as well.
Figure 6.4 shows the adjustment of high-skilled and low-skilled workers in
response to the idiosyncratic productivity shock. If wages would be flexible
the change in the number of high-skilled workers is given by the upwards
sloping solid line, which becomes horizontal if all high-skilled workers are
living in b. The change in the number of high-skilled workers in b if wages
are rigid is shown using the upward sloping dashed line. Together, the lines
indicate that wage rigidities lower the increase in high-skilled workers after a
productivity shock of high-skilled workers in b.
Similarly, the solid decreasing line shows the change in number of low-
skilled workers in b under flexible wages. As more high-skilled workers en-
ter b, house prices in b increase and some low-skilled workers leave b. The
line runs horizontal if all low-skilled workers have left b. However, if wages
174 Chapter 6.

are rigid, less high-skilled workers enter b in general, and therefore less low-
skilled workers leave b, as is indicated with the dashed, less-steeply down-
ward sloping line.

Figure 6.4: Wage rigidities and the change in population in b

The graph shows the situation outlined above where the local amenities, the un-
employment rates of low-skilled labor and housing market elasticities are equal
across both cities. Furthermore, WbH /2s = 4, σ = 500/20 = 25, population size
N = 100 and number of high-skilled workers N H = 50. Positive values on the y-
axis represent the inflow of high-skilled workers into b after a shock to high-skilled
productivity in b (on the x-axis). Negative values on the y-axis indicate the outflow
of low-skilled workers out of b.

6.6 Rent control and wage rigidities combined

The previous analysis showed that rent-control housing vouchers reduce the
inflow of high-skilled workers and the outflow of low-skilled workers if the
vouchers are targeted towards low-skilled workers (see Figure 6.2). For wage
rigidities, I found that wage rigidities reduce the inflow of high-skilled work-
ers and the outflow of low-skilled workers after the increase of productivity
of high-skilled workers in b if unemployment levels are not too far apart (see
Figure 6.3).
As both partial effects point in the same direction and their combined effect
is a multiplicative function, housing market rent rigidities and labor market
6.6. Rent control and wage rigidities combined 175

wage rigidities enforce one another. Figure 6.5 shows this when the initial
productivity levels of high-skilled workers in a and b are equal, 25 percent
of the housing stock is rent-controlled and all rent-control vouchers are allo-
cated towards low-skilled workers. The top panel illustrates the change in
the population of high-skilled workers. Having flexible wages and 25 percent
housing vouchers in the economy allocated to low-skilled workers reduces the
inflow of high-skilled workers about as much as having rigid wages without
rent-control housing vouchers. Also, the gray dashed line shows that hav-
ing both rigid wages and rent-control housing vouchers reduces the inflow of
high-skilled workers the most.
The bottom panel shows the outflow of low-skilled workers after the in-
crease in productivity of high-skilled workers. The effects of rent-control hous-
ing vouchers and rigid wages point in the same direction, and they are similar
in size: In an economy with flexible wages and with 25 percent rent-controlled
housing vouchers allocated to low-skilled workers, the outflow of low-skilled
workers is smaller compared to an economy with rigid wages and no rent-
control housing vouchers. Again, in an economy with both rigid wages and
rent-control housing vouchers, the outflow of low-skilled workers is mitigated
the most.
Finally, Figure 6.6 illustrates a case in which housing and labor market
price rigidities mitigate one another. Here, labor market wage rigidities have
the effect of lowering the inflow of high-skilled into and the outflow of low-
skilled workers out of b: If wages are rigid a larger increase in productivity
is required for a certain change in the number of high-skilled or low-skilled
workers. However, if rent-control housing vouchers are allocated to high-
skilled workers, the negative effect of wage rigidities on labor relocation is
partly offset by increased labor relocation due to housing market rigidities.
Note that wage rigidities do still negatively affect labor relocation, as can
be seen by comparing the situation with and without rent-control housing
vouchers under flexible and rigid wages.
176 Chapter 6.

Figure 6.5: Housing and labor market rigidities combined (I)

All housing vouchers allocated to low skilled:

housing market rigidities amplify labor market rigidities

The graph shows the situation outlined above where the local amenities, the un-
employment rates of low-skilled labor and housing market elasticities are equal
across both cities. Furthermore, α H = 0 (such that all housing vouchers are
owned by low-skilled workers), WbH /2s = 4, σ = 500/20 = 25, population
size N = 100 and number of high-skilled workers N H = 50. Positive values
on the y-axis represent the inflow of high-skilled workers after a shock to high-
skilled productivity (on the x-axis). Negative values on the y-axis indicate the
outflow of low-skilled workers.
6.6. Rent control and wage rigidities combined 177

Figure 6.6: Housing and labor market rigidities combined (II)

All housing vouchers allocated to high-skilled:

housing market rigidities mitigate labor market rigidities

The graph shows the situation outlined above where the local amenities, the un-
employment rates of low-skilled labor and housing market elasticities are equal
across both cities. Furthermore, α H = 1 (such that all housing vouchers are
owned by high-skilled workers), WbH /2s = 4, σ = 500/20 = 25, population
size N = 100 and number of high-skilled workers N H = 50. Positive values
on the y-axis represent the inflow of high-skilled workers after a shock to high-
skilled productivity (on the x-axis). Negative values on the y-axis indicate the
outflow of low-skilled workers
178 Chapter 6.

6.7 Conclusion
I have presented a model based on Moretti (2011) that integrates both the
housing market and labor market in a general equilibrium framework. It has
several new features. First, it is based on the assumption that labor markets
are characterized by labor market search as developed by Mortensen and Pis-
sarides (Pissarides, 2000). As a result, the location decision of households not
only depends on the local wage rate, but also on the local unemployment rate.
This feature makes the model consistent with empirical observations such as
the wage curve (Blanchflower and Oswald, 1995).
The model considers the effect of price rigidities at both the housing and
the labor market on location behavior of workers. On the labor market, the ef-
fect of nominal wage rigidities are modeled. I show that wage rigidities reduce
the adjustment of the labor force with respect to local productivity changes if
the reference rate is not adjusted or both regions have equal initial unemploy-
ment levels. Thus, the increase in the number of high-skilled workers after a
productivity shock to high-skilled labor in b is lowered if wages are negotiated
at a national level without taking local productivity into account. As a result,
the outflow of low-skilled workers is smaller as well.
On the housing market, I introduce housing price rigidities in the form of
rent-control housing vouchers. I show that the sign of the response of high
and low-skilled workers to changes in local productivity depends on the allo-
cation of housing vouchers among high and low-skilled workers. The size of
the rent-controlled housing sector influences the size of the effect, but it cannot
alter the sign. If rent-control housing vouchers are mainly distributed among
low-skilled workers, the adjustment in the number of high-skilled workers af-
ter a productivity shock to high-skilled labor is lower. This occurs because
fewer low-skilled workers leave the city if they possess more rent-control
housing vouchers. As a result, housing prices remain high and the inflow
of high-skilled workers is low compared to the situation without rent control.
6.7. Conclusion 179

Finally, I consider the joint effect of housing market and labor market price
rigidities. I show that housing market rigidities and labor market rigidities
can be amplifying or mitigating, depending on the distribution of rent-control
housing vouchers. From our analysis one could also conclude that although
housing market and labor market rigidities influence the equilibrium alloca-
tion of high-skilled and low-skilled workers across the economy, an equilib-
rium still exists.

In the final chapter, I study the relationship between local housing and labor
markets in the Netherlands from an empirical perspective. The Dutch system
of rent regulation does not provide housing vouchers. Instead, some share
of the local housing stock is declared to be rent-controlled. These differences
in organization of the housing market prevent a one-to-one mapping of the
results in this chapter to the situation in the Netherlands. Nonetheless, the
model provides valuable insights on the interaction between local housing
and labor markets for the Netherlands as well.
The insight that workers in rent-controlled housing do not leave a city if
house prices increase still holds. In addition, given the correlation between
income and education and the fact that rent-controlled housing is mainly al-
located to low-income households, it is expected that rent-controlled housing
reduces the percentage of high-skilled workers in a city.
180 Chapter 6.

6.A Mathematical derivations

6.A.1 f (ucH ) for different values of ucT and δ

Figure 6.7: f (ucH ) for different values of ucT and δ

6.A.2 Derivation rVicTu

Solve rVicTe for VicTe

rVicTe = w T − pc + q(VicTu − VicTe ) + Ac + eic

w T − pc q Ac + eeic
VicTe = + VicTu +
r+q r+q r+q

Substitute into rVicTu and collect terms for ( Ac + eic ) and Vicu

rVicTu = z − pc + θcT m(θcT )(VicTe − VicTu ) + Ac + eic

 T
w − pc

Tu T T Ac + eic q Tu Tu
rVic = (z − pc ) + ( Ac + eic ) + θc m(θc ) + + V − Vic
r+q r+q r + q ic
θ T m(θcT )
 T
w − pc
  
1
rVicTu = (z − pc ) + ( Ac + eic ) 1 + c + θcT m(θcT ) − rVicTu
r+q r+q r+q
6.A. Mathematical derivations 181

Solve for rVicTu

r+q θcT m(θcT )

rVicTu = T T
(z − pc ) + ( Ac + eic ) + (w T − pc )
r + q + θc m(θc ) r + q + θcT m(θcT )

θcT m(θcT )
Define Φ(θcT ) = r +q+θcT m(θcT )
.

rVicTu = (1 − Φ(θcT ))(z − pc ) + Ac + eic + Φ(θcT )(w T − pc )

rVicTu = (z − pc ) + Φ(θcT )(w T − z) + Ac + eic

6.A.3 Derivation rVicTe

Solve rVicTu for VicTu

rVicTu = z − pc + θcT m(θcT )(VicTe − VicTu ) + Ac + eic

1 θcT m(θcT ) 1
VicTu = (z − pc ) + V Te + ( Ac + eic )
r + θcT m(θcT ) r + θcT m(θcT ) ic r + θcT m(θcT )

Substitute into rVicTe and solve for rVicTe

rVicTe = w T − pc + q(VicTu − VicTe ) + Ac + eic

 
q q q
rVicTe = w T + z − p c + 1 − rV Te +
r + θcT m(θcT ) r + θcT m(θcT ) r + θcT m(θcT ) ic
 
q
+ 1 ( Ac + eic )
r + θcT m(θcT )
r + θcT m(θcT ) q
rVicTe = wT + z − pc + ( Ac + eic )
r + q + θcT m(θcT ) r + q + θcT m(θcT )

Rearrange to obtain

r + θcT m(θcT )
rVicTe = (z − pc ) + (w T − z) + Ac + eic
r + q + θcT m(θcT )
 
rVicTe = (z − pc ) + r+q+θrT m(θ T ) + Φ(θcT ) (w T − z) + Ac + eic
c c

6.A.4 Workers located in b in market housing by type

r +q
To ease the notational burden define WaT − WbT = r +q+α Ta
(w T − z) − r+rq++qαT (w T −
b
z). The location decision of rent-controlled tenants does not depend on house
prices, as they pay ρ regardless of location. Therefore, the location decision
182 Chapter 6.

for each skill type is given by:

NbTC − NaTC
s = WaT − WbT + ( Ab − A a )
N TC
N TC h T i
NbTC = Wa − WbT + ( Ab − A a ) + s
2s
N HC   N LC   NC
NbHC + NbLC = WaH − WbH + WaL − WbL + [( Ab − A a ) + s]
2s 2s 2s

The location decision of high-skilled and low-skilled workers in the market

segment is given by:

NbTM − NaTM
s = WaT − WbT + ( p a − pb ) + ( Ab − A a )
N TM

House prices in c are given by pc = ρ + k c Nc . Substitute in pc . Because house

prices are determined by demand for location, the share of workers of skill
type T in b is a function of total workers in b. Thus I have to solve for Nb .
Workers in b can be of the high or low skill type, and both can rent market or
controlled housing: Nb = NbHC + NbLC + NbHM + NbLM .

NbTM
2s = WaT − WbT + k a N − (k a + k b ) Nb + ( Ab − A a ) + s
N TM

Plug this in into the location decision to derive

  
NbTM = φ TM 1 − σ−1 N TC WaT − WbT
0 0 0
 
− φ TM σ−1 N T C WaT − WbT

+ φ TM (1 − σ−1 N C ) (( Ab − A a ) + s) + φ TM k a N
0
− φ TM (k a + k b ) NbT M
N TM
with φ TM = , σ= 2s
k a +k b
2s + (k a + k b ) N TM

0
Fill in the definition of N T M into the definition for N TM .
σ(k a +k b )
Use 1 − (k a + k b )φ TM = 2s+(k +


k ) N TM
to express al parts as functions of
a b
6.A. Mathematical derivations 183

0M 0M
φ TM φ T /N T

0
φ TM φ T M
h i 
NbTM = 0 2s + (k a + k b )( N M − N T ) WaT − WbT
NT M
TM T 0 M 0 0
 
− φ Tφ0 M (k a + k b )( N − N T ) WaT − WbT
N
TM T0 M TM T 0 M
 
+ φ Tφ0 M (k a + k b ) σ − N C [( Ab − A a ) + s] + φ Tφ0 M σ(k a + k b )k a N
N N
T0 M
+ φ TM (k a + k b )φ (k a + k b ) NbTM

0 0
φ TM φ T M /N T M N TM 1
Note that 0 conveniently reduces to 2s+(k a +k b ) N M 2s
. From
1−(k a +k b )φ TM (k a +k b )φ T M
here, it is straightforward to solve for NbHM and NbLM .

h i 
NbTM = N TM µ 1 + σ−1 ( N M − N T ) WaT − WbT
0 0
 
− N TM µσ−1 ( N − N T ) WaT − WbT
 
+ N TM µ 1 − σ−1 N C [( Ab − A a ) + s] + N TM µk a N
1
µ= 2s+(k a +k b )(1−γ) N

Furthermore, I define N C = γN, where 0 ≤ γ ≤ 1, to express the hous-

ing stock that is rent-controlled. The final step is to replace WaT − WbT =

u Ta − ubT W T − z by imposing r + q = 1.

h  i   
NbHM = N HM µ + µσ−1 (1 − γ) N − N H u aH − ubH wbH − z
  
− N HM µσ−1 ( N − N H ) u aL − ubL wbL − z
 
+ N HM µ 1 − σ−1 γN [( Ab − A a ) + s] + N HM µk a N (6.15)

h  i   
NbLM = N LM µ + µσ−1 (1 − γ) N − ( N − N H ) u aL − ubL wbL − z
  
− N LM µσ−1 ( N H ) u aH − ubH wbH − z
 
+ N LM µ 1 − σ−1 γN [( Ab − A a ) + s] + N LM µk a N (6.16)
184 Chapter 6.

Recall that number of rent-controlled high- and low-skilled workers in b are

given by:

N HC  H  N HC
NbHC = u a − ubH (w H − z) + [( Ab − A a ) + s] (6.17)
2s 2s
N LC  L  N LC
NbLC = u a − ubL (w L − z) + [( Ab − A a ) + s] (6.18)
2s 2s

These can be used to compute the total number of high and low-skilled work-
ers in b. Note that µ(k a + k b )(σ + ( N − N C ) = 1. Use N HC = αγN to denote
the number of high-skilled workers on rent control, with 0 ≤ α ≤ 1 such that
αγN H ≤ N H .

NbH = NbHC + NbHM

NH N H − αγN
  H
H N  
Nb = + 1− u aH − ubH W H − z
2 σ + (1 − γ) N 2s
H
N − αγN N − N H  L  
− u a − ubL W L − z
σ + (1 − γ ) N 2s
 H
N − αγN

1
+ NH − N ( A − Aa )
σ + (1 − γ) N 2s b
N H − αγN
 
ka 1
+N − (6.19)
σ + (1 − γ ) N k a + k b 2

NbL = NbLC + NbLM

N − NH (1 − γ) N − ( N H − αγN ) N − N H  L
   
NbL = + 1− u a − ubL W L − z
2 σ + (1 − γ ) N 2s
H
(1 − γ) N − ( N − αγN ) N H    
− u aH − ubH W H − z
σ + (1 − γ ) N 2s
(1 − γ) N − ( N H − αγN ) 1
 
H
+ (N − N ) − N ( A − Aa )
σ + (1 − γ ) N 2s b
(1 − γ) N − ( N H − αγN )
 
ka 1
+N − (6.20)
σ + (1 − γ ) N k a + kb 2
6.A. Mathematical derivations 185

and therefore the number of workers in b is given by:

Nb = NbH + NbL =
N σ NH  H  
Nb = + u a − ubH W H − z
2 σ + (1 − γ) N 2s
σ N − NH  L  
u a − ubL W L − z
σ + (1 − γ ) N 2s
 
σ N
+ ( A − Aa )
σ + (1 − γ) N 2s b
(1 − γ ) N
 
ka 1
+ − N (6.21)
σ + (1 − γ ) N k a + k b 2

6.A.5 Effect productivity of high-skilled workers on their un-

employment rates

The first full derivative of ubH towards ybH equals:

dubH ∂ubH ∂abH ∂ubH ∂abH

= +
dybH ∂abH ∂ybH ∂abH ∂y Hr
!
∂ubH ∂abH ∂abH
= + Hr < 0
∂abH ∂ybH ∂y

with
∂ubH
= −(1 − abH )−2 < 0
∂abH
!
∂abH ∂abH ∂a H
+ Hr = β bH > 0
∂ybH ∂y ∂yb
∂abH h  i 2d−1
1− d µ
d
ycH − (1 − β)y Hr − βz
µ
= 1− d µ h h >0
∂ybH
186 Chapter 6.

The first full derivative of u aH towards ybH equals:

du aH ∂u aH ∂a aH ∂u aH ∂a aH
H
= H H
+ H
dyb ∂a a ∂y a ∂a a ∂y Hr
∂u aH ∂a aH
= >0
∂a aH ∂y Hr
with
∂u aH
= −(1 − abH )−2 < 0
∂a aH
∂a Ha h  i 2d−1
µ d µ H Hr 1− d
= −( 1 − β ) h 1− d µ h y a − ( 1 − β ) y − βz <0
∂y Hr
Chapter 7

Rent control and local skill

1
composition

7.1 Introduction
As rent control prevents the adjustment of housing costs for controlled (and
typically low-skilled) tenants, the growth of cities and their skill composition
could be affected by rent control. This effect on local labor supply might
reduce agglomeration externalities that find their origin in the clustering of
high-skilled workers.2 If so, this provides a new channel by which rent con-
trol distorts outcomes on the local labor market, besides the effect on local
unemployment (Svarer et al., 2005). Additionally, as the main goal of rent con-
trol is to provide affordable housing to workers who are typically low-skilled,
knowledge whether rent control improves the access of low-skilled workers is
informative on its efficacy.3
1 This chapter uses data on the sectoral production and labor force for forty NUTS 3 regions

in the Netherlands over the years 1981-2006. This data has been collected by TNO. I am grateful
to TNO for sharing their data with me.
2 Moretti (2004) presents evidence that clustering of high-skilled workers increases nominal

wages of all skill types. For the Netherlands, De Groot et al. (2013) find that the wage elasticity to
the percentage of high-skilled workers equals 2.9-5 percent (depending on skill-type).
3 Rent control is widespread especially in Europe (Scanlon and Whitehead, 2007, 2008). World-

wide about forty percent of the population lives in rental housing being subject to some controls
(Malpezzi, 1993). Scanlon and Whitehead (2007) mention there is a general public concern in Eu-
rope that without social housing low-income household cannot afford to live in expensive cities.
188 Chapter 7.

Moretti (2011, 2013) has shown how local skill composition is driven by
the local wage rate net of local housing costs. Rent-controlled social hous-
ing distorts this mechanism, as it changes the local housing costs. Under the
assumption that social housing is allocated mainly to low-skilled workers, it
enhances the entrance of low-skilled workers after a generic increase in pro-
ductivity and prevents the outflow of low-skilled workers after a high-skilled
specific productivity increase. Thus rent control is expected to negatively in-
fluence the share of high-skilled workers in a region.
However, the effect of rent control on local skill composition might be
negligible from an economic perspective. This might occur for instance, if
rent-controlled housing is poorly targeted towards low-skilled workers. In
the extreme, if the probability to rent controlled or uncontrolled housing does
not differ between high-skilled and low-skilled workers, increasing the rent-
controlled housing stock will not alter local skill composition.
The effect of rent control on skill composition is analyzed using panel data
on fourty NUTS 3 regions in the Netherlands (in Dutch: COROPs). An advan-
tage of the data is that all NUTS 3 regions in the Netherlands have a substan-
tial share of housing that is rent-controlled. Skill composition, skill premium,
house prices and rent-controlled housing stock for each region have been mea-
sured using the national housing surveys in the years 1981, 1985, 1989, 1993,
1998, 2002 and 2006 (the so-called WBO and WoON surveys). This data is aug-
mented with information on productivity and employment for eight sectors
in these years at the NUTS 3 level.
Of these variables, the local wage ratio and local house prices might be
correlated with changes in unobserved local amenities. This creates an endo-
geneity problem. Using the data on sectoral productivity and employment
and using data on historic housing density these endogenous variables can be
instrumented.
Evidence is found that rent control reduces the percentage of high-skilled
workers in a region. Ceteris paribus a ten percentage point increase of the
rent-controlled housing stock is found to reduce the percentage of high-skilled
workers in a region with 1.5 percentage points. This is equivalent to an in-
crease in house prices with about 8,800 euro.
7.2. Theoretical model 189

This chapter contributes to two branches of the economic literature. First,

it contributes to economic research on the interaction between local housing
markets and local labor markets. Housing markets can create mobility con-
straints that might influence unemployment. Svarer et al. (2005) find that
rent-controlled tenants are more likely to become unemployed. Our contribu-
tion is that the framework of Moretti (2011, 2013) is augmented with a (non-
competitive) rent-controlled housing sector and derive the implications for
local skill composition. This highlights a new labor market distortion created
by rent control.
Second, our research contributes to the literature on economic segregation.
The way rent control affects social segregation depends on local market con-
ditions, see (Glaeser, 2003; Sims, 2011; Öst et al., 2013). In this chapter the
effect of rent control on local composition is studied for an entire country over
a long time-period of 25 years, while controlling for labor demand and local
consumer amenities.
The remainder of this chapter is organized as follows. Section 7.2 presents
our theoretical model how rent-controlled housing might affect local skill com-
position. A description of the data is presented in section 7.3. The estimation
strategy is presented in section 7.4, section 7.5 gives the estimation results.
Conclusions and implications for further research are presented in section 7.6.

7.2 Theoretical model

To see how rent control influences the skill composition in a general equilib-
rium setting, the model by Moretti (2011) is augmented with housing vouch-
ers that provide access to rent-controlled housing. Consider an economy with
two cities a and b and high-skilled workers N H and low-skilled workers N L .
Log indirect utility of workers in city c = a, b is a function of local wages w,
local housing costs pic (that might differ across workers), the pecuniary valu-
ation of local amenities Ac and a worker specific preference for city c.

T
uic = wcT − pic + AcT + eic
T
, T = H, L (7.1)
190 Chapter 7.

For simplicity, it is assumed that high and low-skilled workers work in sep-
arated firms.4 Assume that production is Cobb-Douglas. The wage workers
earn equals the marginal productivity of labor. Assume the supply of capital
is infinitely elastic and capital is applied such that the marginal productivity
of capital equals the rent r. Thus the production function and the marginal
returns to labor equal:5

ycT = XcT + hNcT + (1 − h)KcT

wcT = XcT − (1 − h) NcT + (1 − h)KcT + h

Housing is rented out to workers by profit-maximizing landlords with deep

pockets who reside outside the economy. Housing supply equals the number
of workers in the city and is an increasing function of the price:

pc = ρ + k c ( NcH + NcLm + NcLv )

All housing is rented out against market rents pc , however N v workers receive
a housing voucher which enables them to rent housing at the controlled rent ρ
in any city. Assume housing vouchers are randomly distributed among low-
skilled workers only.6 Thus all N H high-skilled workers and some N Lm low-
skilled workers rent against the market rent. N Lv low-skilled workers rent
against the controlled rent.
Local amenities are valued differently by low-skilled and high-skilled work-
ers. Finally, it is assumed that the relative worker specific preference for city
a over city b is i.i.d. uniformly distributed on the domain −s, s (regardless of
skill).

Equilibrium

Workers locate in city b if uib > uia , which depends on the relative local
wage rate, the relative presence of local amenities and on relative worker spe-
4 Thus, substitution between skill types in production is assumed away. This simplifies the

model, but it is not crucial for its outcome.

5 Similarly, in equilibrium the marginal product of capital equals marginal costs r = X T +
c
hNcT − hKcT + (1 − h).
6 Chapter 5 presents a similar model that shows an extended version of this model in which

rent-control housing vouchers can be allocated among high-skilled workers.

7.2. Theoretical model 191

cific preference. Location depends on local house prices only if the worker
does not have a housing voucher. The larger the difference between the skill-
specific wage net of housing costs and local amenities in the cities, the more
unequal workers of a certain skill will be distributed. The economy contains
three marginal workers who are indifferent between living in city a or b: one
high-skilled worker, one low-skilled worker without a housing voucher and
T − e T is uniformly dis-
one low-skilled worker with a housing voucher. As eia ib
tributed on the domain [−s, s] it is possible to solve for the number of people
of type T in both cities using:7

N Lv h L i
NcLv = (wc − wcL0 ) + ( AcL − AcL0 ) + s (7.2a)
2s
N Lm h L i
NcLm = (wc − wcL0 ) − ( pc − pc0 ) + ( AcL − AcL0 ) + s (7.2b)
2s
NH h H i
NcH = (wc − wcH0 ) − ( pc − pc0 ) + ( AcH − AcH0 ) + s (7.2c)
2s

Solving for wcL and wcH yields labor supply of low-skilled and high-skilled
workers. Equilibrium at the labor market requires that each worker earns
his marginal productivity. Note that the wage rate does not depend on the
local supply of the labor force as capital is attracted such that labor earns its
marginal product.

Labor supply

wcH =wcH0 + 2s
NH
NcH + ( pc − pc0 ) − ( AcH − AcH0 ) − s (7.3)
N Lm
wcL = N Lm2s N L + wcL0 − ( AcL − AcL0 ) − s +
+ N Lv c N Lm + N Lv
( pc − p c0 ) (7.4)

Labor demand
1− h
wcT = 1h XcT + h [(1 − h) − r ] + h (7.5)

Housing supply is a positive function of the housing price. Equilibrium at

the housing market is reached if local housing demand equals local housing
supply. Sum equations (7.2a) to (7.2c) and rewrite as a function of pc to obtain
7 See section 7.A for its derivation. It is assumed that s is large enough such that there are al-

ways workers with a housing voucher living in a and b. Think of this condition as resembling the
“no black hole condition” in New Economic Geography, as it prevents that workers of a particular
skill concentrate in one region (see Fujita et al. 1999).
192 Chapter 7.

inverse housing demand:

housing supply

pc =ρ + k c ( NcH + NcLm + NcLv ) (7.6)

housing demand
   
N Lv N Lv + N Lm
pc = pc0 − 2s
N H + N Lm
Nc + 1 + N H + N Lm
s+ N H + N Lm
wcL − wcL0 + AcL − AcL0
 
NH
+ N H + N Lm
wcH − wcH0 + AcH − AcH0 (7.7)

Figures 7.1 and 7.2 illustrate the equilibrium at the housing and labor market.
If c is the more desirable place to live in for low-skilled workers (wcL + AcL >
wcL0 + AcL0 ), then an increase in the number of housing vouchers in the economy
decreases the number of high-skilled workers and increases the number of low
skilled workers in c.

Figure 7.1: Equilibrium at the housing market

pc 6
@
@ Housing market in c
@
@
@
@
@ S equation (7.6)
@
@
@ @
p∗c 2 @ @
@ @
@ @ D2
@
p∗c
@
@
@
@ D equation (7.7)

-
Nc∗ Nc∗2 Nc
O
Assume that initially N Lv = 0 and equilibrium at the housing market in c is
established at the intersection of D H and S H in the point ( Nc∗ , p∗c ). Then
low-skilled workers with rent-control housing vouchers are introduced (N Lm
does not change). If (wcL + AcL ) > (wcL0 + AcL0 ) this shifts the demand curve
upwards and equilibrium is restored at the intersection of D2 and S in
( Nc∗2 , p∗c 2 ). Thus introducing rent-controlled workers increases the workforce
and house prices in c, and decreases them in c0 .
7.2. Theoretical model 193

Figure 7.2: Equilibrium at the labor markets

wcH 6
@
@ High-skilled labor market in c
@
@
@ @
D H2
@ @
@ @
@ @
@ @
@ @ S H equation (7.3)
@
wcH ∗
@
@ @
@ @
@ @
@ @
S H2 D H equation (7.5)

-
NcH ∗2 NcH ∗ NcH
O

wcL 6
@
@ Low-skilled labor market in c
@
@ D L2
@ @
@ @
@ @
@ @ S L equation (7.4)
@ @
@ @
@
wcL∗
@
@ @

@ @
@@

 D L equation (7.5)

S L2





-
NcL∗ NcL∗2 NcL
O
Initially equilibrium at the labor market for high-skilled workers is reached at
( NcH ∗ , wcH ∗ ). The introduction of low-skilled workers with rent-control
housing vouchers increases prices in c and decreases them in c0 (see Figure 7.1).
This decreases supply of high-skilled workers in c. As supply of capital K is
0
infinitely elastic, capital is increased to K and the new equilibrium at the
high-skilled labor market is at ( N H ∗ 2 , wc ∗ ), with N H ∗2 < N H ∗ .
H

At the low-skilled labor market the equilibrium shifts from ( NcL∗ , wcL∗ ) to
( NcL∗2 , wcL∗ ), with NcL∗2 > NcL∗ . The introduction of low-skilled workers with
rent-controlled housing vouchers decreases the slope and decreases the final
term of the labor supply function (note that N Lv , pc and pc2 change). As a
result firms using low-skilled labor increase the level of capital such that the
labor demand curve shifts to the left.
194 Chapter 7.

The relative wage differential for each skill type depends on the labor intensity
of production h and local general productivity parameter XcT , as in equation
(7.8). Relative housing costs for workers without a housing voucher are a
function of workers in b, all workers in the economy N and the parameters k a
and k b that reflect the price-elasticity of housing supply.

 
wbT − waT = 1h XbT − XaT (7.8)

pb − p a =(k a + k b )( NbH + NbLm + NbLv ) − k a N (7.9)

To solve the model, add equations (7.2a) to (7.2b). Next plug in equations
(7.8) and (7.9) and do the same for equation (7.2c). Express the number of
workers of each skill type with and without a housing voucher as a function
of exogenous parameters and NbL and NbH . Next substitute the equations for
low-skilled workers into that of high-skilled workers (or vice versa) to derive
the equilibrium number of high-skilled workers (equation 7.10).8 For com-
pleteness, the number of low-skilled workers in b is given as well.

  h   i
NbH = σ + N − N H − N Lv N H δ 1h XbH − XaH + ( AbH − A aH )
h   i
− ( N − N H ) N H δ 1h XbL − XaL + ( AbL − A aL )

+ (σ − N Lv ) N H δs + σN H δk a N (7.10)
1 1 2s
δ= , σ= k a +k b
2s σ + ( N − N Lv )

Similarly, NbL becomes:

h   i
NbL = −( N − N H − N Lv ) N H δ 1h XbH − XaH + ( AbH − A aH )
h   i
+ ( N − N H )(σ + N H )δ 1h XbL − XaL + ( AbL − A aL )
h i
+ δs σ( N − N H ) + N H N Lv + σ( N − N H − N Lv )δk a N

Equation (7.10) shows that the number of high-skilled workers in b depends

positively on the relative local productivity of high-skilled workers (XbH − XaH )
and the valuation of relative local amenities by high skilled workers (AbH −
8 See section 7.A for a step by step approach.
7.2. Theoretical model 195

A aH ). Similarly, it depends negatively on relative productivity and valuation

of relative amenities by low-skilled workers. A special case of this model
in which there are no housing vouchers distributed to low-skilled workers
(N Lv = 0). This yields the same equation as Moretti (2011).
Equation (7.10) shows that rent-control negatively influences the number
of high-skilled workers in the most attractive region in terms of productiv-
ity and amenities. Suppose that for both skill types, city b is the more at-
tractive city to live, such that for both skill types T = L, H it holds that
1 T T T T



h Xb − X a + Ab − A a > 0. The fraction that weights the relative valu-
ation of local productivity and amenities by high-skilled workers is positive
and decreasing (concave to the origin) in the number of rent-control vouch-
ers in the economy. Thus rent-control vouchers reduce the increase in high-skilled
workers in b after a positive demand shock for high-skilled workers.
Furthermore, the presence of high-skilled workers in city b depends nega-
tively on the relative valuation of wages and amenities by low-skilled work-
ers. This is because the inflow of low-skilled workers increases house prices
without changing the productivity of high-skilled workers. The fraction that
weights the relative valuation of local productivity and amenities by low-
skilled workers increases with the stock of housing vouchers. Thus rent-control
vouchers augment the increase in low-skilled workers in b after a positive demand
shock for low-skilled workers.
To sum up, in this chapter a general equilibrium model is described that
connects the labor and housing market when rent-control vouchers are al-
located among low skilled workers. This has two important effects on the
presence of low and high-skilled workers. First, as workers with a housing
voucher have equal housing costs in a and b, ceteris paribus rent control hous-
ing vouchers increase the number of low-skilled workers if the attractiveness
of a city to low-skilled workers increases. Second, the same mechanism en-
sures that rent control reduces the inflow of high-skilled workers after an in-
crease in attractiveness of city b to high-skilled workers.
196 Chapter 7.

Application to the Netherlands

The model describes how in relatively attractive cities rent control housing
vouchers reduce the number of high-skilled workers. It is based on several
simplifications. First, in the Netherlands housing vouchers are not allocated
to workers, but a percentage of the housing stock is rent-controlled. These
houses in turn are assigned to workers according to a set of allocation rules
that favors low-income households, who are typically low-skilled. Thus the
model does not reflect that low-income households in controlled housing who
move to another region are not entitled to rent-controlled housing. However,
low-income households do have a larger probability to be allocated controlled
housing.
A second simplification of the model is that it assumes that high-skilled
workers do not receive housing vouchers. If this would be modeled, the
effects of rent-controlled housing on the location decision of high and low-
skilled workers would become smaller without changing sign. This holds as
long as more housing vouchers are allocated to high-skilled workers than to
low-skilled workers.
Despite these simplifications, the model yields useful predictions. As the
majority of rent-controlled tenants is low-skilled and as the share of housing
that is rent-controlled is higher in dense areas, an average negative effect of
rent control on the percentage of the population that is high-skilled can be
expected.

7.3 Data
I test the influence of rent control on local skill composition using longitudi-
nal data for 40 NUTS 3 regions in the Netherlands. The nationwide surveys
WBO and WoON are used to construct measures for local wage premium, lo-
cal housing costs, the percentage of the population that is high-skilled and the
percentage of the population that lives in rent-controlled housing. The waves
conducted in the years 1981, 1985, 1989, 1993, 1998, 2002 and 2006 are avail-
able. All monetary amounts have been converted to 2006 euro using national
inflation figures.
7.3. Data 197

A worker is considered high-skilled if he has obtained a degree from uni-

versity, including universities of applied sciences (in Dutch: has a HBO or
WO-diploma). If not, the worker is considered low-skilled. Based on this vari-
able the percentage of the population that is high-skilled is computed.
Additionally, the survey contains information on the type of landlord if
the worker is renting housing. This allows to indicate whether the landlord is
a social housing organization and is used to construct our measure of social
housing at the regional level.
Next, local skill premia for high and low-skilled workers have been com-
puted based on (gross) income (in Dutch: bruto inkomen) of the households.
To compute the local wage premium in each sample year, income is regressed
on employment status (dummy equal to one if unemployed), household com-
position (single person household, couple with children, other type of house-
hold and couple without children as base household), the number of children
younger than six, twelve and fifteen in the household, and nine age category
dummies (all heads of households are at least aged twenty and the age of the
baseline head of household is between 30 and 39), a dummy-variable equal
to one if the head of household is female and 39 NUTS 3 region dummies
(relative to Zeeuws-Vlaanderen). This is done for high-skilled workers and
low-skilled workers separately. After estimation, the local skill-premium for
each skill type has been computed by adding the coefficients on each of NUTS
3 regional dummies to the constant.
The local house price premium is computed similarly based on the house
price for which homeowners expect to sell their house. For each year in the
sample, the house price is regressed on variables indicating the number of
rooms, the size of the living room, the size of the kitchen, the presence of a
garden (or similar attribute), the bathroom facilities in the house and 39 re-
gional dummies. The baseline house is located in Zeeuws-Vlaanderen, has
four rooms, a living room of 30-40 square metres, a kitchen of five to seven
square metres, and contains a shower and a garden. The local house price-
premium has been computed by adding the coefficients on each of the NUTS
3 regional dummies to the constant.
198 Chapter 7.

Next, the percentage of people that is high-skilled, the percentage of hous-

ing that is rent-controlled and local skill premium and local house prices are
merged with information on regional labor productivity and jobs for eight
sectors.9 For each region, there is information on the number of FTE in each
sector as well as total productivity. This allows to compute the Gross Regional
Product per worker at the sectoral level. Table 7.1 provides summary statistics
of these variables.

Table 7.1: Summary statistics

mean sd min max

Percentage of workers that is high-skilled 16.15 11.48 1.00 44.00
Perc. of housing stock that is rent-controlled 43.35 7.29 24.50 65.60
Local house prices (in e10,000) 10.58 3.39 4.78 25.63
Local high-skill premium (wcH /wcL ) 1.31 0.16 0.49 1.69
Jobs per km2 184.11 201.61 12.02 1249.61
Spatial lag of
Perc. of housing stock that is rent-controlled 42.67 3.95 32.20 52.55
Jobs per km2 177.36 87.69 27.18 481.11

log GDP per worker 10.77 0.34 10.01 11.82

Local productivity share sector 1 (B1 ) 1.49 0.55 0.65 3.67
Local productivity share sector 2 (B2 ) 0.10 0.08 0.00 0.30
Local productivity share sector 3 (B3 ) 1.87 0.35 0.76 2.75
Local productivity share sector 4 (B4 ) 0.23 0.23 0.00 1.39
Local productivity share sector 5 (B5 ) 3.29 0.93 1.03 5.66
Local productivity share sector 6 (B6 ) 1.66 0.43 0.73 3.28
Local productivity share sector 7 (B7 ) 1.50 0.45 0.82 3.23
Local productivity share sector 8 (B8 ) 0.63 0.23 0.23 1.63
Surface 103815.43 70324.12 12818.00 342884.00
Housing density in 1947 81.12 104.93 1.00 574.00
Housing density in 1947 × log GDP per worker 873.63 1135.28 10.29 6499.92

7.3.1 Clustering of high-skilled workers and rent control

Compared to other European countries, the share of housing that is rent-

controlled in the Netherlands is substantial, see Scanlon and Whitehead (2007).
Although the share of rent-controlled housing has been falling over time, about
9 The eight sectors are: banking (in Dutch: banken en zakelijke dienstverlening), real estate (in

Dutch: exploitatie onroerend goed), trade and catering (in Dutch: handel en horeca), agriculture and
fisheries (in Dutch: landbouw en visserij), industry (in Dutch: nijverheid), non profit and other ser-
vices (in Dutch: non-profit en overige dienstverlening), government (in Dutch: rijksoverheid, transport
and communications (in Dutch: transport en communicatie).
7.3. Data 199

one-third of the total housing stock was rent-controlled in 2008.10,11 However,

the share of housing that is owner-occupied is not so different from other
European countries. Thus the Dutch housing market is characterized by a
dual market with owner-occupied (market) housing and rent-controlled (so-
cial) housing. The market for uncontrolled rental housing exists, but is very
small (Scanlon and Whitehead, 2007).
Table 7.2 presents development of high-skilled workers and rent-controlled
housing over time. For each year the average and standard deviation is listed
for the NUTS 3 regions. The average percentage of high-skilled workers in
the population has increased from in 6.8 percent in 1981 to 32 percent in 2006.
The standard deviation over this time period has doubled. Note that between
1989 and 1998 the percentage of the population that is high-skilled has in-
creased from almost 9 to almost 21 percent. This increase is substantial, but
can be explained partly by two cohort effects.
First, the cohorts 1920-1930 and 1930-1940 are skewed towards low-skilled
workers. These workers are expected to retire around 1989 creating a rela-
tively large outflow of low-skilled workers. Second, attendance rates of uni-
versities and universities of applied science have increased exponentially over
the period 1950-1980. This has generated a relatively large inflow of high-
skilled workers. Thus a strong increase around 1989-1993 is expected. How-
ever, the data shows an exceptional sharp increase in the number of high-
skilled workers between 1993 (7.5 percent) and 1998 (20.9 percent). This seems
to be a development that is too large in a too short time-period to be explained
by the cohort effects alone. Therefore, fixed year effects are used to control for
the development of high-skilled workers over time.
The development of the average share of housing that is rent-controlled
is very different. It slightly decreases over time from 46 percent in 1981 to
below 39 percent in 2006, but the change is very gradual. Figure 7.3 shows
10 See Romijn and Besseling (2008); Donders et al. (2010); Regeer and Van Daalen (2011); Van

Daalen et al. (2012) for several recent indicators.

11 Rent-controlled housing distorts the housing market as controlled rents do not reflect the lo-

cal marginal price of housing. This leads to an inefficient allocation of controlled housing (Glaeser
and Luttmer, 2003). On the other hand, the size of the controlled housing stock is arguably not
very inefficient as the marginal willingness to pay for rent-controlled housing quality is close to
its marginal costs (Van Ommeren and Koopman, 2011; Van Ommeren and De Graaf-Zijl, 2013).
200 Chapter 7.

Table 7.2: Share of high-skilled workers and social housing over time

High-skilled Social
housing
mean sd mean sd
1981 6.80 2.58 45.99 7.34
1985 6.58 2.90 47.21 6.83
1989 8.95 3.90 45.32 6.73
1993 7.47 4.29 43.87 6.29
1998 20.90 6.14 42.32 6.27
2002 30.32 5.39 40.10 6.81
2006 32.00 5.15 38.65 6.93

the spatial distribution of high-skilled workers in 1981 and 2006. It shows that
in 1981 the spatial distribution of high-skilled workers over the country did
not follow a clear-cut pattern as regions with the most high-skilled workers
are found in the north, middle and south of the country. Also, regions with
the highest and lowest percentage of high-skilled workers are often located
near each other. However, in 2006 this has changed. High-skilled workers
are now concentrated in the west of the country. In the literature clustering
of high-skilled workers over time has been explained from the agglomeration
externalities and the creation of consumer amenities.
Consider the relationship between high-skilled workers and wages of high-
skilled workers. Figure 7.4 shows a positive relationship in favour of the hy-
pothesis that high-skilled workers have been clustering in regions with high
wages. Also, the clustering of high-skilled might be explained by consumer
amenities for which high-skilled workers have a higher willingness to pay.12
Consumer amenities are not directly observed, but as they are correlated with
house prices insight into this mechanism can be given by plotting the percent-
age of high-skilled workers against local house prices for a standard house. In
line with this, Figure 7.5 indicates that high-skilled workers have been cluster-
ing in regions with high house prices. Figure 7.6 shows the spatial distribution
of rent-controlled housing in 1981 and 2006. In 2006, the share of housing that
is rent-controlled is among the highest in the three largest largest cities Ams-
12 See Brueckner et al. (1999) for this idea applied to the monocentric city model.
7.3. Data 201

Figure 7.3: Share of high-skilled workers by region, 1981 and 2006

Numbers correspond to numbers in Table 7.B

202 Chapter 7.

Figure 7.4: High-skilled workers and local high-skilled wages

Relation between share of high-skilled workers in a region and the absolute wage
of high-skilled workers (measured in 2006 euros) in that region.

Figure 7.5: High-skilled workers and local house prices

Relation between share of high-skilled workers in a region and the local house
prices (measured in 10,000 euros in 2006) in that region.
7.3. Data 203

terdam (5), Rotterdam (21) and The Hague (19). Other regions where the share
of the housing stock that is rent-controlled is high are regions with ‘historic
cities’ like Maastricht (8) and Nijmegen (7).

Figure 7.6: Rent-controlled housing by region, 1981 and 2006

Numbers correspond to numbers in Table 7.B

204 Chapter 7.

Table 7.3 and Figure 7.7 provide rudimentary evidence on the relation-
ship between the share of the workforce that is high-skilled and percentage of
housing that is rent-controlled over time.
The data shows a positive correlation between the share of rent-controlled
housing and percentage of high-skilled workers at the cross-sectional level since
the 1990s. This might have to do with the fact that the share of rent-controlled
housing is higher in more urban regions and these regions are also more likely
to contain a university or university of applied sciences. In contrast, the overall
correlation between the share of high-skilled workers in a region and the share
of housing that is rent-controlled is found to be negative. The change in sign
of the correlation coefficients over time and in the cross-section highlights the
importance of studying the relationship between rent control and local skill
composition from a longitudinal panel perspective controlling for fixed region
effects.

Table 7.3: Correlation high-skilled workers and rent controlled housing

Correlation
coefficientsa
Overall -0.26***
1981 -0.16
1985 -0.07
1989 0.15
1993 0.39**
1998 0.28*
2002 0.46***
2006 0.43***
a Pairwise correlation coefficients are based on 280 observations for the overall
correlation and on 40 observations per year. Significance level denoted with
stars, ∗ p < 0.1, ∗ ∗ p < 0.05, ∗ ∗ ∗ p < 0.01

However, the negative relationship between rent-controlled housing and

the share of high-skilled workers might reflect a spurious relation if the lo-
cal stock of rent-controlled housing is adjusted according to local labor mar-
ket conditions that drive the location of high-skilled workers.13 Figure 7.8
suggests the rent-controlled housing stock is not adjusted in relation to labor
market density. It shows the long run changes in the share of housing that
13 The regression specification controls for this by including fixed region effect, fixed year ef-

fects and explanatory variables that measure local economic conditions.

7.3. Data 205

Figure 7.7: Rent control and high-skilled workers

Relation between share of high-skilled workers in a region and the share of the
housig stock that is social housing in 1981, 1993 and 2006.

is rent-controlled against long run changes in jobs per squared kilometer. No

clear-cut patterns arises. The observations are distributed over virtually the
entire interval of long run changes of rent-controlled housing stock.
206 Chapter 7.

Figure 7.8: Rent-controlled housing and labor market density

Relation between long term changes in the number of jobs per squared kilometre
and the long term change in the share of housing that is social housing.

7.4 Empirical strategy

Equation (7.11) is estimated to investigate the effect of rent control on the pres-
ence of high-skilled workers. The dependent variable is the percentage of the
population that is high-skilled in city c at time t, denoted Hct . Independent
variables are a region specific fixed effect αc , a set of time dummies to ac-
count for non-linear time effects δt . The share of housing that is rented out by
housing agencies Sct and the number of jobs per squared kilometer Jct are in-
cluded as well. Other independent variables are the ratio of local skill-premia
H /w L and local house prices p . A are unobserved local amenities
wct = wct ct ct ct

and ε ct is an idiosyncratic error term.

Hct = αc + β 1 Sct + β 2 wct + β 3 pct + β 4 Jct + δt + νct (7.11)

νct = Act + ε ct

c = 1, . . . , 40, t = 1981, 1985, 1989, 1993, 1998, 2002, 2006

7.4. Empirical strategy 207

However, local amenities Act might influence the high-skill premium wct as
households are willing to accept a lower wage if they live in attractive re-
gions. Local amenities are directly internalized in local house prices pct . Thus
local house prices and wages are endogenous variables that should be instru-
mented. Note that economic amenities like job density and the local wage rate
are included in the specification. The excluded measure for amenities Act
refers to consumer amenities like national parks, open space and so on. Local
house prices and local skill premium can then be instrumented with a variable
that is correlated with local house prices and the local skill premium, but that
is not correlated with the local consumer amenities.
I instrument the local wage premium with the local expected productivity
share Bcst . Bcst is defined using productivity of each of the eight sectors s in
year t, and multiply this measure with the local employment share of this
sector. Denoting employment in sector s in region c at time t with Ecat , and
with C being the number of regions, this measure is defined as:

∑c GDPcst Ecst
Bcst = C ∑s Ecst

Next, local house prices are instrumented with the number of houses per km2
in 1947, multiplied with local GDP.14 The idea is that in historically dense ar-
eas house construction costs are higher because the low-cost options for con-
struction of housing have already been executed. This measure is multiplied
with GDP as to acknowledge that high housing costs only reduce supply if
there is increased demand (see also Diamond (2011)). With S denoting the
total number of sectors in the city, this measure is defined as:

Housing stock 1947 ∑s GDPcst

H1947 =
km2 S
14 There is no official statistic on the housing stock in 1947 for COROP regions, as the regions

have been created in 1971 only. The housing stock in 1947 by COROP-region has therefore been
constructed based on several data sets provided by Statistics Netherlands and manually taking
into account ‘adjustments in municipal boundaries’ (in Dutch: gemeentelijke herindelingen) over the
period 1947-1988. Section 7.B lists the COROP regions and the housing stock in 1947.
208 Chapter 7.

Control for commuting and replacement

As the Netherlands is a small country, workers do not have to work in the

region of residence but can commute to nearby work-locations. Therefore, the
percentage of high-skilled workers in a region is expected to be influenced
by employment conditions in the home-region as well as nearby regions. To
control for commuting by workers measures of the number of jobs in nearby
regions are included.
If rent-control reduces the presence of high-skilled workers, one would ex-
pect that high-skilled workers who cannot enter a city because of its large rent-
controlled housing stock are more likely to settle in nearby cities. Thus the ef-
fect of rent control on the percentage of high-skilled workers, might spillover
to nearby regions as well. To derive consistent estimates, this cross-sectional
dependence should be explicitly modeled.
To control for commuting and spatial spillovers of the social housing stock,
the so-called spatial lag is computed by pre-multiplying the variable with
a spatial weight matrix W.15 Let Wct be the cth row of the spatial weight
matrix W at time t.16 Thus it is tested whether the share of housing that is
rent-controlled in neighboring regions affects the percentage of workers that
is high-skilled by including the spatial lag of rent-controlled housing stock
(Wct Sct ). Similarly, to consider the effect of jobs in neighboring regions on the
share of workers that is high-skilled, the spatial lag of jobs (Wct Jct ) is included.
Thus equation (7.11) is extended with these spatially lagged variables and es-
timate equation (7.12).17
15 W is a N × N matrix that has element w = 1 if j is an adjacent neighbor of i and zero
ij
elements otherwise (including the main diagonal). W is row-normalized such that all elements
on a row sum up to 1. Furthermore restriction on the nonzero elements wij are imposed such that
the spatial interactions fade out with distance. Restrictions are that the row and column sums
of W and ( I − ρW )−1 ,( I − λW )−1 should be bound in absolute value as N goes to infinity. See
Kelejian and Prucha (1998, 1999); Lee (2004).
16 The specification of W influences the parameter estimates on the spatial lags and in empirical

work different specifications of W are often tested. See (Harris et al., 2011) for a discussion on
formulating W. However, according to LeSage and Pace (2010), the effect of the specification of
W on the marginal effects of a (spatially lagged) variable is negligible.
17 I have also included a spatial lag of the independent variable, which was insignificant. Re-

sults available upon request.

7.4. Empirical strategy 209

Hct = αc + β 1 Sct + β 2 wct + β 3 pct + β 4 Jct + δt

+ γ1 Wct St + γ2 Wct Jt + νct (7.12)

νct = Act + ε ct

c = 1, . . . , 40, t = 1981, 1985, 1989, 1993, 1998, 2002, 2006

Measurement of spatial spillovers

It is important to realize that the estimated parameters on the spatially lagged

variable do not necessarily equal the marginal effects of the independent vari-
ables. Therefore, significant parameters are not sufficient to prove the exis-
tence of spatial spillovers (Elhorst, 2010; LeSage and Pace, 2009). Let y be
a vector containing the dependent variable. y is explained by the matrix of
independent variables X, a vector Wy containing the spatially lagged depen-
dent and matrix WX containing the spatially lagged independent variables. ε
is a vector containing an i.i.d. error term with expectation zero and constant
variance. ρ, β and θ are scalars that have to be estimated.

y = ρWy + Xβ + WXθ + ε (7.13)

The reduced form becomes

y = ( I − ρW )−1 X [ β + Wθ ] + ( I − ρW )−1 ε (7.14)

From which it follows that the derivative of y towards any variable xk in X

equals:18

   
∂y1 ∂y1 ∂y1
··· βk w1,2 θk · · · w1,40 θk
 ∂xk1 ∂xk2 ∂xkN   
 ∂y2 ∂y2 ∂y2 
··· w θ · · · w2,40 θk 
 
βk
−1  2,1 k
 ∂xk1 ∂xk2 ∂xkN 
 ..

.. .. .. 
 = ( I − ρW )  .. .. .. .. 

 . . . .   . . . . 
   
∂y N ∂y N ∂y N
∂xk1 ∂xk2 ··· ∂x w40,1 θk w40,2 θk ··· βk
kN

(7.15)

18 Where ∂y a is the derivative of the y in region a to a shock to xk in region b.

∂xkb
210 Chapter 7.

Of which the RHS can be written in matrix notation as ( I − ρW )−1 ( β k I + Wθk ).

As direct and indirect effects are different for each spatial unit, LeSage and
Pace (2009) suggest to compute the average (in)direct effect and consider its
distribution under the null of an (in)direct effect equal to zero. In matrix nota-
tion, the average direct and average indirect effect are given by:

Average direct effect : trace[( I − ρW )−1 Iβ]/N (7.16)

Average indirect effect : i0N [( I − ρW )−1 θW ]i N N ( N1−1) (7.17)

where in is a N-dimensional vector of ones. The direct effect and indirect

effect are divided by N and N ( N − 1) as there are N regions with each N − 1
potential neighbors.

7.5 Results
The model in equation (7.12) is estimated to test whether rent-controlled hous-
ing influences the percentage of high-skilled workers. Table 7.4 shows the re-
gression results. Column one presents the second stage regression, first stage
regression results are presented in columns two an three.
I find evidence at the five percent significance level in favor of a nega-
tive relationship between the share of rent-controlled housing and the share
of high-skilled workers. Ceteris paribus, a ten percentage point increase in the
share of rent-controlled housing reduces the percentage of high-skilled work-
ers with 1.5 percentage points.
I also find evidence that the share of high-skilled workers is larger in re-
gions characterized by high house prices. The coefficient on local house prices
indicates that ceteris paribus an increase in house prices with 10,000 euro in-
creases the share of high-skilled workers with 1.33 percentage points.
Estimates on both the share of rent-controlled housing and local house
prices are credible. A ten percentage point increase in the share of housing
that is rent-controlled has the same effect on the percentage of high-skilled
7.5. Results 211

Table 7.4: Results spatial LSDV

(1) (2) (3)

Percentage Wage House
high-skilled ratio prices
Share of social housing -0.1510** 0.0023 0.1080***
(0.0665) (0.0048) (0.0251)
Spatial lag share of social housing -0.1060 -0.0115 0.0178
(0.1210) (0.0100) (0.0479)
Jobs per squared kilometer 0.0112 0.0004 -0.0024
(0.0096) (0.0004) (0.0069)
Spatial lag jobs per squared kilometer 0.0546** 0.0035 -0.01990**
(0.0246) (0.0023) (0.0082)
Local house prices 1.3310***
(0.3110)
Local high-skill premium -1.2130
(4.4180)
Year 1985 3.8900*** 0.0322 -2.7850***
(0.9860) (0.0580) (0.2370)
Year 1989 4.0090*** -0.0227 -2.0170***
(0.7040) (0.0598) (0.2760)
Year 1993 0.2150 0.0586 -0.7250*
(0.7620) (0.0866) (0.4250)
Year 1998 8.3100*** -0.0573 1.4280**
(1.5010) (0.1470) (0.6890)
Year 2002 17.5500*** -0.0114 1.5000*
(1.6050) (0.1950) (0.8700)
Year 2006 15.1300*** -0.1930 4.2230***
(2.4070) (0.2220) (1.0200)
Local productivity share sector 1 0.1540* 1.8480***
(0.0911) (0.5700)
Local productivity share sector 2 -0.0507 -1.2400
(0.1510) (1.3900)
Local productivity share sector 3 0.1190 -0.5590
(0.1040) (0.7070)
Local productivity share sector 4 -0.2540 -2.0300**
(0.1610) (0.8170)
Local productivity share sector 5 0.1910** -0.9980*
(0.0846) (0.5890)
Local productivity share sector 6 0.1890 -1.9720***
(0.1530) (0.7330)
Local productivity share sector 7 0.2410* -1.9910***
(0.1230) (0.7440)
Local productivity share sector 8 0.1270 -1.6010**
(0.1120) (0.7420)
GDP and historic house density -<0.0001 0.0073*
(0.0003) (0.0038)
Observations 280 280 280
adj. R2 0.949 0.235 0.915
Degrees model 51 58 58
χ2 (F) value 2nd (1st) stage 10873.5 (5.8) (58.7)
Robust score χ2 statistic p-value 0.03
Robust regression F-statistic p-value 0.04
Overidentification χ2 statistic p-value 0.3
Robust standard errors in parentheses
* p < .10, ** p < .05, *** p < .01
Variables not shown: 39 region dummies, constant
212 Chapter 7.

workers as a decrease in local house prices with 8,800 euro.19 Thus the esti-
mated parameters on local house prices and rent-controlled housing (and also
their relative size) are plausible from an economic perspective.
The number of jobs per squared kilometer in neighboring regions is sig-
nificant at the five percent level. I do not find evidence that the the share of
high-skilled workers in a region is influenced by the share of housing that is
rent-controlled in neighboring regions, the local job density or the local high-
skilled wage premium.
As a robustness check the effect of rent-controlled housing on the share
of high-skilled workers is allowed to be nonlinear. Dummies have been gen-
erated for the five quintiles of Sct , and the first four quintiles are compared
with the quintile that has the highest share of rent-controlled housing. Re-
sults in column one of Table 7.5 indicate that the share of high-skilled workers
decreases with rent-controlled housing. The share of high-skilled workers in
the time-region quintile with the lowest share of rent-controlled housing is 2.6
percentage point higher than in the time-region with the most rent-controlled
housing, ceteris paribus. This is about 1.9 for the second and 1.8 for the third
quintile, whereas the fourth quintile is not significantly different from the fifth
quintile, ceteris paribus. This specification provides evidence that the result in
columns one and two is not driven by outliers as all categories are significant
at at least ten percent and they follow a consistent pattern. Under this speci-
fication, the parameter on house prices remains significant at the one percent
level whereas it parameters decreases slightly in absolute value. The signifi-
cance of the parameter on jobs in neighboring regions also slightly decreases,
but remains significant at the five percent level.

Spatial spillover of rent-controlled housing and jobs per km2

This subsection considers the spatial spillover effects using the parameter es-
timates of the first columns in Tables 7.4 and 7.5. One thousand realizations
indexed i = 1, . . . , 1000 for the relevant parameters β and θ have been gener-
19 The estimated coefficient on house prices (measured in e10,000) equals 1.331, while that on

percentage of housing that is rent-controlled equals −0.150. To compare both estimates compute
(1.331) (e10,000)/(-0.15*10) ≈ e8,800.
7.5. Results 213

Table 7.5: Results spatial LSDV (dummy specification)

(1) (2) (3)

Percentage Wage House
high-skilled ratio prices
First quintile share of social housing 2.6430** 0.0734 -1.3750***
(1.0560) (0.0701) (0.4990)
Second quintile share of social housing 1.9230** -0.0127 -0.8850**
(0.8290) (0.0525) (0.3760)
Third quintile share of social housing 1.8180** 0.0426 -0.5740
(0.7990) (0.0489) (0.3510)
Fourth quintile share of social housing 1.4510* -0.0177 -0.2580
(0.7490) (0.0405) (0.2790)
Spatial lag first quintile share of social housing 0.2710 0.1790 -0.2220
(2.4860) (0.1490) (1.0460)
Spatial lag second quintile share of social housing -0.7520 0.2440** 0.4090
(2.1440) (0.1130) (0.7300)
Spatial lag third quintile share of social housing 0.6660 0.1850* 0.6150
(2.1790) (0.1090) (0.7760)
Spatial lag fourth quintile share of social housing -1.7470 0.0772 1.0200*
(1.8890) (0.0885) (0.6050)
Jobs per squared kilometer 0.0136 0.0005 -0.0050
(0.0097) (0.0005) (0.0074)
Spatial lag jobs per squared kilometer 0.0495** 0.0037 -0.0278***
(0.0246) (0.0025) (0.0098)
Local house prices 1.2580***
(0.2760)
Local high-skill premium -0.1750
(3.7480)
Local productivity share sector 1 0.1870** 1.7000***
(0.0931) (0.6400)
Local productivity share sector 2 0.0074 -1.5680
(0.1540) (1.4180)
Local productivity share sector 3 0.1260 -0.9150
(0.0996) (0.7420)
Local productivity share sector 4 -0.2920* -2.3650***
(0.1570) (0.8760)
Local productivity share sector 5 0.2110*** -1.2580**
(0.0810) (0.6350)
Local productivity share sector 6 0.2060 -2.7500***
(0.1470) (0.8260)
Local productivity share sector 7 0.2470** -2.2390***
(0.1200) (0.8470)
Local productivity share sector 8 0.1440 -1.6390**
(0.1080) (0.8000)
GDP and historic house density 0.0002 0.0079*
(0.0004) (0.0042)
Observations 280 280 280
adj. R2 0.950 0.258 0.910
Degrees model 57 64 64
χ2 (F) value 2nd (1st) stage 5.6 51.1
Robust score χ2 statistic p-value 0.03
Robust regression F statistic p-value 0.05
Overidentification χ2 statistic p-value 0.4
Robust standard errors in parentheses
* p < .10, ** p < .05, *** p < .01
Variables not shown: 39 region dummies, 6 year dummies, constant
214 Chapter 7.

ated, where each realization β i , θi has been drawn from an independent ran-
\
b θb and standard deviation se
dom normal distribution with mean β, ( β i ), \
se(θi )
respectively. Next, the direct and indirect spillover effects have been com-
puted according to equations 7.16 and 7.17. Table 7.6 shows the result.
In the preferred specification in which the share of rent-controlled hous-
ing is treated as a continuous variable (column one of Table 7.4 ), I find that
a change in the number of rent-controlled housing with ten percentage points
reduces the number of high-skilled workers in the region with 1.56 percentage
points. However, the indirect spillover effect is insignificant as indicated by
the 95% confidence interval and therefore it is concluded that rent-controlled
housing does not influence the percentage of high-skilled workers in neigh-
boring regions on average. If one allows for a nonlinear effect of the share of
housing that is rent-controlled on the percentage of high-skilled workers, re-
gions with a smaller share of rent-controlled housing are found to have more
rent-controlled workers. However, also here no evidence in favor of spatial
spillover effects to neighboring regions is found.
Next, it is concluded that an increase in the number of jobs per squared
kilometer in a region does not influence the number of high-skilled workers
in a region, but has spatial spillover effects to neighboring regions. The direct
and combined effect are insignificant at the five percent significance level. The
average indirect spatial spillover effect equals 0.0014, which suggests that a
one standard deviation increase in the number of jobs increases the number of
high-skilled in neighboring region with about 0.30 percentage points.20 Using
dummies to allow the effect of the share of housing that is rent-controlled to
be nonlinear does not change conclusions on the spatial spillover effects of the
number of jobs per squared kilometer.

Spatial correlation in the error term

Up to now, spatial dependence of the error term has not been considered.
The robust standard errors developed by White (1980) and Newey and West
20 The standard deviation of the number of jobs in a region equals 201.61 and 201.61(0.0014) ≈
0.28.
7.5. Results 215

Table 7.6: Spatial spillover effects

Table 7.4 , column 1

Variable Typea Mean lb 95%CI ub 95%CI
Direct -0.1561 -0.2900 -0.0221
Share of social housing Indirect -0.0028 -0.0088 0.0032
Combined -0.1589 -0.2931 -0.0246
Direct 0.0105 -0.0089 0.0298
Jobs per squared kilometer Indirect 0.0014 0.0002 0.0026
Combined 0.0119 -0.0076 0.0313

Table 7.5 , column 1

Variable Typea Mean lb 95%CI ub 95%CI
Direct 2.5680 0.4400 4.6960
Share of social housing, first quintile Indirect 0.0053 -0.1178 0.1285
Combined 2.5734 0.4389 4.7078
Direct 1.8639 0.1931 3.5348
Share of social housing, second quintile Indirect -0.0207 -0.1269 0.0855
Combined 1.8432 0.1665 3.5200
Direct 1.761 0.1520 3.3702
Share of social housing, third quintile Indirect 0.0157 -0.0923 0.1236
Combined 1.7768 0.1615 3.3920
Direct 1.3980 -0.1104 2.9063
Share of social housing, fourth quintile Indirect -0.0460 -0.1395 0.0475
Combined 1.3520 -0.1615 2.8654
Direct 0.0129 -0.0067 0.0326
Jobs per squared kilometer Indirect 0.0013 0.0000 0.0025
Combined 0.0142 -0.0055 0.0339
aDirect, indirect and combined effect based on 1000 random and independent realizations
according to equations 7.16 and 7.17.
216 Chapter 7.

(1987) are robust to heteroskedasticity under the assumption that each relation
of the error is independent within and between regions.21 Thus both methods
require that the error term is spatially independent.
Driscoll and Kraay (1998) adjust the Newey and West (1987) standard er-
rors to allow general spatial dependence of the error term. However, esti-
mators that allow the computation of Driscoll and Kraay standard errors can
only handle cross-sectional data or cannot handle endogenous independent
variables. Therefore, it is not possible to control for spatial dependency in the
error term in our setup.
However, no evidence is found that the error term is spatially correlated.
Figure 7.9 shows the spatial distribution of the residual (specification in col-
umn one of Table 7.4) in 2006. Many negative outliers are next to positive
outliers. For all years, a scatterplot of the residuals against their spatial lags
shows no concrete pattern, see Figure 7.10. Also an OLS regression of the
spatial lag of the residual on the residual yields a parameter estimate that
is insignificant at the ten percent level. As there is no concrete evidence for
spatially correlation in the residuals, controlling for heteroskedasticity using
clustered and robust standard errors seems to be valid.

21 It is possible to estimate clustered standard errors that are robust to heteroskedasticity under

the assumption that the error term is only independent across regions.
7.5. Results 217

Figure 7.9: Spatial plot of residuals

Spatial plot of residuals of the specification of column one of Table 7.4 in 2006.
Numbers correspond to numbers in Table 7.B.

Figure 7.10: Spatial dependence in the residuals

Scatter plot of spatial lag of the residual against the residuals of the specification
of column one of Table 7.4.
218 Chapter 7.

7.6 Conclusion
Over the last two decades, high-skilled workers in the United States and the
Netherlands have been found to cluster in cities characterized by high produc-
tivity, high wages and high housing costs. This generates substantial produc-
tivity externalities to low-skilled and high-skilled workers. Rent-controlled
housing distorts this mechanism as it is mainly allocated to workers with low-
income at the time of entry. Also, it reduces skill substitution within a city by
preventing the internalization of housing demand in the controlled rent. This
effect on local skill composition provides a new channel by which rent con-
trol distorts outcomes on the local labor market, besides the effect on worker
mobility and unemployment duration.
In this chapter the framework of Moretti (2011, 2013) is augmented with
a (non-competitive) rent-controlled housing sector. This highlights the mech-
anisms by which rent-control housing vouchers influence local skill compo-
sition in an economy with two regions, two types of labor and idiosyncratic
preference for a region. Next, the influence of rent-controlled housing on the
local skill composition of regions is tested using a longitudinal panel of forty
NUTS 3 regions in the Netherlands over the years 1981-2006.
Evidence is found that ceteris paribus a decrease of the rent-controlled
housing stock with ten percentage points increases the percentage of high-
skilled workers with 1.5 percentage points. Our results are conditional on
fixed region effects and fixed time effects, the local skill-premium, local house
prices and local job density. Local house prices and the local skill-premium
are instrumented using historic housing density and indicators for local labor
demand. Also, the estimation procedure allowed for spatial dependence in
the data, although this has been shown to be of minor importance.
Our results suggest that rent control not only influences local labor markets
as it reduces the mobility of workers, but also alters local skill composition.
This implies that the distortional effects of rent control on the labor market
have been larger than previously thought.
The parameter size on the percentage of rent-controlled housing illustrates
that rent control has an economically significant effect on skill composition.
The estimated coefficients suggest that a decrease of the rent-controlled hous-
7.6. Conclusion 219

ing stock with ten percentage points has a similar effect on skill composition
as an increase of local house prices with about 8,800 euro.
The size of the effect of rent-controlled housing on the percentage of work-
ers that is high-skilled seems limited. Suppose the social housing stock of
Utrecht would be halved from forty to twenty percent. Based on our results,
one would expect the share of workers that are high-skilled to increase with
three percentage points (to 46 percent). An explanation for the limited size of
this effect might be that Dutch entry regulations based on income cannot pre-
vent that high-skilled worker remain living in social housing if house prices
are high.
220 Chapter 7.

7.A Derivations number of workers in a city

To solve the model write the number low-skilled workers (with and without
housing voucher) as

N Lv h L i
NbLv = wb − waL + ( AbL − A aL ) + s
2s
N Lm h L  i
NbLm = wb − waL − ( pb − p a ) + ( AbL − A aL ) + s
2s
N H h H  i
NbH = wb − waH − ( pb − p a ) + ( AbH − A aH ) + s
2s

Therefore the number of high and low-skilled workers in b equals

N H h H  i
NbH = wb − waH − ( pb − p a ) + ( AbH − A aH ) + s
2s
NbL = NbLm + NbLv
N L h L  i
N Lm
= wb − waL + ( AbL − A aL ) + s − 2s ( pb − pa )
2s

Use equations (7.8) and (7.9) to express the number of workers of each skill
type as

NH h 1  H  i
NbH = h Xb − XaH − (k a + k b )( NbH + NbL ) + ( AbH − A aH ) + s + k a N
2s
N L h   i N Lm
NbL = 1
X L
b − X L
a + ( A L
b − A L
a ) + s − (k a + k b )( NbH + NbL )+
2s h 2s
N Lm
ka N
2s

1
Define Z T = XbT − XaT + ( AbT − A Ta ) + s + k a N to reduce the notational


h
burden and rearrange. Note that all parameters that occur on the right hand
side only are exogenous.

NH H NH
NbH = φ H Z − φH (k a + k b ) NbL (7.18)
2s 2s
NL L N Lm N Lv
NbL = φ L Z − φL (k a + k b ) NbH − φ L ka N (7.19)
2s 2s 2s
2s 2s
φL = , φH =
2s + N Lm (k a + k b ) 2s + N H (k a + k b )
7.A. Derivations number of workers in a city 221

Substitute equation (7.19) into equation (7.18) to solve for NbH :

NH H NH NL L NH N Lv
NbH = Φφ H Z − Φφ H (k a + k b )φ L Z + Φφ H (k a + k b )φ L ka N
2s 2s 2s 2s 2s
2s + (k a + k b ) N H
Φ= 2s
2s + (k a + k b ) N H 2s+(k + k ) N Lm
a b

which simplifies to:

  h   i
NbH = σ + N − N H − N Lv N H δ 1h XbH − XaH + ( AbH − A aH )
h   i
− ( N − N H ) N H δ 1h XbL − XaL + ( AbL − A aL )

+ (σ − N Lv ) N H δs + σN H δk a N
1 1 2s
δ= , σ= k a +k b
2s σ + ( N − N Lv )

Plug in the values of δ and σ to derive equation (7.10).

To solve for NbL , substitute equation (7.18) into equation (7.19):

h   i
NbL = −( N − N H − N Lv ) N H δ 1h XbH − XaH + ( AbH − A aH )
h   i
+ ( N − N H )(σ + N H )δ 1h XbL − XaL + ( AbL − A aL )
h i
+ δs σ( N − N H ) + N H N Lv + σ( N − N H − N Lv )δk a N

From which it follows that Nb = NbH + NbL is given by:

h   i
Nb = N H δσ 1
h XbH − XaH + ( AbH − A aH )
h   i
+ ( N − N H )δσ 1h XbL − XaL + ( AbL − A aL )

+ δσNs + ( N − N Lv )δσk a N
1 1 2s
δ= , σ= k a +k b
2s σ + ( N − N Lv )
222 Chapter 7.

7.B Housing in 1947

ID corop housing stock 1947

1 Flevoland 1,281
2 Zuidwest-Drenthe 17,849
3 Midden-Noord-Brabant 47,239
4 Midden-Limburg 19,307
5 Groot-Amsterdam 251,037
6 Noord-Drenthe 18,807
7 Arnhem/Nijmegen 74,935
8 Zuid-Limburg 71,574
9 Zuidwest-Friesland 21,570
10 Zaanstreek 22,166
11 Oost-Groningen 31,692
12 Agglomeratie Leiden en Bollenstreek 42,581
13 Zeeuws-Vlaanderen 20,850
14 Overig Groningen 68,581
15 Zuidoost-Noord-Brabant 56,832
16 Noord-Limburg 22,674
17 Kop van Noord-Holland 39,633
18 Delfzijl en omgeving 9,223
19 Agglomeratie ’s-Gravenhage 146,151
20 Noordoost-Noord-Brabant 45,125
21 Groot-Rijnmond 233,872
22 West-Noord-Brabant 60,596
23 Overig Zeeland 40,179
24 Agglomeratie Haarlem 50,329
25 Achterhoek 44,773
26 Utrecht 120,352
27 Noord-Friesland 61,674
28 Noord-Overijssel 43,581
29 Zuidoost-Drenthe 17,356
30 Het Gooi en Vechtstreek 39,434
31 Zuidoost-Friesland 28,097
32 Veluwe 60,478
33 Twente 71,053
34 Zuidwest-Overijssel 16,989
35 Zuidwest-Gelderland 25,640
36 Zuidoost-Zuid-Holland 48,785
37 Delft en Westland 24,991
38 Oost-Zuid-Holland 33,109
39 Alkmaar en omgeving 22,176
40 IJmond 18,573
Chapter 8

Summary

The objective of social housing is to provide affordable housing to low-income

households. To guarantee the affordability of the dwelling the rent is often
controlled. This changes the costs and benefits of living, which in turn alters
the behavior of tenants of social housing and other agents. These economic
effects of social housing are the subject of this dissertation.
In this dissertation I present two theoretical and four empirical studies.
The first part considers social housing from a microeconomic perspective and
looks at the self-selection into social housing (chapter two), as well as at how
this influences consumption patterns (chapter three), the relocation behavior
of household within and from the social housing sector (chapter four) and the
wage rates workers earn (chapter five). The second part is macroeconomic in
nature and considers the joint effect of housing and labor market price rigidi-
ties on the skill composition of cities (chapter six). I study the effect social
housing on the local skill composition by empirically testing whether a large
social housing stock reduces the percentage of high-skilled workers in the la-
bor force (chapter seven).
224 Chapter 8.

8.1 Summary of this dissertation

Microeconomic framework

Evidence exists to suggest that social housing is not allocated randomly to

(low-skilled) workers. Tenants of social housing with high incomes benefit
more from rent control than tenants with low incomes (Gyourko and Linne-
man, 1989). What’s more, the average incomes of tenants living in social hous-
ing are found to be higher in regions with higher house prices (Van Daalen
et al., 2012). In chapter two I present a microeconomic framework that ex-
plains both observations by the way in which local amenities influence the
self-selection into controlled or uncontrolled housing.
Local amenities such as the presence of jobs, historic scenery or natural ar-
eas (or other local public goods) are internalized into the local house prices,
but not into the controlled rent of social housing. This allows the tenant of
social housing to consume those location amenities for free. Therefore, social
housing in regions with a high level of local amenities will be inhabited by
high-income tenants, whereas tenants with such incomes would rent uncon-
trolled housing in other regions.
The microeconomic framework that is developed assumes that workers
derive utility from consuming housing quality, a composite good and location
amenities. Workers maximize utility by choosing to consume uncontrolled or
social housing. As a result, living in rent-controlled housing (and its quality)
becomes a decision.
The model shows that rent control changes the allocation of the budget to
housing quality and consumption of the composite good. It moreover demon-
strates that workers in social housing can be subdivided into those who con-
sume the optimal quality of social housing given their income and the con-
trolled rent (unrestricted tenants of social housing) and those who would want
to consume social housing of better quality but are unable to (restricted ten-
ants of social housing). Tenants of social housing spend a larger budget share
on the composite good. We show this holds even stronger for restricted ten-
ants of social housing.
8.1. Summary of this dissertation 225

Chapter two points out that the ‘unrestricted tenants of social housing’
are those workers in social housing with income equal or below the income
threshold Y s . The other tenants of social housing make up the ‘unrestricted
tenants of social housing’. Tenants with incomes exceeding another threshold
(Y Al ) will rent uncontrolled housing. The income threshold Y Al varies with
the level of location amenities. This implies that the maximum income level
at which one is willing to live in social housing varies with the local amenity
level.
It is shown that entry regulations based on income are unable to prevent
the self-selection of high-income tenants into social housing if incomes are dy-
namic. This weakens the contribution of rent control to mixed neighborhoods:
In growing cities with many local amenities, rent-controlled tenants will have
a higher incomes, thereby reducing social mixing.
An important contribution of chapter two is its study of the welfare effects
of social housing. Social housing is always inefficient as the costs born by
the lessor always exceed the benefits enjoyed by the tenants. This provides
scope for welfare improvement as winners from the decontrol of social hous-
ing can compensate losers in such a way that both are better off compared to
the current situation. However, such compensation schemes might be politi-
cally unfeasible to accomplish.
The welfare analysis in this chapter shows that the contribution of a worker
to the total welfare loss of social housing increases with income and the mar-
ket rent. This implies that the welfare loss of social housing can be reduced by
reallocating social housing from high-income workers to low-income work-
ers. Such schemes might provide a (second best) welfare improvement that is
politically more feasible.

Consumption differs with housing tenure type

It can be argued that the consumption of tenants of social housing differs from
that of homeowners because of two reasons. First, the property rights that
come with homeownership ensure that movements in house prices influence
(future) wealth of homeowners, but not that of tenants. The propensity to con-
sume out of income might therefore be different for homeowners as compared
226 Chapter 8.

to tenants. Second, the microeconomic framework has shown that rent control
of social housing influences the budget share devoted to the composite good.
Both effects of housing market tenure are studied in chapter three.
I start with testing whether house prices influence consumption as a result
of the increased wealth of homeowners (wealth hypothesis) or whether both
variables move together because of a common driver (common factor hypoth-
esis). Both hypotheses are identified by a different marginal effect of house
prices on consumption for different age groups.
I test whether housing tenure influences the level of consumption by es-
timating a lifecycle consumption model based on data from the Dutch bud-
get survey (in Dutch: Budget Onderzoek) that has been collected by Statistics
Netherlands for the years 1980-1999. The national house price index for the
period 1980-1999 is used to test whether the consumption levels of homeown-
ers and tenants respond differently to house prices.1 Along the lines of At-
tanasio et al. (2009) identification occurs according to the interaction with age,
although propensity score matching (PSM) is applied as a robustness check.
The evidence is in line with the common factor hypothesis: The marginal
effect of house prices on consumption is positive and decreases with age. Re-
sults from the PSM-procedure confirm this for both homeowners and tenants.
Currently, the Dutch economy is characterized by both a low level of house-
hold consumption and decreasing house prices. It has been argued that a re-
covery of the housing market is required in order to raise consumption. How-
ever, my findings suggest that both the low prices on the housing market and
low consumption are caused by other factors.

The second part of chapter three studies the effect of housing market tenure on
the allocation of the budget. Using the same consumption data Engel curves
are estimated for four expenditure categories: ‘Basic goods’, ‘Housing’, ‘Recre-
ation’ and ‘Alcohol and tobacco’. These four categories on average contain 80
percent of total household expenditure.
It is concluded that housing market tenure has an effect on the allocation
of the budget: Tenants in social housing spend less on housing and more on
1 The data does not allow for the consideration of house prices across regions.
8.1. Summary of this dissertation 227

basic goods, recreation and alcohol and tobacco. This difference in budget al-
location remains is PSM is used. The analysis shows that the difference in the
allocation of budget to consumption goods between homeowners and tenants
of social housing has been falling over time. This is in line with the observa-
tion that rents in the controlled housing sector have been increasing more than
the imputed rents in the owner-occupied sector.
Finally, I test consumption patterns of tenants who are similar in charac-
teristics to homeowners differ from those of other tenants of social housing.
Under the assumption that social housing tenants with similar characteristics
as homeowners are restricted tenants of social housing, I find some evidence
pointing to the existence of consumption differences between restricted and
unrestricted tenants. Restricted tenants of social housing are found to spend
less on housing and more on recreation, although the size of the effect is lim-
ited.

Social housing reduces mobility and housing demand

The prediction following from the microeconomic framework is that the in-
come at which workers leave the social housing sector increases with the price
difference between the market rent and controlled rent. Chapter four tests this
hypothesis by considering how rent control benefits influence the decision to
move from and within the social housing sector. Rent control might influence
these types of transitions differently, as moving from the social housing sec-
tor eliminates the benefits of rent control, whereas transitions within the rent
control housing sector do not and might in fact increase them.
Previous empirical research concludes that rent control reduces the overall
propensity to move house.2 However, this research did not make a distinction
between transitions from and within the social housing sector.
Estimation results are based on a unique household panel dataset for the
years 2006 through 2008 that has been constructed from administrative records
provided by Statistics Netherlands. It contains individual household informa-
tion, job characteristics, local housing market characteristics and an indicator
2 See, for instance, Gyourko and Linneman (1989); Ault et al. (1994); Nagy (1995); Munch and

Svarer (2002); Simmons-Mosley and Malpezzi (2006) for empirical results using various estima-
tion methods such as OLS, logistic regression, and duration models.
228 Chapter 8.

of whether the house one lives in is social housing or owner-occupied hous-

ing. Rent control benefits are measured as the median rent at the neighbor-
hood level divided by individual house value.
The results indicate that the effect of rent control benefits on household
mobility is large for transitions within the social housing sector, whereas it is
limited for transitions from the social housing sector. In line with the model
presented in chapter two the probability to move from the social housing sec-
tor decreases with rent control benefits for high-income workers only. In con-
trast, all tenants of social housing reduce the transitions within the social hous-
ing sector if rent control benefits increase.
These results suggest that rent control has a limited effect on the demand
for owner-occupied housing. In the short run rent-controlled housing only
becomes available for new entrants if current inhabitants move. Our results
suggest that less housing becomes available as rent control of social housing
reduces mobility within the social housing sector.

Housing market tenure does not influence wages

On average, workers in owner-occupied housing earn 25 percent higher wages

compared to tenants of social housing. This wage differential might reflect the
sorting mechanism as described in the economic framework. In contrast, as
both homeownership and living in social housing reduce labor mobility (De
Graaf et al., 2009), wages might be affected by housing market tenure.
Although it has been documented that tenants of social housing have a
weaker labor market position than homeowners, the effect of social housing
on wages has not previously been investigated. And while two studies have
investigated the effect of homeownership on wages, however, they are incon-
clusive on the sign and magnitude of its effect. Munch et al. (2008) use Dan-
ish data from administrative records to find that homeowners earn about five
percent higher wages than tenants. In contrast, Coulson and Fisher (2009) use
survey data for the United States and find that homeowners in the United
States earn on average 30 percent lower wages compared to tenants. Both
studies do not distinguish between tenants of private housing and tenants of
social housing.
8.1. Summary of this dissertation 229

Chapter five tests whether homeowners and tenants of social housing earn
different wages compared to tenants of private rental housing. This is done
using a large panel for the years 2006 through 2008 based on administrative
records. This enables controlling for the sorting of low-skilled workers into
social housing.
It follows from the analysis that conditioning on worker skill is very im-
portant: Without controlling for the sorting mechanism of low-skilled workers
into social housing one would conclude that social housing reduces the wage
rate with 17 percent compared to the wage rate of private tenants. Similarly,
homeowners are found to earn seven percent higher wages than tenants of
private rental housing. In contrast, if one controls for sorting according to
skill, there is no evidence that homeowners or tenants living in social housing
earn different wages compared to tenants of private housing.

Macroeconomic framework

The macroeconomic framework of this dissertation is presented in chapter six.

Because workers compete for housing a demand shock to high-skilled labor
increases the number of high-skilled workers in a city and decreases the num-
ber of low-skilled workers in it. Because house prices depend on the price elas-
ticity of the housing supply, the housing market determines how skill-specific
productivity shocks change the composition of a city (Roback, 1982; Moretti,
2011).3 These types of equilibrium models describe the interaction between
local housing and labor markets under the assumption that both markets are
perfectly competitive. In chapter six, I present a macroeconomic framework
that analyzes the location decision of low-skilled and high-skilled workers if
labor and housing markets are characterized by imperfections in the form of
price rigidities.
Price rigidities on the labor market are modeled in the form of nominal
wage rigidities that are the result of a national wage bargaining process. Local
labor markets are characterized by the matching framework in developed by
3 Iflabor demand in a city increases, this might attract labor. If the price elasticity of housing
supply is high, the increase in population will be modest and the increase in local wages and
house prices will be large. However, if the price elasticity of housing is low, many workers can
enter the city and the increase in local wages and house prices will be low.
230 Chapter 8.

Pissarides (2000). I show that these price rigidities in the labor market reduce
the mobility of labor if there are limited spatial differentials in unemployment
rates or if wages are not adjusted in response to local changes in productivity.
This is due to the fact that moving to another labor market would in these
cases not pay off.
Price rigidities in the housing market enter in the form of housing vouch-
ers that enable the holder to rent housing at the rent ceiling. I show that the
effect of these housing market vouchers on the size and composition of cities is
critically dependent on its distribution among workers: If the housing vouch-
ers are held mainly by low-skilled workers, the inflow of high-skilled workers
and outflow of low-skilled workers after a productivity shock to high-skilled
labor will both be lower. In contrast, if rent control housing vouchers are held
mainly by high-skilled workers the inflow of high-skilled workers and out-
flow of low-skilled workers will be larger.
Interestingly, I show the existence of an intermediate distribution of rent
control housing vouchers among high-skilled and low-skilled workers, such
that rent control housing vouchers both increases the inflow of high-skilled
workers and reduces the outflow of low-skilled workers after a productivity
shock to high-skilled labor.
The analysis shows that the effect of housing market and labor market
price rigidities on local skill composition is skill-specific. In general, both
housing market price rigidities and labor market price rigidities have an am-
plified effect on city composition after a shock to the productivity of high-
skilled labor. However, rent control housing vouchers mitigate the effect of
nominal wage rigidities if the productivity of low-skilled workers changes.

Rent-controlled housing reduces local skill level

The macroeconomic framework illustrates how low-skilled workers in social

housing do not leave the city if the productivity of high-skilled workers in-
creases, whereas low-skilled workers in uncontrolled housing do. In addi-
tion, as these low-skilled workers do not leave the city, house prices increase
for high-skilled workers who want to enter. These two mechanisms imply
that the percentage of high-skilled workers in a city decreases with the stock
8.1. Summary of this dissertation 231

of rent-controlled housing. On the other hand, chapter two has shown that en-
try regulations are unable to prevent the self-selection of high-skilled workers
into social housing if the income of a worker might increase over time.
It is important to know whether the social housing stock reduces the skill
composition of local labor markets, as this might reduce the positive exter-
nalities from agglomeration economies.4 If it does, this would be a channel
by which rent control distorts outcomes on the local labor market that has
not been studied before. In chapter seven I test whether the stock of rent-
controlled housing influences the local skill composition of regions and, if so,
to what extent.
It test whether the stock of rent-controlled housing influences local skill
composition using a longitudinal panel of forty NUTS 3 regions in the Nether-
lands. The data is constructed from the WBO and WoON national housing
surveys in the years between 1981 and 2006 and augmented with information
on productivity and employment for eight sectors in these years at the NUTS
3 level for the same period.
I use an instrumental variables approach to control for the endogeneity of
explanatory variables such as the high-skilled wage premium and local house
prices. Rent control is found to reduce the percentage of high-skilled work-
ers in a region. A ten percentage point increase of the rent-controlled housing
stock is found, ceteris paribus, to reduce the percentage of high-skilled work-
ers in a region with 1.5 percentage points. This is equivalent to an increase in
house prices with about e8,800, ceteris paribus. The size of the effect is mod-
est. If the social housing stock of Utrecht would be decreased by half (from
forty to twenty percent of the housing stock), the percentage of high-skilled
workers in Utrecht is expected to increase with three percentage points (to
46 percent). As mentioned the limited effect of the share of rent-controlled
housing on local skill composition might be explained by the self-selection of
high-skilled workers into social housing in attractive cities, the topic of study
in chapter two.
4 Moretti (2004) presents evidence that clustering of high-skilled workers increases nominal

wages of all skill types.

232 Chapter 8.

8.2 Limitations and directions for future research

This dissertation has shown that social housing has many economic effects
on the housing and labor market. However, the economic effects are more
subtle than the canonical “rent-controlled social housing reduces transitions
to uncontrolled housing, which worsens the labor market position of inhabi-
tants of social housing”. In contrast, this dissertation has shown that rent con-
trol mainly reduces transitions to controlled housing, whereas transitions to
the uncontrolled (owner-occupied) housing sector are only limitedly affected.
These results call for the development of dynamic theoretical models in which
household choose to move within or from the social housing sector.
Moreover, although I did not study the direct effect of social housing on
the labor market position, no evidence is found that social housing reduces
the wages of its inhabitants. This is surprising as a negative effect on wages
is expected if social housing worsens the labor market position. Research on
housing tenure type and labor mobility has focused mainly on the effect of
homeownership on the probability to become unemployment, the probability
to leave unemployment, and the probability to switch jobs (Van Ewijk and
Van Leuvensteijn, 2009). An interesting exception is De Graaf et al. (2009),
who find no evidence for a worsened labor market position of tenants of social
housing compared to tenants of private rental housing in the Netherlands. It
remains an open question why ceteris paribus inhabitants of social housing
do not earn lower wages and do not have a worsened labor market position,
when they are found to be less mobile and to commute less.
With regard to economic theory it would be interesting to consider the
joint welfare implications of social housing on the housing and labor markets.
It has been shown that the reallocation of social housing from high-income
to low-income workers reduces the welfare loss created by social housing at
the housing market, but might increase labor market distortions on the labor
market. Future research could focus on further exploring this efficiency vs.
efficiency trade-off and on developing strategies to quantify it.
Also, the models in this dissertation consider the behavior of tenants of
social housing and of tenants of uncontrolled housing, whereas the empirical
sections make a comparison between tenants of social housing and homeown-
8.2. Limitations and directions for future research 233

ers most of the time. The reason is that homeownership influences equity and
therefore the long-term income of households which is more complicated to
model. Although simple and straightforward models are a benefit in itself,
future research on the interactions of the labor and housing markets might
explicitly model the effects of ownership when comparing homeowners to
tenants.
There are several aspects of the Dutch housing market that merit study
in future research. I highlighted why there are so many high-skilled work-
ers to be found in social housing in the economic core. I also explained why
this makes high-skilled workers benefit more from social housing than low-
skilled workers. However, I did not explicitly consider the implications this
self-selection of high-skilled workers into social housing has for redistribu-
tion.
Also, I ignored the effect of the Dutch rent support program on housing
demand. Given the size of the program, both in number of households and
the total amount of subsidy received, it would be interesting to investigate
how this program impacts the demand for housing services. Koning and Rid-
der (1997) conclude that the Dutch rent support program increases demand
for housing services. It would be worthwhile to consider whether individual
rent support affects wages (poverty trap) and whether it influences the self-
selection into social housing, household mobility and labor market position.

The conclusions of any empirical study are restricted by the data that have
been used. This dissertation is no different. In several chapters it would be
interesting to use other types of data to see whether the conclusions of this
dissertation hold. For instance, the study on the consumption of homeown-
ers and tenants of social housing in chapter three uses a pooled cross-section
for twenty consecutive years and a national house price index. With available
panel data on consumption and local house price indices, it would be possi-
ble to control for unobserved worker characteristics (preferences, skills) that
are found to be important drivers of selection into social housing (see chapter
five). Now, while this concern is dealt with using propensity score matching,
panel data techniques are always consistent (and at least more efficient), as is
234 Chapter 8.

explained in chapter three. Local house price indices might clearly be better
able to capture the wealth shocks homeowners face compared to the national
house price index, and are the preferred avenue to improve the efficiency of
this analysis.
In chapters four and five I use a panel based on administrative records. It
would be interesting to extent the panel to include a longer time span. Not
only would this improve the analysis by reducing the incidental parameter
problem, it would also allow for the investigation of the effect of the crisis
on the Dutch (owner-occupied) housing market that started in 2008 and as of
today has not yet come to an end. Since the crisis, house prices have been
falling, thereby reducing the benefits of rent control. This might influence
household transitions and self-selection into social housing as social housing
might be considered a save haven.
The analysis in chapter seven considers the effect of social housing on local
skill composition based on NUTS 3 regions. Because of aggregation issues, a
lower geographical scale such as the municipal level might provide a better
level to measure both the share of the housing stock that is rent-controlled
and the percentage of the labor force that is high-skilled. While it is true that
an analysis at a lower geographical scale presents issues of spatial dependence
of the data, these can be dealt with using spatial econometric techniques.

8.3 Policy implications

This dissertation has shown that social housing has several economic effects
which should be taken into consideration when designing policy for the Dutch
social housing sector. Social housing has been shown to influences consump-
tion and geographical mobility without reducing the earned wage. Also, so-
cial housing somewhat reduces the share of high-skilled workers in the local
labor market, although the magnitude of this effect is limited.

The aim of social housing is to provide dwellings to households who typi-

cally have a low-income. It has been shown that entry-regulations based on
income are unable to prevent that social housing is inhabited by high-income
8.3. Policy implications 235

households. As a result social housing in the Netherlands is badly targeted to

low-income households especially in regions where house prices are high and
keeping housing affordable is most needed. Thus from an equity perspective
regulations that change the allocation of social housing are called for.
A similar conclusion is reached if one is concerned with the efficiency of
the housing market. The contribution to the welfare loss from social housing
has been shown to be larger for high-income workers than for low-income
workers. Thus reallocating social housing from high-income workers to low-
income workers will reduce the welfare loss at the housing market, especially
if this occurs in places where house prices are high.5 As most high-income
workers live in social housing in locations where rent control benefits are high,
social housing will especially become available to low-income households in
those regions where transitions within the social housing sector are lowest and
where entering the social housing sector is hardest.
Thus reallocating social housing from high-income to low-income house-
holds might be considered fair and will reduce the welfare loss at the housing
market. The Dutch administration has taken several steps to adjust the alloca-
tion of social housing away from high-income households.6 The framework in
chapter two suggest that these measures will reduce the share of social hous-
ing inhabited by high-income households. However, targeting of social hous-
ing to low-income tenants is expected to remain less efficient in regions with
high house prices, as long as there exists a wedge between market and social
housing prices and only entry-regulations are imposed.

It should be noted that legislation that ensures that social housing will only
be inhabited by low-income households will increase local labor market dis-
tortions by reducing the share of high-skilled workers. This ‘efficiency vs. ef-
5 Reallocation of social housing from high-income to low-income households is unable to

eliminate the welfare loss. It can be eliminated by decontrolling all social housing and letting
the winners compensate the losers. However, such a policy is likely to be unfeasible.
6 As of July 2011, the rent of new contracts is allowed to increase more in attractive locations in

the country (the so-called schaarstegebieden). As of July 2013 rents in the social housing sector were
made dependent on income. Rents were allowed to increase with the inflation rate plus 2.5, 3.5
and 4 percent maximums for low-, middle-, and high-income households. Here, middle-income
households are defined as earning an annual income between e33,614 and e43,000. Finally, the
government aims to introduce a new pricing mechanism on 1 July 2014 in which rents of social
housing are based on house values, see BZK (2013).
236 Chapter 8.

ficiency trade-off’ is an important mechanism policymakers should be aware

off, if they introduce new policy on social housing.
The ‘efficiency vs. efficiency trade-off’ implies that if legislation that in-
creases the allocation of social housing to low-income households is intro-
duced, then it should be accompanied by a reduction of the social housing
stock: The reallocation will reduce the efficiency loss at the housing market,
whereas the reduced size of the social housing stock will reduce labor market
distortions. If the latter does not occur, the efficiency vs. efficiency trade-
off implies that distortions on the labor market will be amplified. As social
housing is poorly targeted to low-income households, this suggest that low-
income workers need not necessarily be worse off. If so, a more efficient out-
come might be realized without harming the people for whom social housing
is designed.
References

Ai, C. and E. B. Norton (2003). Interaction terms in logit and probit models. Economics Letters 80(1),
123–129.

Alston, R. M., J. R. Kearl, and M. B. Vaughan (1992). Is there a consensus among economists in
the 1990’s? The American Economic Review 82(2), 203–209.

Aoki, K., J. Proudman, and G. Vlieghe (2004). House prices, consumption, and monetary policy:
A financial accelerator approach. Journal of Financial Intermediation 13(4), 414–435.

Arnott, R. (1995). Time for revisionism of rent control? The Journal of Economic Perspectives 9(1),
99–120.

Arrow, K. (1962). The economic implications of learning by doing. The Review of Economic Stud-
ies 29(3), 155–173.

Attanasio, O. P., L. Blow, R. Hamilton, and A. Leicester (2009). Booms and busts: Consumption,
house prices and expectations. Economica 76(301), 20–50.

Attanasio, O. P., A. Leicester, and M. Wakefield (2011). Do house prices drive consumption
growth? The coincident cycles of house prices and consumption in the UK. Journal of the
European Economic Association 9(3), 399–435.

Attanasio, O. P. and G. Weber (1994). The UK consumption boom of the late 1980s: Aggregate
implications of microeconomic evidence. The Economic Journal 104(427), 1269–1302.

Ault, R., J. Jackson, and R. Saba (1994). The effect of long-term rent control on tenant mobility.
Journal of Urban Economics 35(2), 140–158.

Barten, A. P. (1993). Consumer allocation models: Choice of functional form. Empirical Eco-
nomics 18(1), 129–158.

Battu, H. and M. Phimister (2008). Housing tenure, job mobility and unemployment in the UK.
The Economic Journal 118(257), 311–328.

Belot, M. and S. Ederveen (2005). Indicators of cultural and institutional barriers in OECD coun-
tries. CPB Research Memorandum.

237
238 References

Blanchflower, D. G. and A. J. Oswald (1995). An introduction to the wage curve. The Journal of
Economic Perspectives 9(3), 153–167.

Bostic, R., S. Gabriel, and G. Painter (2009). Housing wealth, financial wealth, and consumption:
New evience from micro data. Regional Science and Urban Economics 39(1), 79–89.

Brambor, T., W. R. Clark, and M. Golder (2006). Understanding interaction models: Improving
empirical analyses. Political Analysis 14(1), 63–82.

Brueckner, J. K., J.-F. Thisse, and Y. Zenou (1999). Why is central Paris rich and downtown Detroit
poor? An amenity-based theory. European Economic Review 43(1), 91–107.

Bulow, J. and P. Klemperer (2012). Regulated prices, rent seeking, and consumer surplus. Journal
of Political Economy 120(1), 160–186.

BZK (2013). A new approach to setting maximum rents. News item Ministry of the Interior and
Kingdom Relations 13-09-2013.

Cairncross, F. (2001). The death of distance: How the communications revolution will change our lives.
Harvard Business Press.

Campbell, J. Y. and J. F. Cocco (2007). How do house prices affect consumption? Evidence from
micro data. Journal of Monetary Economics 54(3), 591–621.

Caponi, V. (2008). Centralized wage determination and regional unemployment differences: The
case of Italy. IZA Discussion Paper 3592.

CBS (2013). Wijk- en buurtkaart 2011. downloaded from www.cbs.nl/nl-Nl/themas/

dossiers/nederland_regionaal/publicaties/geografische_data/archief/
2012/default.htm at 01-12-2013.

Chen, N.-K., S.-S. Chen, and Y.-H. Chou (2010). House prices, collateral constraint, and the asym-
metric effect on consumption. Journal of Housing Economics 19(1), 26–37.

Christensen, M. (2007). Heterogeneity in consumer demands and the income effect: Evidence
from panel data. IFS Working Papers 7(16), 1–29.

Combes, P.-P., G. Duranton, and L. Gobillon (2008). Spatial wage disparities: Sorting matters!
Journal of Urban Economics 63(2), 723–742.

Coulson, E. N. and L. M. Fisher (2009). Housing tenure and labor market impacts: The search
goes on. Journal of Urban Economics 65(3), 252–264.

Cristini, A. and S. Almudena (2013). Do house prices affect consumption? A re-assessment of the
wealth hypothesis. IZA Discussion Paper 7576, 1–31.

CSED (2010). Naar een integrale hervorming van de woningmarkt. SER.

References 239

De Graaf, T., M. Van Leuvensteijn, and C. Van Ewijk (2009). Social renting, transfer taxes, and
labour mobility in Europe. In C. Van Ewijk and M. Van Leuvensteijn (Eds.), Homeownership
and the labour market in Europe, Chapter 3, pp. 53–81. Oxford University Press.

De Groot, H., S. Groot, and M. J. Smit (2013). Agglomeration externalities in the Netherlands:
Micro-evidence from wages. In Economic Geography seminar series Utrecht University.

Deaton, A. and J. Muellbauer (1980a). An almost ideal demand system. The American Economic
Review 70(3), 312–326.

Deaton, A. and J. Muellbauer (1980b). Economics and consumer behavior. Cambridge; New York:
Cambridge University Press.

Diamond, R. (2011). The welfare implications of changes in local wages, prices, and amenities
across US cities: 1980 & 2000. MIMEO.

DiPasquale, D. and E. L. Glaeser (1999). Incentives and social capital: Are homeowners better
citizens? Journal of Urban Economics 45(2), 354–384.

Donders, J., M. Dijk, and G. Romijn (2010). Hervorming van het Nederlandse woonbeleid. CPB,
Centraal Planbureau.

Driscoll, J. C. and A. Kraay (1998). Consistent covariance matrix estimation with spatially depen-
dent panel data. The Review of Economics and Statistics 80(4), 549–560.

Eeckhout, J. and P. Kircher (2011). Identifying sortingŮin theory. The Review of Economic Stud-
ies 78(3), 872–906.

Elhorst, J. (2010). Applied spatial econometrics: Raising the bar. Spatial Economic Analysis 5(1),
9–28.

Engelhardt, G. V. (1994). House prices and the decision to save for down payments. Journal of
Urban Economics 36(2), 209–237.

Eurostat (2013). Countries, 2010 -administrative units- dataset. downloaded from http:
//epp.eurostat.ec.europa.eu/portal/gisco_Geographical_information_
maps/popups/references/administrative_units_statistical_units_1 at
01-12-2013.

Fallis, G. (1988). Rent control: The citizen, the market, and the state. Journal of Real Estate Finance
and Economics 1(3), 309–320.

Ferrer-Carbonell, A. and J. C. van den Bergh (2004). A micro-econometric analysis of determinants

of unsustainable consumption in the Netherlands. Environmental & Resource Economics 27(4),
367–389.

Flatau, P., M. Forbes, P. H. Hendershott, and G. Wood (2003). Homeownership and unemploy-
ment: The roles of leverage and public housing. NBER Working Paper 10021, 1–50.
240 References

Friedman, T. L. (2006). The world is flat: The globalized world in the twenty-first century. Penguin
London.

Fujita, M., P. R. Krugman, and A. J. Venables (1999). The spatial economy: Cities, regions and interna-
tional trade. MIT Press.

Gibbons, S., H. G. Overman, and P. Pelkonen (2013). Area disparities in Britain: Understand-
ing the contribution of people vs. place through variance decompositions. Oxford Bulletin of
Economics and Statistics, 1–19. printed online, forthcoming.

Glaeser, E. (2003). Does rent control reduce segregation? Swedish Economic Policy Review 10,
179–202.

Glaeser, E. (2011). Triumph of the city: How our greatest invention makes us richer, smarter, greener,
healthier and happier. Pan Macmillan.

Glaeser, E. L. and J. D. Gottlieb (2009). The wealth of cities: Agglomeration economies and spatial
equilibrium in the united states. Journal of Economic Literature 47(4), 983–1028.

Glaeser, E. L., H. D. Kallal, J. A. Scheinkman, and A. Shleifer (1992). Growth in cities. Journal of
Political Economy 100(6), 1126–1152.

Glaeser, E. L., J. Kolko, and A. Saiz (2001). Consumer city. Journal of Economic Geography 1(1),
27–50.

Glaeser, E. L. and E. F. Luttmer (2003). The misallocation of housing under rent control. American
Economic Review 93(4), 1027–1046.

Glaeser, E. L. and D. C. Mare (2001). Cities and skills. Journal of Labor Economics 19(2), 316–342.

Green, R. K. and P. H. Hendershott (2001). Home-ownership and unemployment in the US. Urban
Studies 38(9), 1509–1520.

Groot, H. L. F., G. Marlet, C. N. Teulings, and W. Vermeulen (2010). Stad en land. CPB, Centraal
Planbureau.

Gyourko, J. and P. Linneman (1989). Equity and efficiency aspects of rent control: An empirical
study of New York City. Journal of Urban Economics 26(1), 54–74.

Harris, R., J. Moffat, and V. Kravtsova (2011). In search of ‘W’. Spatial Economic Analysis 6(3),
249–270.

Hartog, J., E. Leuven, and C. Teulings (2002). Wages and the bargaining regime in a corporatist
setting: The Netherlands. European Journal of Political Economy 18(2), 317–331.

Havet, N. and A. Penot (2010). Does homeownership harm labour market performances? A
survey. MIMEO.

Hughes, G. and B. McCormick (1987). Housing markets, unemployment and labour market flex-
ibility in the UK. European Economic Review 31(3), 615–645.
References 241

Huurcommissie (2008). Puntensysteem zelfstandige woningen vanaf 1 juli 2008. Huurcommissie.

Iacoviello, M. (2004). Consumption, house prices, and collateral constrains: A structural econo-
metric analysis. Journal of Housing Economics 13(4), 304–320.

Jacobs, J. (1969). The economy of cities. Random House, New York.

Kadaster / Centraal Bureau voor de Statistiek (2013). Wijk en buurtkaart 2011. downloaded
from www.cbs.nl/nl-Nl/themas/dossiers/nederland_regionaal/publicaties/
geografische_data/archief/2012/default.htm at 01-12-2013.

Kalwij, A. (2005). Consumption and income around the time of births. Review of Economics of the
Household 3(1), 75–89.

Kalwij, A., R. Alessie, and P. Fontein (1998). Household commodity demand and demographics in
the Netherlands: A microeconometric analysis. Journal of Population Economics 11(4), 551–577.

Kalwij, A. and W. Salverda (2004). Changing household demand patterns in the Netherlands:
Some explanations. Dempatem Working Paper 3.

Kelejian, H. H. and I. R. Prucha (1998). A generalized spatial two-stage least squares procedure
for estimating a spatial autoregressive model with autoregressive disturbances. The Journal of
Real Estate Finance and Economics 17(1), 99–121.

Kelejian, H. H. and I. R. Prucha (1999). A generalized moments estimator for the autoregressive
parameter in a spatial model. International Economic Review 40(2), pp. 509–533.

Kim, J., G. Y. Kim, and J. Yoon (2004). Estimation of the tenants’ benefits residing in public rental
housing with unit size cconstrain in korea. Urban Studies 41(8), 1521–1536.

King, M. (1990). Discussion. Economic Policy 11, 383–387.

Koning, M., R. Nistal, and J. Ebregt (2006). Woningmarkteffecten van aanpassing fiscale behandel-
ing eigen woning [Housing market effects of the adaptation of the fiscal treatment of owner-occupied
housing]. CPB, Centraal Planbureau.

Koning, P. and M. Van Leuvensteijn (2010). De woningcorporaties uit de verdwijndriehoek. TPE

digitaal 4(2), 21–38.

Koning, R. H. and G. Ridder (1997). Rent assistance and housing demand. Journal of Public
Economics 66(1), 1–31.

Kromhout, S., G. Van Daalen, S. Davis, and A. Zandstra (2006). Woonruimteverdeling opnieuw
bekeken: Bouwstenen voor discussie. RIGO Research en Advies B.V & OTB Delft.

Le Blanc, D. and A. Laferrère (2001). The effect of public social housing on households’ consump-
tion in France. Journal of Housing Economics 10(4), 429–455.

Leamer, E. (2007). Housing is the business cycle. Technical report, National Bureau of Economic
Research.
242 References

Lee, L.-F. (2004). Asymptotic distributions of quasi-maximum likelihood estimators for spatial
autoregressive models. Econometrica 72(6), 1899–1925.

LeSage, J. and R. K. Pace (2009). Introduction to Spatial Econometrics. Boca Raton, Taylor and
Francis.

LeSage, J. P. and R. K. Pace (2010). The biggest myth in spatial econometrics. Available at SSRN
1725503, 1–42.

Lind, H. (2001). Rent regulation: A conceptual and comparative analysis. European journal of
Housing Policy 1(1), 41–58.

Lindbeck, A. and P. Samuelson (1977). The political economy of the New Left: An outsider’s view.
Harper & Row.

Lyytikäinen, T. (2006). Rent control and tenants’ welfare: The effects of deregulating rental mar-
kets in Finland. VATT Discussion Papers 385.

Malpezzi, S. (1993). Can New York and Los Angeles learn from Kumasi and Bangalore? Costs
and benefits of rent controls in developing countries. Housing Policy Debate 4(4), 589–626.

Manacorda, M. and B. Petrongolo (2006). Regional mismatch and unemployment: Theory and
evidence from Italy, 1977–1998. Journal of Population Economics 19(1), 137–162.

Marshall, A. (1890). Principles of Economics. downloaded from http://www.econlib.org/

library/Marshall/marP.html at 12-09-2012.

Mas-Colell, A., M. D. Whinston, J. R. Green, et al. (1995). Microeconomic theory. Oxford University
Press.

McCormick, B. (1997). Regional unemployment and labour mobility in the UK. European Economic
Review 41(3), 581–589.

Melitz, M. J. (2003). The impact of trade on intra-industry reallocations and aggregate industry
productivity. Econometrica 71(6), 1695–1725.

MinBZK (2011). Woonvisie. Ministerie van Binnenlandse Zaken en Koninkrijksrelaties.

Minford, P., P. Ashton, and M. Peel (1988). The effects of housing distortion on unemployment.
Oxford Economic Papers 40(2), 322–345.

MinVROM (2000). Nota wonen: mensen, wensen, wonen wonen in de 21ste eeuw. Ministerie van
Volkshuisvesting, Ruimtelijke Ordening en Milieu.

Moretti, E. (2004). Estimating the social return to higher education: Evidence from longitudinal
and repeated cross-sectional data. Journal of Econometrics 121(1), 175–212.

Moretti, E. (2011). Local labor markets. In D. Card and O. Ashenfelter (Eds.), Handbook of Labor
Economics, Volume 4, part B, Chapter 14, pp. 1237–1313. Elsevier.
References 243

Moretti, E. (2012). The new geography of jobs. Houghton Mifflin Harcourt.

Moretti, E. (2013). Real wage inequality. American Economic Journal: Applied Economics 5(1), 65–103.

Muellbauer, J. and A. Murphy (1990). Is the UK balance of payments sustainable? Economic

Policy 5(11), 348–395.

Munch, J. R., M. Rosholm, and M. Svarer (2006). Are homeowners really more unemployed? The
Economic Journal 116(514), 991–1013.

Munch, J. R., M. Rosholm, and M. Svarer (2008). Homeownership, job duration, and wages.
Journal of Urban Economics 63(1), 130–145.

Munch, J. R. and M. Svarer (2002). Rent control and tenancy duration. Journal of Urban Eco-
nomics 52(3), 542–560.

Nagy, J. (1995). Increased duration and sample attrition in New York City’s rent controlled sector.
Journal of Urban Economics 38(2), 127–137.

Newey, W. K. and K. D. West (1987). A simple, positive semi-definite, heteroskedasticity and

autocorrelation consistent covariance matrix. Econometrica 55(3), 703–708.

Öst, C., B. Enström, and W. M. Söderberg (2013). Household allocation and spatial distribution in
a market under ("soft") rent control. Royal Institute of Technology Working Paper 5.

Oswald, A. J. (2009). The housing market and Europe’s unemployment: A non-technical paper. In
C. Van Ewijk and M. Van Leuvensteijn (Eds.), Homeownership and the Labour Market in Europe,
Chapter 2, pp. 43–52. Oxford University Press.

Pagano, M. (1990). Discussion. Economic Policy 5(11), 383–387.

Partridge, M. D. and D. S. Rickman (1997). The dispersion of US state unemployment rates: The
role of market and non-market equilibirum factors. Regional Studies 31(6), 593–606.

Pissarides, C. (2000). Equilibrium unemployment theory. MIT press.

Porter, M. (1998). Competitive advantage of nations. Free press.

Regeer, W. and G. Van Daalen (2011). Eén op de drie woningen eigendom van woningcorpo-
ratie. CBS Webmagazine. downloaded from http://www.cbs.nl/nl-NL/menu/themas/
bouwen-wonen/publicaties/artikelen/archief/2011/2011-3520-wm.htm at 01-
12-2012.

Roback, J. (1982). Wages, rents, and the quality of life. The Journal of Political Economy 90(6),
1257–1278.

Romer, P. (1986). Increasing returns and long-run growth. The Journal of Political Economy 94(5),
1002–1037.
244 References

Romijn, G. and P. Besseling (2008). De economische effecten van regulering and subsidiëring van de
huurwoningmarkt (165 ed.). CPB document. CPB, Centraal Planbureau.

Scanlon, K. and C. Whitehead (2007). Social housing in Europe. LSE London.

Scanlon, K. and C. Whitehead (2008). Social housing in Europe II. LSE London.

Simmons-Mosley, T. X. and S. Malpezzi (2006). Household mobility in New York City’s regulated
rental housing market. Journal of Housing Economics 15(1), 38–62.

Sims, D. P. (2011). Rent control rationing and community composition: Evidence from Mas-
sachusetts. The BE Journal of Economic Analysis & Policy 11(1), 1935–1682.

Sinai, T. and N. S. Souleles (2005). Owner-occupied housing as a hedge against rent risk. The
Quarterly Journal of Economics 120(2), 763–789.

Svarer, M., M. Rosholm, and J. R. Munch (2005). Rent control and unemployment duration. Journal
of Public Economics 89(11), 2165–2181.

Tobler, W. (1970). A computer movie simulating urban growth in the Detroit region. Economic
Geography 46(2), 234–240.

Turner, B. and S. Malpezzi (2003). A review of empirical evidence on the costs and benefits of rent
control. Swedish Economic Policy Review 10(1), 11–60.

Van Daalen, G., W. Regeer, and S. Janssen-Jansen (2012). Ruim een kwart van de huurders van
corporatiewoningen verdient meer dan 33 duizend euro. CBS Webmagazine. downloaded
from http://www.cbs.nl/nl-NL/menu/themas/bouwen-wonen/publicaties/
artikelen/archief/2012/2012-3652-wm.htm at 04-11-2013.

Van Duijn, M. (2013). Location Choice, Cultural Heritage and House Prices. Ph. D. thesis, Vrije
Universiteit, Amsterdam.

Van Ewijk, C. and M. Van Leuvensteijn (2009). Homeownership and the labour market in Europe.
OUP Oxford.

Van Ommeren, J. and M. De Graaf-Zijl (2013). Estimating household demand for housing at-
tributes in rent-controlled markets. Journal of Housing Economics 22(1), 11 – 19.

Van Ommeren, J. and M. Koopman (2011). Public housing and the value of apartment quality to
households. Regional Science and Urban Economics 41(3), 207–213.

Van Vuuren, A. (2007). The impact of homeownership on unemployment in the Netherlands.

CPB Discussion Paper 86.

Vermeulen, W. (2008). Essays on Housing Supply, Land Use Regulation and Regional Labour Markets.
Ph. D. thesis, Vrije Universiteit, Amsterdam.

Vermeulen, W. and J. Van Ommeren (2009). Compensation of regional unemployment in housing

markets. Economica 76(301), 71–88.
References 245

Wang, Z. and J. Kinsey (1994). Consumption and saving behavior under strict and partial ra-
tioning. China Economic Review 4(1), 83–100.

White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for
heteroskedasticity. Econometrica 48(4), 817–838.
Samenvatting

De economie van sociale woningbouw

Sociale huurwoningen stellen mensen in staat om een kwalitatief goede wo-
ning te huren tegen een betaalbare huur. Om ervoor te zorgen dat de huur
betaalbaar blijft, wordt deze gereguleerd. Veel economen menen dat huurre-
gulering nadelige gevolgen heeft. Huurregulering vermindert de welvaart.
Ook zou het leiden tot een vermindering van het aantal sociale huurwonin-
gen en een verschraling van hun kwaliteit. Daarnaast profiteren de bewoners
van sociale huurwoningen van een huur die lager is dan de markthuur. Om-
dat ze dit voordeel verliezen als ze naar een koopwoning verhuizen, zouden
bewoners van sociale huurwoningen minder vaak verhuizen. Dit zou hun
arbeidsmarktpositie kunnen verzwakken en de duur van werkloosheid ver-
lengen.
Het is echter onduidelijk of deze onbedoelde economische gevolgen van
sociale woningbouw optreden binnen Nederland, onder andere omdat de re-
gelgeving op de woningmarkt verder gaat dan enkel het reguleren van de
huur. Zo voorziet de institutionele context van de Nederlandse woningmarkt
bijvoorbeeld in afspraken over het aantal en de kwaliteit van huurwoningen,
waardoor er relatief veel sociale huurwoningen zijn en de kwaliteit goed is
(zie ter illustratie de Figuren 1.1 en 1.2).1 Daarnaast kan een verminderde
arbeidsmarktmobiliteit ook leiden tot een grotere bereidheid om te pendelen
tussen huis en werkplek.
1 In de literatuur wordt naar dit soort institutionele arrangementen verwezen als ‘tweede

generatie huurregulering’ ter onderscheid van een ‘eerste generatie huurplafond’ (Arnott, 1995;
Lind, 2001). Dit soort tweede generatie arrangementen zijn divers (Fallis1988; Lind 2001; Scanlon
en Whitehead, 2007, 2008).

247
248 Samenvatting

Daarom wordt in dit proefschrift onderzocht hoe de sociale woningbouw

de lokale economie beïnvloedt en richt zich daarbij met name op gevolgen
voor de arbeidsmarkt. Het onderzoek bestaat uit een micro-economisch en
macro-economisch deel. In hoofdstuk twee wordt een micro-economisch ka-
der ontwikkeld dat de keuze om in een sociale huurwoning te wonen ver-
klaart uit locatievoordelen. Ook worden de welvaartseffecten hiervan geana-
lyseerd. In hoofdstuk drie, vier en vijf wordt dit model getest: Ik onderzoek
in hoofdstuk drie of huurders in een sociale huurwoning een ander consump-
tiepatroon hebben. In hoofdstuk vier wordt onderzocht of het wonen in een
sociale huurwoning de verhuisgeneigdheid beïnvloedt. Hierbij wordt onder-
scheid gemaakt tussen verhuizingen binnen en uit de sociale huursector. In
hoofdstuk vijf wordt gekeken of het verschil in inkomen tussen huurders van
sociale huurwoningen en huiseigenaren verklaard wordt door de zelfselectie
van mensen met een laag inkomen in sociale huurwoningen of dat het type
woning van invloed is op de hoogte van het inkomen.
In het macro-economische deel worden de gevolgen van sociale woning-
bouw voor de opbouw van de lokale beroepsbevolking besproken. Ik ontwik-
kel een theoretisch kader in hoofdstuk zes dat laat zien hoe loononderhande-
lingen op de arbeidsmarkt en het reguleren van huurwoningen zorgen voor
prijsrigiditeiten, die elkaar kunnen versterken of verzwakken afhankelijk van
het type werknemer dat toegang heeft tot huurregulering. In hoofdstuk zeven
wordt getest of de grootte van de sociale huurwoningvoorraad van invloed is
op het percentage hoogopgeleide werknemers in een gebied.
Deze onderwerpen worden onderzocht vanuit een drietal invalshoeken.
Ten eerste wordt er tijdens het onderzoek rekening mee gehouden dat de po-
sitie op de arbeidsmarkt en de woningmarkt niet onafhankelijk van elkaar
zijn. Mensen hebben verschillende waargenomen kenmerken (zoals leeftijd,
geslacht, gezinssituatie en type baan), maar hebben ook verschillende niet-
waargenomen, ‘zachte’ kenmerken zoals preferenties, vaardigheden en ka-
rakter. Juist deze ‘zachte’ kenmerken zouden heel belangrijk kunnen zijn
voor zowel de zelfselectie in een sociale huur- of koopwoning als de positie
op de arbeidsmarkt en een gedegen analyse dient hiervoor te corrigeren. In
Samenvatting 249

dit proefschrift probeer ik te doen door middel van propensity score matching2 ,
paneldata-schatters en het gebruik van instrumentele variabelen.
Een tweede invalshoek is dat locaties van elkaar verschillen. Economische
activiteit zoals banen en het type bedrijf en industrie zijn niet gelijkmatig ver-
deeld over het land. Dit geldt ook voor locatievoordelen, zoals lokale publieke
goederen (bijvoorbeeld parken, natuurreservaten, open ruimte) en consumen-
tenvoordelen zoals de aanwezigheid van een historisch stadscentrum of re-
creatiemogelijkheden. De nabijheid van zulke locatievoordelen wordt verdis-
conteerd in de marktprijs van woningen, maar niet in de gereguleerde huur.3
Hierdoor profiteren de bewoners van sociale huurwoningen in het economi-
sche centrum meer van huurregulering dan bewoners van sociale huurwo-
ningen in de economische periferie. Dit heeft als gevolg dat de economische
effecten van sociale woningbouw regionaal verschillen.
Een derde invalshoek betreft de ruimtelijke afhankelijkheid van (met name)
geaggregeerde economische variabelen. Ruimtelijke afhankelijkheid wordt
samengevat in de wet van Tobler: “Alles is gerelateerd aan alles, maar zaken
dichterbij zijn meer gerelateerd dan zaken ver weg” (Tobler, 1970). De vele
banen in Amsterdam verhogen bijvoorbeeld niet alleen de woningprijzen in
Amsterdam, maar ook die in nabijgelegen gebieden. Ruimtelijk econometri-
sche technieken worden gebruikt om voor de ruimtelijke afhankelijkheid van
variabelen te corrigeren.

Belangrijkste conclusies

Micro-economisch kader

Sociale huurwoningen worden niet-willekeurig bewoond. Huurders van so-

ciale huurwoningen met een hoog inkomen genieten meer huurvoordeel (Gy-
ourko and Linneman, 1989). Ook wonen er meer huishoudens met een hoog
inkomen in sociale huurwoningen als de huizenprijs hoog is (Van Daalen et
2 Het construeren van een homogene steekproef uit twee verschillende groepen door het kop-

pelen van subjecten in groep a aan vergelijkbare subjecten in groep b op basis van de voorspelde
kans om in groep a (of b) te horen.
3 Dit is recentelijk aangepast. Sinds juli 2011 mag de gereguleerde huur in zogeheten schaars-

tegebieden verhoogd worden met 25 punten als een nieuw contract wordt aangegaan.
250 Samenvatting

al., 2012). In hoofdstuk twee wordt een micro-economisch kader gepresen-

teerd dat deze zelfselectie in sociale huurwoningen verklaart uit locatievoor-
delen: Omdat een sociale huurwoning de bewoner in staat stelt om deze lo-
catievoordelen gratis te consumeren (ze worden immers niet verwerkt in de
huur) blijven huishoudens met een hoog inkomen in sociale huurwoningen
zitten in gebieden met veel locatievoordelen. Hierdoor wordt het bewonen of
verlaten van een sociale huurwoning een keuze die afhangt van de locatie.
Het micro-economisch kader veronderstelt dat huiseigenaren hun nut maxi-
maliseren door de consumptie van locatievoordelen, de kwaliteit van de wo-
ning en alle andere goederen (composite good). Verder kan een ongecontro-
leerde huurwoning iedere gewenste kwaliteit hebben, maar is de kwaliteit
van een sociale huurwoning aan een maximum gebonden. Uit het micro-
economisch kader volgt dat huurregulering de verdeling van het budget over
de goederen beïnvloedt. Ook illustreert het dat de werknemers opgedeeld
kunnen worden in huiseigenaren, en gerestricteerde en niet-gerestricteerde
bewoners van sociale huurwoningen. Niet-gerestricteerde huurders van so-
ciale huurwoningen kunnen de ideale woningkwaliteit consumeren4 , terwijl
gerestricteerde huurders liever een sociale huurwoning van betere kwaliteit
zouden consumeren. Echter, zij zijn beter af in een sociale huurwoning, om-
dat zij het geld dat ze uitsparen aan huur kunnen besteden aan de overige
goederen. De groep werknemers met het hoogste inkomen besluit om onge-
controleerde woningen te huren.
De hoogte van het inkomen waarop men besluit om een ongecontroleerde
woning te huren neemt toe met de locatievoordelen in het gebied, hetgeen
de empirische verbanden tussen het inkomen van huurders van sociale huur-
woningen en de woningprijzen en het huurvoordeel verklaart. Ook laat het
micro-economisch kader zien dat de zelfselectie van werknemers met een
hoog inkomen in sociale huurwoningen niet tegengegaan kan worden met
selectie aan de poort op basis van inkomen als het inkomen aan (positieve)
verandering onderhevig is.
Hoofdstuk twee kan gebruikt worden om de welvaartseffecten van de al-
locatie van sociale huurwoningen aan te geven. De bijdrage aan het welvaarts-
4 Namelijk die woningkwaliteit die het hoogst haalbare nut oplevert gegeven het inkomen.
Samenvatting 251

verlies van huurregulering van de sociale woningbouw is groter voor werkne-

mers met een hoog inkomen dan voor werknemers met een laag inkomen. Dit
impliceert dat het welvaartsverlies van het sociale woningbouw verminderd
kan worden door sociale huurwoningen te herverdelen van bewoners met een
hoog inkomen naar bewoners met een laag inkomen. Deze maatregel is een
second best maatregel vanuit het oogpunt van efficiëntie: het welvaartsverlies
kan er niet door verdwijnen. Echter, het is mogelijk een maatregel die politiek
beter haalbaar is dan de first best welvaartsverhogende beleidsaanbeveling om
de sociale huursector af te schaffen (al dan niet met een compensatieregeling
voor de benadeelden).

Consumptiepatronen verschillen met type woning

Er zijn meerdere redenen om te veronderstellen dat consumptiepatronen van

huiseigenaren anders zijn dan de consumptiepatronen van bewoners van so-
ciale huurwoningen. Bewoners van een sociale huurwoning delen niet in de
groei (of daling) van het vermogen die veroorzaakt wordt door prijsfluctua-
ties op de woningmarkt. Daarnaast heeft het micro-economisch kader laten
zien dat huiseigenaren en bewoners van sociale huurwoningen hun budget
anders over de goederen verdelen.
In hoofdstuk drie test ik of verschillen in consumptie tussen huiseigenaren
en huurders in sociale huurwoningen inderdaad veroorzaakt worden door-
dat huizenprijzen het vermogen beïnvloeden (de ‘vermogenshypothese’) of
dat er sprake is van gelijktijdige beïnvloeding van consumptie en woningprij-
zen door eenzelfde variabele (de ‘gemeenschappelijke-factorhypothese’). Ik
schat daartoe een levensloop-consumptiemodel met behulp van data van het
Budget Onderzoek verzameld door het CBS over de periode 1980 tot en met
1999. De nationale woningprijsindex is gebruikt om te testen via welke hy-
pothese woningprijzen consumptie beïnvloeden.5 Beide hypothesen worden
van elkaar onderscheiden door gebruik te maken van de interactie van wo-
ningprijzen met leeftijd volgens de methode van Attanasio et al. (2009). Als
5 Het is hiermee niet mogelijk om na te gaan hoe lokale woningprijzen consumptie beïnvloe-
den.
252 Samenvatting

een controle op de robuustheid van de resultaten pas ik ook propensity score

matching toe.
De schattingsresultaten leveren bewijs voor de gemeenschappelijke-factor-
analyse: Het marginale effect van woningprijzen op consumptie is positief en
daalt met leeftijd. De resultaten van de propensity score matching bevestigen
deze conclusie. Dit zou betekenen dat het huidige lage niveau van de binnen-
landse besteding niet veroorzaakt wordt door de crisis op de woningmarkt,
maar dat beide beïnvloed worden door een gemeenschappelijke factor.
Tenslotte bestudeer ik in hoofdstuk drie het effect van het type woning op
de allocatie van het budget over consumptiegoederen. Gebaseerd op het Bud-
get Onderzoek 1980-1999 worden Engelcurves geschat voor de categorieën
‘alledaagse goederen’, ‘wonen’, ‘recreatie’ en ‘alcohol en tabak’. Deze vier ca-
tegorieën vormen 80 procent van de totale bestedingen.
Uit de geschatte Engelcurves volgt dat bewoners van sociale huurwonin-
gen minder uitgeven aan ‘wonen’ en meer aan ‘alledaagse goederen’, ‘recre-
atie’ en ‘alcohol en tabak’, maar het verschil is afgenomen over tijd. Deze
conclusie volgt ook uit de robuustheidsanalyse. Tenslotte blijkt uit de ana-
lyse dat bewoners van sociale huurwoningen, die meer op huiseigenaren lij-
ken wat betreft karakteristieken (zoals leeftijd, gezinssamenstelling, inkomen,
etc.), een groter gedeelte van hun budget besteden aan ‘recreatie’ en minder
aan ‘wonen’ vergeleken met de overige bewoners van sociale huurwoningen.
Dit is in overeenstemming met het micro-economisch kader uit hoofdstuk
twee. De grootte van het gevonden effect is echter beperkt.

Huurregulering beïnvloedt de verhuismobiliteit

Het micro-economisch kader voorspelt dat het inkomensniveau waarbij be-

woners van een sociale huurwoning naar de koopsector verhuizen, toeneemt
als het huurvoordeel, het verschil tussen de markthuur en de gereguleerde
huur, toeneemt. Deze hypothese wordt getest in hoofdstuk vier door te schat-
ten hoe het huurvoordeel de kans om binnen of uit de sociale huursector te
verhuizen beïnvloedt.
Samenvatting 253

Eerder empirisch onderzoek heeft uitgewezen dat het wonen in een sociale
huurwoning de verhuismobiliteit vermindert.6 Er is echter geen onderscheid
gemaakt tussen transities binnen en uit de sociale huursector. De verwachting
is dat huurregulering beide transities anders beïnvloedt, omdat bewoners hun
huurvoordeel verliezen als ze naar een koopwoning verhuizen, maar deze
deels behouden (en zelfs kunnen vergroten) als ze verhuizen naar een sociale
huurwoning.
De schattingen zijn gebaseerd op een paneldataset gebaseerd op admini-
stratieve bestanden die beschikbaar zijn gesteld door het CBS. De data bevat-
ten gegevens over het type en de waarde van de woning. Ook bevatten ze
kenmerken van het huishouden, de baan en de lokale woningmarkt. Huur-
voordeel wordt gemeten op basis van de mediane huur in een wijk gedeeld
door de individuele WOZ-waarde van de woning.
De resultaten laten zien dat het effect van het huurvoordeel op de mobili-
teit met name groot is voor transities binnen de sociale huursector, maar dat
het effect op transities naar de koopsector beperkt is. In overeenstemming met
het micro-economisch kader verhuizen alleen bewoners van een sociale huur-
woning met een hoog inkomen minder snel naar de koopsector. Dit geeft aan
dat de verstoring van de koopsector door huurregulering beperkt is en dat de
negatieve effecten zich concentreren binnen de sociale huursector zelf.

Het type woning heeft geen invloed op het loon

Het gemiddelde loon van huiseigenaren is 25 procent hoger dan dat van be-
woners van sociale huurwoningen. Dit verschil in loon kan komen door zelf-
selectie van werknemers met een laag loon in sociale huurwoningen (zie hoofd-
stuk twee). Echter, gezien de verminderde mobiliteit van bewoners in een
sociale huurwoning (zie hoofdstuk drie), is het mogelijk dat het loonverschil
tussen beide groepen werknemers veroorzaakt wordt door het woningtype.7
Empirische studies tonen aan dat bewoners van sociale huurwoningen een
minder goede arbeidsmarktpositie hebben dan huiseigenaren: ze werken va-
6 Gyourko en Linneman (1989), Ault et al. (1994); Nagy (1995), Munch en Svarer (2002),
Simmons-Mosley en Malpezzi (2006) vinden empirisch bewijs hiervoor gebaseerd op schattings-
methoden zoals OLS, logistische regressies en duurmodellen.
7 Eigenwoningbezit vermindert ook de mobiliteit van werknemers. Dit kan ook invloed heb-

ben op de baanmobiliteit van en werkloosheid onder huiseigenaren.

254 Samenvatting

ker in de lokale arbeidsmarkt en zijn vaker werkloos. Er zijn echter geen stu-
dies die onderzocht hebben of deze negatieve arbeidsmarktpositie doorwerkt
in het verdiende loon. In twee studies is onderzoek gedaan naar het loon van
huiseigenaren ten opzichte van alle huurders: Munch et al. (2008) gebruiken
een panel gebaseerd op administratieve bestanden uit Denemarken en conclu-
deren dat huiseigenaren een vijf procent hoger loon verdienen, ceteris paribus.
Daartegenover staat het onderzoek van Coulson en Fisher (2009) gebaseerd op
survey data. Zij concluderen dat huiseigenaren in de Verenigde Staten gemid-
deld 30 procent minder loon verdienen dan huurders, ceteris paribus. Beide
onderzoeken verschillen dus significant in de richting en de orde van grootte
van het effect.
In hoofdstuk vijf wordt onderzocht of eigenwoningbezit of het wonen in
een sociale huurwoning van invloed is op het loon door deze typen werkne-
mers te vergelijken met werknemers in een private huurwoning. De schattin-
gen zijn gebaseerd op een groot paneldatabestand gebaseerd op administra-
tieve gegevens verzameld over de periode 2006 tot 2008. Omdat meer dan één
miljoen werknemers gevolgd worden gedurende drie jaar, is het mogelijk om
te corrigeren voor de selectie van werknemers met veel of weinig vaardighe-
den (‘worker skills’) in sociale huur- of koopwoningen.
De schattingsresultaten geven aan dat het corrigeren voor de vaardighe-
den van werknemers uitermate belangrijk is: zonder de correctie voor de
vaardigheden van werknemers zou men concluderen dat bewoners van so-
ciale huurwoningen 17 procent minder verdienen dan inwoners van private
huurwoningen, doordat ze in een sociale huurwoning wonen. Idem dito zou
men concluderen dat huizenbezitters 7 procent meer verdienen dan huurders
van private woningen, doordat ze in een koopwoning wonen. Echter, na de
correctie voor vaardigheden van de werknemer is er geen bewijs dat het ver-
schil in inkomen tussen huurders in een sociale huurwoning of huiseigena-
ren (vergeleken met huurders van private huurwoningen) veroorzaakt wordt
door het type woning.
Samenvatting 255

Macro-economisch kader

Heterogene werknemers beconcurreren elkaar om dezelfde woningen. Hier-

door hebben veranderingen in de productiviteit van hoog- en laagopgeleide
werknemers invloed op de compositie van de beroepsbevolking: Een produc-
tiviteitsschok voor hoogopgeleide werknemers zal leiden tot een hogere vraag
naar deze werknemers. Gegeven de woningprijs veroorzaakt dit een instroom
van hoogopgeleide werknemers. De prijselasticiteit van het woningaanbod
bepaalt in hoeverre dit leidt tot veel hogere woningprijzen en een kleine in-
stroom van hoogopgeleide werknemers, of enigszins hogere woningprijzen
en een grote instroom.
De verandering in de woningprijzen beïnvloedt ook de locatiekeuze van
laagopgeleide werknemers. Een deel van hen zal de regio verlaten, omdat
de woonkosten stijgen terwijl hun inkomen niet verandert. Veranderingen in
het (opleidingspecifieke) loon en de woningprijs zijn dus van belang voor de
compositie van de lokale arbeidsmarkt en er vinden spillovers plaats tussen
beide typen arbeid via de prijselasticiteit van het woningaanbod.
Dit soort modellen van de arbeids- en woningmarkt is gebaseerd op de
premisse dat de arbeidsmarkt en woningmarkt perfect competitief zijn.8 In
hoofdstuk zes analyseren we hoe de beide markten elkaar beïnvloeden als
deze gekenmerkt worden door marktimperfecties in de vorm van prijsrigidi-
teiten.
Prijsrigiditeiten op de arbeidsmarkt worden gemodelleerd in de vorm van
nominale loonrigiditeiten die het gevolg zijn van nationale loononderhande-
lingen. Lokale arbeidsmarkten worden gekenmerkt door imperfecte koppe-
ling van vacatures aan werkzoekenden (labor market matching) zoals beschre-
ven in Pissarides (2000). Ik geef in hoofdstuk zes aan dat arbeidsmarktrigi-
diteiten de mobiliteit van werknemers beperkt als ruimtelijke verschillen in
werkloosheid gering zijn, of als het loon niet aangepast wordt na een produc-
tiviteitsschok. De verklaring is dat het verhuizen naar een andere regio in
deze gevallen geen hoger verwacht loon oplevert.
Prijsrigiditeiten op de arbeidsmarkt worden gemodelleerd in de vorm van
het recht om woningen tegen een vaststaand tarief te huren ongeacht de markt-
8 Zie Roback (1982); Moretti (2011) voor zulke modellen van de arbeids- en woningmarkt.
256 Samenvatting

huur (rent control housing vouchers). Ik toon aan dat het effect van deze rechten
op de compositie van de arbeidsmarkt (na een toename van de productiviteit
van hoogopgeleide werknemers) afhangt van de verdeling ervan over hoog-
en laagopgeleide werknemers: Als zij in het bezit zijn van met name laagop-
geleide werknemers, leiden ze tot een verminderde instroom van hoogopge-
leide werknemers en een verminderde uitstroom van laagopgeleide werkne-
mers. Als zij daarentegen in het bezit zijn van voornamelijk hoogopgeleide
werknemers, leiden deze rechten ertoe dat de instroom van hoogopgeleide
werknemers en de uitstroom van laagopgeleide werknemers groter is.
De analyse geeft aan dat het effect van rigiditeiten op de arbeids- en wo-
ningmarkt afhangt van het opleidingsniveau. In het algemeen versterken ri-
giditeiten op de arbeids- en woningmarkt elkaar na een toename van de pro-
ductiviteit van hoogopgeleide werknemers. Echter, prijsrigiditeiten op de ar-
beidsmarkt verhelpen deels de prijsrigiditeiten op de arbeidsmarkt als de pro-
ductiviteit van laagopgeleide werknemers verandert.

Sociale woningbouw beïnvloedt het opleidingsniveau van het

lokale arbeidsaanbod

Het macro-economisch kader geeft aan dat laagopgeleide werknemers in een

sociale huurwoning de stad niet verlaten als huizenprijzen stijgen door een
toename van hoogopgeleide werknemers. Daarnaast worden sociale huur-
woningen voornamelijk toegewezen aan huishoudens met een laag inkomen.
Dit zijn twee redenen om te veronderstellen dat sociale woningbouw in Ne-
derland leidt tot een lager percentage hoogopgeleide werknemers. Daar staat
tegenover dat het Nederlandse systeem van het toekennen van sociale huur-
woningen aan huishoudens met een laag inkomen (ten tijde van het tekenen
van het contract), niet kan verhinderen dat sociale huurwoningen op den duur
bewoond worden door werknemers met een hoog inkomen (zoals besproken
in hoofdstuk twee). In extremo kan deze zelfselectie van werknemers met een
hoog inkomen in sociale huurwoningen er voor zorgen dat er een beperkt of
zelfs geen effect is van de sociale woningbouw op het opleidingsniveau van
het lokale arbeidsaanbod.
Samenvatting 257

Het is van belang om te weten of sociale woningbouw gevolgen heeft voor

de opbouw van de lokale beroepsbevolking, omdat dit de positieve externa-
liteiten die het gevolg zijn van agglomeratievoordelen teniet kan doen.9 In
hoofdstuk zeven test ik of de sociale woningbouw van invloed is op het per-
centage hoogopgeleide werknemers in een gebied. Als dit het geval is, toont
dit aan dat de sociale woningbouw de lokale arbeidsmarkt beïnvloedt via een
voorheen niet bestudeerd kanaal.
De schattingen zijn gebaseerd op een longitudinaal panel van de veertig
COROPs in Nederland. De regionale variabelen zijn geconstrueerd op basis
van het WBO en WoON in de periode 1981 tot en met 2006. Deze zijn uitge-
breid met gegevens over de regionale productiviteit en werkgelegenheid voor
acht sectoren over de steekproefperiode.
Uit de schattingen blijkt dat sociale woningbouw een negatief effect heeft
op het percentage hoogopgeleiden in een COROP: Een vergroting van de so-
ciale huurwoningvoorraad met tien percentage punten verkleint het percen-
tage hoogopgeleide werknemers in de regio met 1.5 percentage punten, cete-
ris paribus. Dit effect is even groot als een stijging van de woningprijzen met
e8.800. Bij deze schattingen is gecorrigeerd voor het aantal banen in de CO-
ROP en omliggende COROPs, en de endogeniteit van het relatieve loon van
hoogopgeleide werknemers en de woningprijzen.
De grootte van het effect van sociale woningbouw op de opbouw van de
lokale arbeidsmarkt is beperkt. Ter illustratie: Als de sociale huurwoningsec-
tor in Utrecht gehalveerd wordt van 40 naar 20 procent, dan stijgt het per-
centage hoogopgeleiden in Utrecht met (slechts) 3 percentage punten tot 46
procent.

Beperkingen en richtingen voor vervolgonderzoek

In dit proefschrift worden enkele economische effecten van sociale woning-
bouw behandeld. Ik geef aan dat deze economische effecten subtieler zijn
dan het mantra “sociale huur vermindert de mobiliteit (naar de koopsector),
hetgeen slecht is voor de arbeidsmarkpositie” doet vermoeden. In mijn proef-
9 Zo toont Moretti (2004) aan dat de concentratie van hoogopgeleide werknemers in stedelijke

gebieden zorgt voor hogere lonen voor zowel hoog- als laagopgeleide werknemers.
258 Samenvatting

schrift wordt immers aangetoond dat met name transities binnen de huursec-
tor gereduceerd worden, terwijl transities van huur naar koop amper beïn-
vloed worden. Deze bevindingen vragen om dynamische, theoretische mo-
dellen van het verhuisgedrag van werknemers die de transities binnen en uit
de sociale huursector kunnen verklaren.
In mijn proefschrift wordt geen bewijs gevonden dat bewoners van sociale
huurwoningen een lager loon verdienen, ceteris paribus. Dit is opmerkelijk,
omdat bewoners van sociale huurwoningen minder mobiel zijn op de wo-
ningmarkt en vaker werkzaam zijn in de lokale arbeidsmarkt (ceteris paribus),
hetgeen hun positie op de arbeidsmarkt zou moeten verzwakken. Onderzoek
naar de relatie tussen woningtype en positie op de arbeidsmarkt heeft zich
voornamelijk gericht op de rol van eigenwoningbezit (Van Ewijk en Van Leu-
vensteijn, 2009). Een interessante uitzondering zijn De Graaf et al. (2009).
Zij concluderen dat in Nederland bewoners van sociale huurwoningen geen
slechtere arbeidsmarktpositie hebben dan bewoners van private huurwonin-
gen, ceteris paribus. Het is vooralsnog onduidelijk hoe het mogelijk is dat
bewoners van sociale huurwoningen wel minder mobiel zijn en vaker in de
lokale woningmarkt werken, zonder dat dit nadelige gevolgen heeft voor hun
loon of mobiliteit binnen de arbeidsmarkt.
Ook wordt in dit proefschrift aangegeven dat de allocatie van sociale huur-
woningen welvaartseffecten met zich meedraagt. Zo wordt in hoofdstuk twee
aangetoond dat de bijdrage aan het totale welvaartsverlies van sociale wo-
ningbouw groter is voor huishoudens met een hoog inkomen dan voor huis-
houdens met een laag inkomen. Dit impliceert dat een strengere toedeling
van sociale huurwoningen aan (alleen) huishoudens met een laag inkomen
leidt tot minder welvaartsverlies op de woningmarkt.10 Echter, het strenger
toekennen van sociale huurwoningen aan huishoudens met een laag inkomen
leidt tot een kunstmatige reductie van het percentage hoogopgeleiden, en ver-
groot daarmee de verstoring van huurregulering op de lokale arbeidsmarkt.
Toekomstig onderzoek zou zich kunnen richten op het analyseren en kwantifi-
ceren van deze uitruil van efficiëntie op de arbeidsmarkt en de woningmarkt.
10 Het toekenningsbeleid is aangescherpt sinds de Europese Commissie gebiedt dat 90 procent

van alle vrijgekomen sociale huurwoningen toegekend wordt aan huishoudens met een inkomen
lager dan e34.678 (prijspeil 2014). De EC doet dit om staatssteun tegen te gaan.
Samenvatting 259

De theoretische modellen in dit proefschrift modelleren het gedrag van

huurders van sociale huurwoningen en huurders van private huurwoningen.
Dit gebeurt, omdat het modelleren van eigenwoningbezit (en de daarmee sa-
menhangende gevolgen voor het vermogen in de toekomst) een dynamisch
en dus ingewikkelder model vereist. Hoewel het nuttig is om theoretische
modellen simpel te houden, zouden toekomstige modellen van de interactie
tussen de arbeidsmarkt en de woningmarkt zich kunnen richten op de gevol-
gen van het bezit van de woning voor het gedrag van huiseigenaren.
Er zijn verschillende aspecten van de Nederlandse woningmarkt die in dit
proefschrift wellicht onvoldoende aan het licht gekomen zijn. In het proef-
schrift wordt verklaard waarom, met name in de Randstad, werknemers met
een hoog inkomen in sociale huurwoningen wonen. Dit verklaard waarom
werknemers met een hoog inkomen meer profiteren van huurregulering dan
werknemers met een laag inkomen. De gevolgen die dit heeft voor de herver-
delende functie van de sociale woningbouw komen echter niet expliciet aan
bod.
Ook richt ik mij in dit proefschrift niet op de gevolgen van de individuele
huurtoeslag voor de vraag naar woningen. Koning en Ridder (1997) conclu-
deren dat de (toen nog) individuele huursubsidie de vraag naar woningdien-
sten vergrootte. Gegeven de grootte van het programma (zowel in het aantal
toegekende subsidies als het totale subsidiebedrag) is het interessant om na
te gaan of de huidige individuele huurtoeslag de vraag naar woningdiensten
beïnvloedt en of het gevolgen heeft voor het loon (armoedeval), de zelfselectie
van huurders in sociale huurwoningen en de mobiliteit van huishoudens.

Voor al het empirisch onderzoek geldt dat de kracht van de resultaten afhan-
kelijk is van de kwaliteit van de data. Dat geldt ook voor dit proefschrift. De
analyse in dit proefschrift kan in de toekomst uitgebreid worden door de ana-
lyse op een gedetailleerder niveau uit te voeren. Zo maakt de analyse van het
consumptiegedrag gebruik van gecombineerde cross-sectionele gegevensbe-
standen over een periode van twintig jaar en van een nationale woningprijsin-
dex. Met panel data zou gecorrigeerd kunnen worden voor niet waargenomen
‘zachte’ kenmerken van de werknemer (zoals preferenties en vaardigheden)
260 Samenvatting

en hoe die de selectie in een huur- of koopwoning beïnvloeden. Hoofdstuk

drie probeert dit te ondervangen door middel van propensity score matching.
Dit biedt echter minder zekerheid dan paneldata (en is een minder efficiënte
techniek). Met behulp van lokale woningprijsindices zou het effect van wo-
ningprijzen op vermogen nauwkeuriger geïdentificeerd kunnen worden.
In de hoofdstukken vier en vijf wordt gebruik gemaakt van panel data op
basis van administratieve gegevens die verzameld zijn over de periode 2006
tot en met 2008. Het zou interessant zijn om de gegevens van het panel over
een langere periode te meten. Dit komt de analyse ten goede, omdat er meer
observaties per individu zijn. Ook wordt het dan mogelijk om het effect van
de woningmarktcrisis te analyseren. Gedurende de crisis zijn de prijzen op
de woningmarkt voortdurend gedaald. Dit kan van invloed zijn op de transi-
ties binnen en uit de sociale huursector, zeker wanneer sociale huurwoningen
beschouwd worden als een ‘veilige haven’ waar men beschermd is tegen prijs-
schommelingen op de koopmarkt.
In hoofdstuk zeven wordt het effect van sociale woningbouw op de op-
bouw van de lokale beroepsbevolking bestudeerd op COROP-niveau. Een
lagere geografische schaal zou een nauwkeurigere analyse opleveren, om-
dat COROP-gebieden vanwege hun grootte aggregatienadelen ondervinden.
Daar staat tegenover dat een analyse op een lager aggregatieniveau vermoe-
delijk meer problemen kent door de ruimtelijke afhankelijkheid van de data
(bijvoorbeeld door pendel), maar hier kan voor gecorrigeerd worden door
middel van ruimtelijke econometrische technieken.

Beleidsimplicaties
In het proefschrift is aangetoond dat sociale woningbouw verschillende eco-
nomische effecten heeft die van belang zijn bij het ontwikkelen van beleid. Zo
is laten zien dat sociale woningbouw de consumptie en geografische mobili-
teit van huishoudens beïnvloedt zonder dat het gevolgen heeft voor het loon
dat verdiend wordt. Ook verlaagt het percentage sociale huurwoningen in
een regio het percentage hoogopgeleide werknemers, al is de grootte van het
effect beperkt.
Samenvatting 261

Het doel van de sociale woningbouw is om betaalbare woningen aan te bie-

den aan (voornamelijk) huishoudens met een laag inkomen. Ik heb laten zien
waarom het toekennen van sociale huurwoningen op basis van inkomen bij
het aangaan van het huurcontract niet kan voorkomen dat sociale huurwo-
ningen bewoond worden door huishoudens met een hoog inkomen. Dit heeft
tot gevolg dat er relatief minder huishoudens met een laag inkomen in sociale
huurwoningen wonen juist als de woningprijzen hoog zijn. Dit gaat tegen het
doel van sociale woningbouw in. Vanuit rechtvaardigheidsoverwegingen is
additioneel beleid daarom gewenst.
Deze conclusie wordt ook getrokken als sociale woningbouw vanuit het
perspectief van efficiëntie op de woningmarkt bekeken wordt. In het proef-
schrift wordt aangetoond dat de bijdrage van huishoudens met een hoog in-
komen aan het welvaartsverlies van sociale woningbouw groter is dan dat
van huishoudens met een laag inkomen. Het herverdelen van sociale huur-
woningen van rijke inkomens naar lage inkomens verkleint daarom het wel-
vaartsverlies dat optreedt op de woningmarkt, met name als dit gebeurt waar
woningprijzen hoog zijn.11 De meeste bewoners van sociale huurwoningen
met een hoog inkomen wonen op die plekken waar woningprijzen het hoogst
zijn. Er zullen dan dus meer sociale huurwoningen beschikbaar komen voor
huishoudens met een laag inkomen, in die buurten waar de transities binnen
de sociale huursector het laagst zijn en het betreden van de sociale huursector
het moeilijkst is.
Dus het herverdelen van sociale huurwoningen van huishoudens met een
hoog naar huishoudens met een laag inkomen kan gezien worden als recht-
vaardig en verkleint bovendien het welvaartsverlies op de woningmarkt. De
Nederlandse regering heeft verschillende maatregelen genomen om het be-
wonen van sociale huurwoningen door huishoudens met een hoog inkomen
tegen te gaan.12 Volgens het theoretisch kader in hoofdstuk twee verminderen
deze maatregelen waarschijnlijk het aandeel huishoudens met een hoog inko-
11 Door het herverdelen van sociale huurwoningen van huishoudens met een hoog naar huis-

houdens met een laag inkomen kan het welvaartsverlies nooit elimineren. Dit kan alleen als
sociale woningbouw afgeschaft wordt. De winnaars van deze maatregelen kunnen de verliezers
compenseren. Echter, zo’n ingreep is politiek waarschijnlijk niet haalbaar.
12 Per 1 juli 2011 is de maximum toegestane huur in de zogeheten schaarstegebieden verhoogt.

Per 1 juli 2013 zijn de huren in de sociale huursector inkomensafhankelijk. De huur mag verhoogd
worden met de inflatie plus 2,5, 3,5 en 4 procent voor huurders met een laag, midden- of hoog
inkomen. Huishoudens met een middeninkomen verdienen tussen de e33,614 en e43,000. Ten-
262 Samenvatting

men dat in een sociale huurwoning woont. Echter, zolang de gereguleerde

huur lager is dan de markthuur en inkomenseisen alleen bij het betreden van
de sociale huurwoning gesteld worden, zullen er altijd relatief minder huis-
houdens met een laag inkomen in een sociale huurwoning wonen in gebieden
waar de woningprijs hoog is.

Men dient zich er echter bewust van te zijn dat het strenger toekennen van so-
ciale huurwoningen aan huishoudens met een laag inkomen de verstoring van
de lokale arbeidsmarkt door sociale woningbouw vergroot doordat het aan-
deel hoogopgeleide werknemers hierdoor verkleind wordt. Deze ‘uitruil van
efficiëntie’ is een belangrijk mechanisme waarvan beleidsmakers zich bewust
moeten zijn als zij beleid maken met betrekking tot de sociale woningbouw.
De uitruil van efficiëntie impliceert dat beleid dat ervoor zorgt dat meer
sociale huurwoningen bewoond worden door huishoudens met een laag in-
komen gepaard dient te gaan met een verkleining van de sociale huurwo-
ningvoorraad: Het strenger toekennen van sociale huurwoningen aan huis-
houdens met een laag inkomen verkleint het welvaartsverlies op de woning-
markt, terwijl het verkleinen van de sociale woningvoorraad de verstoring op
de arbeidsmarkt vermindert. Als dat laatste niet gebeurt en de grootte van
de sociale woningvoorraad intact blijft, dan volgt uit de uitruil van efficiën-
tie dat de verstoring op de arbeidsmarkt vergroot wordt. Ik wil benadrukken
dat door de huidige slechte toedeling van sociale huurwoningen aan huishou-
dens met een laag inkomen, deze huishoudens niet noodzakelijkerwijs slech-
ter af zijn als men de sociale huurwoningvoorraad verkleint en tegelijkertijd
de toekenning aan lage inkomens verbetert. In dit geval kan een efficiëntere
woniningmarkt gerealiseerd worden, zonder dat de mensen, voor wie sociale
woningbouw bedoeld is, benadeeld worden.

slotte streeft de regering er naar om per 1 juli 2014 de huur van sociale huurwoningen afhankelijk
te maken van de WOZ-waarde van de woning, zie BZK (2013).
Curriculum Vitae

Mark Kattenberg obtained his master degree in economics at the Free Uni-
versity in Amsterdam in 2008 (cum laude) and his research master degree in
Multidisciplinary Economics at Utrecht University in 2010. In September of
that year he started his PhD at the chair of Applied Econometrics of Utrecht
University School of Economics, the project being cofinanced by the Nether-
lands Ministry of the Interior and Kingdom Relations.

Mark Kattenberg’s research interests are in applied econometrics and policy

evaluation. As of January 2014 he is employed by the CPB Netherlands Bu-
reau for Economic Policy Analysis in The Hague.

263
TKI dissertation series
USE 001 Bastian Westbrock (2010): Inter-firm Networks: Economic and Sociological Perspectives.
USE 002 Yi Zhang (2011): Institutions and International Investments: Evidence from
China and Other Emerging Markets.
USE 003 Ryan van Lamoen (2011): The Relationship between Competition and Innovation:
Measuring Innovation and Causality.
USE 004 Martijn Dröes (2011): House Price Uncertainty in the Dutch Owner-Occupied
Housing Market.
USE 005 Thomas van Huizen (2012): Behavioural Assumptions in Labour Economics: Analysing
Social Security Reforms and Labour Market Transitions.
USE 006 Martijn Boermans (2012): International Entrepreneurship and Enterprise Development.
USE 007 Joras Ferwerda (2012): The Multidisciplinary Economics of Money Laundering.
USE 008 Federico D’Onofrio (2013): Observing the Country: a History of Italian Agricultural
Economics, 1900-1930.
USE 009 Saraï Sapulete (2013): Works Council Effectiveness: Determinants and Outcomes.
USE 010 Britta Hoyer (2013): Network Formation under the Threat of Disruption.
USE 011 Coen Rigtering (2013): Entrepreneurial Orientation: Multilevel Analysis and
Consequenses.
USE 012 Beate Cesinger (2013): Context and Complexity of International Entrepreneurship as
a Field of Research.
USE 013 Jan de Dreu (2013): Empirical Essays on the Governance of Financial Institutions.
USE 014 Lu Zhang (2013): Industrial Specialization: Determinants, Processes and
Consequences.
USE 015 Matthias Filser (2013): Strategic Issues in Entrepreneurship and Family Business
Research.
USE 016 Mikko Pohjola (2013): A Compilation of Studies on Innovation in Firms: Persistence,
Strategies, and Capabilities.
USE 017 Han-Hsin Chang (2013): Heterogeneity in Development.
USE 018 Suzanne Heijnen (2014): Analyses of Sickness Absence.
USE 019 Mark Kattenberg (2014): The Economics of Social Housing: Implications for Welfare,
Consumption and Labor Market Composition.

265