write a phd proposal about the real estate developers and urban development

Research Collection

Doctoral Thesis

The role of real estate developers in the context of land use development and transport

Author(s): Zöllig Renner, Christof

Publication Date: 2014

Permanent Link: https://doi.org/10.3929/ethz-a-010412016

Rights / License: In Copyright - Non-Commercial Use Permitted

This page was generated automatically upon download from the ETH Zurich Research Collection. For more information please consult the Terms of use.

ETH Library

DISS. ETH NO. 22412

THE ROLE OF REAL ESTATE DEVELOPERS IN THE CONTEXT OF LAND USE DEVELOPMENT

AND TRANSPORT

A thesis submitted to attain the degree of

DOCTOR OF SCIENCES of ETH ZURICH

(Dr. sc. ETH Zurich)

presented by

CHRISTOF ZÖLLIG RENNER

MSc ETH Zurich

born on 06.01.1981

citizen of St. Gallen, Switzerland

accepted on the recommendation of

Prof. Dr. Kay W. Axhausen, examiner Prof. Dr. Paul Waddell, co-examiner

2014

To Nina

* * *

In Our Darkest Hour In My Deepest Despair Will You Still Care? Will You Be There?

In My Trials And My Tribulations Through Our Doubts

And Frustrations In My Violence

In My Turbulence Through My Fear

And My Confessions In My Anguish And My Pain

Through My Joy And My Sorrow In The Promise Of Another Tomorrow

I’ll Never Let You Part For You’re Always In My Heart.

* * *

Michael Jackson (1991)

2

Contents

Abstract 13

Zusammenfassung 15

Acknowledgement 19

1 Introduction 21 1.1 Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 21 1.2 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . 23 1.3 Approach . . . . . . . . . . . . . . . . . . . . . . . . . 23 1.4 Scope of dissertation . . . . . . . . . . . . . . . . . . . 24 1.5 Document guide . . . . . . . . . . . . . . . . . . . . . . 25

2 Theoretical background and review of literature 27 2.1 Land use development . . . . . . . . . . . . . . . . . . 27

2.1.1 Urban systems . . . . . . . . . . . . . . . . . . 28 2.1.2 Subsystems . . . . . . . . . . . . . . . . . . . . 29 2.1.3 Interaction of subsystems . . . . . . . . . . . . . 32

2.2 Models of land use development . . . . . . . . . . . . . 35 2.2.1 Updated model systematic of land use transport

interaction . . . . . . . . . . . . . . . . . . . . 36 2.2.2 Description of land use transport interaction (LUTI)

models . . . . . . . . . . . . . . . . . . . . . . 40 2.3 Real estate development . . . . . . . . . . . . . . . . . 51

2.3.1 Spatial development process . . . . . . . . . . . 51 2.3.2 Models of real estate development . . . . . . . 53 2.3.3 Developer types . . . . . . . . . . . . . . . . . . 76 2.3.4 Computational real estate development models

within LUTI models . . . . . . . . . . . . . . . 84 2.4 Conclusions from theory . . . . . . . . . . . . . . . . . 87

Contents

3 Methods 89 3.1 Expert interviews . . . . . . . . . . . . . . . . . . . . . 89 3.2 Agent-based simulation . . . . . . . . . . . . . . . . . . 90 3.3 Discrete choice modelling . . . . . . . . . . . . . . . . 91

3.3.1 A historical introduction . . . . . . . . . . . . . 91 3.3.2 The reference: the basic multinomial logit (MNL) 92 3.3.3 Models considering a heterogeneous structure of

alternatives . . . . . . . . . . . . . . . . . . . . 94 3.3.4 Models considering heterogeneity of preferences 95 3.3.5 Multinomial probit (MNP) . . . . . . . . . . . . 97 3.3.6 Estimation methods . . . . . . . . . . . . . . . . 97 3.3.7 Practical considerations . . . . . . . . . . . . . 100

3.4 Conclusions from methods . . . . . . . . . . . . . . . . 105

4 Analysing Zurich’s real estate development 107 4.1 Theoretical framework and explanatory

strategy . . . . . . . . . . . . . . . . . . . . . . . . . . 107 4.1.1 Different developer behaviours . . . . . . . . . . 109 4.1.2 Expected consequences for spatial development . 111

4.2 Developers and development projects in Zurich . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 4.2.1 Real estate market segmentation . . . . . . . . . 112 4.2.2 Demand . . . . . . . . . . . . . . . . . . . . . . 115 4.2.3 Supply . . . . . . . . . . . . . . . . . . . . . . 115 4.2.4 Products . . . . . . . . . . . . . . . . . . . . . 122 4.2.5 Developers . . . . . . . . . . . . . . . . . . . . 126

4.3 Expert interviews with real estate developers in the Can- ton of Zurich . . . . . . . . . . . . . . . . . . . . . . . 131 4.3.1 Preparation . . . . . . . . . . . . . . . . . . . . 131 4.3.2 Recruitment of interviewees . . . . . . . . . . . 133 4.3.3 Conducting the interviews . . . . . . . . . . . . 134 4.3.4 Analysis of interviews . . . . . . . . . . . . . . 134 4.3.5 Results of expert interviews . . . . . . . . . . . 136 4.3.6 Conclusions for discrete choice modelling . . . . 146

4.4 Discrete choice analysis of real estate development . . . . . . . . . . . . . . . . . . . . . . . . 147 4.4.1 Data preparation . . . . . . . . . . . . . . . . . 147 4.4.2 Use of projects as observations rather than new

buildings . . . . . . . . . . . . . . . . . . . . . 149 4.4.3 Discrete segmentation by developer type . . . . 158 4.4.4 Purpose specific models . . . . . . . . . . . . . 167

4

Contents

4.5 Conclusions from land development analysis . . . . . . 177

5 Simulating Zurich’s land development 179 5.1 Land use transport interaction simulation of the Canton

of Zurich . . . . . . . . . . . . . . . . . . . . . . . . . 179 5.1.1 Data preparation . . . . . . . . . . . . . . . . . 180 5.1.2 Models . . . . . . . . . . . . . . . . . . . . . . 183 5.1.3 Calibration . . . . . . . . . . . . . . . . . . . . 188

5.2 Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . 195 5.2.1 Scenario definition . . . . . . . . . . . . . . . . 196 5.2.2 Results of scenario run . . . . . . . . . . . . . . 196 5.2.3 Conclusions from simulation . . . . . . . . . . . 211

6 Conclusion 213 6.1 Verification of hypotheses and findings . . . . . . . . . . 213 6.2 Discussion of approach . . . . . . . . . . . . . . . . . . 216 6.3 Suggested further research . . . . . . . . . . . . . . . . 218

Bibliography 220

Glossary 247

Acronyms 249

A Appendix 253 A.1 Descriptive statistics of DOCUMEDIA data . . . . . . . 254

A.1.1 Dimensions and format . . . . . . . . . . . . . . 254 A.1.2 Levels of categorical variables . . . . . . . . . . 254 A.1.3 Descriptives . . . . . . . . . . . . . . . . . . . . 259

A.2 Additional models in simulation . . . . . . . . . . . . . 261 A.3 Interview guidelines . . . . . . . . . . . . . . . . . . . . 267 A.4 Land price model . . . . . . . . . . . . . . . . . . . . . 277

Curriculum Vitae 282

5

List of Figures

2.1 A framework of urban systems . . . . . . . . . . . . . . 28 2.2 Feedback cycle between the land use and transport . . . 33 2.3 Model systematics of LUTI . . . . . . . . . . . . . . . . 37 2.4 Real estate developers . . . . . . . . . . . . . . . . . . . 53 2.5 Ball’s functions necessary for the housebuilding process 76

4.1 Framework of developer decisions . . . . . . . . . . . . 108 4.2 Map of the Canton of Zurich . . . . . . . . . . . . . . . 113 4.3 Growth rates . . . . . . . . . . . . . . . . . . . . . . . . 115 4.4 Construction duration . . . . . . . . . . . . . . . . . . . 119 4.5 Construction costs by project type . . . . . . . . . . . . 125 4.6 Newly constructed dwellings per year . . . . . . . . . . 126 4.7 Map of new construction projects . . . . . . . . . . . . . 127 4.8 Developer type share over time . . . . . . . . . . . . . . 128 4.9 Construction costs by developer type . . . . . . . . . . . 129 4.10 Distribution of developers’ professionalism . . . . . . . 130 4.11 Average concentration ratios . . . . . . . . . . . . . . . 132 4.12 Projects by developer type . . . . . . . . . . . . . . . . 159

5.1 Overview of simulation area . . . . . . . . . . . . . . . 181 5.2 Preparation of building data . . . . . . . . . . . . . . . . 182 5.3 Entities and models . . . . . . . . . . . . . . . . . . . . 183 5.4 Validation statistics of demographic model . . . . . . . . 189 5.5 Time series of totals . . . . . . . . . . . . . . . . . . . . 192 5.6 Validation of new buildings per year . . . . . . . . . . . 193 5.7 Validation map main entities . . . . . . . . . . . . . . . 194 5.8 Error densities over municipalities . . . . . . . . . . . . 195 5.9 Main entities over time . . . . . . . . . . . . . . . . . . 198 5.10 Density deviations main entities . . . . . . . . . . . . . 199 5.11 Residential floor capacity and living units . . . . . . . . 200 5.12 Developed living units by purpose . . . . . . . . . . . . 201 5.13 Projects and rent price levels . . . . . . . . . . . . . . . 202 5.14 Time series of projects with high accessibility . . . . . . 204

List of Figures

5.15 Map of projects with high accessibility . . . . . . . . . . 206 5.16 Parcels for development over time . . . . . . . . . . . . 207 5.17 Map of floor capacities by zoning 2015 . . . . . . . . . 208 5.18 Accessibility indexes over time . . . . . . . . . . . . . . 209 5.19 Map of accessibility deviations by mode . . . . . . . . . 210 5.20 Rent price statistics over time . . . . . . . . . . . . . . . 211

8

List of Tables 2.1 Overview multi agent systems . . . . . . . . . . . . . . 42 2.2 Overview representative agent systems . . . . . . . . . . 43 2.3 Characterisation of LUTI models . . . . . . . . . . . . . 44 2.4 Comprehensiveness of selected LUTI models . . . . . . 45 2.5 Considered design factors for real estate development . . 59 2.6 Domains and factors of residential location choice . . . . 64 2.7 Event sequence comparison . . . . . . . . . . . . . . . . 67 2.8 Event sequence comparison (cont.) . . . . . . . . . . . . 68 2.9 Event sequence comparison (cont.) . . . . . . . . . . . . 69 2.10 Comparison of agents included . . . . . . . . . . . . . . 70 2.11 Comparison of agents included (cont.) . . . . . . . . . . 71 2.12 Comparison of proposed influential factors . . . . . . . . 72 2.13 Comparison of proposed influential factors (cont.) . . . . 73 2.14 Typology of developers by McNamara . . . . . . . . . . 79 2.15 Target system of developers . . . . . . . . . . . . . . . . 81 2.16 Typologies of land developers . . . . . . . . . . . . . . 83

4.1 Overview on real estate development datasets . . . . . . 116 4.2 Quality of addresses . . . . . . . . . . . . . . . . . . . . 119 4.3 Completeness of categorical variables . . . . . . . . . . 120 4.4 Quality of numeric variables per projects type . . . . . . 121 4.5 Year built by project type . . . . . . . . . . . . . . . . . 122 4.6 Project and building type . . . . . . . . . . . . . . . . . 123 4.7 Project type and purpose . . . . . . . . . . . . . . . . . 124 4.8 Project inputs and outputs . . . . . . . . . . . . . . . . 124 4.9 Costs per output category by building type . . . . . . . . 125 4.10 Definition of developer types in DOCUMEDIA data . . 133 4.11 Conducted interviews . . . . . . . . . . . . . . . . . . . 134 4.12 Reported main criterion by purpose . . . . . . . . . . . 137 4.13 Ranges of specifically asked criteria by purpose . . . . . 138 4.14 Definition of type professional and unprofessional . . . . 138 4.15 Differences in evaluation methods and information base . 139 4.16 Reported variables parcels . . . . . . . . . . . . . . . . 140

List of Tables

4.17 Reported variables parcel surroundings . . . . . . . . . . 141 4.18 Reported variables municipalities . . . . . . . . . . . . . 142 4.19 Level-of-service measures considered for public service

uses . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 4.20 Differences according to professionalism in search spaces 143 4.21 Tasks of developers . . . . . . . . . . . . . . . . . . . . 144 4.22 Detectability of typology . . . . . . . . . . . . . . . . . 145 4.23 Variable descriptions real estate development model (REDM)152 4.24 Descriptives building location choice model (BLCM) and

project location choice model (PLCM) . . . . . . . . . . 154 4.25 Estimation results BLCM and PLCM . . . . . . . . . . . 156 4.26 Observations by developer type . . . . . . . . . . . . . . 158 4.27 Hypotheses regarding developer variables . . . . . . . . 161 4.28 Variable descriptives DPLCM . . . . . . . . . . . . . . 162 4.29 RPLCM parameter estimates . . . . . . . . . . . . . . . 163 4.30 DRPLCM parameter estimates . . . . . . . . . . . . . . 164 4.31 Variable description of panel linear models . . . . . . . . 169 4.32 Panel linear land price models . . . . . . . . . . . . . . 170 4.33 Variable description of linear price model . . . . . . . . 171 4.34 Linear land price model . . . . . . . . . . . . . . . . . . 172 4.35 Variable descriptives of purpose specific models . . . . . 174 4.36 Comparison of model fit statistic with literature . . . . . 175 4.37 Estimation results segmented by purpose . . . . . . . . . 176

5.1 Overview on models . . . . . . . . . . . . . . . . . . . 187 5.2 Errors in totals of main entities 2008 . . . . . . . . . . . 190 5.3 Validation statistics over municipalities . . . . . . . . . 195 5.4 Scenario effects 2029 . . . . . . . . . . . . . . . . . . . 197 5.5 Descriptive statistics spatial variation . . . . . . . . . . . 198 5.6 Scenario effects 2015 . . . . . . . . . . . . . . . . . . . 200 5.7 Descriptive statistics of spatial variation in supply . . . . 203 5.8 High accessibility developments 2015 . . . . . . . . . . 204 5.9 Land consumption effects 2015 . . . . . . . . . . . . . . 205 5.10 Compactness effects 2015 . . . . . . . . . . . . . . . . . 208 5.11 Rent price effects 2015 . . . . . . . . . . . . . . . . . . 210

A.1 Building types according to Swiss centre for construction rationalisation (CRB) classification . . . . . . . . . . . . 254

A.2 Levels of offer type . . . . . . . . . . . . . . . . . . . . 257 A.3 Levels of construction stage . . . . . . . . . . . . . . . 258 A.4 Levels of project type . . . . . . . . . . . . . . . . . . . 258 A.5 Levels of purpose . . . . . . . . . . . . . . . . . . . . . 258

10

List of Tables

A.6 Descriptives of cleaned DOCUMEDIA dataset . . . . . 260 A.7 Coefficients income regression model . . . . . . . . . . 261 A.8 Coefficients car availability model . . . . . . . . . . . . 261 A.9 Coefficients real estate price model (REPM) . . . . . . . 262 A.10 Coefficients employment location choice model (ELCM) 263 A.11 Coefficients workplace location choice model (WLCM) . 264 A.12 Coefficients household location choice model (HLCM) . 265 A.13 Fixed effects per municipality . . . . . . . . . . . . . . . 277

11

Abstract

Sustainable development is a political goal of Switzerland. Urban devel- opment plays a crucial role in this respect, as a majority of people live in cities where most of the economic value is added. In addition, urban areas are also consuming the most resources, which suggests that an efficient organisation of urban areas is a key element of achieving this goal. The focus of this work is on the behaviour of real estate developers, who play a central role in the transformation of built space (Healey, 1991, p. 224).

The goal of this research is to investigate decision-making of real estate developers and the consequences for spatial development in an urban area. A theoretical argument and empirical analysis of developers’ location choices on a micro level are the prerequisites for a behavioural simulation of spatial development on the macro level. It is investigated, how the behavioural simulation can be used to inform the stakeholders in spatial planning processes.

The developer is defined as the main decision maker in connection with a potential development project. Therefore, the developer is the owner of a property or his representative in most cases.

The methodology comprises a) the study of related literature, b) con- ducting expert interviews, c) empirical analysis of developers’ location decision with discrete choice models and d) simulation of spatial develop- ment using the estimated location choice models. A microsimulation land use transport interaction model is used to achieve a consistent linkage of developers’ decisions, spatial development patterns and transport.

Literature confirms the definition of the owner as the principal decision maker and also notes heterogeneity across developers. Microsimulations on the basis of behavioural models are described as state-of-the-art land use transport interaction models. Most models use a representative agent for real estate supply. UrbanSim (Waddell, 2002) is an example of such a microsimulation model, which is very flexible and has been widely applied. In addition, it can accommodate developer type specific real estate supply models that are the focus of this research.

The analysis of the expert interviews finds that decision-making varies

Abstract

according to project purpose and developers’ level of professionalism. Developers have different decision criteria, different information sources and execute different tasks. The qualitative findings are used to create hy- potheses for discrete choice analysis and add to the general understanding for an interpretation of the results.

Discrete choice models are estimated on development project data with some information on the developers responsible for the project. Intro- ducing submodels according to the purpose of the development allowed an estimation of consistent models. It can be concluded that finding the right segments is critical for successful model estimation. Better estimation results suggest that separating self-providers from commercial developers is important and supports the main hypothesis that developers are of dif- ferent types. The rent price per square meter and the fit of a development project to the parcel’s zoning constraints are found as main explanatory variables for location choice. An increasing rent price level encourages commercial developers and discourages self-providing developers from choosing a location. The positive sign for commercial developers can be explained with expected higher profits. In contrast, self-providing developers avoid areas with high rent price levels, arguably due to taxes they have to pay on property value. More detailed data on developers is needed to apply more advanced techniques, e.g. estimation of latent class models, to investigate heterogeneity.

The simulation shows that questions regarding the development of a real estate industry can be investigated with microsimulation models of transport and land use. However, the simulation is on a proof-of- concept-level that prohibits quantification of effects on a regional scale. The strength of the microsimulation lies in the richness of information produced. The effects can be analysed in their spatio-temporal dynamics on various geographical units of analysis.

Important information unavailable in this study is transaction data of property, which would ideally contain characteristics of buyers and sellers, and the time and price of the transaction. Price models for dif- ferent market segments could be estimated with such information. The ownership structure of parcels would also be clarified and would allow a better assessment of the developers’ strategies. The implementation of an appraisal-based approach similar to the one presented by Foti and Waddell (2014), would allow to model replacement of old structures, which seems important in the light of the densification strategies on the political agenda. The new version of the UrbanSim software (Synthicity team, 2014b) is recommended for such an implementation.

14

Zusammenfassung

Nachhaltige Entwicklung ist ein politisches Ziel der Schweiz. Für die Realisierung spielen urbane Gebiete eine wichtige Rolle, da die Mehrheit der Gesellschaft in urbanen Gebieten lebt und der Grossteil der Wertschöp- fung in ihnen erarbeitet wird. Urbane Gebiete konsumieren aber auch die meisten Resourcen, weshalb ihre effiziente Organisation ein Schlüsselele- ment der nachhaltigen Entwicklung darstellt. Diese Arbeit fokussiert auf Immobilienentwickler, welche eine zentrale Rolle in der Transformation des umbauten Raumes wahrnehmen (Healey, 1991, p. 224).

Das Ziel der Arbeit ist das Entscheidungsverhalten von Immobilienent- wicklern und deren Konsequenzen für die räumliche Entwicklung eines urbanen Gebietes zu untersuchen. Eine theoretische Argumentation und eine empirische Analyse der Standortwahl von Immobilienentwicklern auf der Individualebene sind Voraussetzungen für die verhaltensbasierte Simulation der räumlichen Entwicklung auf der Kollektivebene. Es wird untersucht wie die verhaltensbasierte Simulation die Interessenvertreter in Raumplanungsprozessen unterstützen kann.

Der Immobilienentwickler ist definiert als Hauptentscheidungsträger bezüglich eines Bauprojektes. Folgliche ist der Immobilienentwickler meist der Besitzer der Immobilie oder dessen Stellvertreter.

Die Methodik beinhaltet a) eine Literaturdurchsicht, b) das Durchfüh- ren von Experteninterviews, c) die Analyse von Standortentscheidungen mittels diskreter Entscheidungsmodellierung und d) die Simulation der räumlichen Entwicklung basierend auf den Standortwahlmodellen. Für ei- ne konsistente Abbildung der Zusammenhänge zwischen Immobilienpro- jekten, räumlicher Entwicklung und Verkehr wird eine Mikrosimulation verwendet.

Die Literaturdurchsicht bestätigt die Definition des Immobilienent- wicklers als Hauptentscheidungsträger und weist auf die Heterogenität dieses Akteurtyps hin. Verhaltensbasierte Mikrosimulationen von Flä- chennutzung und Verkehrssystem werden als neuster Stand der Technik beschrieben. Die meisten Modelle bilden das Immobilienangebot über ei- ne repräsentativen Agenten ab. UrbanSim (Waddell, 2002) ist ein Beispiel

Zusammenfassung

eines solchen Modells, welches sehr flexibel ist und weltweit eingesetzt wird. Es können immobilienentwicklerspezifische Angebotsmodelle inte- griert werden, welche in dieser Arbeit im Vordergrund stehen.

Die Analyse der Experteninterviews zeigt die Heterogenität des Ent- scheidungsverhaltens im Bezug auf den Zweck der Projekte und die Professionalität der Immobilienentwickler. Entwickler haben verschiede- ne Kriterien, Informationsgrundlagen und führen verschiedene Aufgaben im Entwicklungsprozess aus. Die qualitativen Ergebnisse dienen der Hy- pothesenbildung der Entscheidungsmodellierung und unterstützen die Interpretation der Modellschätzungen.

Die Standortwahlmodelle werden mit Daten zu Neubauprojekten ge- schätzt. Diese beinhalten Informationen zu den Entwicklern. Die Schät- zung von Teilmodellen bezüglich Projektzweck resultierte in konsistenten Modellschätzungen. Daraus kann gefolgert werden, dass die richtige Segmentierung entscheidend ist für die Modellschätzung. Die Modell- statistiken bestätigen, dass Standortentscheidungen von kommerziellen Entwicklern und von Eigenheimentwicklern separiert werden sollten. Der Mietpreis pro Quadratmeter und die Übereinstimmung von projektierter mit erlaubter Nutzung werden als wichtigste erklärenden Variablen ge- funden. Ein steigendes Mietpreisniveau zieht kommerzielle Entwickler an, während es für Eigenheimentwickler weniger attraktiv macht. De- tailliertere Daten sind nötig um fortgeschrittenere Modelle, wie Latent Class Modelle, schätzen zu können, was zu vertiefter Untersuchung der Heterogenität wünschenswert ist.

Die Simulation zeigt, dass Mikrosimulation von Landnutzung und Verkehr für die Untersuchung von entwicklerspezifischen Szenarien ver- wendet werden kann. Die Simulation kann die effekte auf regionaler Ebene aber noch nicht mit gewünschter Qualität zeigen. Die detailreichen Simulationsergebnisse können für viele Aspekte in Raum und Zeit auf verschiedenen Aggregationsstufen analysiert und dargestellt werden.

Transaktionsdaten von Immobilien sind wichtige Informationen, wel- che in dieser Studie nicht verfügbar waren. Diese enthielten indealerweise Angaben zu Verkäufer, Käufer, Zeitpunkt und Preis. Neben dem Miet- wohnungsmarkt könnten weitere Marktsegmente bei der Preisschätzung berücksichtigt werden. Dies würde auch helfen die Eigentumsverhältnis- se der Parzellen genauer zu untersuchen, welche für die Strategie der Entwickler entscheidend sein kann. Die Implementierung eines Ansatzes gestützt auf Wirtschaftlichkeitsanalysen der Entwicklungsprojekte (Foti and Waddell, 2014) würden es erlauben weitere Projektkategorien wie Ersatzneubau zu berücksichtigen. Die neue Version von UrbanSim (Syn- thicity team, 2014a) wird für eine solche Implementierung empfohlen.

16

Zusammenfassung

17

Acknowledgement I am grateful to Prof. Axhausen and Prof. Waddell who accepted me as doctoral student. It was a very exciting and interesting time. I learned so many things thanks to your support and many discussions.

I thank my collogues Kirill Müller and Patrick Schirmer for their work to set up the parcel based land use transport interaction (LUTI) model for the Canton of Zurich, Timo Horstschäfer, Andri Mani and Reto Fahrni for their help in data preparation. The address matcher coded by Adrian Zaugg was very helpful for this task as well. Thanks also to Michael Heusser for his endurance and good work for the transcriptions of the interviews. Beverly Zumbühl copy edited the manuscript and helped me finding succinct formulations. I thank Liming Wang for is kind advices on UrbanSim and helpful discussions on the model of real estate development. Many thanks also to Erich Renner for his recommendations on scientific working and writing. A special thank you to Martina Koll-Schretzenmayr who was my advisor for the master thesis and encouraged me to write this dissertation. Her advices were always very much appreciated. Further I like to thank Emely Moylan and Fletcher Foti for their support during my stay at UC Berkeley. Special thanks to my wife Nina Renner Zöllig for her proof reading, conversations and strategic advices.

Further I like to thank all data providers (Building Insurance of the Canton of Zurich (GVZ), Cantonal Office for Spatial Development (ARE ZH), Documedia, Swiss Federal Statistical Office (BfS), Zurich Cantonal Statistical Office (SAKZ) and Swiss Housing Association (SVW)) for their generous support with data.

This dissertation is partly founded by the Swiss National Science Foundation (SNF). The support of the national research council of the SNF is very much appreciated. The research has also been partly funded through the SustainCity project (FP7-244557), co-financed by the Euro- pean Union within the Seventh Framework Programme (FP7). The author wishes to acknowledge the Commission for its support of the project, the efforts of the partners and the contributions of all those involved in SustainCity.

Chapter 1

Introduction The starting point for this research is an acknowledgement of the ongoing debate on urban development topics, such as urban sprawl or the energy consumption of settlements. In these debates, there is little quantification of consequences resulting from political decisions or for expected trends. Transparent and comparable assessments would be desirable to inform the planning process and enrich future debates. Therefore, this work investigates the applicability of agent-based modelling to investigate the effect of a more professional real estate industry on urban systems. The case study and the approach are chosen to explore this planning tool for urban regions considered to be complex systems.

The problem and the relevance of the research are discussed first in this introduction, followed by a detailed description of the goals formu- lated as explicit hypothesis and research questions. The approach and methodology are presented with the scope of the project narrowing the research topic to a feasible range. The last section describes the structure of the document.

1.1 Problem Managing urban regions is a process that has gained importance since more and more people are living in cities (Malik, 2013, p. 197). If we want to have better control over the development of urban systems, we have to clarify the processes constituting their evolution, which also means to identify determinants that can be influenced with appropriate policies. Such policies include subsidies for cooperatives, social housing, taxes on real estate, zoning regulations and infrastructure improvements. Sometimes city administrations also try to prevent the decline of neigh- bourhoods by giving incentives for new real estate investments. It is worth mentioning that the first three policies distinguish between different

Chapter 1. Introduction

types of real estate developers. Therefore, it would be ideal to know type-specific reactions in order to better understand the transition of the overall system. This should help planning authorities with their mission to guide and control spatial development. Problems targeted with real estate developer-specific policies include high housing prices and energy issues. A precondition is to learn more about real estate developers and their behaviour, which seems especially important since real estate developers play a central role in the spatial development process.

Land use transport interaction (LUTI) models are tools to study and manage urban systems. Their core idea is to capture the interaction be- tween the land use system and the transportation infrastructures. This re- quires modelling the land development process (described in Section 2.1).

The purpose of real estate development models in LUTI systems is to provide location options for households and firms. Real estate development models describe the evolution of the building stock and thus determine real estate supply at different points in time. The models represent the supply side of real estate markets. They are designed to show the possible results of a given policy, so planners or the voting public can decide whether to implement it or not. Hunt points out that modelling the supply of built space is often the weakest point in land use transport models (Hunt et al., 2005). Haider and Miller (2004) find as well that built space supply is rarely investigated. Literature in real estate research indicates the same (DiPasquale, 1999). Therefore, this work investigates whether the consideration of developer types helps to improve LUTI modelling. The work thus discusses the evolution of the building stock and its modelling with a focus on the actors behind it.

In a spatially and dynamically explicit simulation real estate supply models have to determine when, where and how much of which type of real estate is to be maintained or built. Events that constitute the evolution of building stocks include the construction, alteration, demolition and replacement of buildings. These events are the consequences of decisions made by the owners of the respective real estate. This suggests that real estate development can be modelled by analysing owner choices, which are subsequently simulated. Other events can be the consequences of physical processes, e.g. a building destroyed by an earthquake, but such events are not considered here. The model would ideally provide answers to all decision dimensions at once since these decision variables are considered simultaneously. The decision variables of time and quantity are continuous whereas location and type of real estate are discrete. Hence, a complete model would include explanatory variables from alternatives, decision makers and decision situations to determine the decisions on

22

1.2. Hypotheses

time, location, quantity and type of real estate. For simplification, the work at hand focuses on decision makers’ preferences regarding location choice. The research hypotheses are formulated accordingly.

1.2 Hypotheses In this study, the hypotheses are formulated to the micro-level of individual behaviour and to the macro-level of spatial development because the interdependency of the two is of interest.

1. Micro level (a) There are behavioural differences among real estate develop-

ers. (b) The choice of a real estate developer for a development site de-

pends on the characteristics of the developer. The developer’s resources, such as property, knowledge and money, influences his valuation and thus the choice for a development option.

(c) Heterogeneity in developers’ decision-making can be mea- sured by estimating location choice models for specific devel- oper types.

(d) Specialised professional developers build in central (highly accessible) places.

2. Macro level (a) To simulate the development process more accurately, differ-

ent developer types need to be considered. (b) The consolidation of a real estate industry (having more profes-

sional developers) leads to more efficient spatial development, e.g. less land consumption or less energy use in the transport sector.

These hypotheses require setting up a state-of-the-art LUTI simulation to explain macro level effects with micro level decisions of developers. Furthermore, it is necessary to compare models with type-of-developer considerations and those without. Further elaboration of the hypotheses can be found in Section 4.1.

1.3 Approach The research methodology follows from the hypotheses and theory (Chap- ter 2). To investigate the role of real estate developers for spatial de- velopment we deploy agent-based simulation. This requires an analysis of the individual behaviour of developers as well as their behaviour in

23

Chapter 1. Introduction

the context of spatial development in an urban simulation. Therefore, the methods used to investigate at the level of the individual are expert interviews (qualitative) and discrete choice analysis (DCA) (quantitative). To assess the effects on the urban scale, an agent-based simulation of land use and transport interaction is used. The methods are described in more detail in Chapter 3.

More specifically, we approach the research questions in three main steps. The first step is conducting eleven in-depth interviews with de- velopers active in the study area, the Canton of Zurich in Switzerland (Section 4.3). The qualitative work allows access to the subjects and the development of some intuition about the data available. The second step is to estimate deterministically segmented location choice models (subsection 4.4.3) according to developer information, which comes from data on real estate development projects (subsection 4.2.3.1). In the third step, the estimated models are simulated in a land use transport interaction simulation and the scenario effects are analysed (Chapter 5).

1.4 Scope of dissertation

The focus is on the anthropogenic urban system. The ecological or environ- mental systems are not considered. Here the focus is on the consequences of human action; this can be justified by the fact that humans shape the environment to a large extent. Because many decisions are made in an economic context, it seems appropriate to use economic models such as in DCA.

Concentrating on the questions of where new development projects occur and if there is any distinguishable decision behaviour, the main data source is a set of almost 60,000 records of real estate development project applications in the Canton of Zurich from 2000 to 2010. The records contain contact details of the real estate developers as well as the addresses of the development projects. The dataset is from the firm DOCUMEDIA (Docu Media Schweiz GmbH, 2013), which collects the information to facilitate the formation of construction consortia. The study area and the observations finally used for model estimation are shown in Section 4.2. Regarding the simulations, time horizons have to be chosen based on to the measures assessed. A usual time period covers 20 to 50 years, while here it is a simulation period of 30 years.

24

1.5. Document guide

1.5 Document guide The remainder of this dissertation is structured into five chapters. Chap- ter 2 contains the literature review as well as a general description of the topic. Methodological theory is summarised in Chapter 3 introducing the three main elements: a) a qualitative method for expert interviews (Chapter 3), b) DCA(Section 3.3) and c) agent-based simulation (ABS). Chapter 4 reports on the analysis of development projects and develop- ment decisions while simulation work is treated separately in Chapter 5. The last chapter contains general conclusions and suggested further re- search (Chapter 6).

The real estate developer will be referred to as developer throughout this dissertation for simplicity. All figures, tables, plots and maps are original work by the author unless stated otherwise. North is on top in all maps.

Parts of this dissertation have been the contents of conference papers or have been published as book chapters. All parts are however original work by the author. A list of these references is given below. • Zöllig et al. (2011) • Zöllig and Axhausen (2011) • Zöllig and Axhausen (2012) • Zöllig Renner and Axhausen (2013) • Zöllig Renner and Axhausen (forthcoming) • Zöllig Renner et al. (forthcoming) The dissertation is built on a collaborative effort to set up the LUTI

model for the Canton of Zurich (Schirmer et al., forthcoming).

25

Chapter 2

Theoretical background and review of literature Chapter 2 introduces basic land development processes, the concepts and elements of land use in urban systems and reviews the land development models that are trying to capture these systems and processes. The de- scription of the current situation in the land development sector and its main players, real estate developers, is a particular focus of this review (see hypothesis Section 1.2). It also presents the theoretical background as revealed through the review of literature.

2.1 Land use development

The process of land use development is the topic under investigation, in particular, that of urban developments. However, to understand this process, one has to consider its context, which is the urban system, and its history. Every urban system starts as a non-anthropogenic environment of soil, topography, plants and so on.

Settlements are introduced when human beings build the infrastruc- tures they find convenient for their lives. These infrastructures are the requisites used by people, who often perform their activities in a particular setting; land use is then defined by those activities, e.g. a plot of land planted with corn is agricultural; one with houses is a settlement, etc. The characteristics of a location may make a certain use more probable, but its human use is not yet determined, e.g. an open field could have many different uses, depending on location and need.

People are organised by social structures. The most obvious ones are households and enterprises. There are other institutions, such as extended families, social groups or circles of friends that are also part of the social

Chapter 2. Theoretical background and review of literature

Figure 2.1: A framework of urban systems showing actors, processes as decision sequences and subsystems they constitute

Legend

Very slow

Slow

Fast

Immediate

Housing

Travelling/Transporting

Working

Constructing

TripMobility tool Mode Route

Location

Urban system with locating activities

Activity

Decision

Time

LocationLocation

Persons

EnterprisesHouseholds

System

Actor

Networks Buildings

Deconstructing Location

Buildings

Infrastructure

Regulations Environment

Land use

Society Economy

Transport

Variable

Speed

structure. Social structures are important elements of urban life and each shapes our behaviour to some extent (Frei, 2012; Kowald, 2013).

The process of land use development is affected by non-anthropogenic processes as well. Landslides or earthquakes can affect the develop- ment path as much as the decision to build a new highway. Continuous processes, such as erosion, also influence development in which anthro- pogenic processes are denoted as activities (see Fig. 2.1). The framework is inspired by the work of Wegener and Fürst (1999).

In modern societies, it is not only the physical characteristics that determine the kind of activity, there are also regulations, which are ex- plicitly formulated rules. This thesis is primarily concerned with land use regulations such as zoning plans. However, there are many more laws and regulations that shape the distribution of activities in space, for example, laws of environmental protection or migration.

2.1.1 Urban systems An urban system can be defined as a complex of interacting subsystems, the major parts of which are created by human decisions. Urban systems are also embedded in an ecosystem. Batty (2007a) describes the city as a complex system and Miller and Page (2007) as a complex adaptive system. Complex systems are characterised by non-ergodicity, phase transition, emergent phenomena and universality.

28

2.1. Land use development

• Non-ergodicity means that such systems do not behave in a well- defined way over the long term.

• Due to an external shock, long-term development diverges, leading to a new development path. Such a junction is a phase of transition in which the system may behave in a totally different way.

• Emergent phenomena essentially means that larger, unknown struc- tures arise out of the given details, as expressed in the phrase: “The whole is greater than the sum of its parts.” This not only stresses the importance of a detailed look at systems and their components, but also the significance of the relationship between the components for the overall appearance.

• Universality is the similarity of relationships on different scales. It is often the only constant in complex systems. Fractals are the analogy in geometry, and it is interesting to note that fractal structures can be observed in urban settlement patterns (Batty, 2007b; Batty and Longley, 1994).

2.1.2 Subsystems The holistic concept of sustainability suggests three main subsystems: ecology, economy and society. These can be split up into ever smaller units. This deconstruction can be very detailed, but might not be practical. Thus, it is important to identify subsystems that are relevant and suitable in regard to the research question. The subsystems are identified in Fig. 2.1: a) environment, b) infrastructure, c) land use, d) society, e) economy and f) regulations. A more detailed description of each subsystem follows.

Environment The environment is probably the most complex of the six subsystems. It comprises the ecosystems that provide the basic resources upon which an urban system depends. The primary resource of interest here is land. It also includes natural resources, such as raw materials or ecosystem benefits. However, it becomes evident that the subsystems cannot be separated from each other, since e.g. food actually emerges from agricultural activities. Other parts of the environment are topography, weather, climate, rivers or lakes. All these things are basic conditions and form the foundation of urban systems.

Infrastructure Graaskamp (1981, p. 3) defines infrastructure as ele- ments that provide “economies of scale to be enjoyed through collective action of many parcels, which leads to off-site centralization.” A slightly more general definition is that a system of infrastructures comprises all

29

Chapter 2. Theoretical background and review of literature

installations that facilitate or protect certain activities. The main cate- gories of infrastructures are a) facilities1 (which are mostly for protection and simplification of activity), b) security constructions (such as gal- leries, barriers against floods), c) transport infrastructures (connecting activity locations), d) waste water systems, e) clean water systems and f) communication systems.

Land use Land use is defined by the activities people perform in a given place. Consequently, there are as many land uses as there are activities. Usually, the activities are categorised and summarised. Obvious activity categories are: a) housing, b) working, c) travelling, d) shopping or e) leisure. These categories can be expanded, if necessary.

The land use system orders the types of uses and their spatio-temporal distribution. Depending on the characteristics of a place an activity is more or less likely to be performed at that place because it is more or less suitable. More precisely, the characteristics of a place define its usability2 for a certain activity. These characteristics include properties of the location as well as its relationship to the surroundings.

Some uses can only be imagined on a particular site whereas others might be more flexible. Strictly speaking this does not actually concern the uses themselves, but more the planning of use. This means that users or planners think about the spatial and temporal compatibility of uses. With immediate effects, such as noise, the spatial proximity only matters when the activities happen at the same time.

The quality of a land use system depends not only on the spatial distribution, but also on the ordering of sequences of activities. One sequence of uses might be more feasible than another, e.g. extracting raw materials from an environmentally protected area, although there are examples of the opposite: former raw material extraction sites that are now nature reserves. These examples show that the type of activity and its effects is critical for identifying conflicts and synergy potentials.

Land use can vary in intensity. This is especially important for land uses that exploit resources. The intensity can be so high that the usage cannot be sustained in the long term.

The land use system is not restricted to two dimensions – especially when looking at an urban system, where floor space is nearly as important as the ground. Floor space allows locating activities on top of one another and technological progress allows taller and taller buildings. Scarcity

1The more general term facility seems in this context more appropriate than buildings. 2Alonso (1964) uses efficiency parameters to describe the suitability of land. Märki (2014) describes

it in his model with effectiveness functions.

30

2.1. Land use development

of land pushes development even further in this direction, so it seems important to expand the term “land use” to include floor space as well. If this thought is expanded to include the third dimension, it might be better to describe land use in volumes instead of planes. Then one might speak of “built space” (Farooq, 2010).

Society Society may be defined as people and their relationships. Col- lective living is based on formal and informal norms, habits and traditions. The close relationship to a system of regulations becomes evident here. However, only a fraction of the informal norms ever get formalised in a law or other document. The formal norms are legitimised and accepted through some sort of political process. Also, relationships are more or less formal. While professional relationships are usually defined in a contract, family relationships are based on informal and moral obligations. These norms and relationships are also relevant for the land use development process since they influence the travel and location choices of people (Frei, 2012).

Economy The economy comprises all the actors who are related through trade relations. The place where goods are traded is referred to as a market, which does not necessarily need to be a physical place. Markets can also be delimited by the homogeneity of the traded goods. An example is the real estate market, which has several submarkets, one of which is the market for single-family homes.

Regulations The regulations are the result of the effort of society to or- ganise itself. The subsystem of regulations comprises all laws and bylaws effective for spatial development, which are considered the functional laws of spatial planning (Lendi, 1996, p. 67). The principle nominal law of planning is the Raumplanungsgesetz (Bundesversammlung der Schweiz- erischen Eidgenossenschaft, 1980). Additional land use regulations are cantonal structure plans (Richtplan) and zoning plans (Nutzungsplan). A basic outcome of this regulatory planning is the subdivision of land into parcels by a juridical act. However, there are many more regulations that influence the evolution of the urban system. One example in the real estate market is the limited share of property that can be owned by foreigners (Bundesversammlung der Schweizerischen Eidgenossenschaft, 1983).

31

Chapter 2. Theoretical background and review of literature

2.1.3 Interaction of subsystems

This section describes the connection between the subsystems. The inter- actions between subsystems fall into two categories: physical cause-effect relationships and cognitive cause-effect relationships. Physical cause- effect relationships range from too many cars on a road leading to longer travel times to the destruction of infrastructures due to flooding and, on another level, the emissions from highways that pollute the ecosystems in their surroundings.

Examples of cognitive cause-effect relationships are decisions based on a person’s perception of their environment, such as taking the train instead of the car because of expected local traffic jams.

It can be assumed that the characteristics of any subsystem can become relevant for such decisions at some point. Therefore, the subsystems are all interacting with each other during the considerations before a decision is taken. These decisions can be analysed with statistical methods described in Section 3.3.

In most cases, the effects are then the cause of yet another effect, thus forming chains of effects. In these chains, physical and cognitive cause-effect relationships might alternate and within the system various chains of effects might be identified that are interdependent. A possible way to summarise these chains of effects is with elasticities (Axhausen, 2008, p. 10).

Interaction between land use and transport One central chain of ef- fects important for this dissertation is the feedback cycle (Fig. 2.2) be- tween transport and land use, which are constituted of physical as well as cognitive interactions as described by Wegener and Fürst (1999). Assume a normal working day during which many transport-related decisions are made in addition to the daily commute. However, some decisions were already made before leaving, e.g. the mobility tools are already given for that morning and the decision to go to work has also been made. In most cases, the workplace is also a given. However, departure time, mode of travel and route can be chosen spontaneously. At the end of the day, all these decisions determine the time, distance and monetary costs of travel that day.

The link between land use and transport has been researched for decades. Even though its relevance is theoretically and empirically well understood, there is little rigorous consideration in actual planning practice (Kelly, 1994).

32

2.1. Land use development

Figure 2.2: Feedback cycle between the land use and transport

Generalised costs of travel

Attractiveness of location

Mode choice

Activity locations

Land use

Construction

Route choice

Destination choice

Trip decision

Mobility tool choice

Location decisions of investors

Relocation

Location decision of users

Transport

Link load

Travel time, distance, costs

Accessibility

Departure time choice

Source: adapted from Wegener and Fürst (1999). It is interesting to see that earlier versions of the feedback cycle explicitly assume capacity improvements as the driver of the mechanism. The predict-and-provide paradigm is clearly visible (Stover and Koepke, 1988, p. 2).

2.1.3.1 Markets

Markets are economic systems of interacting agents. The generalised costs for the use of locations (including a time-space slot on a road or in a bus) depend on demand. The decisions of others influence the choice situations of the individual. Firstly, there is a direct effect that reduces the ’comfort level’. One element of this is increased travel or waiting time. Another is crowding or perceived danger. Secondly, there can be a price effect if a market is organised in response to the situation. Markets regulate demand on an abstract level. Rather then having large crowds gather in the same place at once, e.g. a main railway station, someone, a manager or politician decides to organise a market and let the highest bidder have an exclusive right of use. The creation of a market avoids the direct effect of having large crowds in open spaces. This means that the physical interaction of people is to some extent defined by interactions in

33

Chapter 2. Theoretical background and review of literature

markets. In addition to location and time, markets can be distinguished by the

goods traded and the trading agents. Both goods and market participants might be determined by market regulations. An urban system can have a large variety of markets, therefore no breakdown is included here. Impor- tant markets for our work are a) the transport market, b) the land market and c) the built space market (Farooq and Miller, 2012). The housing market is a submarket of the built space market. The built space market is the domain of the real estate developers who are the focus of this study.

2.1.3.2 Accessibility

Accessibility is a central indicator of the attractiveness of a location in this context and can be described as

"... the potential of opportunities for interaction." (Hansen, 1959, p. 71).

There are various alternatives for its calculation (for a thorough review, see Geurs and van Wee (2004)) and a growing body of literature. One possible reason for its popularity may be the generality of the concept, which makes it useful for a variety of disciplines. From an economic point of view, one can argue that accessibility captures the potential of opportunities. The higher accessibility, the higher the welfare indicator since it captures potential utility.

It is important to see that accessibility takes the quality of the trans- port system and the land use system into account at the same time. This means that accessibility can be improved by either modifying the gener- alised cost of travel or by the distribution of activity or housing locations. Accessibility-oriented planning requires a strong integration of transport and spatial planning.

It is also possible to include individual preferences and a time factor (Miller, 1999). This is theoretically favourable, but often difficult in prac- tice due to data limitations. This is one of the reasons why studies seldom use fully specified accessibility measures. The accessibility measure is adapted to the specific context of the problem.

A mathematical formulation that has been used in various studies in Switzerland (Tschopp et al., 2005; Axhausen and Hurni, 2005; Bo- denmann, 2011; Fröhlich, 2008) is given in Eq. (2.1). This formulation considers the land use component by measuring persons or jobs in the locations considered. The transport system is measured using the gener- alised cost of travel between the locations. The distance decay function

34

2.2. Models of land use development

with its parameter β is found by fitting the model to available data. In Switzerland, many studies are referring to Schilling (1973, p. 2.34) and choose 0.2 as the β value. Killer et al. (2013, p. 11) show a slight decrease of this value over time. Their most recent estimate for Switzerland is 0.183 for the year 2000.

Acci = J∑ j

X j e −βci j (2.1)

where i= location of accessibility calculation j= activity location index

X = number of persons and jobs c= generalised travel costs J= number of all activity locations considered β= 0.2 (estimated parameter)

Accessibility as market potential Rephrased in economic terms: Ac- cessibility is the market potential. A producer needs a certain market size to support its enterprise. An adequately defined accessibility measure can thus represent the potential market size for a specific industry or enterprise. A potential location for a consultant for troubled families must offer accessibility to the families. The measure becomes more accurate if family income is also included, since the willingness to pay for such services should be higher with increasing income.

2.2 Models of land use development

This section describes land use development models found in the liter- ature. The body of literature contains model descriptions in different formulations and is quite large. Here the focus is on land use transport interaction (LUTI) models because there is a strong link between trans- port and land use via accessibility as shown very early by Hansen (1959). Subsection 2.2.1 presents the origins of the state-of-the-art land use de- velopment model (LUDM). This gives the background and context for the LUTI models, which are described in more detail in subsection 2.2.2. The description of all LUTI model components is necessary because the focus is on spatial development. Real estate supply is just one element of the complex system described in Section 2.1 and cannot be consid- ered in isolation. However, real estate supply modelling is described in subsection 2.3.2.

35

Chapter 2. Theoretical background and review of literature

2.2.1 Updated model systematic of land use transport interaction

In addition to spatial planning, land use models have been developed in various other disciplines, including geography, urban economics and agriculture and ecology. Based on previous discussions, one can dis- tinguish land use transport models and land use transport environment models; the latter gaining more and more interest due to the sustainability debate. Most LUTI models presented here belong to the first category and focus on the feedback cycle between the two main components. LUTI models are land use change (LUC)3 models, which try to explain why a piece of land is adapted for a particular purpose. The models analysed by Briassoulis (2000) may have a slightly more passive notion. The terms interaction and change show that the models do not aim at describing a static state, but rather interdependent dynamics, i.e. evolution over time. In contrast, land cover change (LCC) models describe the evolution of physical land categories. Urban growth models (UGM) are special cases of LCC models since they model the transition from unbuilt to built land.

Figure 2.3 shows the model systematics of LUTI models. The vertical axis is the time line. Aspects of reality that are reflected in the theories and models are distributed horizontally. The rectangles represent elements of different natures as indicated in the legend. An arrow symbolizes influence. Representative models of the respective groups are shown in capital letters. Predecessors are not included to keep the figure concise.

The first four aspects from the left represent strands of economic theory which formulate the basic principles underlying LUTI models. The dynamics of the urban system was accounted for by using the method of micro-simulation as a

“general approach to the study and use of models” (Orcutt, 1960, p. 893).

The use of parcels became possible with the development of geographic information systems and their data. Parcels are meaningful geographical units of analysis (GUA) because they are the legal entities with the highest spatial and regulatory detail.

Early urban economics One of the earliest economists to write about how space matters was David Ricardo when he introduced the idea of comparative advantage (Ricardo, 1817). His theory basically states that

3This is the terminology of Briassoulis (2000).

36

2.2. Models of land use development

Figure 2.3: Model systematics of LUTI

Lowry (1964): Gravity Models, Wilson (1969): Entropy Maximising Models

1960

1970

1980

1990

2000

Leontief (1936), Isard (1951/56): Input-Output (I/O) models

Alonso (1964), Mills (1967), Muth (1969): Bid-rent-theory of Urban Land Markets

Orcutt (1960): Microsimulation

Spatial Interaction Models: METROPILUS

Spatial I/O Models: MEPLAN TRANUS

McFadden (1978): MNL Discrete Choice Model

Geographic Informations Systems

Hybrid Spatial I/O Models: PECAS 2. Generation

Equilibrium Discrete Choice Land Use Models: MUSSA RELU-TRAN

Dynamic Discrete Choice Land Use Models: HUDS

Spatially Detailed Dynamic Land Use Models: ILUTE ILUMASS UrbanSim TIGRIS XL DELTA

Legend

Theory Computational Model GroupTechnologyMethod

Thurstone (1927): Comparative Judgment Binomial Probit

Marschak (1960): Random Utility Maximization (RUM)

Individual Decisions

Location of activities DynamicsTechnical InteractionSpatial Interaction Land Market Parcel

Source: Adapted from Waddell (2005); Rho and Kim (1989)

some places are more suitable for certain activities and that predetermines their use.

Von Thünen (1826) explains the allocation of production sites for agricultural goods with their location in relation to the city centre. Goods with low transport costs are located further from the city centre. It is one of the earliest theories about activity allocation in a city region.

More than a century later Christaller (1933) examines the spatial distribution of cities on the basis of telephone connections. He investigates empirically his theory of central places. Centrality is measured by the variety of goods and services provided at a given place. Christaller derived a hierarchy of cities for this measure. The explanation for the hierarchy is derived from different market sizes of goods. Under the assumption of rational consumers, producers and homogeneous space, a hexagonal pattern of urban markets should emerge.

Input-output models Leontief (1944) came up with the input-output model of sectors in the economy for the United States. He saw the whole economy as a huge accounting system within which single entities, such

37

Chapter 2. Theoretical background and review of literature

as firms and households, had to be grouped together in sectors to achieve a practical model. It was presented as an extensive table showing input sectors in rows and output sectors in columns. The cells contained the relationships of the sectors in dollars.

In a first step, Isard (1951) expanded Leontief’s model to the spatial dimension. Instead of just distinguishing the sectors, he added an index denominating a certain region.

“. . . if states are designated as regions, Pennsylvania brick becomes a commodity different form New York brick or California brick. (p. 320)

Five years later he published a fully formalised theory and treated space as an explicit factor (Isard, 1956).

Bid rent theory Alonso (1964) introduced the bid rent function as part of an equilibrium framework of urban land markets. The work derives bid functions for agriculture and “urban firms”. The empirical part is limited to the last chapter, which also includes an outline for further empirical research. He was the first to describe the process of housing allocation in an urban environment. The concept behind the bid rent function is that bidders compete in an auction- like process. The auction is won by the highest bid. This mechanism is appealing and solves two problems at the same time. Firstly, the price is determined. Secondly, a use (or user) gets allocated to the site. The third dimension is not considered explicitly, i.e. locations can only vary in size and 2D position.

Muth-Mills model Mills (1967) presents a model that relates size, den- sity and prices in a mono-centric city with three activities, goods produc- tion, housing and transport. The model exhibits a core city where goods (these are summarised to one representative good) with increasing returns to scale are produced. All other activities are assumed to have constant returns to scale (summarised to one representative activity: housing) and are located in the hinterland of the core city.

Transport connects production of industrial and home products to the core city. The model determines the input and output quantities of the three activities, land rents, distribution of residences and the size of the core city. The origin of cities is explained nicely with comparative advantages of certain locations, such as cheap transportation on rivers (p. 198). Substitution of input factors and technology are the core elements of explanation. The work also explains how agglomeration economies

38

2.2. Models of land use development

and economies of scale lead to the emergence of cities, or more generally speaking, agglomerations.

Muth (1969) presents a rigorous housing market analysis in the tradi- tion of the Chicago School of Economics. The first part is a theoretical spatial economic equilibrium analysis. The larger second part is empirical. Compared to the work of Alonso (1964), there is more weight on the distribution of population than on price formation. This work is often referred to as the Muth-Mills model. The model explains decreasing land rents away from the city centre with increasing commuting costs.

Bid-choice Bid-choice theory states that it does not matter whether the demand or the supply side is modelled, because the spatial outcome is similar (Martínez, 1992). In classic economic theory, this is a straightfor- ward process because demand equals supply in a competitive market at equilibrium.

Three waves In Fig. 2.3, the three waves of development as identified by Iacono et al. (2008) are obvious:

1. Spatial interaction and spatial input-output models of the 1960s 2. Econometric models of the 1980s 3. Spatially detailed micro-simulation models at the end of the 1990s

Each of the three waves was triggered by a theoretical and/or techno- logical development. The developments in economic theory, in parallel with the increasing computational power and data availability, led to this development path.

The first LUTI model was devised and implemented (on a computer) by Lowry (1964) on the basis of gravity theory. The theory was borrowed from physics and states that the interaction of two regions is proportional to their size (in terms of jobs) and inversely proportional to the distance in between.

These first models were succeeded by econometric models in the 1980s (second wave). These econometric models formed the group of spatial computational general equilibrium (SCGE) models. These work with representative agents (Table 2.2) and explain the interaction of re- gions based on markets where economic decisions are made. Hence, the empirical work required the analysis of more detailed data than the aggregates per region considered. The observations of economic decision makers therefore had to be collected.

The next step was the development of discrete choice modelling by McFadden (1978, 1981) and the development of information technolo- gies, which made micro-simulation very attractive. The Harvard urban

39

Chapter 2. Theoretical background and review of literature

development simulation (HUDS) was the first large-scale model using micro-simulation (Kain, 1985). HUDS was followed by other projects, such as the transportation and land use model integration project (TLU- MIP) which was then developed into TRANUS (Weidner et al., 2007). The land use scenario developer (LUSDR) (Gregor, 2007) was a result of the TLUMIP project.

Further efforts led to the development of PECAS (Hunt and Abraham, 2003) and UrbanSim (Waddell, 2002). This third wave was triggered by the combination of theoretical developments in discrete choice mod- elling and the technical feasibility of micro-simulation due to increasing computational power, as well as social relevance, because of debates on pollution and climate change. The model requirements now include more ecological indicators such as CO2 emission, land consumption, energy use and air pollution with the trend expanding. Up to now, LUTI models have treated the effects of the anthropogenic system upon the natural environment to some extent. Feedback from the natural environment on the anthropogenic system has been widely neglected, though one exam- ple would be coupling a climate model and nature’s reactions to climate change.

2.2.2 Description of land use transport interaction (LUTI) models 4

LUTI models are models of spatial development with a special emphasis on the relation of transport and land use. The basic idea is to capture the interaction between the land use system and the transportation infras- tructure as explained in subsection 2.1.3. A rather general introduction to the field of urban modelling is given by Batty (2009). Good reviews that go into more detail are Iacono et al. (2008); Timmermans (2007); Chang (2006); Wegener (2004); Hunt et al. (2005); Verburg et al. (2004). Older examples are (Wegener, 1995; Southworth, 1995; Putman, 1975). Wilson (1998) has a more economic perspective. Most of the review articles discuss only selected models. There are good reasons to do so. Firstly, not all models are still relevant. Secondly, too many models make the comparison confusing simply because of the amount of information provided.

Operational LUTI models are calibrated for a specific region and ready to use for policy analysis. A non-operational model might exist as software or mathematical formulation, but has not yet been applied to

4Parts of the section are taken verbatim from Zöllig Renner et al. (forthcoming)

40

2.2. Models of land use development

a specific area. Only operational models are considered in this section. The operationalisation alone can be a work-intensive project (Iacono and Levinson, 2008; Gruber et al., 2000) and keeping it operational as well. The number of applications is a proxy for transferability to other locations that also shows usability and maturity to some extent.

Currently, a number of operational LUTI frameworks exist (Wegener, 2004; Zöllig et al., 2011). Two main groups can be identified with respect to the aggregation level. The first group, which is of main interest here, operates on the level of individual agents (Table 2.1). These models are also referred to as disaggregate micro-simulation models, multi agent systems (MAS) or agent-based model (ABM). The second group uses representative agents (Table 2.2). Therefore, they are also labelled aggre- gate models. The appropriateness of the model will always depend on its purpose. The reasons to use disaggregate models are: • Ability to explain macro level phenomena from a micro level • Grounding in microeconomic theory • Capability to represent complex systems • Flexibility with respect to result evaluation (aggregation levels) • Flexibility in accommodation of different modelling approaches The locations of the first implementation of the models are listed in

the tables because the context of development is important, given the data dependency of the models. The first reference in the footnotes is the principle one. Additional references are given for convenience and to show recent activity. Therefore, maintained websites are also included.

2.2.2.1 Profiling ILUTE, ILUMASS, and UrbanSim

This section compares the characteristics of the three model systems UrbanSim, ILUTE, and ILUMASS. These three model systems were chosen for their microscopic nature, capability to interact with a micro- simulated transport model, explicit representation of time (dynamics) and the representation of social and economic development. The frameworks are micro-simulation models that are able to model disaggregated entities, such as parcels or persons.

A characterisation of the three selected MAS is given in Table 2.3 for easy comparison. It also serves as a structure that is followed in the subsequent description of UrbanSim. A general discussion of the characteristics follows in the next paragraphs. The notations are derived from Wegener (2004); Zöllig et al. (2011).

41

C hapter

2. T

heoreticalbackground and

review of

literature

Table 2.1: Overview of land use transport interaction models with multiple agents

Model name

References Location of first implementation

Nb. of Ap- plications

Urban- Sim

Waddell (2000, 2002); Waddell and Ulfarsson (2004); Waddell et al. (2005); UrbanSimProjekt (2011); UrbanSim Developers (2014)

Eugene, Oregon >1

TRESIS Hensher and Ton (2002); Institute of Transport and Logistics Studies (2009) Sidney >1 ILUMASS Wagner and Wegener (2007); Beckmann et al. (2007); Strauch et al. (2005) Dortmund 1 ILUTE Salvini and Miller (2005); Miller et al. (2004) Toronto 1 TIGRIS XL

Zondag (2007) The Netherlands 1

PUMA Ettema et al. (2007) The Northern Dutch Randstad

1

LUS- DR/TLU- MIP

Gregor (2007); Weidner et al. (2007) Oregon 1

STASA Haag (1990); Pumain and Haag (1991); STASA (2013) Stuttgart 1

42

2.2. M

odels of

land use

developm ent

Table 2.2: Overview of land use transport interaction models with representative agents

Model name References Location of first Nb. of implementation applications

RELU-TRAN Anas and Liu (2007) Chicago > 1 PECAS Hunt and Abraham (2003); Abraham and Hunt (2007); Abra-

ham et al. (2005) Oregon > 1

TRANUS Barra et al. (1984); MODELISTICA (2013) Carracas > 1 DELTA Simmonds (1999); Simmonds and Feldman (2005); Bosre-

don et al. (2009) London > 1

MUSSA Martínez (1996, 1992, 2000); Martínez and Donoso (2010) Santiago > 1 MEPLAN Echenique et al. (1990); Abraham and Ortuzar (1999) Cambridge > 1 METROPILUS Putman (1996) Ohio > 1 RURBAN Miyamoto et al. (1996) Tokyo > 1 METROSCOPE Metro Regional Government (2010) Portland, Oregon 1 BOYCE Boyce and Zhang (1997); Boyce and Bar–Gera (2003) Chicago 1 CUFM Landis (1994); Landis and Zhang (1998b,a) California 1 POLIS Caindec and Prastacos (1995) San Francisco 1 IMREL Anderstig and Mattsson (1991) Stockholm 1 LILT Mackett (1991) Leeds 1 KIM Kim et al. (1989); Rho and Kim (1989) Urbana, Illinois 1

43

Chapter 2. Theoretical background and review of literature

Table 2.3: Characterisation and comparison of selected LUTI models.

Criteria UrbanSim ILUTE ILUMASS Purpose Assist regional Experimental tool for Develop full model of

planning at local investigating practicability microscopic land use, and state level of microsimulation transport and environment

Study world wide Toronto Dortmund regions (Canada) (Germany) Theoretical Foundation Decision rule RUM RUM RUM Speed of equilibration lag possible lag possible lag possible Perception full asymetric asymetric Resolution Spatial grid cells, zones, grid cells, zones, grid cells resolution parcels parcels Temporal annual flexible, annual resolution depending on

simulated subsystem

Functioning Scope of equilibrium partial partial partial Simulation of time iterative iterative iterative Model structure composite composite composite Transport model microsimulation microsimulation microsimulation Modelling concept hybrid hybrid hybrid Usability Calibration statistical statistical statistical technique Calibration model sub-model sub-model scope Data requirements micro objects, micro objects, micro objects,

observed behaviour observed behaviour observed behaviour

Source: Adapted from Wegener (2004); Zöllig et al. (2011)

Study regions The number of applications gives an idea of the model’s ease of application. Some are applied once, i.e. for one study region, such as ILUTE and ILUMASS, while UrbanSim has been applied several times to metropolitan areas worldwide. UrbanSim is probably the most frequently used micro-simulation model. However, the ease of application is also very dependent on data availability and data requirements.

Comprehensiveness Comprehensiveness is shown separately in Ta- ble 2.4. The first part of the table shows the comprehensiveness with respect to real world subsystems, as presented in the framework in Sec- tion 2.1. The second part represents the detail with which processes and their interactions are considered. The possible detail in both components is related, e.g. preferences of households for upper level living units can only be considered if this information is available in the representation of the building stock.

44

2.2. Models of land use development

Table 2.4: Comprehensiveness of selected LUTI models.

Criteria UrbanSim ILUTE ILUMASS Sub-systems Persons, households, cliques yes, yes, no yes, yes, yes yes, yes, no Jobs, firms yes, yes yes, yes yes, yes Land use yes yes yes Network, buildings exogenous, yes yes, yes yes, yes Regulations yes yes yes Environment no no yes Processes Demography yes yes yes Firmography yes yes yes Housing yes yes yes Working yes yes yes Travelling exogenous yes yes Transporting exogenous yes yes Constructing yes yes yes

Source: Adapted from Wegener (2004); Zöllig et al. (2011)

Modelled sub-systems What was named “society”5 in our framework, is modelled here as a general population. The models also exhibit the structure in the population to some extent. Gender, age, income and household structure are often considered. The involvement with mobility tools and social networks are less frequently considered, despite their im- portance regarding travel behaviour. The economy is represented together with jobs and firms. Again, the models differ in the degree of detail in the sense that some represent the structure of jobs in firms and others do not.

In the models, the land use sub-system is the outcome of location choices for activities. The available alternatives are buildings or facilities. Thus, it is evident that buildings are prerequisites for many final activity locations.

The transport network is another important infrastructure in the con- text of transport planning. It is used to calculate more realistic impedances between activity locations that avoid using approximations, such as Eu- clidean distances. Regulations are introduced to the model as constraints. An example on the land use side are development constraints that make sure that no residential building is located on a site dedicated for agricul- tural use. The representation of the environment is reduced to aspects that are found to be influential for the decisions modelled. Examples are

5The meaning of the term is very broad, making it evident that the models’ representations are minimalistic compared to the system in the real world.

45

Chapter 2. Theoretical background and review of literature

the representation of green spaces or lake views that increase a location’s attractiveness for housing.

Processes The models selected consider all the processes noted in the framework (Fig. 2.1). An exception is UrbanSim, which relies on exter- nal transport models. The LUTI models differ in the modelling of the processes that constitute the development of the urban system. Some processes are captured by modelling discrete decisions, others are transi- tion models that update certain quantities on the basis of assumed rates. Discrete choice models capture the decision behaviour (decision rules and preferences) of the actors represented. These models are especially suitable for markets of discrete goods. Consequently, discrete choice theory (DCT) can be applied to location choice of activities, to travel and transport demand, construction and some parts of demography. The last case looks at the choice of partners.

There are currently no models that show the modification of regu- lations and ecosystems. Given the complexity of these processes, it is reasonable to work with assumptions here.

Theoretical foundations The urban simulation models selected are based on DCT (Domencich and McFadden, 1975), which is used to model the demand side of markets in which goods, such as jobs, land, housing or transport services, are traded. The usual assumption for the decision rule is random utility maximisation (RUM). Different assumptions about the speed of equilibration in the modelled markets have also been identified. The models presented in Table 2.3 also allow independent model variables, such as supply and demand, to adjust to equilibrium with some delay. The equilibration process takes multiple time steps in such cases. Micro-simulation models can vary in their assumptions about the agents’ perception of their environment. In UrbanSim, it is assumed that market participants have full information. In ILUTE and ILUMASS, the information is assumed to be asymmetric in the markets, i.e. agents have individual knowledge and search spaces.

Resolution Table 2.3 shows the supported spatial units of the respective model and the temporal resolution in terms of a typical simulation period. The temporal resolution is an artefact of discrete simulation of time. The temporal resolution is a year in UrbanSim and ILUMASS. ILUTE allows specifying the time steps of sub-processes.

46

2.2. Models of land use development

Functioning In all three models, a partial equilibrium is calculated, i.e. equilibrium is computed separately for each submarket. This is a fundamental difference to models calculating a general equilibrium, such as the SCGE models (subsection 2.2.1). Lagged equilibration is also possible if partial equilibria are calculated.

Dynamic models represent time explicitly, which allows investigating the speed of effects. Thus, dynamic models show the development of the system over time. A dynamic model allows, for example, giving evidence on the timeframe that has to be expected for a desired transformation. Such an analysis is not possible with cross-sectional models. The selected models are simulating the evolution of the urban system over time by cal- culating a sequence of time steps. The modellers discretised the evolution and calculate cross-sections of the system in an iterative way.

The model structure of each framework is classified as composite. This means that they consist of loosely coupled submodels, where each submodel has its own independent internal structure (Wegener, 2004). A composite structure allows the integration of different model types. All selected models include or are able to be coupled with a micro-simulation transport model.

The modelling concept distinguishes between input-output models (I/O-models) and MAS. If both concepts are combined, the model is called hybrid (Zöllig et al., 2011). I/O models formulate relationships between areal units on an aggregate level. MAS explicitly simulate these relationships via the behaviour of agents in space.

Usability All submodels are calibrated with statistical methods. To the author’s knowledge, an approach to assess uncertainty for the overall model is only shown for UrbanSim (Ševčíková et al., 2007). A possible drawback of micro-simulation models is their extensive data requirements. The modeller needs data for the base year that describes the starting point and data about the behaviour of agents modelled. The latter can be derived from surveys or observations of real world behaviour, e.g. route choices can be inferred from GPS data. In many situations, however, it is possible to get a first model from reduced and partially heuristic data sets, which can be extended later if the need arises.

2.2.2.2 UrbanSim

UrbanSim (Waddell, 2000, 2002; Waddell and Ulfarsson, 2004; Waddell et al., 2005; UrbanSimProjekt, 2011) is an extendable, agent-based urban simulation model developed by Paul Waddell and his team, first at the

47

Chapter 2. Theoretical background and review of literature

University of Washington, Seattle, and later at the University of California, Berkeley. UrbanSim was initially developed in 1996 as part of TLUMIP for the Oregon Department of Transportation (Waddell, 2002). In 2005, UrbanSim was reimplemented as part of the open platform for urban simulation (OPUS). In the following, for simplicity’s sake, no distinction is made between UrbanSim and OPUS.

UrbanSim aims at simulating interactions between land use, transport, the economy and the environment for large-scale metropolitan areas and over a long time span, typically 20–30 years. The motivation for Ur- banSim is to assist integrated land use and transportation planning at the regional level within the context of growth management policies carried out at both the state and local level (Waddell, 2002). It is designed to explore and analyse the effects of policies at a disaggregated level as a scenario evaluation system (Waddell, 2011a). It is intended to support modellers and decision makers in government. UrbanSim has been ap- plied in several metropolitan areas such as the Puget Sound Region, the San Francisco Bay Area and, as part of the SustainCity project, the Canton of Zurich, the Greater Brussels Area and Île de France.

Main components and structure UrbanSim is not a single model, rather it is a tool for the integration of several models aimed at the simula- tion of urban development. UrbanSim consists generally of six models reflecting the decisions of households, businesses, developers and gov- ernments (as policy input) as well as their interactions in the real estate market (Waddell, 2002). The responsible models are transition models, relocation models, location choice models, the real estate development model, and the real estate price model. Transition, relocation and loca- tion choice models exist for employment and households in analogous, independent versions; in this chapter these are presented jointly for sim- plicity. UrbanSim does not model transport itself. To update traffic conditions, it relies on the interaction with external transport models (Wegener, 2004). As part of the SustainCity project MATSim, an agent- based travel model, and METROPOLIS, a dynamic transport model, are integrated with UrbanSim– detailed descriptions are given in Hurtubia et al. (forthcoming). Moreover, external macroeconomic models can be integrated. The scheduling and implementation of events, meaning read and write access to the database of these individual model components, is managed by a coordinator module. The UrbanSim models are described according to their processing sequence during the simulation. The se- quence of model calls does not necessarily indicate an interaction between successive models. A comprehensive model description is provided in

48

2.2. Models of land use development

(Waddell, 2002, 2000).

Accessibility model The accessibility model is the link between land use and transport. It takes the output data provided by the external transport model and maintains an accessibility pattern for the internal UrbanSim models. Models that make use of travel model output are the household location choice model (HLCM) and the real estate price model real estate price model (REPM).

Household and employment transition models The household transition model (HTM) simulates births and deaths in the population. These can be specified by providing population control totals, e.g. by income groups or age. Analogously, the employment transition model (ETM) simulates the creation and loss of jobs. Newly created households and jobs have no location. The location assignment follows later through the household and employment location choice models.

Household and employment relocation models These models sim- ulate whether households or jobs relocate. Such households or jobs are placed in a queue and receive a new location from the location choice models that are described next. If a household or job moves their cur- rent location becomes vacant. Thus, they change the real estate vacancy conditions, which are used in the real estate development and price model.

Household and employment location choice models Using a three- step process, these models select a location for each household and a job that has no current location. For households, a random sample of vacant residential units is selected first. In the second step, the selected units are evaluated for their desirability by a multinomial logit (MNL) model based on the variables and estimated coefficients included in HLCM. Finally, households pick their most desired location. The employment location choice model (ELCM) approach is very similar.

Real estate development model Developer decisions, such as new construction, renovation and reconstruction of existing structures and the type of development, is simulated by the real estate development model (REDM). The software is flexible as regards GUA. Grid cells, zones or parcels might be used as possible locations for developments. A layer containing regulations allows control over what or how much development is possible. The return on investment (ROI) is calculated

49

Chapter 2. Theoretical background and review of literature

for a set of sampled locations and generated development proposals. The alternatives, including the possibility of no development, then gets chosen by a choice model on the basis of the ROI.

Real estate price model The REPM predicts the prices of each property or GUA based on location characteristics, such as neighbourhood accessibility and policy effects. The resulting land values are used as input in the next UrbanSim iteration in the Household and Employment Location Choice Models and the Real Estate Development Model.

Key features OPUS is a framework for urban land use, transport and environmental modelling and aims to provide a shared platform that can be easily extended by developers or users and adapted for different ap- plications. Therefore, the software was released as open source software under the GNU general public license (GPL). The implementation and maintenance burden of the model infrastructure is taken on by the Urban- Sim developers. This approach enables developers and users to focus on experimenting with and applying models (Waddell et al., 2005).

The system is easily extendable, either by creating an individual OPUS package or by coupling external models via dedicated interfaces. This approach eliminates several sources of inefficiency and inconsistency, such as implementing complex data exchange methods, handling incompatible data formats and software languages or having problems accessing internal algorithms when coupling external models or adding new OPUS packages (Waddell et al., 2005).

A particular focus of the OPUS framework is on the computational performance. It is implemented in Python and takes advantage of high performance C and C++ libraries (Waddell, 2011a; Waddell et al., 2005). Another important aspect of the OPUS software is its usability by a wide group of users and modellers, without a solid expertise in software development, by providing a graphical user interface (GUI) (Waddell, 2011a).

UrbanSim provides various visualisation techniques to present model input, processes and simulation results as charts and coloured static or animated 2D maps (Waddell et al., 2005; Vanegas et al., 2010). Fur- thermore, OPUS provides integrated model estimation functionality that allows keeping the model specification consistent between estimation and simulation runs. The visualisation and analysis functionalities are being extended significantly in a closed source software called UrbanCanvas (Synthicity team, 2014a). UrbanCanvas is designed for a high perfor- mance 3D visualisation of spatial data – of which UrbanSim simulation

50

2.3. Real estate development

data is but one example. UrbanSim supports three different GUA (UrbanSimProjekt, 2011, pp.

93). These are parcels, zones and grid cells with a configurable resolution. The UrbanSim models simulate the evolution of the data store in annual steps (Waddell, 2002).

Data requirements and preparation The input to the UrbanSim mod- els includes the base year data, access indicators from the external trans- port model, and control totals derived from external macro-economic forecasts. In UrbanSim, the base year data store contains the initial state of a scenario. It represents chosen attributes of persons, jobs, real estate and locations and the mapping among these attributes. Typically, the database includes a) geographies, b) initial household information and c) job information for a given base year. The geographic layer represents administrative boundaries. Households are represented as individual ob- jects, including the requisite attributes in order to model location choice decisions. Persons are attached to households and exhibit attributes rel- evant in terms of travel behaviour. Finally, the database includes job entries, which incorporate the employment sector and represent employ- ment (Waddell, 2002). The primary sources of the base year data are usually surveys or censuses. If disaggregate information is not available, the population synthesizer in OPUS can be used instead. The synthesizer is included and gives a snapshot of the PopGen algorithm developed under the SimTRAVEL research initiative (Ye et al., 2009).

2.3 Real estate development This section describes the process of spatial development and goes into detail about the dynamics. There are three distinct sub-processes within the spatial development process: the land development or acquisition process, including laws and infrastructure elements, real estate develop- ment, which determines its future use, and the actual use of the buildings by people (Fig. 2.4). The model overview moves from general to more specific models.

2.3.1 Spatial development process The first sub-process is land development, which prepares an area for construction. One elementary task is setting the subdivision parameters that define the size and geometric form of parcels. In addition to providing

51

Chapter 2. Theoretical background and review of literature

the necessary infrastructure, roads, water and sewer systems, etc., it is mainly the zoning regulations that regulate further development. This is a standard requirement in modern countries. The actors responsible for the regulations and infrastructure are usually public bodies. Theoreti- cally, regulations cannot be changed by real estate developers, however, in practice there is some space for negotiation concerning construction regulations6.

Real estate developers construct and provide buildings, facilities and housing once the regulations, zoning and infrastructure specifications are in place (second sub-process). The real estate development process is a series of steps, each requiring several decisions. Decisions that affect the stock of real estate directly and lead to its transformation over time are of primary interest for this research. Development projects constitute the choice set for such decisions. A project is roughly characterised by its location, the time of construction, the type(s) of built space that should be provided and the quantity of each. The choice is theoretically of a discrete-continuous nature since decisions about categorical and continu- ous variables have to be made. Development decisions are typically based on prerequisites related to land development and in anticipation of the land use development, which is the next sub-process.

In the third sub-process, people, households and firms use supplied real estate according to their needs. The built environment gets used, which results in a socio-economic system (society). Land use development happens after real estate development when the members of society make use of the spaces provided that facilitates their activities. Thus, land use development is the evolution of the land use system over time.

Defining the role of real estate developers For this project, real estate developers are defined in relation to spatial development processes as the decision makers who provide built spaces of various kinds through their decisions to take on and complete construction projects. In reality, it might be difficult to clearly identify a single actor, since there are a number of institutions and persons involved in the real estate development process (See Subsection 2.3.4). From the legal point of view, it is plausible to assume that the owner of a parcel takes the ultimate decision on how to develop the site within its given constraints. Most of the time, however, there is a development consortium at work. In many situations, a developer might be referred to as the specialist who coordinates the development project and prepares the case for the decisions ultimately taken by the

6Rybczynski (2007) describes this process in much more detail on the basis of some examples in the USA.

52

2.3. Real estate development

Figure 2.4: Definition of real estate developers in respect of the develop- ment process

S p

a tia

l d e

v e

lo p

m e

n t

Sub-processes

Land use development

Real estate development

Land development

Actors

People, Firms

Real estate developer

Spatial planner, developer

LAND

Output

Society

Buildings

Parcels, infrastructure

owner. Other examples of specific actors are the builders who carry out the construction or the marketing experts who sell the final products, but there are many more (See Table 2.10). In such cases, it might be necessary to subsume relevant characteristics of the consortium into a representative decision maker.

2.3.2 Models of real estate development 7

The review of literature in this section covers real estate development in detail because the research is primarily concerned with this part of LUTI modelling. Various attempts have been made to simplify the development process in conceptual models. In her meta-study, Healey (1991) identifies four approaches to modelling the development process, which provided guidance for comparing the descriptions found in various other sources, because the identified model types stress certain aspects. The next section contains a comparison of the concepts according to described event se-

7Parts of this section are taken verbatim from Zöllig and Axhausen (2011) and Zöllig and Axhausen (2012).

53

Chapter 2. Theoretical background and review of literature

quences, involved agents, definitions of developer agents and information of interest to the developers. An introduction to Healey’s four categories follows:

Equilibrium models are based on neoclassical economic theory. The core idea is that development activities are structured by signals of eco- nomic demand. There may be supply constraints introduced, such as those imposed by a planning system. The development process is seen as un- problematic. The concept of ’rational expectations’ is also applied. Such models are usually computable and are thus treated in subsection 2.3.4. A critique includes: a) failure to explain market creation, b) that demand is not diversified (e.g. user, investor), c) assuming certainty in assessing future gains, d) no differentiation of valuation methods and e) oversimpli- fying the development process itself.

Event-sequence models outline the development as a sequence of processes and actions, shown in Tables 2.7 to 2.9. The main drawback of these models is that sequences are fixed and, consequently, there is a lack of an explanation for changing sequences.

Healey describes agency models with a focus on actors and their roles in the development process. Events may occur in parallel as well as in sequence. Such models allow the consideration of interests and strategies of actual entities and to link them in a broader context that may shape the behaviour of an actor. On one hand, these models open up complexity, but on the other, they cannot highlight critical elements or relationships with respect to the overall outcome. A main reason has been identified as the lack of driving forces in the description.

For structural models, it is emphasised that these recognise the im- portance of real estate as a financial asset and thus the dependency on financial markets that determine capital flow into the production of real estate. One example discussed is Harvey (1985), who conceptualised his idea by postulating three circuits of capital: the production circuit, the consumption circuit and the social expenditure circuit. In this framework, the social expenditure circuit is mainly dependent on state functions. It has been noted that these models hardly treat the interactions between agencies that are necessary to explain development in a specific place. Consequently, the claim is that empirical analysis must enter into the details of agency relationships.

A second meta-study by Gore and Nicholson (1991) is very similar to the one of Healey (1991). The main difference is that other model classes are chosen. Gore and Nicholson (1991) do not mention equilib- rium models, instead they split the structural models into the two classes of production based approaches and structures of provision models. Pro-

54

2.3. Real estate development

duction based approaches focus on construction as a process of putting together the input factors to produce the commodity of built property. The capital flow in and out of sectors is considered crucial in such concepts. Socio-economic relationships are added as an important factor to the production based approaches. These relationships are denominated as structures of provision and are very similar to socio-economic networks. This approach was pioneered by Ball (1983, 1985, 1986b,a) and launched a strand of research also labelled institutional analysis (Healey, 1992; Ball, 1998; Guy and Henneberry, 2000; Ball, 2002; Guy and Henneberry, 2002).

In his review, Diaz (1999) discusses the distinction between norma- tive models and behavioural models. The behavioural approach assumes non-rational decision making that is, for example, biased by an anchor point (prospect theory (Kahneman and Tversky, 1979)). He cites mul- tiple studies that found psychological effects in property market agents’ behaviour.

2.3.2.1 Event sequence and agency concepts

The literature discussed in this section describes the behaviour of devel- opers and the environment (development processes) they act in. Most authors focus on the management and tasks of the real estate development process. This sort of literature is more oriented towards practitioners and thus the concepts are not expressed explicitly as theoretical models. Nevertheless, the texts hold concepts about the development process and describe involved actors as well as their behaviour. The nature of these descriptions is more normative and may also be referred to as applied literature.

Ratcliffe et al. Ratcliffe et al. (2004) describe the development process in the United Kingdom. The introduction is a history of urban planning. The second part describes the organisation of urban planning, acknowl- edging the high degree of regulation in the real estate market. After the discussion of three current issues in urban planning, the authors turn to a description of the real estate development process that details four specific sectors of retail, office, industrial and residential use. Only the description of the general real estate development process is summarised, since the other aspects are specific to the study area. A similar description of these aspects is given in Section 4.2.

The development process is described by the necessary tasks and involved agents. It is a summary of how good real estate development

55

Chapter 2. Theoretical background and review of literature

should be done and also addresses pitfalls. The tasks are listed in the respective column in Table 2.7.

Regarding the participants (agencies) in the development process, reference is made to the three main groups defined by Graaskamp (1981)8. The authors further describe a list of specialists in a development team Table 2.10. The following is their definition of a developer:

“. . . a developer is an entrepreneur - someone who can identify the need for a particular property product and is willing to take the risk to produce it for a profit.” (Ratcliffe et al., 2004, p. 349)

The notion of risk taking can be interpreted to mean that responsibil- ity for major decisions is taken by the developer. This position of the developer as the key decision maker suggests investigating the actors’ heterogeneity first to see if the development process can be simplified to one decision.

The authors also note a more general interpretation of the term devel- oper some pages earlier:

“The generic term ’developer’ embraces a wide heteroge- neous breed of agencies, from central government at the one extreme to the small local house builder at the other.” (Rat- cliffe et al., 2004, p. 343)

In the following, the authors also outline a typology of development agencies (see the respective columns in Table 2.16 and explanations in subsection 2.3.3). It mentions that in all cases, there is some sort of assessment on whether the returns (measured with whatever indicator) are worth the investment, i.e. there are always trade-offs of some sort. When the authors focus more specifically on the issue of location choice, they list the factors that developers would consider (Table 2.12).

Ashworth A very similar book to the previous one was published by Ashworth (2008). The author takes a life-cycle approach to the devel- opment process and structures it as shown in Table 2.8. The author also lists various participants in the development process (Table 2.10). In comparison to the previous book, the landowner here has a prominent introduction. It seems that Ashworth sees the development process domi- nated by the three top-listed agencies. There is no clear definition of the developer’s role, but it is again said that different objectives are involved

8See subsection 2.3.2.2

56

2.3. Real estate development

(see subsection 2.3.3), of which profitability is key in the private sector. The major factors identified to affect real estate demand and thus its value, which would influence directly the profitability assessment and thus the decision to develop, are similar to those listed by Ratcliffe et al. (2004).

Alda and Hirschner The book describes the real estate development process and the situation in Germany (Alda and Hirschner, 2011). The process is generally defined as combining location, project idea and capital for the creation of profitable facilities. The authors draw from their experience, case studies and literature.

The authors distinguish between a short-term development process and a long-term development process. The long-term development pro- cess covers the entire life-cycle of an estate and includes, in addition to the initiating and design phase, the phases of realisation, usage, use conversion, modernisation and demolition. The short-term real estate development process is of primary interest here. Its sequence is shown in Table 2.7.

In comparison to most other sources, there is an outline of agencies that includes consumers as well (Table 2.10). A closer look reveals that some of the consumers are more like investors. There is no further description of the agencies and thus also no developer definition.

Instead of distinguishing between developer types, two essential forms of projects are identified. The first being investment projects for tenants and the second projects for private owners. Valuation of property, fund raising and project development in line with demand are further tasks that are discussed. The factors that determine adequate projects – and thus the probability of an appropriate development decision – are structured similarly to Schalcher et al. (2009), but the listing is less extensive. It can be argued that all factors ultimately determine a project’s profitability. The timing factor mainly means that developers are considering price trends or expectations around demand and supply of a targeted product sector. Indicators mentioned to be useful are: rents, returns, vacancies and the estimate of new supply planned. It also refers to the cyclic behaviour of prices that suggests anti-cyclical development activity.

Coles The study by Coles (2012) investigates whether there are different developer types in Germany. In order to present the context, she defines project development as:

“. . . interdisciplinary overall co-ordination in the areas of de- sign, economy and law/organisation . . . aiming at the realisa-

57

Chapter 2. Theoretical background and review of literature

tion of economically meaningful and environment friendly real estate projects.” (Coles, 2012, p. 40, translated9)

Further, she describes the development process by referring to the model categories proposed by Healey (1991). For her interpretation of the development process as a value creation chain, she makes use of an event-sequence model and adds possible exit points available to developers. Compared to other descriptions, the exit point elements are similar to tasks. However, it is an interesting notion that a developer has the possibility to sell the project at various stages of completion and that one could theoretically identify markets for each of these completion stages. It steams from this notion that the marketing task is seen as parallel to all project development phases in her publication.

In terms of participating agencies, she relies on other studies and concludes that most of the time, more or less the same agencies get identified. She also notes that public interest groups are rarely mentioned, even though they have been identified as possible opposition to project realisation. Neither the event sequence nor the agency listings are shown here since they are not original contributions to her publication.

Developers are identified by their main task, which is described as the coordination of processes and the management of an efficient collaboration of all participating agencies. The decision behaviour of developers is described quantitatively by analysing the answers to forty-nine written surveys. Stated weights that the respondents attribute to a priori defined target dimensions (Table 2.15) and the factors considered for project development and investment decisions are analysed. The weights on the target dimensions are used to perform a cluster analysis described in subsection 2.3.3. It is unclear how the factors queried are related to the target dimensions. The author structures the factors on a first level according to the design of the built structure and profitability, while noting that design factors influence profitability. On a second level, the factors are separated according to their relevance to location choice and investment decisions (Table 2.5). Coles reports the average weight for each determinant graphically.

Wallbaum et al. The book by Wallbaum et al. (2011) describes the context and approaches for sustainable real estate development. The strong interaction between the capital market and the real estate market is

9. . . fachübergreifende Gesamtkoordination in den Aktionsfeldern Gestaltung, Wirtschaft und Recht/Organisation . . . mit dem Ziel, wirtschaftlich sinnvolle und umweltverträgliche Immobilienprojekte zu realisieren.

58

2.3. R

ealestate developm

ent

Table 2.5: Considered design factors for real estate development

Design Investment Location

Factor 1, urban development Legal and political factors Land price Design quality Construction and planning laws Land price Excellent architectural quality Preservation orders Transport aspects Urban diversity Environmental laws Transit connection External relations of buildings Energy bylaw Proximity to highway Good integration into neighbourhoods Urban politics Proximity to airport Convincing impression of ensemble Tax law Level of immissions Reference to genius loci Strict design guidelines Low immissions Design of open space Strict city planning targets Existing land development Compatibility of scale Insufficient co-operation of administration Established location Exploitation of allowed density Economic factors Pioneer location

Factor 2, architectural design Land costs State of estate Language of form and colours Construction costs Unbuilt parcel Choice of materials Funding conditions Revitalisation object Individuality Complexity of construction Object of stock Originality Lack of suitable parcels

Factor 3, ecology Restrictive building land provision Usage of renewable energies Insufficient demand Low energy consumption of building Insufficient profit estimates Resource friendly components Insufficient tax reduction Lowering of CO2-emissions Missing subsidies

Factor 4, realisation Level of interest rates Floor plan quality Internal factors Quality of construction Insufficient equity capital Excellent details Raising operating costs Workmanship Loss from previous projects Architectural experiments Lack of personal ressources Flexibility of buildings

Factor 5, practicability Organisation of rooms Suitability of building Usability of end user

Source: Coles (2012, p. 201, 220, 211 respectively)

59

Chapter 2. Theoretical background and review of literature

discussed. Because real estate is a relatively secure investment, it is said to have a regulatory function in asset management. It is a kind of retention vessel in the capital market in economically unstable times. However, the process of real estate development is hardly discussed. The processes described are shown in Table 2.9 for comparison.

The actors in real estate development are structured into clients, con- tractors and further stakeholders (Table 2.10). No explicit definition of developers was found, nevertheless, from the association to the client group and the description of business cases, it can be concluded that agen- cies with commission power and a develop – sell strategy are identified as developers, who construct a building to sell it after completion.

There is little description of the behaviour of players in the devel- opment process, but tools are described that can be applied for certain tasks. Some of them are like guidelines and thus provide normative models according to which developers may take action. Others are soft- ware programs that can be interpreted as partly implemented behaviour since information processing formerly done by humans is now left to the computer.

The following is a translation of the authors’ classification of these tools:

• Simple planning support tools – Concepts, guidelines – Recommendations, norms, standards – Check lists, explanatory leaflets – Recommendation and disqualifying criteria – Product declarations – Quality and environmental labels for products – Element catalogue – Tendering support tools – Certificates

• Advanced planning and valuation tools – LCA-/LCC-tools – Simulations – Environmental certificates for buildings – Comprehensive planning and valuation instruments

• Foundations

• Methods • Databases • Laws and by-laws

60

2.3. Real estate development

Schalcher et al. The textbook edited by Schalcher et al. (2009) provides a comprehensive overview of real estate management in Switzerland. It is comprehensive because the entire life-cycle of a building is described. Concerning the real estate management process, the chapters subsequently treat project development, planning, realisation and management.

The event-sequence model underlying the description is mostly taken from the Swiss Norm: SN 508 112 (SIA, 2001). The first part on project development is a conceptualisation taken from a company that demon- strates that developers implicitly or explicitly develop their own frame- work for the real estate development process. Examples of other explicit concepts are discussed in Strohm (2012) or Cramer (2008, p. 60). The structure found in Schalcher’s publication can be summarised as shown in Table 2.9. The authors state, however, that the sequence is an idealisation of the process and that in reality it can be modified. Also, the marketing and communication tasks are seen to be parallel to the other tasks and last from the beginning to the end of the process.

The developer is defined as a project developer who creates the prin- ciple idea of the project. It is thus argued that he deals comprehensively with the situation at a very early stage going through the phases of project development depicted in Table 2.9. The authors state that developers take all risks in this early phase only being joined by other agencies when the idea is mature enough.

This is compared to the situation in the UK and the USA where devel- opers are supposed to carry the project solely all the way to completion and hand-over to the end-costumer. This aspect is also taken as charac- teristic to distinguish developer types (see subsection 2.3.3). The authors share the view of the developer being central to the development process, making reference to Graaskamp (1981). From the more detailed descrip- tion of the tasks and the fact that landowners are listed separately, it seems that the authors perceive the developer primarily as being without land resources.

Important agencies dealing with the developer are only mentioned on the side and in organisational charts. They are summarised as a listing in Table 2.10. An exception is made in the case of independent advisors in real estate matters who are described in more detail since it is a new, up-and-coming profession. Interestingly, all organisational charts show the owner as decision maker. The charts further reveal that diverse organisations (structures of provision) exist.

The behaviour of the developer that is of most interest in this work is designated as real estate research. The purpose of this task is to be able to make an adequate evaluation of available information. Described methods

61

Chapter 2. Theoretical background and review of literature

include the analysis of primary and secondary data as well as visiting the site. Different types of analysis are also described and the importance of their interpretation for successful project outcomes is stressed. The main emphasis of the description is on the factors that are under consideration. Besides the hard factors listed in Table 2.12, the authors point to soft factors that determine the image of a final product.

Weiss Weiss (1966) describes the development decision early with a micro-economic, neo-classical approach. She develops a conceptual model of the residential land development process (Table 2.8) and a descriptive model of the location decision process. Interviews with devel- opers in Greensboro, North Carolina form the basis for this descriptive model. Developers, landowners and consumers are identified as main agencies in the process (Table 2.11). A mathematical formulation for residential development is then given in the fourth section. In this, the developer is defined as follows:

“In the language of microeconomic theory, let us view the de- veloper as the entrepreneur of a development firm, a technical unit that transforms production inputs into saleable outputs.” (Weiss, 1966, p. 62)

Her formulation is based on the assumption that profitability P (Eq. (2.2)) is the key criterion that ought to be maximised under the constraints of production expressed by a production function F (Eq. (2.3)).

Pd,m = Rm (y1, . . . , yK ) − Cs (x s,1, . . . , x s,Q, xQ+1, . . . , x N , cd, f ) (2.2)

Where:

R : Revenue depending on product characteristics y m : Subscript reflecting consumer groups C : Costs depending on site characteristics xs and other characteristics

cd,f : Fixed overhead cost per residential package for developer type d

Fd,s (y1, . . . , yK , x s,1, . . . , x s,Q, xQ+1, . . . , x N ) = 0 (2.3)

The model hypothesises factors that structure these three categories context, actors and site to be influential for profit and thus the developer’s location decisions. Characteristics listed by Weiss (1966) are shown in

62

2.3. Real estate development

Table 2.6. The empirical work that followed is presented in Kaiser (1968); Kaiser et al. (1968); Kaiser and Weiss (1970)).

Kaiser Kaiser (1968) presents an empirical study of the importance of certain factors for residential development subdivision location choice. The empirical part of the study considers two time periods (1958–1960, 1961–1963) in Greensboro, North Carolina. Observed is the development type of 333 ft2 grid cells. The development types are differentiated along the variables of developer type (large-scale, others) and the price range of produced residential units. Another category comprises the ’no develop- ment event’ and ’non-residential development’. The relative importance of the explanatory variables is measured using the association index of Goodman-Kruskal (Goodman and Kruskal, 1959). The independent vari- ables investigated are listed in Table 2.13. From the strong associations of location characteristics with the scale of the developer’s operation, the authors conclude:

“Such a finding supports the hypothesis about expecting to find observable differences in locational behavior between different types of developers.” (Kaiser, 1968, p. 361)

Kaiser and Weiss Starting from the theoretical model developed in Weiss (1966), the authors investigate empirically the decisions of pre- development landowners to hold or sell their land10 and the developers’ location choices. The analysis is restricted to the residential sector (Kaiser and Weiss, 1970).

Regarding the decision of pre-development landowners, the authors find that the decision is mainly subject to the landowner’s estimation of future cash flow and the present or future market value of the land. It is stated that negative cash flows are mainly subject to taxation and more important than revenue. Furthermore, the decision agent characteristics are identified to be key to land sales predictions, also more important than property characteristics. The paper finds that present and future land values are influenced by: • Locational characteristics

– Prestige level – Accessibility – Institutional characteristics ∗ Zoning protection ∗ Availability of public services

10The study is describe in more detail in Kaiser et al. (1968)

63

C hapter

2. T

heoreticalbackground and

review of

literature

Table 2.6: Domains and factors of residential location choice

Context Actor Site

Socio-economic factors Geographic location Economic structure and growth prospects Community leadership Local housing market Local development industry Concentration Competition

Psychology of the times Public Policies Federal State Local Investment and service Transportation Water and sewer Schools Community maintenance

Regulatory Subdivision regulations Zoning Land use plan Annexation

Developer Type of firm Scale of operation Entrepreneurial approach Life cycle of firm

Landowner Place of residence Type of landowner Financial position Reason for holding land

Consumer Life cycle Family status Education Income

Physical Tract size Soil conditions Ground cover

Locational Social location Accessibility to urban activity places Proximity to existing development Visual quality of approach Proximity to incompatible uses

Institutional Governmentally imposed boundaries Water and sewer service Zoning regulation Subdivision regulation School district

Land ownership patterns Size of parcels under individual ownership Market availability of parcels Terms of availability

Source: Weiss (1966, p. 13, 15, 18 respectively)

64

2.3. Real estate development

∗ Subdivision regulations • Contextual characteristics

– Spatial distribution of decision agent income – Importance of non-pecuniary motives for holding land

In terms of non-pecuniary factors decreasing the probability of selling the authors find the following items: • Pre-development owner characteristics

– Living on property – Working – Single ownership – Pre-development ownership between 10 and 40 years

• Property characteristics – Not contiguous to development – Not located at fringe of urbanized area

The authors find that effects of property context and developer charac- teristics on marketability are dominant for location choice, i.e. variables affecting expected revenue are more important than those affecting costs. This is explained with more homogeneous production costs. Similar to the landowner’s decision, the developer’s characteristics are found to be important as this quote in the context of location choice demonstrates:

“But, just as in landowner and consumer decisions, the devel- oper’s characteristics affect his reaction to property charac- teristics and contextual factors.” (Kaiser and Weiss, 1970, p. 33)

The characteristics affecting marketability are listed in Table 2.13. A linked decision agent model system is presented as a framework at

the end. The model features three types: a) pre-development landowner model, b) single-family subdivision developer model and c) residential mobility model to estimate supply which is balanced via a ’residential choice model’ with demand. Demand is modelled by the residential mobility model and a demographic system.

The framework has been implemented on a computer “based on the mathematical form of the discriminant function” (Kaiser and Weiss, 1970, p. 36) and predicts subdivision probabilities for geographies. Parameters have been calibrated and predictive capacity is shown in percentages (between 50.1 and 92.4%) of correctly classified sample cells.

The authors conclude that the decision agent approach can be very useful because one examines the components of the development process more closely, which leads to a better understanding of it. It is also stated

65

Chapter 2. Theoretical background and review of literature

that the approach fits the problem of planning well since individual and organisational decisions are targeted.

Size of developments Some literature is dedicated to the specific case of large-scale developments in the USA (Johnson, 2008; Schmitz, 2004; Rybczynski, 2007; Weiss, 1987; McKeever, 1973). Johnson (2008) de- scribes the development tasks specifically for large-scale developers and thus also the development process from this actor’s perspective as well as its behaviour. It is a description of good practice on the basis of experience and case study analysis.

The scope and style of the handbooks by Schmitz and McKeever are very similar to the one by Johnson. They are more specific regarding different land uses that can be planned. Schmitz focuses on the residential case, whereas McKeever covers multiple development types. McKeever’s first section deals with residential developments, the second with spe- cial developments and the third with retail developments. Weiss (1987) describes the transformation of the community builders’ industry from 1890 to 1940 in the state of California. This analysis focuses on the institutional aspects and is an example of a structural analysis. These outlined specialities are less relevant in the context of this work since such large-scale real estate development is rare in Switzerland.

Norms: Royal Institute of British Architects (RIBA), SIA Another model of the real estate development process is outlined by the Royal Institute of British Architects (RIBA). It has been published in several versions, the most recent being the RIBA Plan of Work 2013 (Royal Institute of British Architects, 2013). This outline is the perspective and scope of architects in a development project. Consequently, acquisition of land does not show as a task (See Table 2.9). A similar guide can also be found for Switzerland (SIA, 2001). It has a similar sequence to the UK publication (See Table 2.8). In the USA, professional associations provide such guides (McKeever, 1973).

2.3.2.2 Structural concepts

Graaskamp One of the earliest and most influential researchers of real estate markets is Graaskamp. His seminal work at the beginning of the 1980s (Graaskamp, 1981) describes the development process and the work of developers from a financial perspective. Main determinants are the consumers, the financial environment and regulations, which also includes regulations on capital.

66

2.3. R

ealestate developm

ent

Table 2.7: Event sequence comparison

Ratcliffe et al. (2004) Alda and Hirschner (2011)

Concept and initial consideration Site appraisal and feasibility study Initial study of costs and returns, Knowing and preparing stake holders Find right fund (finance)

Detailed design and evaluation Assemble the professional team Prepare a brief describing the project Preliminary design plan Submission to local authority and other interested parties Make necessary changes and get final approvals from all concerned

Contract and construction Decide for contractor scheme and sign contracts Establish management structure Install appraisal system to monitor project viability Ensure checks on all delivered components Supervise all contractual affairs

Marketing, management and disposal Plan marketing campaign Decide on marketing strategy Establish management for handover Maintain security and safety Monitor (marketing) agents’ performance Reorganize financial arrangements

Project initiation Identification of starting point Location seeks capital and idea Idea seeks location and capital Capital seeks location and idea

Preparation of design brief Simple project appraisal Design phase Feasibility study Market analysis Location analysis Use concept analysis Competition analysis Risk analysis Cost analysis

Profitability analysis Project management Planning and controlling of costs Deadlines and quality

Marketing

67

C hapter

2. T

heoreticalbackground and

review of

literature

Table 2.8: Event sequence comparison (cont.)

Weiss (1966) Ashworth (2008) SIA (2001)

Decision to consider land for purchase Marketing approach Contact approach

Decision to purchase land Economic feasibility study Land engineering study Marketability study Evaluation process

Checks with financial intermediaries Checks with public officials Investment decision

Decision to develop land

Inception phase Appraisal Strategic briefing Feasibility and viability

Design phase Outline proposal Detailed proposal Final proposal Production information Tender documentation Tender action

Construction phase Mobilisation Construction to practical completion

Occupation phase Demolition phase

Strategic planning (Phase 1) Definition of needs Determine solution strategy Tendering of planning work

Prestudy (Phase 2) Feasibility study Definition of project framework

Detailing plans (Phase 3) Pre-project Construction project Permit procedure

Tendering (Phase 4) Realisation (Phase 5) Project for realisation Construction Opening Handover

Management (Phase 6) Operation Maintenance

68

2.3. R

ealestate developm

ent

Table 2.9: Event sequence comparison (cont.)

Royal Institute of British Ar- chitects (2013)

Schalcher et al. (2009) Wallbaum et al. (2011) Graaskamp (1981)

Strategic definition Preparation and brief Concept design Developed design Technical design Construction Handover and close out In use

Project development – Acquire (Founda-

tions) – Organise information

(Analysis) – Evaluate (Synthesis) – Create (Idea) – Produce (Project) – Marketing – Communication

Planning (SIA phases 1-3) Realisation (SIA phases phases 4-5) Management (SIA phase 6)

• Strategic planning / Appraisal • Preparatory study • Detailed design • Tendering • Realisation • Handover / opening

Feasibility analysis Organizing funding Risk assessment

69

C hapter

2. T

heoreticalbackground and

review of

literature

Table 2.10: Comparison of agents included

Ratcliffe et al. (2004) Ashworth (2008) Schalcher et al. (2009) Alda and Hirschner (2011)

Developer Project manager Construction manager Architect Engineers Structural engineer Geotechnical engineer Building services engineer Environmental consultant

Quantity surveyor Builder / contractor Real estate agent Valuer Solicitor

Developers Landowners Statutory bodies Professional advisers Architects Surveyors Engineers Builders and contrac- tors Planners Tax advisers Accountants Economists

Developer Know-how partners Planer Architect Engineers Specialists Craftsmen

Investor Landowner Advisor Users Public Neighbours

Costumers Public institutions Private investors Enterprises Institutional investors Insurances and pension funds Closed-end real property funds Open-end real property funds Leasing companies Real estate investment trusts

Suppliers Developer Institutional investors Builders and contractors Banks Architects Engineers Estate agents Consultants Communities

70

2.3. R

ealestate developm

ent

Table 2.11: Comparison of agents included (cont.)

Form (1954) Weiss (1966) Graaskamp (1981) Wallbaum et al. (2011)

• Real estate and building business • Larger industries, businesses and

utilities • Individual home owners and other

small consumers of land • Local governmental agencies

Landowner Developer Consumer

Consumers Producers Infrastructure providers

• Developer, Investor, Building owner (Bauherr), Owner

• Architects, Planners • Constructors • Future users, local administration,

neighbourhood organisations

71

Chapter 2. Theoretical background and review of literature

Table 2.12: Comparison of proposed influential factors

Ratcliffe et al. (2004) Schalcher et al. (2009)

Planning policy and practice Planning documentation The planning application Consultations with other bodies Planning obligations and planning gain

Economic climate for development General market conditions Overall economic climate Business cycles Urban structure theories (proximity to markets) Economic needs (urban economic model(s) used) Local markets (its conditions) Market delineation (geographically)

Demand for development Catchment area Population Employment Labour supply Rents and values Vacancy rate Informal enquiries Taxation Special incentives Interest rates Local amenities Leisure facilities Environment

Supply of development Anticipated supply Existing and planned supply Competitors Land availability The planning register Neighbouring markets Absorption and capture rates Informal enquiries Infrastructure costs Land assembly problems Land-holding issues Capital investment programs Available grants and subsidies Urban regeneration and economic development projects

Site survey and analysis Legal considerations Ownership Land assembly Boundaries and obligations Covenants Planning permission Planning and preservation Environmental protection of the site

Physical considerations Site measurement Ground conditions Topography Archaeological remains Building surveys (of neighbourhood)

Functional conditions Transportation Main services (gas, electricity, water, communication) Social amenities

Macro analysis Market Political organisation Political orientation Economic development Tax rate development Employment market structure Real estate market potential Absorption per year Benchmark Rent / Ownership Land prices

Supply Approved new construction Buildings under construction New buildings Vacancy rate Building stock according to construction period Building stock according to building type Building stock according to owner type Living units stock according to nb. of rooms Living units stock according to building type

Demand Population structure Population growth Purchasing power Education level Unemployment rate Family types Household types Target groups

Parcel analysis Location Exposition View Neighbourhood quality Image of location Immissions Shopping opportunities Schools Post office, bank, restaurant Recreation facilities Tax burden

Transport Accessibility with public transport (PT) Accessibility with car

Parcel Relics Main servicies Geology / soil Topography

Construction law Zoning legally valid District plan necessary Design plan necessary Buildings under preservation Building lines Existing servitude Allowed density

Land register Easement Security interests in real property Use transfers Building restrictions Contracts (rent / lease)

72

2.3. R

ealestate developm

ent

Table 2.13: Comparison of proposed influential factors (cont.)

Kaiser (1968) Kaiser and Weiss (1970)

Location Proportion of marginal land Proportion of poor soil (not suitable for on-site sewage) Socio-economic rank of the location Distance to nearest major street Distance to nearest elementary school Distance to employment opportunity areas Distance to central business district Amount of contiguous residential development Amount of recent contiguous recorded subdivisions Availability of public utilities Zoning protection

Developer Size [developed lots per year]

Product Price segment

Property context Locational Social prestige Accessibility to School Recreation Shopping Employment

Institutional Availability of urban services School district affiliation Stability of regulations (security of investment)

Developer characteristics Capital in firm with corresponding need for financing Size of firm [developed lots per year] Entrepreneurial approach Nature of the production process used Targeted price market

73

Chapter 2. Theoretical background and review of literature

Consumers, producers and public infrastructure are identified as the three main actor groups in real estate development processes (See Ta- ble 2.11). Cash solvency is argued to be the main goal for all actors. Therefore, the actors ’think’ in cash cycles. ’The cash cycle of infras- tructure’ also explains the financial situations of public administrations. The point is made that it is hard to install a fair tax system where each consumer pays the costs he is responsible for.

Notably, he distinguishes land and location. Land is the raw resource. Location incorporates the position of a site relative to points of interest for the activity at the site. Each activity has links to other activities, i.e. an exchange of some sort that is associated with costs. Activity-specific accessibility expresses something very similar, but Graaskamp probably had existing links in mind, where accessibility would also include potential links.

The important role of property rights is acknowledged, i.e. property can only be used as allowed by law. The value of a property is highly dependent on the regulations in place. Therefore, changes in property rights can have massive consequences for the actors’ cash cycles. The commons in England are mentioned as an example. Therefore, Graaskamp separates the most fitting use (theoretical ideal) from the most probable use (closest to the ideal that is feasible). It is argued that in most cases it is not the ideal that is installed but the most probable.

Having outlined the fundamentals of the development process, the risk management in development is discussed. It is argued that the developer faces a special situation in terms of risk, since he cannot change location. Time is also identified as critical risk element. Six possible risk manage- ment techniques are listed: a) Statistical research, b) improve forecasting by scale of operation (e.g. build 100 instead of 4 units), c) shifting risk by insurance contract, d) shifting the risk by two-party contract, e) limit liability for losses through the form of ownership and f) hedging.

Graaskamp lists three types of feasibility problems: • “The search for the most fitting site for a use(s). • The search for the most fitting use(s) for a specific site. • The search for the most suitable investment by investors.” He then explains the three approaches of feasibility calculation: • “Loan to Cost Ratio Approach (Frontdoor Approach)” • “Debt Cover Ratio Approach (A Backdoor Approach) Lender’s

Point of View” • “Default Ratio Approach (Another Backdoor Approach) Devel-

oper’s Point of View” The first approach starts from the costs of a suggested project and

74

2.3. Real estate development

calculates the rents. Starting from expected rents, the maximum expense for the land is calculated in the second approach. Compared to the second approach, more money comes from equity investment that results in an- other capital cost structure. The principle parameters from the feasibility analyses are converted to cash flow projections by the financial analyst. This is necessary because regulations influence these projections. The author concludes that regulations on capital, e.g. rules for pension funds, have a strong influence on real estate investments. The paper also dis- cusses market analysis, which is concerned with finding the competitive edge through a careful analysis of the potential costumers. This analysis is considered to be very important since it leads to basic assumptions about what can be sold.

Form An early example of a model focusing on the structure of provi- sion can be seen in the contribution of Form (1954). He argues that there is a need to explain land use change as an effect of sociological forces rather than purely economic ones. He identifies four groups of major interest in an urban environment (see Table 2.11). • Real estate and building business • Larger industries, businesses and utilities • Individual homeowners and other small consumers of land • Local governmental agencies Form acknowledges the importance of zoning in the last part of his

publication. It becomes clear that the author sees the land use development process as strictly guided by zoning regulations and that his interest is on the political process of drafting and issuing regulations.

Ball Ball’s work picks up the ideas of Form (Ball, 1983, 1985, 1986a,b, 2003). He focuses on institutions and the networks they constitute in the real estate and construction industry. Ball considers networks to be structures of provision as he states in his review paper on methods for investigating institutional structures (Ball, 1998, p. 1513). The main interest is on how to explain the emergence of the observed structures of provision. According to his findings, institutional networks are path- dependent and convergence in the networks is not obvious. In terms of the development process, Ball identifies three overlapping ’functions’ as shown in Fig. 2.5. These functions are like events, but are not fixed in a sequence.

75

Chapter 2. Theoretical background and review of literature

Figure 2.5: Ball’s functions necessary for the housebuilding process

Housebuilding

House sales Land assembly & development

Housing and mortgage markets

Land market, planning system, finance, strategy

Labour, materials, sub- contracting, management, finance, regulations

Source: Ball (2003, p. 903)

2.3.3 Developer types Research has shown that differences between actors in the urban develop- ment process exist (van Wezemael, 2005; Healey, 1994; Coiacetto, 2001). It is also noted that they can play an important role in explaining urban development phenomena (Diappi and Bolchi, 2006). For the explana- tion of urban phenomena, possible categorisations also depend on data availability.

Swiss urban context At least three typologies of developers have been developed for Switzerland (Schüssler and Thalmann, 2005; van Weze- mael, 2005; Friedrich, 2004). All three studies focus on housing. While Schüssler, Thalmann and Van Wezemael concentrate on the total produc- tion of housing, Friedrich concentrates on the treatment of the housing stock on the perimeter of Zurich.

Schüssler and Thalmann (2005) focus on the objectives of developers. They conclude that the difference in behaviour of developer originates from their business model. Either a developer is a promoter or he is an owner-occupier. Developers in the first category want to sell the product after development, while the latter keeps the building and manages it.

76

2.3. Real estate development

This distinction includes other considerations of the time horizon for evaluation. Promoters will consider a shorter time horizon than owner- occupiers. Within the second category, one can also make distinctions based on the main management goal. One group is mainly interested in the financial aspect. Developers in this category see their engagement as an investment. The main goal of a second group is utility maximisation of the community. This concerns public housing developers and cooperatives to some extent. A third group tries to sell the development as soon as market conditions are good enough. Further distinctions can be found in terms of motive, required conditions for a development start, information considered, frequency of development (proxy for professionalism), size (proxy for available resources) and evaluation of profitability.

Van Wezemael (2005) identifies three ’lines of differentiation’. In a first line, which can be called purpose, he identifies commercial and public authorities. A second line concerns the strategy where institutional developers follow portfolio management strategies to achieve their pri- marily financial objectives and other developers follow an object-oriented management strategy. In an object-oriented strategy, the option of selling is not present. In the size dimension, a third line of differentiation be- tween larger and smaller players can be found. Larger players have more resources they can draw on.

In her study, Friedrich (2004) uses a typology with three categories: institutional, public and private developers. The typology is not clearly differentiated since public developers are institutions as well. Different levels of building activity are noted.

Widler Widler (2013) investigates the potential to activate real estate owners to renew their stock. She uses three classes of owners: private, pro- fessional and cooperative11. These categories and findings are very similar to the study of Schüssler and Thalmann (2005). The author writes about the idea that professional developer are better prepared for redevelop- ment within the built environment, which makes the development process more complicated. The reasoning is that professional developers have more resources and a better overall awareness of real estate (especially on the financial side). The work subsequently analyses two governmental planning efforts that tried to activate real estate owners. Widler’s study concludes that activation is possible if the government is prepared to show win-win situations.

11Public institutions are deliberately excluded.

77

Chapter 2. Theoretical background and review of literature

Coiacetto In his analysis, Coiacetto (2001) identifies six developer types according to their behaviour in semi-structured, flexible and interactive interviews in two regions in eastern Australia. The distinction is made between a) passive local property-owning developers, b) means to a mission developers, c) specialised client developers, d) showpiece devel- opers, e) ’eye on the street’ developers and f) value-adding opportunity developers.

Ruming The typology of Ruming (2010) is based on the size of the developer company and the area of activity. His types are 1) small local, 2) medium local/regional and 3) large regional developers. Informal associations with local governments are found to be very important in facilitating the approval process.

Mc Namara McNamara (1983) defines the development process first to derive factors that are relevant for a developer typology. The development process is identified as a fusion of land (according property rights), labour (skills necessary for development) and capital aiming at material change of a site for its intended use. His definition of the developer role reads as follows:

“Developers have been described as "impressarios" orches- trating development [. . . ], bringing the land rights, labour and capital together at a particular place and time.” (McNamara, 1983, p. 92)

In empirical data on land rights purchases from Edinburgh, McNamara finds different patterns of developer interest in a site over time. He finds that these differences not only occur across available developer classifications, but also within each of them. The author concludes from this finding that the available classification is arbitrary and thus ’difficult to establish or defend’ (p. 91).

In section 4, the author proposes a typology (Table 2.14) according to the purpose of development that is defined by the strategy regarding ownership of the site. The strategy is measured with three characteristics: the time the developer holds the site before, the time the developer holds the site after the development, and the use type (he distinguishes the two categories lease and occupy). The understanding of the before develop- ment situation can be referred to as an endowment, whereas the latter is a purpose. Developers derive the strategy according to their endowment and purpose.

78

2.3. Real estate development

Table 2.14: Typology of developers by McNamara

Length of own- ership

After develop- ment

Short Long term Long term Before develop- ment

(leasing out) (owning & occu- pying)

Short Entrepreneurial builder

Land developer- investor

Developer-user

Long term (leas- ing out)

Asset clearing, probably invest- ment switch

Property im- prover / rentier

Expanding developer-user

Long term (own- ing & occupy- ing)

Capitalising as- sets

Change in re- turns from prop- erty

Owner- occupier / developer

Source: Adapted from McNamara (1983, p. 91)

It is notable that the classification assumes ownership. This means that land tenants are not supposed to do any development at all, which is explicitly mentioned in the text:

“In a society based on the concept of private property rights [. . . ], one must obtain certain rights over land before being able to develop it.” (McNamara, 1983, p. 89)

From the perspective of discrete choice modelling, this statement can be interpreted to mean that the owner of a site is the final decision maker. It is also of interest that the importance of the site varies over time. This suggests that valuation of site attributes would be subject to this fluctuation too, which would have consequences for their measurements. A further central conclusion from the paper is that a classification according to purpose becomes possible since all developments have a purpose.

Ratcliffe et al. Ratcliffe et al. (2004) write in their textbook that at least two broad types of developers can be distinguished in terms of objectives:

“In the private sector of the property industry the overriding objective is unashamedly one of profit maximization. In the public sector, other objectives apply, depending on the raison d’être of the development organization - it might , for example, be seeking to remove people from a housing waiting

79

Chapter 2. Theoretical background and review of literature

list or provide some infrastructure as its ’return’.” (Ratcliffe et al., 2004, p. 331)

The authors further detail different developers as listed in Table 2.16. Property development companies are described to be very diverse in size, specialisation, activity space and tasks they fulfil. What they have in common is that the results are based on one man’s initiative and their goal is to maximise their profit.

In comparison, financial institutions have more capital strength. They are supposed to take a longer view and to be more cautious, which leads to more conventional behaviour. Construction firms are supposed to find their ’competitive edge in the bidding process for land’ by integrating building and development. This means that the builder also acts as the developer and is able to distribute profitability over multiple tasks in the development process. Large land owners who become developers when faced with development decisions concerning their property are also men- tioned as a developer type. The authors further distinguish developers of business concerns as being firms that develop for their own purposes, however, they need to work closely with a professional property developer. The last type of developer is found in public sector agencies. They are characterised by high accountability, higher degree of participation and community consultation and diverse objectives. The objectives include provision of shelter (social housing), generation of employment and en- vironmental protection (sustainability). Regional development agencies, urban development agencies as quasi-public agencies and agencies for one-off projects (e.g. Olympic Games) are some examples.

Aside from the objectives, the authors do not explicitly name the characteristics these types are based on. The descriptions above allow the conclusion that other constituting aspects are specialisation (difference be- tween development companies and construction firms) and an endowment with land as its primary resource (large landowners).

Ashworth Ashworth (2008) presents two typologies. The first only dis- tinguishes between investor developers and merchant developers. Investor developers are supposed to keep the respective property after completion, whereas merchant developers sell the project. The finer typology is shown in Table 2.16.

Occupiers realise projects for their particular needs and are said to be less interested in market valuations. Property companies are argued to be quite exclusively interested in maximum profit. This also includes the decision about which tasks shall be carried out in house or bought

80

2.3. Real estate development

Table 2.15: Target system of developers

Target category Sub target

Financial targets Securing of liquidity Striving after profit Striving after net operating margin Maximum profitability with increased risk Moderate profitability with risk minimisation

Performance targets Quality of supply Costumer satisfaction Increase of market share Opening up of new markets Securing of competitiveness

Social / non monetary targets Striving after design activity Securing of employment Social targets of employees Ethical and social efforts Responsibility for built environment Striving after political or social influence Positive firm image in public Positive firm image of partners Long term stable relationship with administration Open communication to partners

Source: Coles (2012, p. 185)

from external specialists. Specialisation is described in terms of location, project type or type of work (new construction, conservation, refurbish- ment). Investors are argued to have a long-term view and interested in more stable and secure profits. It is also noted that such investments are usually made into portfolios that allows for diversification and thus risk minimisation.

The next type of developers, builders and contractors are described as enlarging their business strategies upstream and downstream from ’pure realisation work’. This essentially means that they engage in property selling and marketing. Finally, Ashworth also mentions public sector developers with very diverse aims. He argues that such projects would not be profitable in the private sector, but that they offer some sort of benefit to the community.

Coles Coles (2012) derives a topology by factor and cluster analysis of forty-nine responses to a written survey sent to 369 developers (active nationwide). Three developer types (see Table 2.16) are identified based on assumed dimensions of their target system (Table 2.15).

81

Chapter 2. Theoretical background and review of literature

Schalcher et al. The typology of developers described by Schalcher et al. (2009, p. 66) is based on the degree of risk-taking and the business strategy. The three types are listed in Table 2.16.

Kaiser The number of yearly lots provided is used to define the two developer types of 1) large scale developers and 2) others in Kaiser (1968). The threshold for defining large scale is chosen at 100 lots a year (See Table 2.16).

Dong and Gliebe Dong and Gliebe (2010) compare a MNL model with partially exogenous market segmentation (via interaction of explanatory variables with characteristics of developers and projects), random param- eter logit (RPL) models and latent class (LC) models with endogenous market segmentation. The external segments for the MNL and RPL models are single-family household projects and multi-family household projects. The latent classes are based on the variables of project size, contract type, developer size and specialisation12. It is worthwhile noting that the two last variables describe developers. The findings are clear taste variations across developers in which the taste for housing projects varies as well, indicating specialisation for certain projects. However, in a subsequent version of the paper, these findings were presented differently (see end of subsection 2.3.4), i.e. the developer-based segmentation is dropped due to practical considerations in forecasting (Dong and Gliebe, 2011).

Waddell Waddell (2011b) identifies fee developers and speculative developers according to the predetermined aspects of a project, namely land and tenant. In his definition, fee developers are characterised by the fact that the land where the development should take place is a given. He sees four different types of development opportunities:

1. Build to suit (a known customer) 2. Government sponsored 3. Listed land 4. Hot market These development opportunities can also be seen as market seg-

ments. The first two opportunities are the same with the exception of the government as customer in the second case, but the product sold is the development service. The last two opportunities are distinguished

12Specialisation is measured in terms of whether of not the developer developed also multi family housing (MFH) besides single family housing (SFH)

82

2.3. Real estate development

by the type of land acquisition. In the first situation, the land is offered by the current owner. In the second situation, the developer first has to convince the owner to sell his land. The developer can be active in all market segments.

Table 2.16: Typologies of land developers

Author(s) Types Classification characteristics

Ratcliffe et al. (2004)

Property development companies Financial institutions Construction firms Public sector agencies Large land owners Business concerns

Objectives Specialisation Size

Ashworth (2008)

Occupier Property companies Investors Builders and contractors Public sector developers

Purpose [Sell, hold] Type of institution

Weiss (1987)

Community builders Builders

Size [of development]

Wallbaum et al. (2011)

Developer Investor Building owner Owner

Business strategy / tenure type

Schalcher et al. (2009)

Trader-Developer Investor-Developer Service-Developer

Purpose (Business strategy) Risk taking

Schüssler and Thal- mann (2005)

Promoter Often developing Seldom developing Work provider Non-work provider

Owner-occupier Investors Work provider

Business strategy [Sell, hold] Purpose [Work, non-work] Frequency of developments [Of- ten, seldom]

Van Weze- mael (2005)

Private Public Portfolio Object-oriented Big, Small

Legal status Scope of profitability Size

Continued on next page

83

Chapter 2. Theoretical background and review of literature

Author(s) Types Characteristics used for classifi- cation

Coiacetto (2001)

Specialised client developers Passive local property owning developers “Eye on the street” developer Value adding opportunity developers “Means to a mission” developers Show-piece developers

Search strategy Endowment with property

Ruming (2010)

Small Local Purely profit Altruistic

Medium Local and Regional Large Regional

Size [Nb of developments] Activity space [Geography]

Waddell (2011b)

Spec. developer Fee developer

Purpose

Dong and Gliebe (2010)

Small project developer Mid-size project developer Large project developer

Project size [Nb of units]

Dong and Gliebe (2011)

3 Segments Size [Average nb of units built] Project size Contract type Specialisation

Coles (2012) Risk minimisers (Risikominimierer) Pluralists (Wertepluralisten) Rationalists (Kalkülgeleitete)

Table 2.15

Kaiser (1968)

Large scale developers (> 100 lots / year) Others

Size [Nb.. of developed lots]

Friedrich (2004)

Private persons Company Construction company Investment fund Cooperative

Legal status (Purpose proxi)

2.3.4 Computational real estate development models within LUTI models

This section concentrates on real estate development models. The purpose of these models within LUTI systems is to provide real estate options for households and firms. Real estate development models describe the evolution of the building stock over time and thus determine real estate supply at different points in time.

84

2.3. Real estate development

The first LUTI models did not explicitly model real estate supply. They related spatial distribution of residents directly to employment distribution (Lowry, 1964; Putman, 1974), following a gravity approach. With the second wave (Iacono et al., 2008) of LUTI models, choice modelling provided the principles for land use allocation. From then onwards, the processes were understood as resulting from market interactions and could be modelled on the basis of discrete choice analysis as demonstrated by McFadden (1977). In a first category of models, regional economic models determine land prices via a market clearing mechanism. Real estate supply, in terms of floor area, adapts as a reaction to price change. Examples of such models are TRANUS (de la Barra, 1989) and PECAS (Hunt and Abraham, 2005). A second category of models exhibits more detail in land market models by describing both demand and supply. Anas and Arnott (1993) develop a model predicting construction and demolition probabilities according to real estate related costs and expected market prices. Demand is modelled with a nested logit choice model. A similar approach is taken in DELTA (Simmonds, 1999) where the amount of newly built space is determined by calculating a ratio of current rent levels and construction costs. Allocation to zones is based on profitability.

Martínez (1992) further developed the bid-rent theory formulated by Alonso (1964) and introduced the bid-choice approach in his model MUSSA. This framework simulates an auction process simultaneously determining price and allocation of households. All the models discussed at this point assume equilibrium to determine prices. These models work on ever more disaggregated populations, i.e. considering household types and land use categories, but do not simulate individual economic actors.

With increasing computational power, the trend of disaggregation leads to microsimulation models where the behaviour of individual actors is modelled and simulated. This trend is accompanied by relaxing the assumption of equilibrium in the markets. Waddell (2002) already used a disequilibrium approach in early versions of UrbanSim. A hedonic real estate price model (Rosen, 1974) is used to determine prices. The price is then used in choice models to determine real estate supply. However, the framework also allows forcing equilibrium as shown by Wang and Waddell (2013). The authors’ conclude that this approach produces more realistic results than aggregate equilibrium models or dis-equilibrium microsimulation models. The supply side is represented by a development proposal choice model. Proposals are selected with a choice model that weights the alternatives with the expected ROI. The quantity of real estate provided gets determined according to vacancies. The most recent version implements an appraisal guide line called ’pro forma’ for ROI calculation.

85

Chapter 2. Theoretical background and review of literature

Zhou and Kockelman (2008) show a microsimulation on the basis of bid-rent theory and market equilibrium. Supply and demand of single- family housing are explicitly modelled. Real estate is built on a vacant piece of land that maximises profit. Cost of land purchase and construc- tion are imposed on the developer and selling prices are determined by the highest bid of a household at equilibrium, i.e. all households are allocated to a home. An interesting feature of the model is the treatment of large undeveloped parcels that are subdivided according to an empirically de- rived size distribution. This is a statistical parcelling model. Three years later, the same authors present another model. The choice of development type (home, apartment, retail, service, undeveloped), intensity (FAR) and building quality (price per square foot developed) is modelled as a joint decision with an MNL model. Prices for land units are determined at level of a traffic analysis zone (TAZ). The developers are assumed to have perfect knowledge of the market. They anticipate regional growth rates of households and firms, but can only adjust supply by a maximum of ±10%. There is no competition among the five uses. The market clearing is achieved by simulating location choices of households and firms. If an option has been chosen more than once, its price is adjusted within an allowed price range. If the price reaches the boundaries of the price range, a randomly chosen remaining actor gets assigned. The subdivision of parcels is no longer modelled (Zhou and Kockelman, 2011).

Hurtubia et al. (2012) show a quasi-equilibrium model for the housing market based on the bid-choice approach. Supply is exogenous. Another disequilibrium approach for microsimulating housing markets is proposed by Farooq and Miller (2012). They use game theory and random utility theory to model price formation. In ILUTE (Salvini and Miller, 2005) the housing market is modelled with a disequilibrium framework where a real estate unit can remain unoccupied. Real estate supply is determined with an economic model that predicts the number of housing starts and four location choice models that predict in which zone the housing start will occur. The location choice models are stratified according to project type (Haider and Miller, 2004). A similar approach is described by Dong and Gliebe (2011). The quantity of new housing supply is determined with a time series model. The type of project to be located is determined with Monte Carlo drawing from a fitted empirical distribution. Three models for spatial distribution of single-family housing projects are compared. The first is a MNL model without market segmentation. The second model is an MNL model with exogenous market segmentation and the third model is a latent class model, i.e. endogenous market segmentation. The authors find that the models with taste heterogeneity are theoretically more

86

2.4. Conclusions from theory

appealing. However, predictive power is not improved due to practical issues. Even though the authors use the term ’developer heterogeneity’, they do not actually discriminate developments on the basis of developer attributes. They use project size as the discriminating variable, which describes the project rather than the developing actor.

The first model to recognise the central role of the developers is the California urban futures model (CUFM) (Landis, 1994). The notion of heterogeneous actors on the supply side of housing markets is also acknowledged by Martínez and Donoso (2010) in the context of aggregate models.

2.4 Conclusions from theory

The chapter started out with a description of the object under investigation and identified six relevant subsystems for this research: a) environment, b) infrastructure, c) land use, d) society, e) economy and f) regulations. The interactions of the subsystems are of a physical and an economic nature. In this thesis, the economic interaction is more important as the economic actors are part of every subsystem. Consequently, the subsystems evolve according to the decisions of the economic actors.

Accessibility is introduced as an important indicator for location at- tractiveness because it relates the quality of transport infrastructure with the quality of the land use system. This indicator is of special interest in the spatial analysis.

Spatial development is roughly divided into a sequence of three phases: a) land, b) real estate and c) land use development. The owner of the land is the principal decision maker for the built space. Therefore, the owner is of special interest in the following research and is referred to as the developer.

The real estate development process is described in many publications. Often the concepts are presented as one of three types a) equilibrium, b) event sequence and agency or c) structural. The structural concepts are part of the focus of this research and therefore presented in more detail here. Structural concepts stress the influence of input factors, in particular of capital markets (production-based concepts) and institutional networks (structures of provision). Econometric analyses of developers are few so far. The main categories of factors to influence development decisions are identified as the a) market situation, b) location and c) decision maker.

Developers are described as central actors in the development team. The main conclusions from the literature review regarding developer

87

Chapter 2. Theoretical background and review of literature

typologies are that a) there is a lack of knowledge in respect of the supply side of land development, b) heterogeneity among real estate developers is shown in several studies c) a majority differentiates among developers according to the purpose of development, but ultimately d) there is no established typology. In the following analysis (Chapter 4) the owner is investigated as principal decision maker, i.e. developer. The purpose of a development is used to distinguish developer types.

Conceptual models go into more detail than can be captured empiri- cally, partially due to data limitations. The conceptual models describe the subject more extensively and establish the context for the empirical modelling phase Section 4.4, which is important for the interpretation of the results. One example is that the development process, which actually consists of several decisions, is simplified to one decision in real estate development models, thus the development process is reduced to one decision at a single point in time.

There is a long history of land use models in various disciplines. Re- cently, new interest in microsimulation models has grown due to increased computation power and data availability. UrbanSim is a state-of-the-art land use microsimulation software and has been implemented at various places all over the world. This software is also used for the implementation of the LUTI model as presented in Chapter 5.

Literature on computational models of real estate development shows that a) most models are market-based, b) sub-markets are identified by real estate type and c) supply is almost always modelled with a representative agent.

88

Chapter 3

Methods This chapter discusses methods and techniques of interest for this research. The focus is on the econometric methods of discrete choice analysis (DCA), discussed in Section 3.3. Further methods used for this research are the expert interview and agent-based simulation (ABS) introduced in Section 3.1 and Section 3.2 respectively. The application of the methods is described in the following parts: a) expert interviews (Section 4.1 and Section 4.3), b) discrete choice analysis (Section 4.4) and c) agent-based simulation (Chapter 5).

3.1 Expert interviews For a better understanding of decision makers, it can be helpful to inter- view them. A methodology for the qualitative analysis of expert interviews is proposed by Gläser and Laudel (2004). The suggested methodology covers the entire research process with these main steps and methods:

1. Preparation (theoretical developments) (a) Formulation of research questions (b) Decision on explanatory strategy, based on theoretical frame-

work (relating independent with dependent variables) 2. In-depth personal interviews (data collection) 3. Qualitative analysis of content (data analysis)

(a) Extraction of content i. Creation of extraction raster, based on hypothesised inde-

pendent variables ii. Extraction of characteristics of independent and depen-

dent variables (reported relationships are also extracted) (b) Preparation of content (organising extracted information so

the material can be analysed) (c) Analysis of content

Chapter 3. Methods

4. Interpretation The core of the methodology is the extraction and analysis of relevant

information gained from in-depth personal interviews. Agreements and differences to the a-priori defined theoretical framework are filtered out in the analysis. Expert interviews are conducted here to inform DCA.

3.2 Agent-based simulation

A method that has been put forward to study complex systems, such as urban development, is agent-based modelling (Batty, 2007b). The principle of this method is to reconstruct the phenomena of interest by modelling components of the system and their interactions. Economic actors are the object of first interest, given the idea that spatial development is the consequence of people’s interdependent decisions (Section 2.1). This thesis analyses real estate developers using discrete choice modelling (DCM) and ABS to assess effects of their behaviour on the urban scale. Abelson (1968) recognised very early that this method offers new potential. Ostrom (1988) introduces simulation as a third symbol system to express and communicate ideas. Simulation is the appropriate tool if qualitative results are not enough and the problem is hard or impossible to solve with analytical methods. Traffic assignment is the prototypical example in the transportation field and in DCA when estimating parameters of more complex models with simulation.

In the field of transport and land use, simulations have been developed to assist in urban policy-making, as discussed in Section 2.2. Typically, ur- ban models calculate equilibria, but there are also examples that calculate sets of possible development paths. The evolution of urban development is simulated in such systems in time steps. Discrete choice models are applied to locate people, firms and buildings.

Agent-based models (ABMs), also called multi agent systems (MAS), can be distinguished from cellular automata (CA), although both are mi- crosimulations. Benenson and Torrens (2004) describe these two concepts as the basis for geosimulation. CA models concentrate on the behaviour of spatial units determined mainly by neighbourhood effects. MAS is more flexible as it is also able to capture entities that are moveable in space, such as households, cars or companies. Combining these two concepts results in models of free agents on a cellular space (FACS)) (Portugali, 2000). Benenson and Torrens (2004) further stress the importance of ’truly geographic representations in automata models’ that are results in their framework of geographic automata systems (GAS).

90

3.3. Discrete choice modelling

3.3 Discrete choice modelling This section presents econometric method of DCA. More model types than actually used are presented in order to acknowledge available methods. Other statistical methods used here are explained briefly.

Discrete choice modelling (DCM) is the abstraction of economic actor’s decisions and represents individual demand. As in the general scientific process, there is a theoretical choice model at the beginning (discrete choice theory (DCT)), i.e. an idea of how actors make their decisions. This idea contains a) a population of decision makers, b) the objects of choice (availability must be defined) and c) a decision rule that identifies one alternative, given the choice set and the attributes of the alternatives and the decision maker. When performing a DCA, the researcher already has a DCT in mind. DCA gives the necessary steps to estimate a discrete choice model, i.e. to determine the parameters of the utility function. DCA is the empirical part of DCM. DCA allows inferring the overall demand from an observed sample because choice probabilities can be calculated.

3.3.1 A historical introduction DCM theory was developed from the late 1920s onwards. Daniel Mc- Fadden identifies the paper by Thurstone (1927) as the starting point. Thurstone’s paper describes an experiment in which respondents had to choose the brighter light source. A model is developed to explain the choices. Today, it is known as the binomial probit model. In this earliest case, the perceived quantity was light. Marschak (1960) was the first who transfered the concept to the perception of utility into the field of econometrics. This work also introduces the generally applied random utility maximisation (RUM) decision rule.

Luce (1959) contributes the discussion of axioms1 to the DCM theory. The first contribution of McFadden (1974) was the formulation of the MNL model, which he introduced as a conditional logit model because it represented demand distribution, given feasible alternatives and their attributes. In further work, McFadden showed that the MNL model is consistent with the RUM assumption, if the error terms are independently, identically distributed following an Extreme Value Type I distribution.

The nested logit (NL) model was introduced by Ben-Akiva (1973). The log sum formula is also derived in his work. The nests allow the

1Axiom I is the independence from irrelevant alternatives (IIA) assumption behind multinomial logit (MNL) models.

91

Chapter 3. Methods

consideration of the correlation of error terms in nests. The first step was taken to reduce the somewhat restrictive assumptions for the MNL model. In the following years, various models were proposed to further relax the assumptions. These include a) generelised extreme value (GEV) models (McFadden, 1978), b) multinomial probit (MNP) models (Hausman and Wise, 1978; Thurstone, 1927) and c) mixed multinomial logit (MMNL) models (Cardell and Dunbar, 1980; Revelt and Train, 1998). All these models maintain consistency with the RUM principle.

In the 1980s, more steps were taken, including a) analysis of experi- mental data (mostly in marketing labs), b) endogenous sampling (Manski and McFadden, 1981), c) a framework for the analysis of dynamics of decision-making (Heckman, 1981), d) joint discrete continuous models (Dubin and McFadden, 1984) and e) development of methods for estimat- ing models with simulation. In the 1990s, research contributions pushed the analysis of stated preference (SP) data (McFadden, 2000) forward.

3.3.2 The reference: the basic MNL

The classical discrete choice model is the MNL model (McFadden, 1974). It is widely used and often serves as a reference model due to its neat, closed-form solution for choice probabilities.

Mathematical formulation To derive the MNL model it is assumed that a decision maker n assigns a random utility U (Eq. (3.1)) to each discrete alternative j. The random utility is composed of a systematic V and a random component �. The systematic utility can be regressed to a vector of known attributes X .

Utility functions The utility is composed of a systematic (representa- tive) and a random (individual) part (Eq. (3.1)).

Uj n = Vj n (X ) + � j n (3.1)

92

3.3. Discrete choice modelling

Where:

U : Probabilistic utility V : Systematic utility j : Index of alternatives n : Index of decision maker X : Vector of attributes � : Error term following an assumed distribution

In the basic MNL the systematic utility function is linear in parameters, i.e. the utility of an alternative is described by the sum of K weighted attributes (Eq. (3.2)). The weights of the attributes are the parameters β, which capture preferences.

Vj n = β ∗ ~X j n = β1 ∗ x1 + β2 ∗ x2 + ... (3.2)

Where:

V : Systematic utility j : Index of alternatives n : Index of decision maker β : Vector of parameters ~X : Vector of attributes

When the RUM decision rule is introduced, there is also the assump- tion that the choice (denoted by ∗) will be the alternative that will provide maximum utility to the decision maker. The decision rule includes the utility function as Eq. (3.3) shows. There are DCMs with other decision rules, like regret minimisation, which picks up the idea of prospect theory, i.e. effects of anchor points2 (subsection 3.3.7.8).

j∗ = F ( j ∈ C|U ( j∗) = ma x (Uc )) (3.3)

If it is further assumed that the error terms � are identically and independently distributed following an extreme value distribution type I3,

2This is a concept from decision makers’ experiences. 3Sometimes also referred to as independent and identically distributed (IID) property. Mostly the

random term is assumed IID Gumbel distributed.

93

Chapter 3. Methods

the probability of a decision maker n to choose alternative j∗ becomes

Pn ( j ∗) =

eVj∗n J∑

j=1 eVjn

(3.4)

The MNL model is most often used for its closed form solution for probabilities.

Properties From the assumption of the error term distribution, it follows that choice probabilities have the IIA property. IIA says that the probabil- ity ratio of two alternatives is independent from other alternatives and their attributes, which is the assumption that the error terms are uncorrelated. The ratio is not affected (independent) by other (irrelevant) alternatives than the ones considered in the ratio. The sample covariance matrix of the residuals with the explanatory variables should be 0 (with maximum likelihood and MNL) (Train, 2009, p. 62). However, this assumption can be problematic with similar alternatives in the choice set, because these have (in reality) an influence on the ratio of probabilities (a well-known problem is the red bus, blue bus problem). The IIA property is therefore an assumption about substitution patterns.

3.3.3 Models considering a heterogeneous structure of alternatives

GEV If the error terms of an MNL model are correlated, the researcher can either try to find a better specified MNL model to capture the cor- relation in the deterministic utility or he can account for the correlation structure by using one of the following models: • NL • Cross nested logit (CNL) • Pairwise cross nested logit (PCL) • Ordered general extreme value (OGEV) These models are special cases of the GEV model (McFadden, 1978),

characterised by the assumption that the unobserved utility of the alterna- tives jointly follows a generalised extreme value distribution that allows correlations among alternatives. This is the logit family of discrete choice models. Thus, the assumption of uncorrelated random utilities is relaxed. The assumption about the distribution of the error terms determines the model structure.

94

3.3. Discrete choice modelling

NL The NL (Ben-Akiva, 1973) recognises that some alternatives have correlated attributes and thus belong to the same nest. The choice model is thus nested, which comes down to combining a MNL model for the nests and a MNL model for the alternatives in the nests. The result is the choice probability of an alternative i as the product of the choice probability of the nest and the choice probability within the nest.

Pi = PSm Pi|Sm (3.5)

with

PSm = eλm Im∑M

m=1 e −λl ll

(3.6)

Pi|Sm = e

Vi λm∑

j∈Sm e Vj λm

(3.7)

and

Im = l n( ∑ j∈Sm

e Vj λm ) (3.8)

Where:

λm= independence parameter Im= logsum or inclusive value

(3.9)

Therefore, it is not surprising that within the nest and across the nests, the same conditions apply as for a regular MNL model. However, the properties have been named according to the context. The independence from irrelevant nests (IIN) property corresponds to the IIA property of MNL models, but it refers to nests rather than alternatives.

3.3.4 Models considering heterogeneity of preferences

Options to consider heterogeneity in preference parameters are:

95

Chapter 3. Methods

• Deterministic representation of heterogeneity (link different tastes to observed attribute)

– Basic interactions can be realised with two basic procedures: ∗ Separate models for discrete segments of population ∗ Separate parameters within the model

– Continuous interactions (in case of continuous variables, e.g. de- veloper size in terms of turnover)

• Stochastic representation of heterogeneity – estimate mixed logit (ML) (also MMNL and random coeffi-

cient logit (RCL)) models (assumes distribution of tastes and estimate distributions of parameters)

– estimate latent class (LC) models (a finite number of homo- geneous segments is assumed, parameters are estimated per segments)

• Behavioural mixing (assumes different underlying decision paradigms) (Hess et al., 2012)

ML model Cardell and Dunbar (1980) came up with the first ML model. The concept was generalised by Walker and Ben-Akiva (Walker, 2001; Walker and Ben-Akiva, 2002). This type of model overcomes three main drawbacks of MNL models by allowing for a) random taste variation among decision makers, b) correlation between alternatives and c) correla- tion of unobserved explanatory variables over time (Picard and Antoniou, 2011, p. 29).

The probabilistic utility function of such models is specified as:

Uj n = X j n β + σjνj n + � j n (3.10)

Where:

j= Index of alternatives X = Vector of attributes describing decision situation β= Vector of estimated parameters ν= Gaussian distributed error term with mean zero and standard deviation σ �= Error term following an extreme value distribution with IID

(3.11)

The error term ν can follow any distribution (Picard and Antoniou, 2011, p. 29). Revelt and Train (1998) developed estimation methods that make these models applicable in a wide range of cases. The computational effort needed is substantially higher than for an MNL model. McFadden and Train (2000) showed that ML models are capable of approximating

96

3.3. Discrete choice modelling

any RUM consistent choice model. The general formulation for the choice probabilities is

PC (i) =

1∫ 0

. . .

1∫ 0

eZiα(�)∑ j∈C e

Zjα(�) d� (3.12)

Where:

P= choice probability conditional to choice set C α(�)= vector of polynomial functions of the uniform random vector �

Zj = vectors of polynomial functions of observed characteristics of both consumer and alternative (3.13)

LC model The LC models are a special case of ML models. Unlike other ML models, heterogeneity is captured by discrete classes and not as a continuous distribution (Hess et al., 2011). The modeller does not know the classes beforehand. However, he has to determine the number of classes (Picard and Antoniou, 2011, p. 30). LC models are also GEV models.

3.3.5 Multinomial probit (MNP)

The MNP model was introduced by Hausman and Wise (1978). Earlier, Thurstone (1927) developed the special case of the binomial probit model. The defining characteristic of these models is an assumed random term that is jointly normal distributed, i.e. N (0,Ω) (Train, 2009, p. 18). The main advantage is the handling of correlations over alternatives and time. The probit function maps an index (in this context, the utility) to a probability value of the normal distribution. The probit model is restricted to the normal distribution. Probit is the short form for probability unit.

3.3.6 Estimation methods

There are two basic estimation techniques for model estimation. The tra- ditional maximisation of the likelihood function and Bayesian procedures (Train, 2009, p. 282). The work described here relies on the traditional estimation described in the next paragraphs. In the following, a brief note is included on Bayesian procedures for the sake of completeness.

97

Chapter 3. Methods

Maximum likelihood estimation The traditional estimation maximises the log-likelihood (Eq. (3.14)) function (McFadden, 1976).

L L ( ~β) = N∑ n

l n(Pn, jn ( ~β)) (3.14)

Where:

L L : Log-likelihood function n : Index of decision makers

jn : Index of decison maker’s n alternatives ~β : Model parameters

The parameters are estimated so that the probability of the observed choices is highest. At a maximum, the first derivative is zero (Eq. (3.15)). The decision maker’s specific alternatives index jn reflects the situation that not all decision makers necessarily face the same choice set.

∂L L ( ~̂β)

∂ ~̂β = 0 (3.15)

Where:

~̂β : maximum likelihood values

Estimation statistics The adjusted ρ2 statistic is commonly used as goodness-of-fit measure. It compares the log-likelihood of a model with all parameters assumed to be zero and the model with the estimated parameters. This measure can be compared across different models if the models are estimated on the same dataset. An increase of the measure indicates a better model. The likelihood ratio test has the same purpose, however, another test statistic is used. The test statistic is −2(L ( β̂R) − L ( β̂U )) where index R refers to a restricted, an U to an unrestricted model. The test statistic is χ2 distributed with (KU − KR) degrees of freedom. KU and KR are the number of parameters of the respective models.

Additional statistics of interest are the standard deviation σ (Eq. (3.16)),

98

3.3. Discrete choice modelling

t−r atio (Eq. (3.17)) and p−value (Eq. (3.18)) of the estimated parameter with the following definitions. The standard deviations are the diagonal elements of the reciprocal Hessian matrix −H−1 (Eq. (3.20)).

~σ = I − H−1 (3.16)

Where:

I : Identity matrix

t k = βk σk

(3.17)

Where:

k : Index of attribute

pk = 2(1 −Φ(t k )) (3.18)

Where:

Φ : Cumulative distribution function of standard normal distribution

Significance is usually reported on levels of 10%, 5% and 1%.

Normalisation Discrete choice models need to be normalised before an estimation is possible (Train, 2009, p. 16). Otherwise, the model is not identified, i.e. estimation equations cannot be solved. The reasons are that only utility differences among alternatives matter and that the scale of utility is irrelevant. For MNL and NL models, the strong assumptions ’include’ normalisation. Normalisation becomes relevant when parameters of different models are compared. The models are normalised by defining them relatively to an alternative. If IID is assumed, normalisation is not

99

Chapter 3. Methods

needed, since it is assumed that the error components follow an identical distribution.

Alternative specific constants For alternative specific constants, the consequence is that models with J constants given J alternatives are not identified, i.e. there is an infinite number of constants fulfilling the estimation equations. Since only a difference in utility matters, only differences in constants can be estimated. J−1 constants can be estimated, if J alternatives are given.

The situation is similar for variables describing the decision maker (socio-economic variables), since these variables remain with the use of alternatives. Socio-economic characteristics can only be introduced for J − 1 alternatives or they have to interact with variables of the alternatives. Otherwise, the model is not identified.

Bayesian procedures The concept of the Bayesian approach is to link an a priori distribution of parameters (assumed by modeller) with an a posteriori distribution of parameters. The posteriori distribution is the result of the researcher’s adaptation based on some observations. The method no longer requires calculation of choice probabilities (Train, 2009, p. 282).

3.3.7 Practical considerations Practical considerations are presented in the following to give some in- sight on how models can be developed. The discussion includes relevant considerations for this thesis and is not exhaustive.

3.3.7.1 Interpretation of parameters

The sign of the parameters show if the attribute is positively or negatively influencing choice probability. If the variables have been normalised, the absolute values show the strength of the influence. Estimated βs are such that predicted averages of the explained variables are equal to observed averages of the explained variable in sample (N ).

3.3.7.2 Modelling techniques

General modelling techniques are: • Use of constants (for alternatives that can be given a name) • Use of categorical variables

100

3.3. Discrete choice modelling

– Dummy variables – Effect coding

• Use of alternative, decision maker and context variables • Variable transformations • Interactions

Techniques can be combined, e.g. interactions with transformed variables.

Use of constants The alternative specific constant makes sure that pre- dicted shares equal observed shares (estimates are correct on average) (Train, 2009, p. 66). Therefore, they are also useful for model calibration.

Use of categorical variables The use of dummy variables in the case of categorical variables is explained by Louviere et al. (2000, p. 86). The resulting estimates show the deviation from the reference category that is not integrated in the model. If a categorical variable is introduced using effect codes, the resulting estimates show the deviation from the overall mean (Louviere et al., 2000, p. 87).

Variable domains The variables in the utility function are from three principle domains; either they describe the alternatives, the decision maker or the context of the decision. Variables in the alternatives domain are usually hypothesised to be more relevant, which is why they are included first. Variables of the other domains are then added.

Transformations It can be helpful to transform the variable in order to capture the relationship of the variable with utility most appropriately. While scatter plots can help get an idea of meaningful transformations in the case of linear regression models, the researcher has fewer options for exploration in DCA. Careful thinking about possible relationships of the explanatory variables with utility should precede their testing.

Interactions Interacted variables capture the combination of character- istics in the form of a function to be defined. The function then enters the linear utility function as a summand.

3.3.7.3 Panel effects

If a dataset of observations contains multiple decisions of a decision maker, then one has to assume panel effects. Panel effects are the correlation of decisions due to unobserved characteristics of the decision maker. Panel effects can be captured with a a) Jackknife, b) Bootstrap, c) Sandwich,

101

Chapter 3. Methods

d) MMNL and e) error component approach. (Daly and Hess, 2013) Analysts are warned to use the theoretically favourable methods, MMNL and error components, first because they can bring potential pitfalls in specification. The Sandwich method is recommended for a simpler and more direct treatment of panel effects.

The sandwich4 estimator is defined as shown in Eq. (3.19):

S = (−H)−1B(−H)−1 (3.19)

Where:

H : The Hessian matrix Eq. (3.20) B : The Berndt-Hall-Hall-Hausman matrix Eq. (3.21)

The Hessian matrix (Eq. (3.20)) is the second derivative matrix of the likelihood function (Train, 2009, p. 186)

H = ∇2 L ( ~β∗) (3.20)

Where:

H : The Hessian matrix ∇ : Nabla operator ~β∗ : Estimated parameters (these maximise the likelihood function)

The Berndt-Hall-Hall-Hausman matrix is defined with the matrix elements as shown in Eq. (3.21) (Daly and Hess, 2013, p. 7).

B = ∑

n

L j n L k n (3.21)

Where:

B : The Berndt-Hall-Hall-Hausman matrix L jn : Derivative with respect to model parameter j of the contribution to the log likelihood function

from observation n

4Also denominated as the robust variance-covariance matrix. (Bierlaire, 2012)

102

3.3. Discrete choice modelling

3.3.7.4 Choice set formation problem

Choice set formation is necessary if the choice set is not clear. Therefore, it is mainly relevant when working with revealed preference data. Some- times the theoretically available alternatives are known to the researcher, but he does not know which alternatives have actually been considered for the observed decision. In other cases, it is mostly about deciding the relevance of an alternative.

In case of sampling for choice set formation, a sampling correction is introduced to the standard logit formula (Eq. (3.4)) to consider the selection bias (Guevara and Ben-Akiva, 2013, p. 33). In the case of an MNL model, the sampling correction is

ln P( An| j ) (3.22)

Where:

A : sampled choice set n j : chosen alternative

The model collapses to a standard logit if the sampling correction is the same for all alternatives, which is the case with a sampling protocol of random sampling.

Choice set sampling is another way to account for a similarity of alternatives. This means it is an alternative, for example, to nested logit models to account for the fact that some alternatives are more likely to be less relevant and thus have a lower utility. Weighted sampling can be applied in this context. In spatial choice situations, such as location choice or destination choice, the weight can be a Euclidean distance.

3.3.7.5 Spatial similarity of alternatives

In the literature, there are three concepts for spatial similarity: compet- ing destinations (Fotheringham, 1988), dominance (Cascetta and Papola, 2009a) and agglomeration (Bernardin et al., 2009). A particularly good review by Hunt et al. (2004) on spatial choice modelling shows that the competing destinations approach is a special case of choice set modelling. The more general approach of implicit availability/perception random utility (IAPRU) is described in Cascetta and Papola (2001).

103

Chapter 3. Methods

3.3.7.6 Endogeneity

Considering endogeneity relaxes the assumption that explanatory variables are not correlated with the error term. Endogeneity is present if the �’s are not independently distributed. Techniques to handle endogeneity are 1. Berry, Levinsohn and Pakes (BLP) approach (fixed-effects procedure to solve market level endogeneity) (Train, 2009, p. 318), 2. control function approach (instrumental variables are the special case of a control function, when λ = β (Guevara and Ben-Akiva, 2006, p. 61)) (Train, 2009, p. 334) and with a 3. full maximum likelihood approach (Train, 2009, p. 340).

3.3.7.7 Heteroscedasticity

Heteroscedasticity is the characteristic of a random variable that different dispersions can be found in subpopulations. In the context of discrete choice models, this can be the case for distributions in subpopulations of observed choices, e.g. the variance of the unobserved utility components have another variance for observations in Zurich than in Winterthur. To find heteroscedasticity, one has to analyse the distribution of the unob- served utility component. If subpopulations of the observations have different variances of the �’s heteroscedasticity is given. Heteroscedas- ticity can be accounted for by estimating scale parameters for given subpopulations. Another method to tackle heteroscedasticity is the error components approach (Walker, 2001).

3.3.7.8 Behavioural mixing

Behavioural mixing5 points out discrete choice models, which allow for different decision rules among the observed decision makers. The re- searcher assumes in such cases that decision makers in the dataset applied different decision rules for their decisions. The decision rules6 are accom- modated in a latent class framework. Some of the more common decision rules in literature are a) RUM, b) lexicographic, c) multiple reference points (Dugundji and Walker, 2005), d) prospect theory (Kahneman and Tversky, 1979), e) elimination by aspects (EBA), f) dominance variable / ranking (Cascetta and Papola, 2009b) and g) random regret minimisation (RRM) (Chorus et al., 2008). By far most often applied is utility max- imisation. Hess et al. (2012) demonstrate the method using a mixture of RUM, lexicography, multiple reference points, elimination by aspects and random regret minimisation.

5The models are also called multi-paradigm models. 6also named decision paradigms

104

3.4. Conclusions from methods

While most studies focus on heterogeneity within one main component of discrete choice models, there are a few recent exceptions in combining techniques (Teye-Ali et al., 2013). Their paper presents a simultaneous application of latent classes in a nested logit model.

3.3.7.9 Testing the specification

The different choice models are derived under a specific assumption about the distribution of the unobserved utility components �. Hence one can check if the right model specification has been chosen by testing the distribution of the �.

If alternatives are similar, then the � is correlated. If IIA is assumed for a set of alternatives, then the �’s within that set of alternatives are uncorrelated. IIA can be tested with the Hausman test (Hausman and McFadden, 1984)).

3.4 Conclusions from methods DCT proposes a variety of models to capture heterogeneity in populations of decision makers. DCT further suggests to start with an MNL model which serves as a reference. This is done in subsection 4.4.2. Introduced practical consideration are used for model development. Models with deterministic representation of heterogeneity are used in subsection 4.4.3. More sophisticated models with stochastic representation of heterogeneity and behavioural mixing are not applied due to data limitations discussed in subsection 4.4.1.

105

Chapter 4

Analysing Zurich’s real estate development

The chapter describes the analysis of real estate development in the Canton of Zurich. First, a theoretical framework and explanatory strategy is introduced. A section on available data and their descriptions follows. The third section covers the qualitative study with expert interviews, followed by a description of quantitative model estimations. Several of the models developed are shown, and their advantages and disadvantage discussed.

4.1 Theoretical framework and explanatory strategy1

In this section, a model with heterogeneous real estate developers is em- bedded in urban economic theory, followed by a description of why real estate developer types are expected to behave differently on the basis of discrete choice theory. The expected consequences are discussed on a more aggregated level that relates the work to classical urban economics. The following postulated arguments give the background for the interpre- tation of discrete choice model estimations and simulation results.

The theoretical framework pictured in Fig. 4.1 is derived from the literature review, discrete choice modelling theory and the data at hand. The independent components and assigned variables are shown with light grey boxes and dependent components and variables are shown in dark grey boxes. The independent variables are related to the dependent variables through the decision process, which is shown in the centre of the

1Parts of the section are taken verbatim form Zöllig and Axhausen (2012).

Chapter 4. Analysing Zurich’s real estate development

Figure 4.1: Conceptual basis for the expert interviews.

Real estate developer ●Professionalism

● Legal form ● Size

● Number of projects ● Workers ● Turnover

●Purpose ●Specialisation

● Range project size ● Uses built ● Market segments aware of ● Market segments active in

●Endowment ● Portfolio ● Portfolio size ● Portfolio structure

Development event

Development decision

Real estate developer

Alternatives

Conditions

Decision making ●Strategy

● Portfolio strategy ● Diversification ● Good option ● Favoured projects ● Evaluation method ● Search strategy

● Search space ● Origin development option(s)

●Perception of locations ● Parcel ● Municipality ● Spatial distribution ● Neighbour activity

●Structures ● Mix of uses ● Number of buildings

●Criteria ● Main criterion ● Amortisation ● Profitability ● Pre selling

●Decision unit ●Networking

● Own Function(s) ● Partners ● External Functions ● Equity ratio

●Information basis ●Timing land acquisition

Development decision •What

•Size •Project sizes •Number of buildings

•Use(s) •Mix of uses

•Where •When

Spatial development

Decision making

figure. The rough sketch of the process starts with the general conditions that impact all the alternatives and the developers. This could be an economic decline, a natural hazard or a change in urban policies. The confrontation of the developers with the alternatives results in a decision situation, which is depicted as filtering the alternatives framed by the trapezoid. Alternatives are described with the same dimensions as the development decision that is the chosen alternative. The consequence of the decisions is the development event, which constitute the evolution of the building stock.

The main components that constitute the development process are the general conditions, the alternatives and the developer with his decisions. These components are described in more detail with the variables shown in the grey boxes on each side of Fig. 4.1. Using the terms of Gläser and Laudel (2004), the definition of the variables (first level, bullet list) is given by their dimensions (second level, bullet list), which can also be detailed in terms of indicators (third level, bullet list).

In this framework, the developer is seen as the entity that takes the final development decision. This definition assumes that the developer has the relevant information at his disposal. This does not necessarily mean

108

4.1. Theoretical framework and explanatory strategy

that the decision maker does the preparatory work himself. The actor who carries out most of the development work in terms of planning and project management can also be a service provider. Thus, it is possible to have the distinction of an owner and development service provider that have a development service contract. Therefore, one can also refer to such actors as contractors. However, it is the owner who ultimately takes the decisions (Schalcher et al., 2009; McNamara, 1983).

In the evaluation process, the developer sorts out his favourite alter- natives through iterative decisions. The alternatives are combinations of development sites and structures in a certain area, in this case, in the Canton of Zurich. The development sites, which are parcels in the study area, are characterised by attributes, such as geometry, slope and radiation index. In addition, they can be attributed with characteristics of their sur- roundings. The structures are the physical elements under consideration to be built. The perception2 of these attributes is part of the behaviour. The discrete choice analysis (DCA) in Section 4.4 quantifies these perceptions. A project is the combination of structures, location and the timing of realisation.

4.1.1 Different developer behaviours In the city of Zurich, cooperatives contribute 18% of the residential build- ing stock, non-profit organisations (public housing, foundations) hold 13% and private owners 50%, while bigger companies and pension funds have a share of 19% (Stadt Zürich, Stadtentwicklung, 2008). 25% of the residential building stock is assigned to non-profit housing. These numbers stand out in national and international comparisons. The Canton of Zurich shows a different composition to the city of Zurich: only 15% of the overall residential building stock is owned by non-profit organisations (Zaborowski et al., 2001). It can be assumed that different developers effect the composition of the building stock and contribute to the specific character of the built environment of a certain place or region.

The developer is an actor with a certain degree of professionalism, a purpose (profit or non-profit driven), a given specialisation and a certain endowment (see also subsection 2.3.3). Based on their individual char- acteristics, the decisions they make can be supposed to be different. For instance, a developer with high level of professionalism (public limited company, dozens of projects each year and millions in turnover) should be able to invest more resources in the search for new locations and be active in a wider market. In other words, they should have different cost struc-

2In discrete choice modelling (DCM) the perception is estimated in the parameters.

109

Chapter 4. Analysing Zurich’s real estate development

tures than other competitors in the market. In addition, such a developer will probably have a good information base in terms of data, in-house know-how or via business cooperations.

Purpose The purpose of the construction is particularly interesting be- cause it is likely to determine the perception and weighting of (location) factors (Kaiser and Weiss, 1970, p. 33). There are two broad categories: developers doing the project for their own use and developers who build for a real estate market. One can assume that personal use developers are less interested in the market conditions, since they do not intend to market their property in the near future. For these types of developers, factors supporting their intended use on the plot are more important (Arentze and Timmermans, 2007).

Comparative advantages Development opportunities are unique be- cause of two aspects: • The site itself is unique, each parcel has a fixed location. • Each developer has individual characteristics that influence choices

regarding projects (Kaiser and Weiss, 1970). Some combinations of site and developer fit better than others. A large- scale developer with a lot of overhead needs development opportunities with appropriate returns to investment. On the other side of the spectrum are households that develop their customized houses. The location choice for a real estate development project will be in accordance with these prerequisites. Property ownership is a special prerequisite in this context.

Another aspect of developer heterogeneity is their individual history, i.e. the situations when an owner decides to do a project (and thus becomes a developer (McNamara, 1983, p. 89)) come about in various ways. a) offer of an owner who has to sell his property, b) inheritance, c) active search for buying and reselling or d) maintenance work to maintain the property value.

Selective regulations Along with the legal form of a developer, come regulations that can only be fulfilled with a certain professional experi- ence. The applicable regulations can have direct consequences for the final decision when some alternatives are excluded by these regulations (Bundesversammlung der Schweizerischen Eidgenossenschaft, 1983).

110

4.1. Theoretical framework and explanatory strategy

4.1.2 Expected consequences for spatial development Alonso’s bid-rent model The bid rent theory of Alonso (1964) explains how the comparative advantages of a location can be capitalized into the land price, e.g. lower transportation costs allow the budget constrained household to make a higher bid for a central location3. On the supply side, the developer needs a piece of land as primary resource. This piece of land will be more expensive at central places due to the higher bids, i.e. the primary input factor is more expensive. The increase in land price leads to factor substitution of land into construction technology and know-how. This relationship is the core of Muth and Mill’s theory (Mills, 1967; Muth, 1969). On the supply side, the requirements with respect to know-how, financial resources and technology increase. Consequently, a certain level of specialisation is necessary to build in central locations.

Spatial effects Reasons to expect more professional developers at cen- tral locations are a) the market size is larger at central locations. Higher demand requires suppliers capable of providing the quantities needed. b) When financial agents are present, capital can be borrowed. Presum- ably, there are lower capital costs for professional large-scale developers since they can provide more securities. This favours professional develop- ers at central locations that require more capital input.

Parcel size Parcel size is the result of the ownership history of the land and of any attached buildings. The consolidation of parcels is in itself a costly and lengthy process. Therefore, the existing parcels cannot easily be adjusted. The existing pattern of parcels of different sizes offers opportunities for developers of different scales.

Trends in the supply industry The DOCUMEDIA data on develop- ments in the last ten years show that developers doing a project once constitute ever smaller shares to the yearly total of development projects (Fig. 4.8). They might lack the necessary skills and connections and can- not compete with the more efficient professional developers. The effect in the long run is a consolidation of the real estate development industry towards more professional suppliers.

Propensity to boom and bust Professional developers who sell their projects afterwards are focused on profit in the short run (Schüssler and

3Centrality is measured in terms of accessibility in this case.

111

Chapter 4. Analysing Zurich’s real estate development

Thalmann, 2005; McNamara, 1983). They are interested in current con- struction prices and sale prices in the short-term (McNamara, 1983). Anyone who builds for their own use is interested in long-term prices (Wallbaum et al., 2011; Ashworth, 2008). For single-family housing de- velopers, it can be argued that individual preferences are more important than market conditions. Thus, a population of non-professional real estate developers is hypothesised to be less prone to booms and busts.

Consolidation and built structure Assuming market consolidation and increasing returns on investment, an increase in average project size (mea- sured in terms of investments) is to be expected. Residential developers can achieve this goal by building housing estates. This type of develop- ment seems more common in the USA compared to Europe and certainly happens on a totally different scale. However, we can also find hous- ing estates in Switzerland and developers specialised in their provision (Fassbind and Göhner AG, 1960).

4.2 Developers and development projects in Zurich4

The environment in which real estate developers act is the real estate market of which space, time, commodities, consumers, suppliers and price are the primary elements5. The conditions of the observed market can be described briefly using these core elements. The description includes an identification of submarkets6. With the description, a more complete picture is drawn, within which the developer is but one element. The description is based on the literature reviewed and on real estate development data at hand.

4.2.1 Real estate market segmentation The study area is the Canton of Zurich, which is a busy centre in northeast- ern Switzerland. In the following, Zurich’s real estate market is described along the dimensions of space, time, commodities, consumers and suppli- ers. Fig. 4.2 shows the Canton and its settlements in black. The two major

4Parts of this section are taken verbatim from Zöllig and Axhausen (2011). 5Price is considered a secondary element because it results from the interaction of customers and

suppliers. 6What submarkets can be found and how it can be done is summarised by Watkins (2001). His

analysis does not cover time and suppliers in the market. Space, commodities and consumer are identified as submarket dimensions.

112

4.2. Developers and development projects in Zurich

Figure 4.2: Map of the Canton of Zurich

Data: c© 2013 swisstopo (JD100042)

cities are Zurich at the north end of Lake Zurich and Winterthur, 20 km to the northeast. The Canton’s real estate market is a spatial sub-market of the national real estate market.

In previous studies by Rey (2009, 2011), spatial segmentation was done on the regional level distinguishing four large regions: city of Zurich, lakesides of Lake Zurich, agglomeration and peripheral areas. These regions are similar to drawing concentric circles around the city of Zurich. On a local scale, 12 subspaces are defined. A similar spatial subdivision is used by Kubli et al. (2008). While not specifically mentioned, it can be assumed that the areas have been defined ad hoc according to similar real estate price levels. More sophisticated methods, such as multi-level hedonic transaction price models are presented by Leishman (2009).

Real estate markets are dynamic, i.e. they constantly change over time. The dynamics are difficult to deal with in real estate development because

113

Chapter 4. Analysing Zurich’s real estate development

of the long production process, which requires anticipation of future market conditions. Therefore, it is crucial to analyse the development of the market over time. The study period of this study is from 2000–2010.

Products can be either commodities or services. The former can be split into different categories, e.g. building types that are closely related to intended use. Real estate services include e.g. construction, facility management or demolition. In the DOCUMEDIA data we find project types, which are construction services.

Segmentation according to building type is the most common. The residential market is best observed and most discussed in publications. Sub-segments within the residential submarket are single-family houses, terraced houses, multiple-family houses or single apartments. The remain- ing uses are hard to track because of their small numbers and diversity. Therefore, often, only two use segments, housing and commercial can be found (Ball, 2006). In the case of Zurich, commercial use and mixed use are also important categories. Demand segments are of interest to the developers since they reflect the variety of preferences and needs.

A further important aspect specific to real estate commodities is tenure type. Two common forms are ownership and lease. Segments such as condominium ownership, are a combination of building type and tenure type, more specifically the product is the combination of an apartment with tenure type ownership. For the Swiss market, Schüssler and Thalmann (2005) noted a tendency towards condominium ownership during the nineties. This is also visible in cantonal data (Rey, 2009). The shift towards condominium ownership has continued since 1999 but at a lower speed.

Prices for residential units have been rising since the year 2000. This is true for renting and selling. The quality corrected index of residential property prices went up from 177 index points to 250, an increase of 41% (Zürcher Kantonalbank, 2011). This index also shows the previous 20 years, which were characterised by a steep price increase during the 1980s which was then followed by a price decrease up to 2000. There are studies explaining the prices cross-sectionally with hedonic regression methods (Löchl, 2006; Haase, 2011; Kubli et al., 2008), of which Haase focuses on commercial real estate. Characteristics of the unit and the location are used to explain prices.

Another important metric in real estate markets is the vacancy rate. For the residential segment, it is stable around 0.5% in the study area (Fig. 4.3). As expected, it develops in reverse to the population growth rate, which is always positive for the observation period. Between 2003 and 2006, it increased due to relatively high residential production compared to

114

4.2. Developers and development projects in Zurich

Figure 4.3: Rates of population growth, dwelling supply growth and va- cancy in the Canton of Zurich from 2000 to 2010

Data: Zurich Cantonal Statistical Office (SAKZ) (2014)

population growth. From 2007 onwards, the population growth increases significantly and surpasses residential growth in three out of four years. It almost seems that supply did not have to react during 2008 – 2010 due to existing stocks.

4.2.2 Demand The prospering economy led to population growth based on immigration, which kept demand high. Properties for sale as well as rent are absorbed quickly by the market. Rental apartments are on average no longer than 20 days on the market. Properties for sale are approaching this absorption level, but still remain about 10 days longer on the market. Interest rates for 5-year mortgage loans were about 4% before 2008. In 2008, interest rates dropped below 3% (Würth and Meier, 2014) and have been slowly falling until they reached the current level of 2%. The average net return from a rental object was 4.9% in 2010 (Bröhl et al., 2011).

4.2.3 Supply The data used for the analysis of the supply side is introduced next, followed by descriptive statistics of real estate supply. Special attention is

115

Chapter 4. Analysing Zurich’s real estate development

Table 4.1: Overview on real estate development datasets

Data provider Entities Comment

DOCUMEDIA Projects Attributes on the constructed build- ings and some on the actors work- ing on the projects.

Federal building and housing register (GWR)

Buildings, living units, projects

The entities are related via keys.

Building insurance Can- ton Zurich (GVZ)

Buildings Information on volume, value of buildings and ownership.

given to the developers.

4.2.3.1 Development data

The datasets available are listed in Table 4.1. References to the datasets use the acronyms of the data provider. There are basically two different types of records: projects and buildings, as the Entities column shows. If a project is about constructing buildings, it can consist of one or more buildings. This relationship is only encoded in the federal building and housing register (GWR) data which is kept in a relational structure, i.e. the entities of buildings, dwellings and projects are related via key vari- ables. All datasets cover the observation period from 2000 to 2010. The descriptives of the variables can be found in the appendix (Appendix A.1).

The DOCUMEDIA dataset is used as a primary source of information because it contains more development project records than the GWR data and it contains more attributes of the developers undertaking the project, including names and addresses. The dataset was bought from the private company DOCUMEDIA. Development project announcements for the Canton of Zurich are recorded from 2000 to 2010. The observations are not transactions, but offers for the formation of a construction consortium.

Location choices by the developers are implicit to the observations, as are the choices for project type and time. It can be argued that once an investor announces his project, it is at a mature planning stage and the next step is realisation. Another assumption is that many decisions have already been made.

The GWR was launched as recently as 2004. Theoretically, it con- tains the entire population of developers because a building permit is required for almost all projects. However, under certain limits, construc-

116

4.2. Developers and development projects in Zurich

tion projects do not have to be reported. The relational structure of the GWR not only links projects to build-

ings, it also links buildings to living units. In theory, this allows the buildings and living units that are part of the same project to be iden- tified. However, data preparation (subsection 4.3.1) showed that fewer observations remain using the GWR projects.

The data from the Building Insurance of the Canton of Zurich (GVZ) is split in two datasets. The first is a cross-section of all buildings in the year 2000, the second contains all buildings built from 2001 to 2010. It is unclear from the meta-data, whether the set also includes buildings built and torn down in this period. This is, however, unlikely. Attributes of special interest are building volume, estimated value and owner identifica- tion, since similar information is not contained in the DOCUMEDIA or GWR data. The GVZ data is generally regarded as most accurate, since it is the basis of the compulsory building insurance.

4.2.3.2 The DOCUMEDIA data

The DOCUMEDIA data is assessed here with descriptive statistics to show the quality of the dataset. Descriptive statistics are generated with R. An ISO-8859 encoded text file is delivered. Except for the project ID, the first 60 attributes are contact details of the developer, planner and engineer of the development consortium. These are essentially the planners of a construction project. The entire development consortium would include craftsmen, representatives of the government, financiers and possibly further consultants (e.g. for energy issues). The data only contains partial information on the actors participating in the development. All planners are identified by an ID and their address. Attributes 61 to 90 are details on the projects. Attribute 91 includes the purpose of the development and represents three levels: renting, selling or private use. This attribute captures part of the strategy of the developers by giving information on how the developer is going to use the development. Attributes 92 to 181 contain details on the built structure following the classifications of the Swiss centre for construction rationalisation (CRB)7.

Data editing The data has to be edited before the analysis. In a first cleaning step, 983 records with a duplicated key variable (objektnr) are removed. The records with more useful attributes are kept for further analysis. Attributes 94 to 181 are discarded because they are not of interest

7http://www.crb.ch/crbOnline/

117

Chapter 4. Analysing Zurich’s real estate development

for this research. Then, individual attributes are checked for consistency and implausible attribute combinations.

Implausible by definition Undefined category codes are replaced with not available (NA). Zeros are replaced with NA in the baujahr attribute, e.g. an implausible entry stating construction year 200. It is also implausible that projects take place on parcels with zero square feet. These zeros are replaced by NA. Two variables contain process durations (bewdauertge, bauzeitmte) which cannot be negative. Thus, negative values are also replaced by NA.

Implausible combinations Examples of related attributes are the durations of approval and construction as well as their dates. In some cases, it is possible to compute missing durations from the start and end dates of construction in order to complete the dataset. Therefore, the dates have to be checked first for plausibility. In a few cases, it is obvious that the year has not been entered correctly. These cases are corrected so, that the dates fall into the observed time period. It can be seen that the durations have been calculated from these dates because of the negative durations found. The biggest bias comes with zero values, which are not observed values (average 0 in boxplot of raw data8, Fig. 4.4). Consequently, the zeros are replaced with NA values and durations are derived where possible, which results in a more plausible distribution.

Completeness The analysis for completeness of address fields shows that the construction site is known in 99.9% of all cases. The developer addresses are quite complete with 99.2%. The combination of both entities still yields 58,555 records, which is 99.1%. When planners of the development consortium are considered, the number of complete records drops to 62.0%. Only 4.5% complete records remain if all three consortium entities are considered (Table 4.2). More important though, is how many of the developments can be geocoded, i.e. the construction site address is more important. The address details are not of much interest for descriptive statistics and are not investigated further here. They are however of interest for the qualitative discussion described in Section 4.3.

Completeness of the categorical variables is shown in Table 4.3. Only the attribute Offer type is always available. Building type 01 has only a few missing values for New construction, otherwise, it is complete.

8Outliers are omitted. Values being less than LO B = Q1 − 1.58 ∗ IQ R/ √

n and greater than UO B = Q3 + 1.58 ∗ IQ R/

√ n are defined as outliers.

118

4.2. Developers and development projects in Zurich

Figure 4.4: Boxplots of raw and imputed construction durations

0

5

10

15

Raw Imputed Variable

C o

n st

ru ct

io n

d u

ra tio

n [ m

o n

th s]

Data: DOCUMEDIA

Table 4.2: Quality of addresses

Nb_records %

Developer 58606 99.2 Planer 36922 62.5 Engineers 3904 6.6 Construction site 59018 99.9 Developer and construction site 58555 99.1 Developer, planer and construction site 36617 62.0 All 3 consortium entities 2674 4.5 All 4 entities 2669 4.5

Data: DOCUMEDIA

119

Chapter 4. Analysing Zurich’s real estate development

Table 4.3: Completeness of categorical variables with respect to project type

Offer type Construction

stage Purpose Building type 01

Project type Count % Count % Count % Count %

New construction 21384 100.0 21271 99.5 17191 80.4 21317 99.7 Extension 14430 100.0 14316 99.2 11416 79.1 14430 100.0 Rebuilding 18095 100.0 17806 98.4 14959 82.7 18094 100.0 Renovation 789 100.0 779 98.7 578 73.3 789 100.0 Demolition 4375 100.0 4354 99.5 1934 44.2 4375 100.0

Data: DOCUMEDIA

Construction stage is almost complete for all project types. Unfortunately, the information is not of the same quality for project purposes. The availability of the attribute Purpose ranges between 44.2% and 82.7%. Location choice models are only estimated for New construction. For this category the availability of the purpose attribute is second highest (80.4%). The attributes Offer type and Construction stage are not used in DCA, but they are listed here for completeness.

In Table 4.4, the percentages of available values per numeric vari- able are given. The first two columns show that the approval and the construction period are not documented in the same quality. While the approval period is known for 74.6% of the records, the construction pe- riods are only known in 23.3%. Starting and ending dates of the two processes are of equal quality since the duration variables and dates have been made consistent. The construction costs are the best documented (99.9% of all cases). The quality is equal for all project types. This shows the importance of this information to potentially interested construction service contractors. The number of buildings is fairly complete and it only has a share of 5% of zeros9. The rest of the values are more difficult to interpret because the share of zeros is between 35–68%. Parcel area is reported in only 0.1% of the projects, which makes the information almost non-existent.

A look at how complete the variables are per project type reveals that renovation projects have the most complete documentation. This is plausible since the structures are already in place. Second best is the project type New construction, which is helpful since these are the projects to be used in the DCA. The deconstruction projects are less well

9The shares of zeros are reported in Table A.6

120

4.2. Developers and development projects in Zurich

Table 4.4: Quality of numeric variables per projects type [counts and % of available values]

Application duration

Construction duration

Construction cost

Project type Count % Count % Count %

New construction 16141 75.5 6682 31.2 21354 99.9 Extension 10099 70.0 2709 18.8 14425 100.0 Rebuilding 13346 73.8 3918 21.7 18082 99.9 Renovation 677 85.8 256 32.4 789 100.0 Demolition 3807 87.0 62 1.4 4374 100.0 (all) 44070 74.6 13627 23.1 59024 99.9

Buildings Dwellings Floors

New construction 20841 97.5 19150 89.6 18975 88.7 Extension 14284 99.0 11918 82.6 11990 83.1 Rebuilding 17866 98.7 15500 85.7 14735 81.4 Renovation 789 100.0 789 100.0 788 99.9 Demolition 4303 98.4 3281 75.0 2777 63.5 (all) 58083 98.3 50638 85.7 49265 83.4

Basements Parking lots GFA

New construction 18434 86.2 16981 79.4 16356 76.5 Extension 10970 76.0 10778 74.7 10809 74.9 Rebuilding 12980 71.7 12786 70.7 12657 69.9 Renovation 788 99.9 788 99.9 788 99.9 Demolition 2742 62.7 2742 62.7 2742 62.7 (all) 45914 77.7 44075 74.6 43352 73.4

Footprint Parcel area Volume

New construction 16388 76.6 37 0.2 18332 85.7 Extension 10856 75.2 10 0.1 10861 75.3 Rebuilding 12687 70.1 10 0.1 12699 70.2 Renovation 788 99.9 0 0.0 788 99.9 Demolition 2739 62.6 0 0.0 2745 62.7 (all) 43458 73.6 57 0.1 45425 76.9

Data: DOCUMEDIA

documented. The attribute Year built is not useful, because 85% of the values are

missing (Table A.6). In addition, it is unclear what it refers to as shown in Table 4.5. Year built seems to represent the year of planned construction for new construction projects and the year of construction of the existing structure for other project types. With this interpretation, all values for the attribute Year built in New construction-projects smaller than the attribute baubeginn are implausible, which is the case for 14 observations

121

Chapter 4. Analysing Zurich’s real estate development

Table 4.5: Descriptives of attribute Year built by project type

Project type Min Q25 Mean Median Q75 Max Sd

New construction 1400 2001 2003 2003 2006 2014 10.8 Extension 1586 1950 1968 1977 2002 2011 46.8 Rebuilding 1259 1907 1930 1950 1980 2011 88.8 Renovation 1732 1950 1949 1960 1966 2004 54.1 Demolition 1640 1930 1946 2000 2008 2014 96.7

Data: DOCUMEDIA

(e.g. minimum 1400 in Table 4.5). A look at the description of the projects suggests that nine of these cases have been incorrectly classified and should have been under type Rebuilding. Five other cases are probably reporting the building year of the structure that is being replaced. Due to this unreliability, the attribute is not used in further analysis. The year of construction is instead derived from the dates and duration of approval and construction processes.

4.2.4 Products In this section, the product specification is discussed along with the deci- sion dimensions of what, when and where.

What Table 4.6 shows the number of projects and their shares according to project type and building type10. It is not surprising that in terms of project number most work is done in the single family housing (SFH) cat- egory (43%); that it has even more projects than for all the non-residential projects, is more surprising. However, the dominance of the residential projects is strong because multi family housing (MFH) (18%) and mixed- use11 (1%) projects could be added to the segment. The small number of projects concerning mixed-use is notable. It is interesting that half of the extension projects are concerned with non-residential structures and half of rebuilding projects is done for SFH. It seems that the need for internal structural changes is faster for housing than for other uses. The fact that rebuilding is more popular in SFH than in MFH is probably a consequence of higher owner occupancy and fewer complicated decision processes. Summing up over all building types, the numbers show the

10The same building type segments are used in the DCA, Section 4.4. 11Such buildings are mostly used for residential purpose with a ground floor for other uses.

122

4.2. Developers and development projects in Zurich

Table 4.6: Project and building type

SFH MFH Mixed-use Project type Count % Count % Count %

New_construction 8054 14.0 4972 8.4 158 0.3 Extension 5629 9.5 1497 2.5 213 0.4 Rebuilding 9448 16.0 2988 5.1 222 0.4 Renovation 214 0.4 394 0.7 6 0.0 Demolition 2126 3.6 752 1.3 14 0.0 (all) 25471 43.0 10603 18.0 613 1.0

Non-residential Provisional (all)

New construction 8175 14.0 25 0.0 21384 36.0 Extension 7088 12.0 3 0.0 14430 24.0 Rebuilding 5431 9.2 6 0.0 18095 31.0 Renovation 175 0.3 0 0.0 789 1.3 Demolition 1478 2.5 5 0.0 4375 7.4 (all) 22347 38.0 39 0.1 59073 100.0

Data: DOCUMEDIA

highest share for New construction (36%) followed by 31% rebuilding and 24% extension (Table 4.7). Demolition and renovation have together a share of less than 10%.

It is problematic that datasets use different classifications for building types. The uses allowed inside the building are not clear. Only living units are known, whereas floor area dedicated for other uses is not recorded.

10% of the projects are made for tenants (Table 4.7). Interestingly, 5.2% of rental projects are Rebuilding compared to only 2.9% of New construction. Counting up the shares of Extension (1.6%), Renovation (0.3%) and Rebuilding projects, a 7.1% share of projects allow owners to raise rents. Surprisingly, few projects are carried out for sale (7.2%). The share of new construction for sale (6.8%) is a bit more than a third of the new construction for own use (19%). This suggests that the classical business case12 of a professional developer is only practiced to a limited extent in the study area. Across all projects, most are done for own use (61%). The shares of New construction, Extensions and Rebuilding are around 19% within that subset of projects. The 3.2% of Demolitions for own use can be interpreted as the first phase of a replacement. It can be

12The classical business case of a commercial developer is to buy a lot, improve the structure and sell it to a user. With this definition, it becomes impossible to identify the classical business case in the data because there is no information on buying or selling property. The classical business case is overrepresented in the expert interviews since the ten most active developers have been sampled purposely.

123

Chapter 4. Analysing Zurich’s real estate development

Table 4.7: Project type and purpose

Letting Sale Own use Project type Count % Count % Count %

New construction 1685 2.9 4011 6.8 11495 19.0 Extension 970 1.6 71 0.1 10375 18.0 Rebuilding 3045 5.2 173 0.3 11741 20.0 Renovation 195 0.3 13 0.0 370 0.6 Demolition 8 0.0 7 0.0 1919 3.2 (all) 5903 10.0 4275 7.2 35900 61.0

Data: DOCUMEDIA

Table 4.8: Project inputs and outputs in terms of buildings, units and costs by project type

Construction cost Buildings Dwellings

Project type Mio. CHF % Count % Count %

New construction 80447 73.8 32843 45.1 116598 67.8 Extension 11090 10.2 14373 19.8 16489 9.6 Rebuilding 16425 15.1 19317 26.6 32699 19.0 Renovation 789 0.7 1089 1.5 4173 2.4 Demolition 305 0.3 5125 7.0 1908 1.1

Data: DOCUMEDIA

argued likewise for the other purposes. For another 22% of the projects, the purpose is unknown (not shown in table).

Construction activity is not only recorded in terms of projects. That new construction is still the dominating project type for the transition of the building stock can also be seen from construction costs and the number of concerned buildings and dwellings (Table 4.8). Construction costs are most concentrated on new construction (73.8%).

The construction costs are the measured input for generating the output. Figure 4.5 shows the box plots of construction costs according to project type. It is plausible that costs for new construction are highest, that renovations rank second, rebuilding third, extensions fourth and demolitions fifth. The distributions are all left-skewed, i.e. there are more extreme observations with high costs.

124

4.2. Developers and development projects in Zurich

Figure 4.5: Construction costs by project type

2

4

6

New_construction Extension Rebuilding Renovation Demolition Project type

C o

n st

ru ct

io n

c o

st s

[M io

. C

H F

]

Data: DOCUMEDIA

Table 4.9: Construction costs per output category by building type

Building type CHF m3

Mio. CHF Building

Mio. CHF Dwelling

MFH 1063.3 3.9 0.5 Mixed-use 1508.2 8.6 0.6 NonResidential 1035.8 2.8 NA Provisional 9570.3 0.6 2.2 SFH 841.2 0.8 0.6

Data: DOCUMEDIA

When The construction activity in the residential sector has been in- creasing over the last ten years. In 2000, about 6400 dwellings were built compared to 11,000 in 2011 (Fig. 4.6). The numbers in Rey (2010) show that this trend is not yet so clear. Production thus follows population growth, but with a lag of approximately two or three years (Fig. 4.3). Given the relatively high population growth between 2007 and 2010, high production can be expected from 2010 onwards.

Where Half of the new apartments are built in the two major cities Zurich and Winterthur (Rey, 2013; Kubli et al., 2008). This main emphasis is also visible in Fig. 4.7. New construction further tends to concentrate

125

Chapter 4. Analysing Zurich’s real estate development

Figure 4.6: Newly constructed dwellings per year

7000

8000

9000

10000

11000

2000 2004 2008 Year

D w

e lli

n g

s

Data: DOCUMEDIA

along lake Zurich, in the Limmatthal and around the airport.

4.2.5 Developers Hauri and Steiner (2006, p. 44) come to the conclusion that the recession at the end of the 1990s forced developers to carry out tasks that are up or downstream to the actual task of producing houses. These ’big players’ try to cover more of the development process and function rather as coordinators. In this role, they engage in a lot of subcontracting and thus bring together the necessary resources in a very fragmented industry.

In the last decade, Zurich’s real estate market has been characterised by stable demand increase, moderate production and rising prices. Therefore, it is not surprising that professionally operating international developers are entering these promising markets (Cramer, 2008). Cramer (2008) lists the nine major market players, six medium market players and seven foreign market players.

In the Swiss housing market, non-profit housing providers are found to have influence (Kemeny et al., 2005). The study classifies the Swiss rental market as unitary, which is defined as a market in which barriers to non-profit providers competing in the rental market are removed. This

126

4.2. Developers and development projects in Zurich

Figure 4.7: Map of new construction projects

New construction projects

Lakes

Canton Zurich

Legend

Data: c© 2013 swisstopo (JD100042), DOCUMEDIA

supports not only public housing construction, but also the cooperatives, which are especially strong in the city of Zurich. 19% of all dwellings are built by cooperatives (Schmid et al., 2007, p. 9).

Following the definition in Table 4.10, 6% of promoters, 17% of developers with a portfolio and 77% developers without portfolio were active during the last 10 years. 33,146 of the developers only built one of any project type.

127

Chapter 4. Analysing Zurich’s real estate development

Figure 4.8: Developer type share of annual construction costs over time by project type

New_construction Extension Rebuilding

Renovation Demolition

0%

25%

50%

75%

100%

0%

25%

50%

75%

100%

1998 2001 2004 2007 2010 1998 2001 2004 2007 2010

Year

In ve

st m

e n

ts Developer type O1

Om

Smc

Data: DOCUMEDIA

Figure 4.8 shows the shares of projects carried out by Developer Type13 and Project Type over time. In the segment New Construction, the share of developer with one project for own-use (O1) developers is shrinking whereas the share of developer with multiple projects for own-use (Om) developers is increasing. In the other project type segments, the shares are relatively constant.

The box plots of construction costs (Fig. 4.9) by developer type show clearly that developer with one or multiple projects developed for sale (Smc) developers are constructing larger projects.

Professionalism Professionalism is a complex variable, i.e. it is com- posed of multiple factors (Gläser and Laudel, 2004). In this thesis, it is measured in terms of development frequency and total construction cost of the realised projects. The distribution of these two factors is shown in Fig. 4.10.

If we plot the logarithm of the size, measured in number of announce- ments, against the number of developers, there is a heavily right-skewed

13The definition of the developer types is given in Table 4.10.

128

4.2. Developers and development projects in Zurich

Figure 4.9: Construction costs by developer type

3

6

9

O1 Om Smc Developer type

C o

n st

ru ct

io n

c o

st s

[M io

. C

H F

]

Data: DOCUMEDIA

distribution (Fig. 4.10(a)). This confirms the results of Coiacetto (2009), who finds that the real estate industry is oligopolistic with many oppor- tunities for small-scale firms. He also states that the highly dynamic industry is not competitive and that it is likely to concentrate further. The oligopolistic structure is also observed by Farooq (2010). Figure 4.10(b) confirms this finding in terms of construction costs. The decline towards zero is an artefact of the minimum size of construction. This relates as well to the reporting, in the sense that no permit is required for small projects.

In most cases, it is possible to find the developer in the central com- panies index if the developing entity is not a private person or public institution. Ten of twelve large developers can be identified in the central companies index. There is one public institution and one developer that has established several companies for individual projects. This seems to be common practice when large projects endanger the existence of developing companies. This contributes to the large number of developers who only build one project.

In a first analysis 108 general constructors, 136 cooperatives and 839 architects are identified. This shows that cooperatives are very present in the construction market.

129

Chapter 4. Analysing Zurich’s real estate development

Figure 4.10: Distribution of developers’ professionalism

(a) Log-log scatter diagram of the num- ber of developers by development fre- quency

●

●

●

●

●

● ●

●

●●● ●

●●

●●●●

●●

●

●

●

●●●

●

●●●

●●●

●●

●

●

●

●●●●●●●●●●●●

●

● ● ●

10

1000

1 10 100 Freqency of developement

N b.

o f D

e ve

lo p

e rs

(b) Log-log density plot of developer size in terms of construction cost [Mio. CHF]

0.01

1.00

0.1 10.0 1,000.0 Sum of construction costs

D e

n si

ty

Data: DOCUMEDIA

On the basis of c/o signs (in care of) it can be seen that some home builders hire a professional to take care of their building project. This in- dicates that investors or owners mandate the execution of the development buying development services. In such cases, the name of the investor with a c/o is entered in the branch field. The contact details then refer to the development service provider.

Specialisation A developer can specialise in all dimensions that char- acterise development activity for a more rational production. The di- mensions available in the data are: type of work, purpose, building type and location. As research has shown, developers specialise in spatial submarkets (subsection 2.3.3). Developers can also specialise regarding a clientele, e.g. developers only or mostly developing for public institutions. The downside of specialisation is that ever fewer opportunities must be expected since no construction site is identical to another.

Specialisation can be measured with multiple metrics that have been developed for various disciplines. Here the concentration ratio (CR) is

130

4.3. Expert interviews with real estate developers in the Canton of Zurich

used; this is a metric from the economic literature (Zeebroeck et al., 2005; Cifarelli and Regazzini, 1987). It is usually used to assess specialisation of regional economies, but it is possible to adapt it to the context of individual specialisation. The difference between specialisation and concentration is that specialisation happens by choice, whereas concentration is rather the outcome from a ’macro’ mechanism. More sophisticated metrics for individual specialisation have been developed in ecology since they must take the environment of the species into account as well.

Specialisation has two qualities. Firstly, it is of interest in what cate- gory of a dimension a developer is most specialised (e.g. developing SFH or MFH). Secondly, the degree of specialisation is of interest because it indicates how strong the specialisation is. It is proposed to calculate the latter as the maximum concentration ratio of the developer’s cate- gorical choices. In analogy, this metric is referred to as CR1 since it is the ratio of the most popular alternative. The ratio of the two most frequently chosen alternatives could be calculated to become CR2. The speciality is consequently the alternative with highest CR. The analysis is limited to developers with more than one project because it is not much of specialisation if a developer only builds once.

The average of all considered dimensions is chosen as a metric for specialisation over multiple dimensions (average concentration ratio over the considered dimensions of specialisation. (AvCR)). In this particular case, type of work, building type and purpose are considered dimensions. The box plots of the AvCR by developer type (Fig. 4.11) show that Om developers are more specialised, which is plausible because they know their needs and worry less about the diversity of demand.

4.3 Expert interviews with real estate develop- ers in the Canton of Zurich14

This section describes in detail how the qualitative research approach briefly introduced in Chapter 3 was put to practice in this study.

4.3.1 Preparation As multi-agent models of transport and land use predominate, most of the explanatory strategy is given in Fig. 4.1. The research questions of this section follow from the decision to explain the development process on the basis of developer agents. The particular research questions are:

14Parts of this section are taken verbatim from Zöllig and Axhausen (2012).

131

Chapter 4. Analysing Zurich’s real estate development

Figure 4.11: Average concentration ratio (AvCR) by developer type

●●●●●●

●

●●●●●●

●

●●●●●●

●

●●●●●●●●●●●●●●●

●

●●●

●

●●●●

●

●●●●

●●

●

●

●●●●●●●

●

●●●●●●●●●●●●●

●

0.4

0.6

0.8

1.0

Om Smc Developer type

A vC

R

Data: DOCUMEDIA

1. Is it justified to assume that developer types exhibit different be- haviours as defined in Table 4.10?

2. What are the behavioural differences in respect of decision criteria, underlying information and considered alternatives?

3. Can a useful typology be applied to the data collected? 4. Is the typology based on the data, the same as the typology based

on the interviews?

Following on from the literature review and the data for the planned quantitative analysis, the research strategy is to confirm different be- haviours of developer types according to the variables and indicators found in the data collected. The qualitative work has to be seen in the con- text of the quantitative data analysis. The interviews should help explore different developer behaviour, in particular, for the hypothesised typology shown in Table 4.10, which is applicable to the data from DOCUMEDIA. Thus, the interviews investigate the behavioural differences between these developer types. The point is to relate the attributes of the developers with attributes of their decision-making.

132

4.3. Expert interviews with real estate developers in the Canton of Zurich

Table 4.10: Definition of developer types in DOCUMEDIA data

Developer type code

Developer type name Purpose Number of projects

O1 Self-owning without portfolio strategy

Own use or lease

1

Om Self-owning with portfolio strategy

Own use and or lease

Several

Smc Commercial developer / Promoter

Sale Several or 1

4.3.1.1 Creation of interview guidelines

The interview guidelines are based on the explanatory strategy and target both independent and dependent variables. The interview guidelines are organised into five sections:

1. Characteristics of the developer 2. Decision process 3. The type of projects 4. Location choice for projects 5. Assessment of market conditions

Because of the different vocabularies, it seemed appropriate to design a questionnaire for commercial developers and one for private persons (home-builder). Otherwise, some questions could have been confusing (Appendix A.3).

4.3.2 Recruitment of interviewees The aim was to interview ten developers and at least one developer of each assumed developer type (Table 4.11). Following the classification of Mayring (2002), the survey method is a problem-centric interview. Therefore, the contact details of the DOCUMEDIA data are grouped according to the developer type definition (See Table 4.10). From these groups, 20 addresses were randomly sampled for recruitment. The first sampling is thus a stratified random. In addition, the ten developers with the most projects during the studied period were selected in order to question the most relevant ones for the study area in terms of frequency. The second sampling can be classified as purposive in terms of intensity (Patton, 1990). Other options would have been to order the choices by the number of buildings or construction costs. However, since the goal was

133

Chapter 4. Analysing Zurich’s real estate development

Table 4.11: Conducted interviews

Pages of Case Legal form Type Mode transcription Duration

1 Inc. Om Face-to-face 16 01:09 2 SME O1 Telephone 7 00:32 3 Private O1 Telephone 9 00:41 4 SME Smc Telephone 12 00:49 5 Inc. Smc Telephone 11 00:49 6 Public Om Face-to-face 20 01:29 7 Inc. Smc Face-to-face 16 01:34 8 SME Smc Telephone 9 00:45 9 SME Smc Telephone 9 00:29

10 Cooperative Om Face-to-face 9 00:31 11 Institution Om Telephone 13 01:03

Sum 131 09:51 Average 12 00:53

to analyse choices, the choice was to use frequency. To reach developers with fresh memories of their development projects, selection was limited to projects submitted during the most recent year15 of the observation period. Projects not of the type new development were excluded.

Contact details of the owner or his development service provider were used for the recruitment call. The decision of whether to send the question- naire for commercial developers or for private persons was made using information gained from the recruitment call and the DOCUMEDIA data.

4.3.3 Conducting the interviews The interviews are open and semi-standardised. They are either conducted face-to-face (4) or by Skype (7). All 11 interviewees agreed on being recorded which is essential for further processing. Table 4.11 lists the analysed interviews, their mode, duration and the type of the developer.

4.3.4 Analysis of interviews The analysis of the interviews has two main parts. The first part is the transcription from recorded audio documents, and the second part is the

15gesuchvom > 4.12.2009

134

4.3. Expert interviews with real estate developers in the Canton of Zurich

text analysis.

4.3.4.1 Transcription of interviews

All transcriptions are done by the same assistant. The interviewer checks the transcription for parts that were not understood by the assistant and complements the text if possible. The recommendations of Gläser and Laudel (2004) for interviews were followed, including the use of their set of transcription rules on page 188. In addition, the following transcription rules were defined:

1. Parts that are difficult to understand are marked with red brackets, which indicates the runtime of the interview recording. (E.g. [13:45])

2. Names that are critical regarding anonymisation are marked in red font.

3. Special interview situations, such as misunderstandings are indi- cated in brackets, e.g. [misunderstanding].

The transcriptions are not anonymised, this is done during the extractions step. The transcription produced 131 pages of text, which are analysed in the following.

4.3.4.2 Analysis of content

Creation of search raster The idea is to filter the interview text with the extraction raster. The search raster is created on a theoretical basis and extracts information on the dependent and independent variables. A spreadsheet is used as a tool in which the rows are assigned to a variable and the columns are assigned to an interview. To make use of the structured interview guideline, the sequence of the variables follows the order of the questions that the variables targeted.

Extraction The characteristics of the variables of interest were extracted according to their definitions. Generally, the variable dimensions con- sidered have a nominal scale and are open, which means that additional characteristics can be added. The collection of distributed information is done with the extraction by assigning relevant contents directly to the designated variable. Reported relations of variables are extracted as well.

The questions target a dimension of a variable that suggests using the answers as a unit of analysis. In the case of follow-up questions, it is especially important to reflect on whether a new dimension or even a new variable must be added to the extraction raster and consequently as well to the theoretical framework.

135

Chapter 4. Analysing Zurich’s real estate development

In the direct interviews, the case that the interviewees brought informa- tion material on paper and referred to them during the interview occurred three times. This information is included in the extraction. Reference to this sort of information follows the citation principles.

Preparation of content The main task of preparing the extracted infor- mation is to reorganise the information pieces to follow the theoretical framework of explanation. The use of a spreadsheet was convenient for this task.

Analysis The qualitative analysis follows the analysis strategy for few cases (Gläser and Laudel, 2004, p. 243). Because of its purpose in the overall study, individual characteristics were compared to confirm the identifiable developer types. For the analysis, all eleven cases are compared on the basis of selected independent variables contrasting them with selected dependent variables. The dependent variables are checked for similarities according to the independent variables that indicate a relationship. The focus is on the independent variables according to which the developers in Table 4.10 were grouped. The primary interest is in the relationships of types that are based on similarities of relationships according the levels of relationships defined in Gläser and Laudel (2004, p. 241). The other two levels of relationships would be reported relationships and relationships within a single case.

4.3.5 Results of expert interviews The results section is organised according to the research questions for- mulated in subsection 4.3.1. The first part of the results show that the assumption of different behaviour can be confirmed with the comparison of the criteria, the information used for decision-making and the char- acteristics of alternatives. The second part of the results compares the typology based on DOCUMEDIA data with the typology gained from the interviews with additional information (Table 4.22). This is like a spot check of the semantics.

4.3.5.1 Different behaviour of developers

The following shows that the decision-making of developers varies along the independent variables of the assumed typology. The three variables purpose, professionalism and endowment are of primary interest.

136

4.3. Expert interviews with real estate developers in the Canton of Zurich

Table 4.12: Reported main criterion by purpose

Own use, letting Sale

Availability of affordable land Net present value > 0 Conservation of value Profit opportunities Cost-benefit ratio positive Evaluation report positive Nb of housing unit > 100 Fit demand Location and profitability Gut feeling

Differences in decision criteria When asked for decision criteria, the interviewees seldom reported conditions, more often citing the variables of alternatives they look at. In the analysis, however, the distinction between criteria being conditions and attributes being characteristics of distinction is made.

Table 4.12 shows the main criteria mentioned by the interviewees when asked for their main criterion according to different purposes. None of the interviewees only relied on one criteria. Most mentioned that criteria are necessary, but not sufficient in relation to a development decision. Thus, the main criteria mentioned most often is the predominant aspect within the trade-off process. Most criteria and attributes are not mentioned on an operational level. This indicates that most of these actors do a lot of their evaluation work in a qualitative way.

When asked for specific criteria the answers have been on an opera- tional level. The three examples are shown in Table 4.13. One interviewee reports that he would calculate with an amortization of 0.5% from the eleventh year after construction. Which results in a payoff time of 210 years. Generally expected payoff time is reported to be between 20 and 50 years. The reported profitability targets range between -20% and 10%. The negative profit is reported from the public institution which does not aim at profitable developments. They introduce a measure of efficiency gains per invested money unit for their assessment. The criterion regard- ing the pre-sold share of units of a project is only relevant with purpose Sale. It reflects a form of risk assessment.

The responses of the interviewees confirm that developers with sale criteria are more profit-oriented and take more risks. The higher risk is taken into account and consequently a higher profitability and a shorter pay-off time is required. In case three, the interviewee noted that the high risk of development is likely to be compensated by good margins within the project realisation. This was said to be a danger when uniting the entire development process in one firm. This is probably a reason why

137

Chapter 4. Analysing Zurich’s real estate development

Table 4.13: Ranges of specifically asked criteria by purpose

Criterion Own use, letting Sale

Payoff time 10 – 210 years 1 – 10 years Profitability -20 – 5.3% 5 – 10% Pre selling N/A 30 – 70%

Table 4.14: Definition of type professional and unprofessional

Dimension Unprofessional Professional

Legal form Private person Company Number of projects <5 >5 Number of employees 0 >0 Turnover Small >1 Mio.

firms tend to separate the development unit from the construction unit. In cases of selling, the pay-off time for developers seems similar to usual project durations.

The following list shows additional decision criteria mentioned by the interviewees: • Public transport (PT) within walking distance • No contaminated sites • Tax savings higher than rent price • Lake within walking distance • Zoning with more than 3 stories • Floor area ratio 0.6 • Min. size of lot 6000m2

• PT station within 300m • Population of municipality > 3000 • more than 1% population growth during last 5 years • Highway exit closer than 2 kilometres • New supermarket openings • Influx of young adults

Differences in underlying information To show behavioural differ- ences according to the variable professionalism, developers are cate- gorised as non-professional or professional as shown in Table 4.14.

The differences in evaluation methods and the underlying informa-

138

4.3. Expert interviews with real estate developers in the Canton of Zurich

Table 4.15: Differences according to professionalism in evaluation meth- ods and information base

Unprofessional Professional

Evaluation methods

Study advertisements Looking around Ask around Scouting expeditions Compare with neighbouring projects

GIS-Tools Price calculators Optimising budget and parcel Location analysis Market analysis Demographic analysis Consultation of ratings IFRS component approach Sustainability tool Portfolio review

Information basis

Press Personal situation Conditions of parcel Internet Local knowledge Opinion of trusted persons Professionals

Press Zoning Online markets Own market data Local knowledge Professional reports Prepared data Professional tools Statistical offices

tion base for the development decision are shown in Table 4.15, which shows that the evaluation methods of professional developers are more data-oriented. Professional developers must rely on these data sources in order to manage their work. This also relates to the size of the port- folios of renting developers and the activity space of selling developers (Table 4.20).

Professionals are also mentioned as sources of information. This shows that non-professional developers are relying on professional ser- vices. In addition, professionals rely on information from other specialists, such as consulting firms or banks, which shows that some actors are spe- cialised in selling development services while others sell development as a product. A comparison of the various tasks performed supports this reasoning (Table 4.21).

More sophisticated evaluation methods of professional developers result in a wider information base. Professional developers extend the non-professional information base by using data and data mining methods. They also buy prepared information from specialised service providers.

139

Chapter 4. Analysing Zurich’s real estate development

Table 4.16: Reported variables for parcels

Unprofessional Professional

Size Size Zoning Zoning

Floor area ratio Floor area ratio Soil condition

Contamination Development potential

Slope Form

The information generated from such data processing is in most cases compared to each other, which increases reliability.

Differences in considering alternatives The attributes mentioned by professional and non-professional developers are compared to judge their importance for location choice. The attributes are sorted in separate tables according to the described spatial unit. Parcels (Table 4.16), parcel surroundings (Table 4.17) and municipalities (Table 4.18) are considered units. As long as the investment decision concerns an empty parcel, no building attributes can be evaluated. In cases of redevelopment, the conditions of existing objects do matter.

The attributes mentioned were expected to a large extent. Two new interesting indicators were mentioned by professional developers. One reported that the opening of supermarkets is a good sign for a location. Not only that the location gets more attractive, but it is expected that the retailers make well-informed decisions. Another developer pointed out that immigration of young households is considered an especially good sign.

The number of attributes shows that professional developers analyse the situation in more detail. The general aspects, however, are also covered by the non-professional developers.

Some attributes are linked to administrative boundaries, which justifies the use of such administrative attributes because people identify with city quarters or municipalities. Another reason is that a lot of data considered by developers is associated with administrative spatial entities. Another argument is that administrative names occur in the media, which creates an image for a location.

The responses showed that the attributes under consideration depend

140

4.3. Expert interviews with real estate developers in the Canton of Zurich

Table 4.17: Reported variables for parcel surroundings

Unprofessional Professional

Accessibility public transport, car Accessibility public transport, car Accessibility of population

Accessibility schools Accessibility schools Access job Access jobs

Access shop Accessibility shops Access lake Access lake

Access recreation area Access church

Access children’s playground Access leisure facilities

View View Sunshine duration Sunshine duration

Noise Noise Electro smog Air pollution

Aesthetics neighbourhood Zoning neighbouring parcels Zoning neighbouring parcels

Designated development zones Centrality

Uses on neighbouring parcels Nb of families in neighbourhood

Socio economic structure Share of foreigners

Projects in neighbourhood Projects in neighbourhood Image of neighbourhood

on the planned uses in a project. Most interviewees answered from the perspective of housing. However, there were a few respondents who distinguished considered attributes according to planned uses. For public uses, the level of service that a particular location makes possible is reported to be crucial. How the level-of-service is estimated depends on the facility installed. Some of the examples mentioned are listed in Table 4.19. Models such as those presented by Arentze and Timmermans (2007) are appropriate for the uses that base location choices on catchment area analysis.

The comparison of search strategies and search space is shown in Table 4.20. Again, it seems that the methods are more ad hoc in the case of non-professional developers. Developers with regional or national search spaces use spatial analysis tools.

141

Chapter 4. Analysing Zurich’s real estate development

Table 4.18: Reported variables for municipalities

Unprofessional Professional

Population size of locality Population size of locality Tax level Tax level

Infrastructure Infrastructure Infrastructure improvements

Vacancy rate Vacancy rate Structure of building stock

Population growth Population growth Immigration of young households

Rate of absorption Time of absorption

Large scale developments Image of municipality

Existence of lake Opening(s) of supermarkets

Price trend Price level Share of foreigners in schools

Finances of municipality Close to economic centre

Table 4.19: Level-of-service measures considered for public service uses

Use Level-of-service measure

Retirement home Coverage of neighbourhoods Schools Catchment area walking distance children Protection and rescue Catchment area 10 min drive

The search space of non-professional developers is found to be local. This is consistent with findings of household location choice studies that show a strong attachment to home locations. Inheritance of property or childhood attachment seem to be present in location choice in Case two.

Professional developers can also profit passively from offers for de- velopment opportunities. In such cases, they are contacted by a property owner who wants to sell its property or have it developed. This passive way of obtaining estates for development is only possible for actors who are known for their development skills. In Case seven, it is mentioned that active searches are increasingly important, which in that case was the consequence of running out of development opportunities within their

142

4.3. Expert interviews with real estate developers in the Canton of Zurich

Table 4.20: Differences according to professionalism in search spaces

Unprofessional Professional

Search strategy Looking and asking around Use local knowledge Read news

Construction sites offered (passive) Systematic search with spatial analysis (active) Activate network of agents

Search space Local, regional Local, regional, interna- tional

own portfolio.

Differences in tasks Table 4.21 shows all eleven cases and the tasks they reported to fulfil during the development process. The table is organised in such a way that developers developing for their own use are on the left side and selling developers are on the right. Characteristics of the purpose attribute are at the bottom of the table because the tasks are ordered according to the development process. The purpose variable (sale, lease, own use) is extended to reflect the finding that commercial developers can either sell a finished property or sell development services.

Thus the tasks being carried out are quite heterogeneous. One similar- ity is that developers with the purpose own use are engaged in financing. They represent the demand side in a market where development services are traded.

On the opposite side are the development service providers who all have the coordination task of construction management in common. This confirms the findings of Healey (1991), which defined the developer as a coordination actor, which is comparable to a development service provider.

Most of the commercial developers are also characterised by tasks they carry out optionally, which indicates that they are pursuing multiple business cases adapted to certain situations. Nevertheless, they specialise in a main purpose according to which they define their spectrum of tasks. Tasks not covered by their main purpose would have to be organised on respective markets.

The condition that all tasks have to be carried out to realise a develop- ment and that none of the developers cover the full task spectrum shows that in all cases a network of actors must exist. Table 4.21 shows the

143

Chapter 4. Analysing Zurich’s real estate development

Table 4.21: Comparison of covered tasks in development process

Task 2 1 10 11 6 4 5 7 8 3 9

Financing xa x x x x x (x)b x Search for location x x (x) x x x Buy property (x) x (x) x Concept of use x x (x) x x x x x Design x x x Construction management x x x x x x Engineering x Construction x x x Marketing x (x) x Sell property (x) x x x x Sell service x x Lease x x x x x x x Own use x x

aTask fulfilled bTask optional

position (in terms of tasks carried out) of the developer in such a develop- ment network. The complete network is unknown, however, there must be assumed that these networks are essential for the explanation of particular developments. These developer networks are dynamic in the sense that they change from project to project and sometimes even during a project. However, statements from the interview allow the conclusion that certain parts of such a developer network can be more stable because social and business contacts have been established. Interviewees 3 and 7 mention such relationships when talking about their development division inside a holding or a group that acts as feeder for the subsequent production pipeline.

The same concept also works on a smaller scale. Case 4 is an architect who privately holds a real estate portfolio. For such actors, it is reasonable to use the synergies in the sense of complementary needs, e.g. if orders can be carried out by a self-owned company.

Buying property can be optional in cases where property assets are at hand (existing portfolio). Typical examples are large industrial firms with other core businesses than real estate that are developing their unused land banks. Interviewee 7 reported that his current business resulted from a large industrial company folding. These portfolios also serve as security for financing. This example also shows that endowment, especially with

144

4.3. Expert interviews with real estate developers in the Canton of Zurich

Table 4.22: Detectability of typology on the basis of DOCUMEDIA data and interviews for all 11 cases

1 2 3 4 5 6 7 8 9 10 11

Typecast a priori (DOCUMEDIA) O1 O1 Om Smc Smc Om Smc Smc Smc Om Om

Typecast a posteriori (interviews) Om O1 Smc Smc

Om Smc Om Smc

Om Smc, Om

Smc, Om

Om Om

Typecast consistent No Yes No (Yes) Yes Yes (Yes) (Yes) Yes Yes Yes

an existing portfolio, creates a completely different situation in terms of development opportunities. It is to note that the real estate business was not the main interest when the property was bought. An interesting example in this realm is gardening firms that tend to buy property on the outskirts of cities. With city growth, the gardening firms eventually sell their properties with profit and move further away where they once again follow their business approach.

4.3.5.2 Detectability of developer types in the DOCUMEDIA data

To check the semantics of the DOCUMEDIA data, the typology is com- pared based on the interview information in Table 4.22.

The typology does not correspond in two cases. In Case 1, a project was realised outside of the perimeter covered by the DOCUMEDIA data. Consequently, classification failed in the dimension Number of Projects. In Case 3, the classification failed in the variable Purpose, why is unclear.

In three cases, the a posteriori typology is not unique. In Case 4, the interviewed person is also privately engaged in the real estate business and thus has a different typology in the private domain. In Cases 7 and 8, the firms actually consist of multiple business units. If all business units are subsumed, the developer type is Om. If the development unit is considered separately, it is of type Smc. In Case 4, the interview was conducted with a development service provider (in this case, an architect). It was possible to some extent to extract whether the details on the developers belong to the development service provider or to the owner because the address details contain a c/o that indicates that a development service provider represents the owner (compare with definition in Section 4.1).

145

Chapter 4. Analysing Zurich’s real estate development

4.3.6 Conclusions for discrete choice modelling

The decision-making of developers varies according to the variables pur- pose and professionalism. Developers have different decision criteria, different information sources and execute different tasks. It is worth- while to build models for heterogeneous decision makers (multinomial logit (MNL) with interactions, latent class (LC), mixed logit (ML)) in quantitative modelling.

Developers with the purpose Sales, see their profit either in providing efficient development services that costumers are willing to pay for or in anticipating the preferences of a finished product they can bring on the market profitably. Therefore, they expect shorter pay-off periods than developers with ’own use’ as their purpose. Developers with purpose leasing have reduced expectations for profitability in the short run, which is explained by their preferences for secure and long-term investments. Non-profit developers satisfy public needs (social housing, schools, etc.) or in the case of cooperatives are oriented towards cost-covering rents.

Professional developers have more resources and know-how, which allows them to exploit more information than non-professional developers. This information asymmetry allows them to realise their margins. It is also found that professional developers have a wider activity space. In terms of considered attributes of locations, developers are similar.

The behaviour also varies in terms of tasks carried out within the development process. The business cases they have are related to their endowment in terms of portfolio and skills. The endowment with a real es- tate portfolio is found to be important because it provides better conditions for getting loans and land resources as development opportunities.

Rather than a single developer as a clearly defined entity, development networks are often found to be active in a development process. Thus considering characteristics of the development networks might help to explain development events. This is problematic since during this project, obtaining the necessary data seemed difficult. Furthermore, it is unclear how to include collective decision making in DCA. A starting point is (de Palma et al., forthcoming) work on decision-making in couples.

146

4.4. Discrete choice analysis of real estate development

4.4 Discrete choice analysis of real estate development16

To assess behavioural differences quantitatively, location choice models are estimated. Location choice is only one facet of developers’ behaviour since a development decision is not only about location, but also about timing, funding and quality of the built space. Experts, however, point out the importance of the location for the real estate product (Alda and Hirschner, 2011, p. 65). In addition, the interest lies on spatial effects due to the transport context. Therefore, a location choice model seems to be a meaningful starting point for this investigation. It is chosen to estimate location choice models because spatial preferences can directly be observed.

It is generally recommended to start model development with a simple specification (Train, 2009). Therefore, a building location choice model (BLCM) is estimated first by using single buildings from the GWR. The problem with this approach is a neglected correlation due to the projects themselves, i.e. in some cases the decision to build is actually made for multiple buildings at once. The influence of this correlation is discussed in subsection 4.4.2. The basic project location choice model (PLCM), used for comparison to the BLCM, is the starting point for the investigation of developer heterogeneity. The next two sections report two approaches for assessing the influence of developers on the location decision. In the first approach, deterministic basic interaction is used to find differences in attribute valuation. A second approach using segments according to development purpose is undertaken because the results in the first approach were not satisfactory due to omitted variables and resulting unexpected signs (positive/negative). These final models are presented in subsection 4.4.4, but first, the data preparation is described.

4.4.1 Data preparation The observations documented by DOCUMEDIA on real estate develop- ment are the most important source of information used in this analysis. The raw data has been described in subsection 4.2.3.1. At this point, the additional data used for estimation and how it has been prepared is described. The available data allows two approaches to model estima- tion: a) single buildings as development observations and b) projects as

16Parts of this section are taken verbatim from Zöllig Renner and Axhausen (2013) and Zöllig Renner and Axhausen (forthcoming).

147

Chapter 4. Analysing Zurich’s real estate development

development observations (can also be multiple buildings).

4.4.1.1 Preparation of buildings

For the first approach, all buildings from the building stock in 2010 with a construction year between 2000 and 2010 are filtered into a separate table. The table now contains the history of buildings developed in that period. Otherwise, these buildings are prepared exactly as the ones for the base year in 2010 (subsection 5.1.1).

4.4.1.2 Preparation of projects

The preparation of projects as observations of development events includes matching project data to the building data from the GWR, removing data points with missing content and dismissing implausible data points, i.e. development events outside construction zones. The combination of the DOCUMEDIA data and GWR data is desired because the latter data provides more detail on the constructed buildings and living units. The information added from the projects data is the information about joint construction, the developers’ frequency of development, the development purpose and the classification from the qualitative study (Table 4.10). In addition, the development projects have to be linked to buildings and living units because at the end of the study, the development model needs to add new buildings and living units to the stock of built space.

Matching development projects to constructed buildings is achieved with spatio-temporal matching. Therefore, the projects are matched to parcels by address matching, since this is the only location information in the dataset. At this stage, 43% of all recorded projects must be discarded since they cannot be matched to parcels. The reasons are not clearly identified, but it can be assumed that unrealised projects are responsible for part of the unmatched projects. Another reason is incomplete addresses in the development project records. 74% of the unmatched projects do not have a house number. From the located projects, only new construction is selected since it can be expected that these are the most relevant events for the land use transportation system. This reduces the sample by another 32,392 cases (87%). For 73% of the remaining projects, no building can be matched, so they are also removed. Buildings are assigned to a project when they are located on the same parcel and when the Year Built attribute shows registration after the project was finished. After further removing the projects located on parcels not inside the construction zone, the result is a dataset of 1301 projects comprising 1576 buildings and 5454 living units. The reduction of the number of observations limits the flexibility of

148

4.4. Discrete choice analysis of real estate development

modelling. The projects are located in 114 of the 151 municipalities in the Canton of Zurich (Fig. 4.12). Apart from address matching, which is a bash script, all the data is converted with PostgreSQL.

The parcels, which are the alternatives to be chosen by developers, are enriched with context data. The preparation process is described in subsection 5.1.1 and in more detail by Schirmer et al. (2011). The result is a database comprising parcels with planning constraints, buildings with dwellings and households with persons and jobs. All entities are related to each other via their relative spatial position. Households are located in dwellings, dwellings are associated to buildings and buildings are standing on a parcel. This abstraction of spatial reality is the context for the development events, which are the observations used for model estimation.

Each parcel also has its accessibility calculated using the Multi-Agent Transport Simulation (MATSim) (Balmer, 2007) implemented for Zurich. The use of the highly detailed transport simulation allows calculating individual accessibility for each parcel for different modes, as described in Nicolai and Nagel (forthcoming). Here, the car and PT accessibilities to jobs are used.

The final datasets for estimation are created in UrbanSim when run- ning a model estimation. The estimation dataset is different for each estimation, since random sampling of alternatives is applied. The vari- ables used in the utility functions are partly calculated outside UrbanSim using geographic information system (GIS). Such variables are included as primary variables that are directly attached to a dataset during data preparation, e.g. lake view or exposure to evening sunshine. The vari- able definition in UrbanSim uses python modules and a domain-specific modelling language (Borning et al., 2008). An overview of the variables used in the location choice models of the following experiments (subsec- tion 4.4.2, subsection 4.4.3 and subsection 4.4.4) is given in Table 4.23. Further descriptions are given in the text when the variable is used the first time.

4.4.2 Use of projects as observations rather than new buildings

This section investigates the effects of using the different data sources available. This is a practical issue, since data availability varies from application to application. As described earlier, there are two sources of information on land-use development events at hand. The first source is the GWR and the second source is the DOCUMEDIA dataset on

149

Chapter 4. Analysing Zurich’s real estate development

development projects. The first source is more comprehensive and richer in detail in terms of building description and also allows development events to be extracted by the Year Built attribute. However, the second source is theoretically preferred since, in reality, development decisions are not necessarily taken for a single building but rather per project, which are also composed of multiple buildings. In this work, the term project denotes a prepared plan for building new built space that materialises in one or more buildings. Hence, it is more precisely a real estate project. Such a construction project is a process with many decisions and complex dynamics, but in this context, it is limited to the main decision to proceed with construction.

Project data allows verification of the buildings built as part of the same construction project and hence to estimate the proposed land use development location choice model with or without the information on actual projects. In the following, the first case is referred to as BLCM and the second as PLCM. Thus, when estimating BLCMs on the basis of single buildings as observations, there is an implicit assumption that the development decisions are made for each building separately, which is not actually the case. For the PLCMs, projects are taken as observations that can be composed of multiple buildings. By comparing the estimation results of the BLCMs to the results of the PLCMs, the bias imposed when estimating BLCMs can be assessed.

4.4.2.1 Model specification and estimation

Using MNL models allows the application of random sampling of alterna- tives to generate the choice set (McFadden, 1978). This is necessary since considering full choice sets would generate very high computational costs. The available template to formulate agent location choice models is used. The agents are, in this case, development projects that choose to locate on parcels as alternatives. The models are estimated with version 4.4.0 of UrbanSim, which employs the B-triple-H algorithm (Berndt et al., 1974) for maximum likelihood estimation.

Each observed location of a project is completed with a random sam- ple of thirty parcels. A parcel is a valid alternative if it has capacity for further development. Land use regulations define floor area ratios, which limits the capacity of parcels in terms of allowed floor space. Separate MNL models are estimated for residential and non-residential develop- ment. Estimating separate models is necessary in order to consider the planning restrictions in the sampling of alternatives. With separate mod- els, it is possible to exclude certain alternatives from the sampling, e.g. a

150

4.4. Discrete choice analysis of real estate development

residential building is not allowed on a parcel zoned for industrial uses. Formulating an overall model with dummies is therefore not an option. Further segmentation is introduced within the residential model by distin- guishing between SFH, MFH and mixed-use developments. The segments reflect intended uses for the developments. It was not possible to consider sub-segments within the non-residential segment due to the small number of observations. The utility functions are linear in parameters and use the variables listed in Table 4.24 with their descriptive statistics. The variables are described in Table 4.23 and the following paragraphs.

Literature review in subsection 2.3.2 and the interviews with devel- opers in Section 4.3 give an idea of what variables are important for the location of development events. The two sources show that characteristics of the parcel itself, plus externalities from the neighbourhood are con- sidered. Examples of attributes mentioned are construction costs, legal situations, variables that describe the land market (provision of same built space type, absorption rate), condition of soil (contamination demand- ing cleaning before development), population growth or influx of young adults (newcomers). The consideration of variables is limited by data availability.

All variables that do not characterize the parcel directly are location externalities for which the extent of their spatial reach has to be defined. A similar problem is discussed by Guo and Bhat (2007) for the USA. They try to pin down the neighbourhood concept or, more generally, to determine to what spatial extent the endowment of the vicinity is perceived. For operationalisation with circular units, they come up with a ’neighbourhood radii’ of 0.4 km, 1.6 km and 3.2 km. They also find that socio-economic variables ’have significantly smaller spatial extent of influence than the land-use variables’ (p. 44). A quick analysis of the areas given a name from the cadastre data in the Canton of Zurich shows a median size of 4.17 hectares, which suggests a radius of 115 meters. As might be expected, this is small compared to US numbers. A sample of ad-hoc measurements on a city map of Zurich yield neighbourhood areas of 3.5 to 35 hectares suggesting radii of 105 to 334 meters. On this basis, radii of 150 to 300 meters for neighbourhood variables are chosen for this study.

The accessibility variables are calculated using Eq. (4.1).

Acci = l n( J∑ j

X j e −βtti j ) (4.1)

151

C hapter

4. A

nalysing Z

urich’s realestate

developm ent

Table 4.23: Description of variables used in real estate development models (REDMs)

Variable Description Unit Data sources

Accessibility car Accessibility of jobs by car according to Eq. (4.1). Travel times are calculated with MATSim.

[-] Cadastre, employment census 2000, MATSim road network

Accessibility PT Accessibility of jobs by public transport according to Eq. (4.1). Travel times are calculated with the cantonal transport model (Vrtic et al., 2005).

[-] Cadastre, employment census 2000, impedance matrix from cantonal travel model

Distance to closest school

Euclidean distance to next school facility. [m] Cadastre, GWR

Fit of development to par- cel constraints

Step function (Eq. (4.2)) of the difference between permitted floor area (F Ap) on the considered parcel and floor area of the development project (F Ad ).

[m2] Cadastre, zoning plans, DOCUMEDIA, GWR

New neighbouring build- ings

Number of buildings with year built later than 1995 within 150 m.

[-] Cadastre, GWR

Newcomers in neigh- bourhood

Number of residents which reported a different address five years ago within a radius of 300 m.

[-] Cadastre, census 2000

Land price per permitted floor area

Price per permitted square meter floor-space. [ C H F m2

] SAKZ, Cantonal Office for Spatial De- velopment (ARE ZH) zoning plans

Slope Slope of parcel. [%] Cadastre, digital terrain model Share of recreation are in zone

Share of land area dedicated to recreational use within the traffic analysis zone of the development.

[%] Cadastre, Zoning plans

Mean m2 price of living unit in zone

Average rent price per sqm in associated municipality [ C H F m2

]

Profitability proxi Approximation of profitability as given by Eq. (4.3). [-] Comparis asking prices, GVZ, SAKZ Visible lake area Lake area visible from the parcel. [ha] Digital terrain model of swisstopo Index of evening sun- shine exposure

Index for sunshine exposure in the evening. [-] Digital terrain model of swisstopo

Distance to CBD Euclidean distance to the city centre of Zurich. [m] Cadastre Municipal tax index Tax index of respective municipality. [%] SAKZ

152

4.4. Discrete choice analysis of real estate development

The opportunities X are approximated by the number of persons and jobs in a travel analysis zone j. Each of the contributions is multiplied by a negative exponential weight based on travel time tt, which is an approximation of generalized travel costs c. The cantonal transport model was used for calculating travel times. In the case of car accessibility, the travel times are calculated on the basis of the street network and in the case of PT on the basis of the public transport network. β is set to 0.2 as discussed in subsection 2.1.3.2. Applying the logarithm can lead to negative accessibility values for very remote locations (Table 4.24).

An example of an access variable is the distance to school. Access is a proximity measure to the closest satisfying option. In comparison to the accessibility variables, it does not capture potential alternatives. In addition, this variable is only based on Euclidean distance, which is less accurate than using network distances. Here Euclidean distance is used because it can be computed with parcel resolution, which is not possible for the accessibility variables with the implementation used.

The variable Fit of development to parcel constraints interacts the project size with the allowed capacity on the parcel (Eq. (4.2)). Utility is drastically reduced if the project is larger than the allowed floor space, i.e. there is no hard capacity constraint assumed. What this reflects in reality is that negotiations are possible by allowing projects to exceed allowed densities. The logarithmic formulation in case of unexploited floor area capacity reflects that there is a decreasing marginal utility for unexploited building capacity.

f it (F Ap, F Ad ) = {

l n(F Ap − F Ad ) if F Ap − F Ad > 0 4 ∗ (F Ap − F Ad ) if F Ap − F Ad < 0

(4.2)

The newcomers’ variable is motivated by a developer’s statement that they would analyse demographic development of candidate areas with respect to population growth. Thus, this is an attempt to capture upcoming areas by measuring the influx of people.

The price per permitted floor space is calculated by multiplying the average land price per square meter in the respective municipality by the land area of the respective parcel, divided by the permitted floor space according to the zoning constraints. A parcel with a lower price per permitted floor space is expected to be attractive for development.

The slope variable indicates locations on hillsides. It is thus a proxi for the View attribute.

153

Chapter 4. Analysing Zurich’s real estate development

Table 4.24: Descriptives of variables used in BLCM and PLCM

PLCM BLCM Parameter Name mean sd min max mean sd min max

SFH Newcomers in neighbourhood 310.2 268.3 0.0 2774.0 312.2 269.5 0.0 2867.0 Distance to school 527.2 361.9 4.0 2868.0 524.1 348.4 3.0 2868.0 Accessibility car 9.7 0.5 0.0 11.0 9.7 0.5 0.0 11.0 Accessibility PT 10.7 1.8 -19.0 13.0 10.7 1.9 -19.0 13.0 Fit of development to parcel constraints -894.5 2438.3 -50099.6 10.5 -801.6 1185.6 -18563.6 10.5 Price per permitted floor area 2414.2 1494.1 359.3 11015.0 2429.9 1502.9 325.0 11015.0 New neighbouring buildings 5.5 9.3 0.0 86.0 5.4 9.2 0.0 107.0 Slope 5.7 4.4 0.0 31.1 5.7 4.4 0.0 31.1 MFH Newcomers in neighbourhood 312.5 269.9 0.0 2851.0 317.5 278.1 0.0 2867.0 Accessibility car 9.7 0.5 0.0 11.0 9.7 0.5 0.0 11.0 Accessibility PT 10.7 1.8 -19.0 13.0 10.7 1.8 -19.0 13.0 Fit of development to parcel constraints -3979.4 5583.3 -90548.8 10.7 -3460.7 3307.1 -31857.0 11.1 Price per permitted floor area 2442.0 1536.6 299.1 11015.0 2412.4 1508.8 299.1 11015.0 New neighbouring buildings 5.3 9.0 0.0 107.0 5.4 9.2 0.0 107.0 Slope 5.7 4.5 0.0 31.1 5.7 4.4 0.0 31.1 Mixed-use Fit of development to parcel constraints -5858.3 8360.3 -52911.0 10.7 -5344.3 5313.0 -23478.6 11.1 New neighbouring buildings 5.3 9.2 0.0 66.0 5.3 8.9 0.0 86.0 Accessibility car 9.7 0.5 8.0 11.0 9.8 0.5 8.0 11.0 Accessibility PT 10.7 1.6 -19.0 13.0 10.8 1.6 -19.0 13.0 Price per permitted floor area 2389.4 1563.7 299.1 11015.0 2364.1 1432.5 299.1 11015.0 Non-Residential Newcomers in neighbourhood 291.8 283.7 0.0 1967.0 306.3 292.7 0.0 1957.0 Accessibility car 9.6 0.6 0.0 11.0 9.6 0.5 8.0 11.0 Accessibility PT 9.9 3.2 -19.0 13.0 10.0 3.0 -19.0 13.0 Fit of development to parcel constraints -8572.5 22477.2 -173320.0 12.8 -7096.6 16178.9 -92838.0 12.9 Price per permitted floor area 1165.4 668.5 196.9 5188.6 1154.2 654.6 196.9 5188.6 New neighbouring buildings 3.3 5.3 0.0 61.0 3.4 5.4 0.0 65.0 Slope 3.8 3.3 0.0 27.3 3.7 3.1 0.0 27.3

4.4.2.2 Estimation results and comparison

The estimation results of all four models are shown in Table 4.25. Parame- ters estimated on the basis of projects are to the left of those estimated on the basis of single buildings. Below the parameter estimates of each model, the corresponding model statistics are shown. These show small numbers of observations for the models With Side Use and Non-Residential. The maximum number of variables that would still be possible to estimate was used to reduce chances for omitted variables. A consequence is that the models for mixed-use and non-residential development end up with few observations relative to the number of estimated parameters. For the model of non-residential developments, the situation of having few observations is even more unfortunate since there is a considerable variety of uses in this segment that cannot be accounted for. The statistic ρ2 is consistently higher for estimates for single buildings. The higher fit is probably a result of the higher number of observations that are partly found at the same locations. Thus, the variance of observed location choices is lower relative to the observations that makes them easier to predict. From the

154

4.4. Discrete choice analysis of real estate development

theoretical point of view, this is misleading since the number of decisions in reality is probably closer to the number of observations in the case of project location choice models.

All parameters show consistent signs that indicate that estimation with single buildings is a viable option. The only exception is the slope variable in case of non-residential buildings. Variables hypothesised to be sensitive to a consideration of project information are Fit of development to parcel constraints and New neighbouring buildings.

The variable Fit of development to parcel constraints shows a positive sign in all models, indicating that developments locate on parcels with more floor space permitted than used. This shows that development reserves are valued, which confirms previous research (Thalmann, 2009). The bias as a result of single buildings as observation is expected to be a higher estimate because the chosen parcel is observed without the additional buildings of the same project. This hypothesis is confirmed for the segments SFH and MFH. Why the bias is not visible in the other two cases is unclear.

All models show positive signs for the variable measuring new devel- opments in the vicinity. This confirms the spatial inertia of land devel- opment found in previous studies (Haider and Miller, 2004; Dong and Gliebe, 2011). Possible explanations for spatial inertia are that a) settle- ments grow at their borders, b) the intention of planning authorities is to concentrate development or c) developers tend to develop in areas they are familiar with. It is assumed that explanation b) is most influential. The interviews in Section 4.3 and other studies (Ruming, 2010) also give evidence for the last argument (c). The expected bias due to multiple observations for the same choice is in reality a more significant estimate that is confirmed in all four models. The effect itself is expected to be the same since the buildings of the same project are not counted in the neighbourhood variable.

A negative sign for accessibility variables is unexpected. Here it shows in all residential submodels for car accessibility and in the case of SFH, also for PT. The estimate for car accessibility is only significant17 in the MFH model. One part of the problem is probably endogeneity due to a poorly measured price variable so that part of the price effect is captured in the accessibility estimates. The positive estimates for the land price variables point at the same issue. Further indication is found in a strong correlation of the accessibility variables and the land price variable. This is not surprising insofar as price and accessibility are both general measures for attractiveness.

17An estimate is considered significant at the 95% level.

155

Chapter 4. Analysing Zurich’s real estate development

Table 4.25: Estimation results of BLCM and PLCM

PLCM BLCM

Parameter name Estimate t-Value Estimate t-Value

SFH Accessibility car -0.3707 -5.17 -0.3740 -5.52 Accessibility PT -0.0142 -0.65 -0.0216 -1.00 Fit of development to parcel constraints 0.0010 11.22 0.0012 10.95 Distance to school -0.0005 -3.47 -0.0005 -3.94 New neighbouring buildings 0.0493 13.24 0.0512 14.17 Newcomers in neighbourhood -0.0025 -7.61 -0.0024 -7.77 Price per permitted floor area 0.0001 5.21 0.0001 5.38 Slope 0.0366 3.34 0.0349 3.29

Observations 501 523 LL(0) -1703.9999 -1778.8262

LL(conv.) -1455.2932 -1493.9691 Adj.ρ2 0.141 0.156

MFH Accessibility car -0.1261 -1.04 -0.0800 -0.67 Accessibility PT 0.1628 3.63 0.1728 3.92 Fit of development to parcel constraints 0.0005 17.27 0.0006 16.87 New neighbouring buildings 0.0339 5.74 0.0338 6.33 Newcomers in neighbourhood 0.0009 3.50 0.0008 3.63 Price per permitted floor area 0.0002 6.78 0.0002 7.97 Slope 0.0115 0.80 0.0107 0.78

Observations 405 445 LL(0) -1377.4849 -1513.5328

LL(conv.) -1168.1905 -1230.9202 Adj.ρ2 0.147 0.182

Mixed-use Accessibility car -0.7055 -1.80 -0.4893 -1.27 Accessibility PT 0.4154 2.30 0.3273 2.19 Fit of development to parcel constraints 0.0005 4.69 0.0005 4.87 New neighbouring buildings 0.0111 0.31 0.0148 0.50 Price per permitted floor area -0.0004 -2.59 -0.0004 -2.75

Observations 54 65 LL(0) -183.6647 -221.0778

LL(conv.) -143.4228 -157.4724 Adj.ρ2 0.192 0.265

Non-Residential Accessibility car 0.4569 1.42 0.6646 2.23 Accessibility PT 0.1762 2.04 0.3245 4.03 Fit of development to parcel constraints 0.0005 12.96 0.0005 13.56 New neighbouring buildings 0.0534 2.50 0.0584 3.64 Newcomers in neighbourhood -0.0001 -0.15 -0.0012 -2.97 Price per permitted floor area 0.0007 4.74 0.0008 5.52 Slope 0.0152 0.39 -0.0230 -0.64

Observations 84 114 LL(0) -285.7006 -387.7365

LL(conv.) -196.0938 -261.6918 Adj.ρ2 0.289 0.307

156

4.4. Discrete choice analysis of real estate development

The variable Newcomers in neighbourhood measures the attractive- ness of a neighbourhood. Positive signs are expected for residential developments. There are significant estimates for SFH, MFH projects and non-residential developments. Latent demand cannot be captured with this variable for SFH. If available, absorption rates might be a better choice. The negative sign for non-residential development is not significant. It can be interpreted that firms like proximity to other workplaces due to agglomeration economies. This is very general and shows the need for more specific knowledge about intended uses.

The slope variable is only significant for single-family developments. The positive sign shows a preference for locations on hillsides. The model could be improved by considering aspect and solar exposure.

The distance to school is only included for SFH because it can be expected to be relevant for households with children, which are likely to locate there. The expected negative sign shows in both cases and the estimates are significant. Proximity to a school is thus an appreciated amenity in the case of SFH.

For the price variable, a negative sign would be expected in general, indicating that developers tend to buy and develop land that is inexpensive. However, the signs in our models show positive signs for SFH projects, MFH projects and non-residential projects and buildings. The effect of the price variable is surprisingly small. As discussed earlier, the permitted floor space price variable might be insufficiently observed and thus the models suffer from an endogeneity problem as described by Guevara and Ben-Akiva (2006).

4.4.2.3 Intermediate summary

In this section, the viability of extracting pseudo-development events from building register data is investigated. The comparison of model estimates on the basis of single buildings with model estimates using development project information shows relatively little variation in the case of the Canton of Zurich. This suggests that building register data could be used for creating land-use development models when information about projects is missing.

However, one should take biased results into consideration due to the theoretically inadequate observations when taking newly registered build- ings as development events. Variables such as the Fit of development to parcel constraints are more sensitive to the observations used. Therefore, it is still advisable to use information on development projects. The draw- backs are less severe in regions with just a few multi-building projects.

157

Chapter 4. Analysing Zurich’s real estate development

Table 4.26: Number of observations by developer type

Developer type SFH MFH

O1 281 54 Om 148 139 Smc 150 274

Total 579 467

Hence, consulting aggregate information on the size of projects, in terms of number of buildings per project, is recommended before implementing a land-use development model as a BLCM.

4.4.3 Discrete segmentation by developer type In this section basic interaction is used in combination with determin- istically identified developer types to address heterogeneity among de- velopers. The aim is to confirm the findings from the qualitative study with the data at hand. Deterministic basic interaction has the advantage of being more easily interpretable and is also applicable in a case with few attributes characterising the decision maker. The typology assumed follows from the available data (Table 4.10). The accuracy of type assign- ment is assessed in the qualitative study (Table 4.22) and exhibits that in some cases the derived type is not correct.

The sample sizes per developer and building type that remain after data cleaning and integration are shown in Table 4.26, their spatial distribution in Fig. 4.12. In this particular investigation, only projects with residential use are considered. The samples of mixed use and non-residential projects are too small to allow an investigation of heterogeneity. The map shows that most development projects in the sample happen in highly attractive municipalities along lake Zurich. Real estate prices are traditionally high in these municipalities (Salvi et al., 2004). Few developments are observed inside the city boundaries of Zurich.

4.4.3.1 Model specification and estimation

Firstly, two reference models that do not use information on real estate developers are developed. The PLCM shown in Table 4.25 is taken as a starting point. Two submodels for SFH and MFH projects are specified. Various specifications of linear models were tested searching for variables

158

4.4. Discrete choice analysis of real estate development

Figure 4.12: Observed real estate development projects in the study area according to developer types

Legend Developments

O1 Om Smc

Municipalities

Lakes

Data: c© 2013 swisstopo (JD100042), DOCUMEDIA

159

Chapter 4. Analysing Zurich’s real estate development

that are significant, show the expected sign and are weakly correlated with each other. A correlation of less than 0.2 is considered weak. The descrip- tion of the remaining explanatory variables is included in Table 4.23 and respective data sources are given. The variable Share of recreation area in zone is introduced to capture the hypothesised attractiveness of remote locations that potentially bias the car accessibility variable. The share is calculated as the fraction of land area dedicated to recreational use within the traffic analysis zone (TAZ) of the considered parcel. The data sources used to calculate are the cadastre and the zoning plans. The land price variable is removed because the positive sign persists in all formulations. The variable is probably measured too roughly. It is unclear which devel- opers actually have to consider the land price in their evaluation, since they eventually already own the property. The variable Newcomers in neighbourhood is discarded due to its correlation with the accessibility variables. This makes the coefficient for the variable Distance to school become positive, so it is removed as well. One accessibility variable per model is discarded, also due to correlations. The variable with the higher impact is chosen for each submodel. More detail about the data sources can be found in Schirmer et al. (forthcoming).

Secondly, building on the reference model, each variable was tested for developer type-specific estimation results. The significance of param- eter difference is tested with a two-sample t-test (Cressie and Whitford, 1986). The variables that show significant differences between developer- specific estimates are introduced as basic deterministic interaction terms in the developer-based model (Table 4.30). All models are estimated in UrbanSim as described in subsection 4.4.2.1. The descriptive statistics of the choice set are given in Table 4.28.

The qualitative study confirmed that location characteristics are con- sidered among other factors. Only a few of the other potentially valuable explanatory variables18 are found in the data. Thus, the focus moves to different evaluation of location attributes. Where a different evaluation is expected, the hypotheses in relation to the variables are shown in Ta- ble 4.27. They are derived from literature and the results of the interviews. The first two hypotheses are based on two aspects: Firstly, accessibility is a very complex and comprehensive indicator that captures a potential. As it is measured here, it is quite general and thus assumed to be even more relevant for professional developers who build for an unknown client. Secondly, findings from the interviews (Table 4.15) and literature (Wallbaum et al., 2011) indicate that professional developers use more sophisticated tools, which suggests that they would be more capable of

18Compare with Fig. 4.1

160

4.4. Discrete choice analysis of real estate development

Table 4.27: Hypotheses regarding developer type specific parameter esti- mates

Variable Hypothesis

Accessibility car We expect professional develop- ers to have stronger preference for high accessible parcels (βSmc > βOm > βO1).

Accessibility PT We expect professional develop- ers to have stronger preference for high accessible parcels (βSmc > βOm > βO1).

Fit of development to parcel con- straints

We expect stronger preference for optimal fit for professional devel- opers (βSmc > βOm > βO1).

Share of recreation area in zone We expect professional develop- ers to have weaker preferences for neighbourhood endowment with recreation area (βSmc < βOm < βO1).

Slope We expect professional developers to have weaker preference for view (βSmc < βOm < βO1).

assessing accessibility. Due to professional developers’ focus on profit (Ratcliffe et al., 2004), it can be expected that they also exploit the permit- ted capacities more thoroughly. Regarding the endowment of a location with possibilities for recreation, the interviews gave some evidence that professional developers might give this aspect less weight (Table 4.17). It has also been reported that self-providers are more emotionally attached to their property (Schüssler and Thalmann, 2005). A similar argument can be made for the Slope variable.

4.4.3.2 Estimation results

The parameter estimates, t-values and estimation statistics of the reference model are shown in Table 4.29. The only remaining unexpected sign was found for car accessibility for SFH. Again, this might be due to the limited number of parameters and resulting endogeneity problems (Guevara and Ben-Akiva, 2006). Quietness and remoteness are potentially omitted

161

Chapter 4. Analysing Zurich’s real estate development

Table 4.28: Descriptives of variables in sampled choice sets

Variable Mean St. dev. Sum Min Max Single family housing (reference) Fit of development to parcel constraints -818.00 2234.90 -14208700 -50127 10.3 New neighbouring buildings 5.08 8.93 88299 0 107.0 Accessibility car 13.74 0.23 238707 12.9893 14.4 Slope 5.71 4.39 99217 0 30.9 Multi family housing (reference) Accessibility PT 12.30 0.66 172343 9.77058 13.7 Fit of development to parcel constraints -5144.44 9353.75 -72073600 -130918 10.7 New neighbouring buildings 5.00 8.87 69996 0 99.0 Share of recreation area in zone 1.11 0.49 15502 0.668107 14.2 Single family housing Fit of development to parcel constraints -813.25 2250.04 -14126100 -50145.8 10.4 Accessibility car for O1 developers 6.67 6.87 115833 0 14.4 Accessibility car for Om developers 3.51 6.00 61024 0 14.4 Accessibility car for Smc developers 3.56 6.02 61828 0 14.4 New neighbouring buildings for O1 2.52 6.75 43768 0 86.0 New neighbouring buildings for Om 1.29 4.96 22357 0 86.0 New neighbouring buildings for Smc 1.33 5.03 23031 0 106.0 Slope 5.75 4.41 99898 0 32.3 Multi family housing Accessibility PT for O1 developers 1.42 3.93 19884 0 13.8 Accessibility PT for Om developers 3.67 5.64 51357 0 13.7 Accessibility PT for Smc developers 7.22 6.08 101184 0 13.9 Fit of development to parcel constraints -5130.26 9369.33 -71875000 -130903 10.7 New neighbouring buildings 5.00 8.70 70102 0 106.0 Share of recreation area in zone for O1 0.13 0.39 1779 0 7.0 Share of recreation area in zone for Om 0.32 0.56 4541 0 14.2 Share of recreation area in zone for Smc 0.64 0.64 9006 0 14.2

variables. A possible interpretation is that the variable captures preference for remote locations. However, the parameter estimate also remains negative when the variable recreation area is tested. The adjusted ρ2 is quite low with 0.091 in the case of SFH. For MFH, the goodness-of-fit is better with 0.155.

The results of the models with developer type-specific parameters are shown in Table 4.30. Since it is not possible to introduce effect coded variables, here, it can only be mentioned if a variable is more or less considered by a certain developer type. The different estimates of developer-specific parameters can be interpreted as follows with respect to SFH projects:

Accessibility car The accessibility variable is interpreted against the background of the endogeneity problem. The results indicate that in cases of self-owning developers, the potentially omitted variables (noise, pollution) superimpose the accessibility effect most. This effect is less strong for self-owning developers with multiple projects and weakest for developers selling the project.

162

4.4. Discrete choice analysis of real estate development

Table 4.29: Estimated parameters of residential project location choice models (reference models)

Variable Estimate T-value

Single family housing Fit of development to parcel constraints 0.0009 12.06 New neighbouring buildings 0.0329 8.32 Accessibility car -1.7301 -9.24 Slope 0.0575 6.03

LL(conv.) -1785.62 LL(0) -1969.29

Adj. ρ2 0.091 Observations 579

Multi family housing Accessibility PT 0.3574 3.73 Fit of development to parcel constraints 0.0005 18.58 New neighbouring buildings 0.0180 2.80 Share of recreation area in zone 0.4449 11.24

LL(conv.) -1337.90 LL(0) -1588.36

Adj. ρ2 0.155 Observations 467

New neighbouring buildings The estimated parameter is insignifi- cant for type Om. There is no significant difference for the significant parameters. This confirms the hypothesis that areas with construction activity are more attractive for all developer types.

In case of MFH projects, the developer-specific parameters are interpreted as follows:

Accessibility PT Stronger preference for good public transport is found for self-owning developers with multiple projects. Consequently, public transport accessibility is valued more. In contrast, selling develop- ers also have to expect car users to buy their estates. The insignificance for O1 developers may be a consequence of the small sample of 54 obser- vations.

Share of recreation area in a zone Self-owning developers with one project value green space the most. It seems like more attention is paid to that aspect if just one project is built. This is probably due to the fact that these developers live in this building, which would confirm more emotional attachment.

163

Chapter 4. Analysing Zurich’s real estate development

Table 4.30: Developer specific parameters of residential project location choice models (developer based models)

Variable Estimate T-value

Single family housing Fit of development to parcel constraints 0.0009 9.61 Accessibility car for O1 developers -2.3598 -8.76 Accessibility car for Om developers -1.2414 -2.78 Accessibility car for Smc developers -1.0018 -2.67 New neighbouring buildings for O1 0.0440 8.86 New neighbouring buildings for Om -0.0001 -0.01 New neighbouring buildings for Smc 0.0348 3.95 Slope 0.0566 5.98

LL(conv.) -1769.37 LL(0) -1969.29

Adj. ρ2 0.097 Observations 579

Multi family housing Accessibility PT for O1 developers -0.1500 -0.76 Accessibility PT for Om developers 0.5717 2.85 Accessibility PT for Smc developers 0.3526 2.66 Fit of development to parcel constraints 0.0005 17.63 New neighbouring buildings 0.0203 3.41 Share of recreation area in zone for O1 0.5775 3.69 Share of recreation area in zone for Om 0.4652 7.97 Share of recreation area in zone for Smc 0.4890 8.25

LL(conv.) -1329.13 LL(0) -1588.36

Adj. ρ2 0.158 Observations 467

The developer-specific model shows better goodness-of-fit in case of SFH projects and MFH projects compared to the PLCM without segmentation. However, a second estimation showed that the improvement depends on the estimation run and thus on the sampling of alternatives. Parameter estimates also have deviations up to 50%. These findings suggest that the assumptions for an MNL model do not hold and thus random sampling biases parameter estimates. The model should be estimated on the full choice set.

Another concern is a potential panel effect coming from the data. In this work, it is assessed with the ’sandwich estimator method’ as described by Daly and Hess (2013). Using biogeme (Bierlaire and Fetiarison, 2003)

164

4.4. Discrete choice analysis of real estate development

for estimation because it computes the necessary robust statistics, the analysis shows that the classic and robust estimations differ, which sug- gests that there are issues with the model specification. However, the geometric means of the robust t-ratios are smaller for both models if the panel information is considered for estimation. The conclusion is that a panel effect is present. The reduction in the biogeme means of the t-ratios are 5.8% and 2.2% for the SFH and MFH model respectively. Compared to the results of Daly and Hess (2013), the effect is small. One reason is that only 9.6% and 8.4% of the observations are of the same respondent for SFH and MFH projects respectively.

4.4.3.3 Conclusions from developer type specific estimations

The analysis gives further evidence that real estate developers do behave differently in terms of location choice for their projects in the Canton of Zurich. The multinomial location choice models show significant differ- ences between the three investigated developer types for some variables when formulating models with basic interaction variables. This shows that behavioural differences across developer types can be relevant for real estate supply modelling. The insights in behavioural differences of real estate developers are: a) Developers who only develop once have the weakest preference for central locations. This indicates that professional developers tend to build at central locations. b) In case of SFH projects, developers who keep a portfolio of estates are less concerned about the development projects of others. This is arguably a consequence of the properties they already have that seem to fix development activity to some extent in space. c) In the case of MFH, proximity to recreation areas is valued higher by O1 developers. It can also be interpreted that other developer types underestimate the importance of this attribute. These results are potentially biased due to endogeneity.

The model estimations reveal the behavioural differences between developer types, but the better fit to the data shown in the results is not a robust finding. The conclusion is that the considered information is insufficient to assess the heterogeneous preferences among real estate developers with enough detail. This is not only true for the estimation of the models, but even more so for the simulation of scenarios. The needed synthesis of an entire population of real estate developers would add additional uncertainty to simulation results. However, the relevance of considering developer heterogeneity also depends on the scenarios that need to be evaluated.

Goodness-of-fit depends on the estimation run and thus on the sam-

165

Chapter 4. Analysing Zurich’s real estate development

pling of alternatives. This indicates that the multinomial model and the chosen sample size might not be fully appropriate for the given modelling task. A possible reason is that the alternatives (in terms of parcels) avail- able to a developer are limited in the land market. This means that choice set formation should be revisited.

The negative sign regarding car accessibility is unexpected. This issue is probably due to an omitted variable, which biases the parameter estimate (Guevara and Ben-Akiva, 2006). Potentially omitted variables include noise, pollution and land price.

Further options for model improvement are to consider model formu- lations that capture correlations of alternatives in space and time. Spatial correlations can be accommodated in ’competing destination formula- tions’ (Fotheringham and Curtis, 1992). While their work is in the context of migration, it seems suitable to investigate the activity spaces of real estate developers.

In this study, inter-temporal decision-making is neglected. However, real estate developers consider rather long-term horizons compared to other decision makers. Therefore, it seems relevant to estimate models considering dynamic optimization (Train, 2009, p. 169), i.e. optimising choice over multiple time periods.

The quality of the observations in the DOCUMEDIA dataset is lim- ited for the purpose of discrete choice modelling. A lot of information is not included because the data has been collected for another purpose in the first place. The dataset itself comprises little information on the real estate developers themselves. Further reasons are incompleteness and low agreement with other datasets. The latter issue is important if the observations are to be enriched with information from other sources.

The reconstruction of the decision situation is clearly imperfect. Only one cross-section (in the year 2000) has been generated. One can argue that this is less of an issue when observing long-term decisions. However, it is still preferable to derive the context of the observations from a more detailed spatio-temporal database. In addition to neglecting the update of context information, all data sources contain errors that influence the results as well.

Furthermore, different data availability makes it difficult to model the entire supply spectrum in comparable quality. The residential sector is much better documented than the non-residential sector.

166

4.4. Discrete choice analysis of real estate development

4.4.4 Purpose specific models

All models developed so far are not satisfactory insofar as unexpected signs point at endogeneity problems. This section reports on the attempts to resolve this issue. Firstly, experiments with different imputations for land prices and a specification using an approximation of profitability in a PLCM are performed because a price variable has to be integrated from a theoretical point of view. Secondly, the segmentation into submodels is revised and separates the observations according to the observed pur- pose. It is aimed at a clearer separation of self-suppliers, i.e. households constructing their own house, and commercial developers. It is hypoth- esised that these two major groups are fundamentally different in their behaviour, which could be a reason for the implausible coefficients esti- mated. Segmentation according to purpose seems a promising direction since literature points out its importance (Table 2.16, McNamara (1983)).

4.4.4.1 Specification and estimation

To find a model with a concept of market competition, expected signs and significant estimates, the PLCM presented in Table 4.25 is taken as a starting point. In the following, only residential projects are used because there are sufficient observations.

The first experiment replaces the price per floor area with the square meter land price for a parcel. This variable disaggregates the average square meter land price in the respective municipality to the parcel. The estimated coefficients are positive, which points again at insufficient ac- curacy of measurement, i.e. omitted variables. Another explanation can be seen in relation to the point in time of acquisition of the parcel. If the developer owns the land parcel already, it is plausible that it is developed at high land prices because they reflect some of the possible profit. If the developers anticipate a stronger price increase in the future, it makes also sense to buy highly priced locations since profitability will be higher. This suggests that developers are favouring high-priced markets because they expect them to yield higher profits.

In a second experiment, the accessibility variables are removed due to a considerable correlation19 with the land price variable and high correla- tion20 with the rent price variable. The latter is also tested to focus only on revenues. Dropping the accessibility variables lowers the adjusted ρ2

by approximately 0.05.

19Correlation coefficient over 0.3. 20Correlation coefficient between 0.6 and 0.76.

167

Chapter 4. Analysing Zurich’s real estate development

Versions for land price derivation To improve the land price variable, four approaches for its derivation are tested: a) Estimation of a simple linear regression in time, b) estimation of a panel linear model, c) esti- mation of a linear regression on location factors and d) combining the panel linear model with linear regression. The approaches are aimed at a better representation of price patterns over time and in space. For the estimation of the first two approaches, the yearly median land prices per municipality between 1996 and 2012 are used because the time series overlaps the period of observed developments. The third approach uses asking prices for parcels parsed between 2010 and 2014 to assess spatial patterns in more detail.

Simple linear models per municipality Linear time trends are es- timated for each municipality with a simple linear regression of the yearly median square meter land price to the time variable. The approach is problematic since there are municipalities with less than two observations, which prohibits model estimation. A second problem is that some esti- mated price trends result in negative predictions, which is implausible. The first approach is therefore not very meaningful and not discussed further.

Panel linear models In a second approach, a panel linear model with fixed effects per municipality is estimated in R using the plm-package (Model 2, Table 4.32). The variables used are described in Table 4.31. The mortgage interest rate was taken from the Website of the house owners association (HEV). All other variables are downloaded from the SAKZ. The variables are specific to municipalities, except for the reference mortgage rate, which is the same for all municipalities in the Canton. All variables are also time series from 1996 to 2012.

The panel data is first tested for the presence of individual, i.e. mu- nicipalities, and time effects (Honda, 1985). The test results suggest that the effect per municipality is more probable which supports fixed effects per municipality. Model 1 (in Table 4.32) is estimated by disregarding the panel nature of the data, i.e. all observations are pooled together, thus ignoring the fact that the same entities are observed multiple times. Compared to Model 1 with fixed effects, Model 2 shows changing signs for the estimates of the variables Vacancy rate and Reference mortgage rate (Table 4.32). In the first case, the change is unexpected because va- cancy rates are usually negatively correlated with land prices. The second change is desired since low mortgage rates can be expected to increase land prices. The signs of the other variables are expected. Generally,

168

4.4. Discrete choice analysis of real estate development

Table 4.31: Variable description of panel linear models

Variable Description Unit Data source

Built construc- tion zone share

Share of built construction zone in municipality

[%] SAKZ

Vacancy rate Vacancy rate of living units in municipality

[%] SAKZ

Public in- vestment per capita

Public spending per capita of in- habitants in municipality

[ C H Fr esident ] SAKZ

Municipal tax index

Tax rate in municipality [%] SAKZ

Reference mortgage rate

Weighted mean reference mort- gage interest rate per year in Can- ton

[-] HEV

the significance statistics for the explanatory variables and the overall model fit (R2) are reduced because a lot of the variation is captured in the fixed effects per municipality. This shows that the explanatory power of the temporal variation with the considered variables is actually limited. Eleven of the fixed effects deviate significantly from the overall intercept. The municipalities thus have, in most cases, a similar land price level despite the eleven exceptions. The fixed effects are shown in the appendix (Table A.13).

Even though the explanatory power of the fixed effects model is lower, it is preferred for the imputation because municipal price levels are individually represented. Whenever the yearly average square meter price is not available in the statistics, it is derived from the panel linear model. The result is a complete dataset of all municipalities with time adjusted yearly averages of square meter land price. The right price is then applied in the location choice model estimation by specifying an interaction variable with the development project, which filters out the respective price according to construction year and municipality.

Linear regression on location factors The spatial detail of the panel data discussed above is not corresponding with the observed parcels chosen by the developers. To account for this shortcoming, a linear re- gression model is estimated on asking prices for land parsed between

169

Chapter 4. Analysing Zurich’s real estate development

Table 4.32: Panel linear land price models

Model 1 Model 2

(Intercept) 1040.59∗∗∗

(129.93) Built construction zone share 9.54∗∗∗ 4.21

(1.11) (2.45) Vacancy rate −4.99 17.00∗∗

(6.51) (6.34) Public investment per capita 0.10∗∗∗ 0.04∗∗

(0.01) (0.01) Municipal tax index −12.44∗∗∗ −4.95∗∗∗

(0.49) (1.43) Reference mortgage rate 36.82∗∗∗ −37.93∗∗

(11.06) (12.84)

R2 0.38 0.08 Adj. R2 0.38 0.07 Num. obs. 1839 1839 ∗∗∗ p < 0.001, ∗∗ p < 0.01, ∗ p < 0.05, (St. Err.)

Data: SAKZ, HEV

2010 and 201421. The focus lies on the influence of parcel characteristics. Temporal effects are not considered because the number of observations is relatively small (233 observations) and the panel data described above provides longer time series.

The model is estimated in R and in UrbanSim yielding the same estimates. Table 4.34 contains the estimated coefficients of the land price model finally used for price derivation. It is chosen because it has the highest model fit (R2 of 0.36). In comparison to other studies, it is low (Table 4.36). Explanations can be the estimation of asking prices and too few observations. The variables are explained in Table 4.33. Unexpected signs are found for the variable Municipal tax index and for the dummy of centre zones. The reference category of the Zone Dummy variables is the residential zone, which is why a positive sign is expected. The negative signs for the other dummies seem plausible given available statistics (Rey, 2009). The tax index is not adjusted for the year of observation, which might cause the positive estimate. The model shows a similar estimate for the constant as in the panel linear model, which means that the overall

21The observation period is a consequence of data availability. Comparis provides land price observa- tions only recently.

170

4.4. Discrete choice analysis of real estate development

Table 4.33: Description of variables used in linear price model

Variable Description Unit Data source

Distance to CBD

Euclidean distance to the city centre of Zurich

[m] Cadastre, GWR

Mean federal tax revenue per capita

Mean federal tax revenue of mu- nicipality residents

[ C H F year ] SAKZ

Visible lake area

Lake area visible from the par- cel.

[ha] Digital terrain model of swis- stopo

Index of evening sun- shine exposure

Index for sunshine exposure in the evening.

[-] Digital terrain model of swis- stopo

Municipal tax index

Tax index of respective munici- pality.

[%] SAKZ

level of the observations in both datasets is comparable.

Combination of both approaches An approach to exploit all avail- able information is to combine the estimates from the panel data and from the asking price data. The variable used in the location choice model esti- mation uses the observed mean land price in the municipality whenever possible as a constant and applies the effects from parcel characteristics found in the linear regression on parcel data. If no observation for the con- stant is available, the yearly mean land price per square meter is derived from the panel linear model with fixed effects per municipality. For the implementation in UrbanSim, another interaction variable is formulated. The contributions of the parcel characteristics to the average land price are added as linear terms.

Approximation of profitability Approximations of profitability are tested in further experiments. Both the literature and the interviews showed that this variable is key for commercial developers. Profitability is calculated by subtracting the costs from the revenues. The normalised form is divided by the costs as shown in Eq. (4.3).

Revenue − Cost s Cost s

(4.3)

171

Chapter 4. Analysing Zurich’s real estate development

Table 4.34: Linear land price [CHF / m2] model estimated on parcel ob- servations

Model 1

(Intercept) 1153.70∗

(572.22) Dummy residential and commercial zone −34.75

(183.43) Dummy centre zone −154.73

(148.88) Dummy industry and commercial zone −341.12

(239.62) Dummy open space zone −393.71

(283.51) Dummy undefined zone −963.19

(783.39) Distance to CBD −0.04∗∗∗

(0.01) Mean federal tax revenue per capita 0.08∗∗

(0.03) Visible lake area 0.25∗∗∗

(0.04) Index of evening sunshine exposure 15.61

(13.99) Municipal tax index 5.78

(5.57)

R2 0.38 Adj. R2 0.36 Num. obs. 233 ∗∗∗ p < 0.001, ∗∗ p < 0.01, ∗ p < 0.05, (St. Err.)

Data: Comparis 2014

Here the monthly revenue is estimated as the average rent of living units in the municipality, multiplied by the number of planned units. The monthly revenue is extrapolated to an annuity, which is then discounted over 40 years. A problem with this approximation is that the value of the proposed development is multiplied by a locally estimated average unit price, which is based on the built space characteristics existing at the considered location. Therefore, it does not allow modelling the effect of newly introduced built space characteristics at a given place, which is clearly an advantage of the model presented by Wang (2009, 31–58). The

172

4.4. Discrete choice analysis of real estate development

considered costs are those for the land and construction of the planned structure. For the land costs, all derivations mentioned above have been tested. The construction costs are the estimates for the replacement value of the building by the GVZ.

The estimates for the land price variables as well as for the profit variables persistently showed unexpected signs. It seems like the land price information is still too approximate to allow for meaningful model estimates. Furthermore, it would be crucial to know when the developers bought the land for an appropriate calculation of the costs.

Submodels according to purpose The next experiment is to separate the observed projects according to the purpose the developer had in mind. In the data are the three levels of lease, sale and own-use. The last category thus comprises projects of self-suppliers, the first two are those of commercial developers.

For the commercial developers, the variables measuring location amenities are discarded because they are assumed to be profit-oriented. Therefore, only profitability or the rent price level as a measure of demand is kept together with the variable accounting for development constraints. In the case of developers developing for their own use, the price level variable is used to capture the reaction towards market price. The vari- ables for amenities are maintained to measure how the attractiveness of a location is composed. The Slope variable is replaced with the more precise measurements of Lake View and Evening Sunshine. The variable measuring recreational areas in the zone of the parcel shows the right sign, but is not included in the final model because it is not significant. Estimation using UrbanSim is explained in subsection 4.4.3.1.

4.4.4.2 Estimation results and interpretation

PLCM 4 is an example of the experiments with land price variables and profitability variables (Table 4.37). As in this case, the profitability vari- able shows an unexpected negative sign in all experiments. The sign for the land price variable comes out positive, which is also unexpected, but interpretable with omitted variables. A possible explanation is that high-priced markets allow for higher margins for the developers. Inter- estingly, the profitability variable is highly significant and the model fit is considerably increased (0.27 – 0.33). The calculated impact for the profitability variable is also very high, which supports the hypothesis that profit is a key target variable for commercial developers.

When using the variable of the average rent price per municipality,

173

Chapter 4. Analysing Zurich’s real estate development

Table 4.35: Variable descriptives of choice sets for purpose specific esti- mation

Variable Mean Sd Sum Min Max

Lease Fif of development to parcel constraints -5836.32 11960.39 -26438500 -130911.00 10.30 Mean m2 price of living unit in zone 7.36 1.39 33319 1.71 12.56 Sale Fif of development to parcel constraints -4067.40 6558.93 -47710600 -70678.00 10.31 Mean m2 price of living unit in zone 7.34 1.41 86100 1.92 12.56 Own-use Tax index 107.30 15.59 1828360 69.00 131.00 Distance to closest school 530.69 353.12 9043010 3.38 2800.42 Index of evening sunshine exposure -0.13 3.85 -2256 -8.30 20.59 Mean m2 price of living unit in zone 7.32 1.41 124662 1.71 12.56 Fif of development to parcel constraints -1729.92 5177.12 -29477800 -93473.80 10.70 Visible lake area 598.51 1315.74 10198600 0 7532.00

three submodels are found that yield significant estimates with expected signs for all variables (Table 4.37). The rent price variable is positive for projects with a commercial purpose. This is expected since higher revenues are the consequence. In the case of projects for sale, it would be better to consider sales prices since the two markets can have different price levels. In case of self-providing developers, the negative sign for price can be expected because the owners have to pay taxes on the deemed rental value22 of their property. These developers thus have more of a consumer perspective. Another explanation can be that self-providing developers are outbid by commercial developers and thus end up with parcels in low price areas.

The variable that captures the fit of a project to the development constraints is important in all submodels. A positive sign is expected and shows in all cases. An interpretation is that developers tend to exploit the allowed density regardless of the purpose.

Distance to the next school is negative, as expected. Generally, it can be assumed that locations close to education facilities are more attractive for residential use. The estimated impact of the variable is low (0.05% average utility share). It would be interesting to interact this variable with the self-providing developers’ family status because the presence of children of school age would be expected to raise the relevance of this aspect.

As suggested by other studies (Salvi et al., 2004), the tax index is

22In Switzerland, homeowners are required to pay income tax on a notional rental value for the home they use themselves, either as primary or as holiday home. This notional value is called "deemed rental value" or ’Eigenmietwert’. The deemed rental value is on average 70% of the potential market rent. Usually, the deemed rental value is estimated by the tax authorities.

174

4.4. Discrete choice analysis of real estate development

Table 4.36: Comparison of model fit statistic with literature

Study adj. R2

PLCM Own models 0.1 – 0.204 Dong and Gliebe (2011, p. 84) 0.097 – 0.142 Haider and Miller (2004, p. 154) 0.101 – 0.207

LPM Own model 0.36 Waddell and Ulfarsson (2003, p. 18) 0.76 Hartmann (2013, p. 61 – 73) 0.482 – 0.641 Kuster-Langford (1989, p. 84) 0.62 – 0.79

very influential. It can be explained by an argument similar to the one in the context of the land price models. To save taxes, self-providers prefer locations with low tax rates.

The results suggest that the amenities of lake view and sunshine in the evening are appreciated, but of little relevance. This is in contradiction to the map in Fig. 4.12. It is possible that some of their effect is captured in the price variable. It seems an appropriate ranking that the tax index is most important, followed by the rent level variable and the fit to building constraints.

The estimation results can be summarised as follows: Commercial developers look for locations with high revenue potential and exploit the allowed density as much as they can. Self-providing developers look for places with a low tax burden, which includes low price levels due to the taxes they have to pay on the value of their property (Eigenmietwert). They also make use of the allowed built space volume, preferably at locations close to education facilities and lakes.

The R2 is on the same level as similar published models (Table 4.36). The simulation results could be compared on the basis of root mean square errors (RMSEs) of predicted versus observed projects to assess the predictive quality of the models (Dong and Gliebe, 2011).

4.4.4.3 Conclusions from purpose specific models

Introducing submodels according to the purpose of the development finally allowed an estimation of consistent models. It can be concluded that finding the right segments is critical for successful model estimation. The typology chosen for the first models seems to be inadequate. The better

175

Chapter 4. Analysing Zurich’s real estate development

Table 4.37: Estimation results segmented by purpose

Variable PLCM 4 PLCM 7

Estimate t-Value Estimate t-Value

Lease Mean m2 price of living unit in zone 0.2770 5.10 Fit of development to parcel constraints 0.0004 7.41 0.0006 13.38 Profitability proxi -17.0611 -12.70

Observations 151 151 LL(0) -513.58 -513.58

LL(conv.) -343.37 -406.60 Adj.ρ2 0.328 0.204

Sale Mean m2 price of living unit in zone 0.2051 5.63 Fit of development to parcel constraints 0.0003 6.52 0.0005 18.76 Profitability proxi -21.4444 -22.48

Observations 391 391 LL(0) -1329.87 -1329.87

LL(conv.) -970.22 -1178.38 Adj.ρ2 0.269 0.112

Own-use Fit of development to parcel constraints 0.0007 13.77 0.0008 12.58 Mean m2 price of living unit in zone -0.2452 -7.10 -0.2313 -6.63 Distance to closest school -0.0005 -3.90 -0.0005 -4.01 Visible lake area 0.0001 4.18 0.0001 4.12 Municipal tax index -0.0216 -6.03 -0.0183 -5.66 Share of recreation area in zone 0.1044 1.18 Index of evening sunshine exposure 0.0489 3.98 0.0577 4.72

Observations 568 568 LL(0) -1931.88 -1931.88

LL(conv.) -1735.54 -1732.40 Adj.ρ2 0.098 0.100

estimations suggest that separating developers developing for own-use from commercial ones is important and supports the main hypothesis that developers are of different types. LC models could help find the appropriate types if more information on the developers were available.

It is possible to have approximations of profitability in PLCMs of Ur- banSim. This allows relating it to other location factors. Experiments with profitability variables indicated that it is highly significant for commercial developers, but it did not show the expected sign. It is assumed that the derived land price information is inaccurate causing the unexpected sign. If influential variables are not available with enough quality, model estimation becomes difficult. This might be more often the case when working with revealed preference data.

The rent price variable is used in the final models. It allows incor-

176

4.5. Conclusions from land development analysis

porating the market signal. An increasing rent price level encourages commercial developers and discourages self-providing developers. The positive sign for commercial developers can be explained with expected higher profits. In contrast, self-providing developers avoid areas with high rent price levels, arguably due to taxes they have to pay on property value. The property value is determined by market prices.

Model fit can be increased in the case of self-providing developers by adding other location variables. This seems plausible since this type of de- veloper is going to live at the respective site afterwards while commercial developers are mainly interested in the revenues they can achieve.

4.5 Conclusions from land development anal- ysis

The descriptive statistics revealed different developers active in the Canton of Zurich. An oligopolistic structure is found for their population. The share of developers who built one project for their own use is shrinking in the observation period from approximately 40 percent to 20 percent. This indicates consolidation in the real estate supply industry.

The typology based on development frequency and purpose does not consistently match reality as found by interviewing the respective de- velopers. The interviews further revealed that a lot of factors vary with developers’ purpose and level of experience. Several of potentially influen- tial factors could not be considered due to a lack of data availability, such factors include ownership structure, capital costs, date of land acquisition, land prices and endowment with other properties (portfolio). Property ownership is especially of interest since it determines beneficiaries of land rents. The limited information available reduces possible typologies and prohibits an estimation of LC models.

Estimating location choice models proved to be a viable option for investigating developer heterogeneity. The approach with separate param- eter estimates for developer types is difficult due to unclear categories. Relying on the purpose of development yields better results. This may be due to using an arguably important characteristic23 of the developers directly, which provides a more meaningful discrimination.

Building location choice models can be a viable option if no informa- tion on projects is available. An assessment of the quantity of multiple- building projects present in the study area suggests estimating potential bias.

23Compare to Table 2.16.

177

Chapter 4. Analysing Zurich’s real estate development

A ceiling of about 0.2 is revealed when all model fit statistics (ρ2) are compared over all estimated models. This suggests that the explanatory power of the data is fairly well exploited with respect to location choice. The exceptions are models with a profitability approximation included (Table 4.37), but the variable shows the wrong sign. It is probable that this is related to the cost component of the variable since multiple attempts to derive and use land prices failed.

The model developed by Wang (2009) is favourable since the expected value of a project on a considered site can be estimated with the price models. For the purposes of this study, it could not be used because necessary price data from developers’ evaluations were not at hand.

Estimating submodels according to building type is not very useful since one cannot filter the alternatives with an interaction variable, which would allow zoning restrictions on building types to be considered. The estimation of separate models would allow for building-type-specific filters. While building such segments makes the models more accurate, it reduces the sample sizes, which might prohibit their estimation.

The purpose specific models are used for simulation. Further mod- els of the models system and the simulation results are presented and discussed in the next chapter.

178

Chapter 5

Simulating Zurich’s land development This fifth chapter reports on the simulation of scenarios with developer- specific real estate development models using UrbanSim. The first section describes the implementation of the land use transport interaction (LUTI) simulation for the Canton of Zurich, which also includes the description of the reference scenario. Calibration and validation are also treated. The results of the simulation are in the second section, with the scenario definition outlined first. Then the results are presented, discussed and conclusions drawn.

5.1 Land use transport interaction simulation of the Canton of Zurich

For the investigation of developer type specific scenarios, a reference scenario that represents the ’business as usual’ case is needed. This base- line is the development path that is most likely to happen without any modifications to the expected development. Its definition and set-up as a LUTI simulation with UrbanSim requires: a) the creation of the base year with all entities depicted in Fig. 5.3 as ovals, b) assumptions about their evolution, either in terms of control totals or by defining appropriate transition models, c) selection of geographical units of analysis (GUA), d) estimation of choice models to define the entities’ behaviour, e) esti- mation of a hedonic real estate price model, f) coupling with a transport model, g) implementing regulations, h) integration of relevant environ- mental data and i) a compilation of already approved projects and political measures becoming effective during the simulation period.

The creation of the base year containing the integration of environmen-

Chapter 5. Simulating Zurich’s land development

tal data is briefly described in subsection 5.1.1. More emphasis on that process is given in Schirmer et al. (2011). An overview of the estimated models (rectangles in Fig. 5.3), and the real estate development model (REDM), presented in Table 4.37, follows in subsection 5.1.2.

Parcels are chosen as the most detailed GUA because the data is high quality and it is also the legal unit for land use regulations. The parcel data is not available for the scarcely populated municipalities in the southeast of the canton (Fig. 5.1), which reduces the number of municipalities from 171 to 151. It would have been ideal to use a functional region defined by commuting patterns (Killer, 2011; Gmünder et al., 2010) as a study area, but time and budget constraints have limited the simulation area to these 151 municipalities.

The shades of grey show the quality of data preparation for the base year 2000. The percentage is the average of relative differences in the number of main entities per municipality. For most municipalities, a quality above 80% is achieved. Separate analysis of each main entity yields the information that jobs contribute the most to the average of relative differences. There is not enough space to allocate all reported jobs in these municipalities because the survey includes all jobs, even those not occupying workplaces. It is plausible that many of these jobs are associated with the airport, which is located in the white area in the centre (Fig. 5.1).

Further GUAs of interest are traffic analysis zone (TAZ), municipal- ities1 and the total study area. TAZs are considered appropriate for the use of traditional transport models and their output. Municipalities are important because much of the data, e.g. tax levels, is associated with this level of administration. It is also an appropriate geography for evaluating and communicating the results, since the voting public decides on policies at this level.

5.1.1 Data preparation Data preparation is an important task for the set-up of a LUTI simulation and requires a fair amount of work. It consists of integrating relevant available information into the format required by the simulation software. The process consists of: a) obtaining the data, b) backing it up, c) checking the data, d) cleaning it, e) combining and integrating various datasets and f) converting it to the required format. To be able to repeat the process, in case there are necessary modifications, it was decided to automate it as much as possible. The entire process is coded with a

1Grey and black poligons in Fig. 5.1

180

5.1. Land use transport interaction simulation of the Canton of Zurich

Figure 5.1: Simulation area of the Canton of Zurich

Quality of base year < 60% 61% - 70% 71% - 80% 81% - 90% 91% - 100%

Lake Border of Canton Zurich

0 5 10 15 km

Source: Schirmer et al. (forthcoming)

combination of scripts that can be evoked by a main shell script. However, it is also possible to execute single scripts that are designed as tools for specific tasks. First, a set of scripts import the original data from the various file formats into a PostgreSQL database. This is because much of the information has to be related to spatial attributes and the UrbanSim developers have made a structured query language (SQL) definition of the data model available for the simulation. Further reasons to choose PostgreSQL were its licence, reliability and performance. A second set of scripts converts the original data into the format requested by UrbanSim. This step includes cleaning, completion and complementation to achieve a consistent database that is as complete as possible. This step is not as modular as the previous one due to interdependencies. Most operations have been coded in SQL. An exception is the derivation of car ownership and income from the micro-census of travel behaviour, which is done in R. Computation is done in parallel for municipalities in order to improve performance. More details can be found in Schirmer et al. (2011).

The Fig. 5.2 shows five data sets contributing to the building entities finally used in UrbanSim. The building dataset (gwr_buildings) from the

181

Chapter 5. Simulating Zurich’s land development

Figure 5.2: Preparation of building data for use in UrbanSim

Household

Living unit

Parcel

Building

Resident

arv_parcels parcel_id

arv_soil_coverage building_type_id land_area geom

gwr_apartments residential_units m2_per_unit

gvz_buildings improvement_value

gwr_buildings building_id

building_quality_id stories year_built

by address

by coordinates

by egid

by coordinates

federal building and housing register (GWR) were used as a reference because the entities are geocoded and records from the Building Insurance of the Canton of Zurich (GVZ) are not. The GVZ dataset contains an estimate of the buildings’ replacement value, which can be added by matching the addresses. The land coverage dataset of the Cantonal Office for Spatial Development (ARE ZH) contains geometries (geom) of building footprints, and consequently also its area and useful building categories. This data is joined to the GWR record whenever its coordinates lie on the respective polygon. The parcel_id of a building is determined using the same mechanism. Joining the apartments is simple since the GWR data is already related via unique identifiers (Federal building identifier (egid)).

Data preparation also includes the determination of categories and the categorisation of entities. Relevant categorisations found here are a) employment sectors, b) land use types, c) building types and d) zoning plan types. Many categories are predefined in the data but cause difficulties when combining datasets if the categories are not identical. If possible, which categories correspond to one another should be defined.

182

5.1. Land use transport interaction simulation of the Canton of Zurich

Figure 5.3: Entities and models of the LUTI model of the Canton of Zurich

Project location choice

Household location choice

Employment transition Employment relocation

Employment location choiceReal estate price

Building transition

Household relocation

Workplace choice

Household

Geographies Plan types (development constraints) Environment data (e.g. topography)

Agents / entities

UrbanSim model

Living unit

Parcel

Building

Demography Income and car ownership update

Resident

Departure time choice

Mode choice Route choice

Accessibility

MATSim model Modgen model

Network

Job

Parcel

Building

5.1.2 Models Three software packages are used for the entire simulation: Modgen (Statistics Canada, 2009, 2011) for simulating demography, Multi-Agent Transport Simulation (MATSim) for simulating transport and UrbanSim for the simulation of land use in space. The following covers the integrated models briefly to present a complete picture. A list of references is provided in Table 5.1. The sequence of model execution is presented in the following and describes how the simulation works.

5.1.2.1 Demography

The purpose of the demographic model is to update the population over the course of the simulation. The demographic evolution is microscopically simulated in advance and then fed to UrbanSim for location choice. Most of the demographic models are rate-based and require transition proba- bilities as input. The detail is considerable and requires many population segment specific parameters (a total of 1,660,696) to be set. Simulated demographic events are ageing, migration, labour participation, house-

183

Chapter 5. Simulating Zurich’s land development

hold formation (marriage, devorce, leaving parent household) births and deaths.

5.1.2.2 UrbanSim

UrbanSim distributes land uses in space. The main entities considered are households, employment and real estate projects that are located in dwellings, buildings or parcels, respectively. The locations are connected by transport networks, i.e. streets and public transport (PT) infrastructures.

Population update Before the core models can be applied, it is nec- essary to update the population with characteristics used by additional models. Income and car availability are two such cases in the current sim- ulation. Income is derived with a regression to the level of education, the number of cars in the household and the size of the household (Table A.7). Car availability is simulated with a binomial choice model where chances of not having a car decrease with education level. Chances of having a car increase with household size, distance to the central business district (CBD) of Zurich and income (Table A.8).

Building transition model (BTM) This model determines the number of units per considered submarkets. Nine non-residential submarkets are considered in addition to the residential market. The submarkets correspond to the employment categories (Table A.10) in the employment location choice model (ELCM). The number of units is calculated on the basis of market-specific vacancy rates that are entered in the simulation as assumptions. Here, a 0.66% vacancy for the residential and 4.02% for non-residential markets are assumed, based on cantonal statistics. The rate is low2 compared to other regions.

Whenever the simulated vacancy in a market falls below the respective target vacancy, the necessary number of projects, including buildings and associated living units, is sampled from a pool of development projects to relax the constraint again.

Project location choice model (PLCM) The sampled projects are then located on parcels by the PLCM. Its specification and estimation is de- scribed in subsection 4.4.4. Thirty alternatives are sampled and evaluated for simulation, which is the UrbanSim default. Due to computational

2Besides the tightness in the local market, it also relates to different practice of statistical offices. The officially reported rate only includes units offered in the market (Thalmann, 2012). This is not considered in the simulation.

184

5.1. Land use transport interaction simulation of the Canton of Zurich

constraints, it is not possible to include all the alternatives. The eligibility of a parcel is determined by plan type and associated density levels for respective uses. Extension, rebuilding, replacement and demolition are not simulated.

Real estate price model (REPM) A hedonic real estate price model de- termines the rent price for living units on the basis of living unit character- istics and location variables (Table A.9). Price models for non-residential units are not available so far.

Employment transition model (ETM) Next, employment transition is simulated based on assumed control totals per sector. The model creates or deletes the requested number of jobs for each simulation year. A trend continuation, as observed between 1996 and 2003, per sector is assumed for the simulation period.

Employment relocation model (ERM) The number of relocating jobs is calculated based on exogenous relocation rates per sector. Sampled jobs are left to the ELCM to be relocated in a building with remaining capacity.

Employment location choice model (ELCM) This model simulates location choices of jobs. The estimates of the ELCMs show that jobs tend to cluster in highly accessible places (Table A.10). Differences between the sector-specific submodels occur for both highway access and centrality. Unlike other sectors, jobs in hotels and restaurants (HR), service (Srv) and health (Hlt) tend to locate away from highway access points. Service and health jobs tend to locate centrally whereas jobs of other sectors do not.

Workplace location choice model (WLCM) Employed persons are linked with a job by the WLCM. It is designated as a WLCM because the job is already located when the person chooses it. The choice currently depends on the distance between the worker’s residence and the job location. Chances to find a job decrease exponentially with distance between the two locations. The model has been fitted against population census data (Table A.11) for the year 2000.

Household relocation model (HRM) Since population development has already been simulated, no transition model is needed for households.

185

Chapter 5. Simulating Zurich’s land development

To determine relocating households, relocation rates distinguished by income and age of household head are applied. Analogue to the ERM, selected households are located by the household location choice model (HLCM).

Household location choice model (HLCM) The HLCM locates house- holds and associated persons in an available living unit. The multinomial logit (MNL) model features non-linear interaction terms in the utility function for distance to workplace and previous residential location. The strongest effects come from these two variables and the rent-income ratio (Table A.12). An increase in all the three variables decreases the chances of a household to consider the living unit.

5.1.2.3 MATSim

MATSim is used to calculate travel times and accessibilities for each parcel. Accessibility influences rent price, job and household location choice. Also location choice for building projects is influenced by accessibility via the rent price variable.

MATSim is a dynamic, activity and agent-based microsimulation of travel demand. This means that daily plans for activities of an initial population are simulated on networks resulting in scored plans after their execution. An iterative evolutionary algorithm calculates a relaxed state of the system such that agents cannot significantly improve the score of their plans.

As part of their improvement strategies, the agent can make choices regarding mode of transport (PT or car), departure time and route. Destina- tion choice is not included since only activity chains of type home-work- home are considered. Origin and destination of the trips are thus given by the travellers’ residence and workplace location. Due to performance limitations3, only 10% of the population are actually simulated and the transport simulation is only run for every fifth year of simulation. The resulting travel indicators are attached to parcels (workplace accessibility) and persons (mode, travel time, travelled distance), which are fed back to UrbanSim influencing land use choices. Departure and arrival times at activity locations are not exchanged with UrbanSim because there is no use for this information at this point.

3In this case a 10% MATSim run takes approx. 4 hours.

186

5.1. L

and use

transportinteraction sim

ulation of

the C

anton of

Z urich

Table 5.1: Overview on models in the LUTI model

Model Abbrev. Dependent Independent Model type Reference

Demography Size pop. Seg- ments

Various person attributes and rates

Microsim., rate based

Turci et al. (2012)

Income update Income Table A.7 Regression Schirmer et al. (forth- coming)

Car availability update Car ownership Level of education, HH income, HH size

Binary choice Schirmer et al. (forth- coming)

Building transition BTM Nb of buildings Residential vacancy rate Transition Subsection 5.1.2.2 Project location choice PLCM Parcel of projects Table 4.37 MNL Subsection 5.1.2.2 Real estate price REPM Price living unit Table A.9 Regression Schirmer et al. (forth-

coming) Employment transition ETM Nb jobs Vacancy per sector Transition Schirmer et al. (forth-

coming) Employment relocation ERM Nb relocating

jobs Sector Rate based Schirmer et al. (forth-

coming) Employment location choice

ELCM Building of jobs Table A.10 MNL Schirmer et al. (forth- coming)

Workplace location choice

WLCM Job of person Distance to job MNL Schirmer et al. (forth- coming)

Household relocation HRM Nb HH relocating HH income, age head HH and according rates

Rate based Schirmer et al. (forth- coming)

Household location choice

HLCM Living unit of HH Table A.12 MNL Schirmer et al. (2013)

Transport MATSim Transport indica- tors

Locations HH and job, car own- ership, employment, network

Microsim., ac- tivity based

Balmer (2007)

187

Chapter 5. Simulating Zurich’s land development

5.1.3 Calibration Batty (2009, p. 51) defines calibration in the context of urban modelling as:

“The process of dimensioning a model in terms of finding a set of parameter values that enable the model to reproduce characteristics of the data in the most appropriate way. Cali- bration is not the same as validation which seeks to optimize a model’s goodness of fit to data, but often, these processes are equivalent.”

Description of calibration steps Model estimation is a precondition for calibration, but does not concern the whole system. Therefore, a second step is needed to calibrate the overall system to observed development. It is basically a comparison and manual adjustment to improve a match to the statistics of interest. For some parts of that problem, automated methods have been proposed (Flötteröd et al., 2012, 2011; Flötteröd, 2009). The author is not aware of a similar approach to the use of LUTI models. For the simulation at hand, the demography model and the travel model are manually calibrated as follows.

Demography model calibration The simulation is calibrated against the overall population size of the cantonal statistics with the following steps: • Multiply emigration probabilities by 1.3 • Multiply immigration numbers by 1.5 • Multiply mortality by 1.8

The individual statistics for overall population dynamics such as births, deaths and migration counts are not fitted (Fig. 5.4). The simulation produces an ageing population that has an effect in HRM and HLCM since they include the age of the head of household as an independent vari- able. Sensitivity of the travel model towards ageing is not implemented, i.e. models of travel-related decisions do not include the agent’s age.

Calibration of MATSim Travel model calibration is done against travel times of the cantonal travel model (Vrtic et al., 2005) and previously calculated workplace accessibilities (Löchl, 2010). The travel times are reproduced approximately with the initial parameter set and thus left as they are. The parameter of the distance decay function Eq. (2.1) for accessibility calculation is set to 0.2 (Killer et al., 2013, p. 11).

188

5.1. Land use transport interaction simulation of the Canton of Zurich

Figure 5.4: Validation statistics of demographic model

0

20

40

60

2000 2004 2008 Year

R e

la tiv

e E

rr o

r [%

]

Statistic

Total

Births

Deaths

Emigrants

Immigrants

Saldo

5.1.3.1 Validation

Validation is the assessment of the model performance after calibration. In a dynamic simulation, a validation period has to be defined, in this case, from 2000 – 2010. According to Gilbert and Terna (2000, p. 66), valida- tion can be done at four levels, which depends on simulation performance and detail of validation data. • Level 0: Behaviour of simulated agents mimics the one of observed

object • Level 1: Qualitative agreement of simulation with empirical macro-

structures • Level 2: Quantitative agreement of simulation with empirical macro-

structures • Level 3: Quantitative agreement of simulation with empirical micro-

structures The base line can be validated up to level 3. Different data should be used for validation than for estimation and calibration. The estimation data is describe in subsection 4.4.1. The results are validated against official statistics of the Zurich Cantonal Statistical Office (SAKZ). Some of the indicators are not available in the official published statistics. Com- plicating issues are different categories and irregular time series of the

189

Chapter 5. Simulating Zurich’s land development

Table 5.2: Errors in totals of main entities 2008

Indicator [%] Abs. Deviation

Persons -1.26 -15191 Jobs 0.88 6774 Living units 4.79 28948 Buildings 1.42 3229

SAKZ-data. Validation of scenarios cannot be done against measured data because the developer population cannot be replaced in reality to observe what happens. Therefore, the scenario can only be validated up to level 0. Results on the system level can however be examined for plausibility (subsection 5.2.2).

The three main dimensions of time, space and content have to be covered. For each indicator, dynamics and distribution in space can be assessed with longitudinal analysis (time series) and cross-sectional analysis (spatial patterns). The main options for analysis are calculation of statistics and visualisation for better context-related interpretation. The methods used for visualisation are time series plots and maps. Animations would be especially suitable for spatio-temporal dynamics, but cannot be used in this printed document.

In the following, the main simulated entities (persons, jobs and projects) are checked. It has to be kept in mind that calculated accuracy measures depend on the detail of analysis, e.g. the deviations on the level of munici- palities are lower than those on the level of TAZ. Firstly, after eight years of simulation, there is a search for errors, and secondly, the simulated data is analysed longitudinally and then spatially.

Deviation after validation period Table 5.2 shows the errors regarding the totals of the main entities after eight years of simulation. The choice of the year for comparison is 2008 because validation statistics for jobs are only available for that year. The totals show an underestimation for number of persons and an overestimation for all other quantities. The simulation especially overstates built space production (4.8% in living units).

Assessment of dynamics Figure 5.5(a) shows the development of the totals of the main entities for simulation and in the validation data. The coefficient of variation (CV) is used as comparative statistic (Fig. 5.5(b)).

190

5.1. Land use transport interaction simulation of the Canton of Zurich

The curve for persons shows the main result of the demographics model. Accuracy for the total is high (CV smaller than 0.01), which is also a consequence of calibration. The development of households shows a similar slope as for persons. Consequently, average household size is constant. The validation statistic for the number of households is not available since the census is only done every ten years. The total of jobs is the assumed control total and thus grows linearly. Higher growth in the first simulation year is an artefact of data preparation for the base year, where some jobs are discarded. The parallel development of buildings and living units is the consequence of the vacancy mechanism triggering built space production. The highest CVs are found for buildings and living units as the simulation progresses (CVs of more than 0.4). The employment census is available for the years 2001, 2005 and 2008, which leads to the dip in the curve.

Figure 5.6 shows the comparison of simulated new construction to the validation statistics in terms of buildings, i.e. the first derivative of building stock development. The validation data shows regular increments of buildings. Simulated increments are more volatile. The CV ranges between 0.05 and 0.5 over the validation period, but does not consider values from 2010. This shows that accuracy is variable over time and much lower compared to the metrics of the totals where CV varies between 0.005 and 0.05. It also reflects that changes are relatively small compared to the total building stock.

Spatial assessment Figure 5.7 shows that jobs are the least accurately predicted per municipality. The number of jobs is especially underesti- mated in the Glatttal, which is located east of Zurich. It is conspicuous that jobs in the city of Zurich are overestimated while those in Winterthur, the other large city, are underestimated. This effect might be attributable to the variable Distance to Zurich CBD, which attracts jobs to the city of Zurich. The errors for number of persons are similar for the two big cities and the municipalities in the Glatttal. Population growth is overestimated along the Lake of Zurich and the northern part of the canton. For some municipalities, it seems to be a consequence of the distribution of living units. Underestimation in the city of Zurich and in the Glatttal seems to have other causes.

Table 5.3 shows the descriptive statistics of the relative error of the respective indicator over all municipalities. The metrics show that build- ings are most accurately predicted regionally (per municipality) (standard deviation (SD) of 4.9%). Persons follow with 12.8%, which is almost an equivalent accuracy to living units (13%). The two entities are strongly

191

Chapter 5. Simulating Zurich’s land development

Figure 5.5: Time series of main entity totals

(a) Totals

250000

500000

750000

1000000

1250000

2000 2002 2004 2006 2008 2010 Year

[− ]

Run

Validation statistic

Baseline

Entity

Persons

Households

Jobs

Living units

Buildings

(b) Coefficient of variation

● ● ●

● ●

● ●

●

●

●

●

● ● ●

● ● ● ●

● ●

●

●

0.00

0.01

0.02

0.03

0.04

0.05

2000 2002 2004 2006 2008 2010 Year

C o

e ff ic

ie n

t o

f va

ri a

tio n

Entity ●● Persons Households Jobs Living units Buildings

Validation data: SAKZ and GWR 2000 – 2010

192

5.1. Land use transport interaction simulation of the Canton of Zurich

Figure 5.6: Validation of new buildings added to the building stock per year

1000

2000

3000

4000

2000 2002 2004 2006 2008 2010 Year

[− ]

Indicator

New buildings per year

Run

Validation statistic

Baseline

Validation data: GWR 2000 – 2010

related since households choose living units to locate. It is expected that the number of persons is distributed with a quality similar to the living units. Employment shows the largest standard deviation, which suggests that these location choice models need the most improvement. Minimum and maximum errors per municipality can be large, as the example of jobs shows (140%). Comparison of mean and median indicate whether the distribution is skewed to the left or right. If the median is lower than the mean, the distribution is skewed positively, i.e. the longer tail is to the right of the centre (Fig. 5.8). It means that there are more small errors than large ones, which is the case for all indicators shown in Table 5.3. Because each observation that shows a different error than the overall total has to be ’compensated’ by another municipality, it can be concluded that allocation concentrates on a few municipalities at the expense of many others.

193

Chapter 5. Simulating Zurich’s land development

Figure 5.7: Relative error of simulated versus surveyed counts of main entities per municipality 2008

Persons Jobs

Living units Buildings

Deviation [%]

20 10 5 2.5 1 0 −1 −2.5 −5 −10 −20 −46

Data: c© 2013 swisstopo (JD100042), SAKZ

194

5.2. Scenario

Table 5.3: Validation statistics over municipalities for selected indicators in 2008

Indicator Min Q1 Mean Median Q3 Max Sd

Persons −20.83 −5.14 3.59 1.16 9.89 58.30 12.79 Jobs −46.05 −21.50 −1.10 −8.85 9.84 139.68 30.80 Living units −13.68 2.16 11.11 9.14 18.30 65.09 13.00 Buildings −14.93 −2.70 0.49 0.48 3.52 13.74 4.89

Figure 5.8: Smoothed error densities over all simulated municipalities

−50 0 50 100 Deviation [%]

Indicator

Persons

Jobs

Living units

Buildings

5.2 Scenario

Scenario simulation is done after calibration and validation. Here only one scenario is simulated as proof of concept because the quality of the simu- lation is not yet on a satisfactory level. Simulation and its evaluation are very time-consuming and thus did not allow to run multiple experiments, which is actually desirable.

195

Chapter 5. Simulating Zurich’s land development

5.2.1 Scenario definition

For the scenario, only one factor is changed at a time. This factor is the share of projects according to the purposes of lease, sale or own-use, which are set to 20%, 60% and 20% respectively in the development project pool. In the reference scenario, the respective shares are 13%, 32% and 53%. The share of projects for sale and for own-use are consid- erably changed, whereas the share for lease only changes by 7%. This modification assumes a transition in the development industry towards commercial actors.

5.2.2 Results of scenario run

The scenario results are presented by focusing on several aspects as listed below: • Main entities • Supply by developer type • Centrality of developments by developer type • Land consumption • Compactness • Rent prices

The following structure is used for the discussion of each aspect. Firstly, the effects of the entire simulation period are discussed. Secondly, the dynamics are investigated, which leads to the outcome by analysing the simulated data longitudinally. Thirdly, the spatial patterns of the simulation are assessed by showing maps. Simulation data from the baseline is the reference for the comparison.

Main entities Table 5.4 shows the relative and absolute deviation of the scenario to the baseline. There are more persons in the scenario than in the baseline, which is surprising since the demography data is the same. The difference comes from the fact that not all households, and consequently persons, can be located. Persons in a household without a living unit ID are not counted in this indicator. The same happens with jobs, but there are only 23 unplaced jobs. Similarly, some buildings and living units cannot find a suitable parcel. These buildings remain unplaced. There are fewer buildings in the scenario than in the baseline, but more living units. The lower number of buildings can be explained by a higher average of living units per building.

196

5.2. Scenario

Table 5.4: Scenario effects 2029 regarding totals of main entities

Indicator [%] Abs. Deviation

Persons 6.87 94513 Households 5.61 41135 Jobs 0.00 -23 Living units 5.56 41085 Buildings -2.22 -5735

Dynamics Fig. 5.9 shows the evolution of population totals during the simulation. Households and persons show the predicted linear increase as simulated by the demographic model. The knees in the curves of households, persons and living units originate from unallocated residential projects due to a scarcity of suitable parcels (Fig. 5.16). Consequently, it is only meaningful to evaluate results up to the year 2015. The exponential increase of jobs is an assumption implemented via control totals. All jobs are allocated because non-residential projects can find enough viable parcels. The close match of living unit and household curves is the consequence of the vacancy mechanism, which controls the number of new living units provided. The assumed fractions of purposes in the development project pool for the scenario reduces the number of buildings slightly.

Spatial variation Figure 5.10 does not show an obvious spatial pattern for households or for jobs. The comparison of households with living units shows that households follow the available living units, which is a consequence of the very low vacancy assumed (0.4%). Vacancies for jobs are high, which imposes few availability constraints on location choice for jobs. In all municipalities, fewer buildings are built, which is due to larger projects in the scenario pool (Fig. 5.10).

A comparison with the remaining residential capacity according to zoning shows somewhat complementary maps (Fig. 5.11). This indicates that more residential units are built in municipalities where more capacity is remaining. The likelihood of finding alternatives to these municipalities in the choice set is higher.

The distribution of living units is skewed negatively (Table 5.5, median larger than mean). For a majority of municipalities, this means more living units are produced in the scenario. A minority of municipalities receives fewer living units, but the percentage of deviation is higher in these cases.

197

Chapter 5. Simulating Zurich’s land development

Figure 5.9: Development of main entity populations in the simulation area over the simulation period

500000

1000000

1500000

2000 2010 2020 2030 Year

[− ]

Entity

Persons

Households

Jobs

Living units

Buildings

Run

Baseline

Scenario

Table 5.5: Descriptive statistics of distributions of deviations [%] over municipalities in 2015

Indicator Min Q1 Mean Median Q3 Max Sd Household density -33.3 -2.1 2.2 1.6 6.1 41.7 9.5 Jobs density -44.4 -5.8 1.5 0.0 6.3 105.7 15.2 Living units density -33.3 -2.0 2.2 1.5 5.8 41.7 9.5 Buildings density -11.6 -6.0 -4.3 -4.0 -3.0 0.0 2.6

A few municipalities receive considerably fewer living units, favouring many municipalities that receive more.

Supply by developer type The results in Table 5.6 show that an as- sumed consolidation of the real estate industry leads to fewer projects, which is explained by the larger size of the projects. Developments and living units provided for commercial purpose increase. This is a direct consequence of the scenario definition.

198

5.2. Scenario

Figure 5.10: Deviations in main entities’ density per municipality 2015

Households Jobs

Living units Buidings

Deviation [%]

20 10 5 2.5 1 0 −1 −2.5 −5 −10 −20 −44.4

Data: c© 2013 swisstopo (JD100042)

199

Chapter 5. Simulating Zurich’s land development

Figure 5.11: Residential floor capacity and living units 2015

(a) Remaining residential floor capacity

Deviation [%]

14.9 10 5 2.5 1 0 −1 −2.5 −5 −10

(b) Living units built

Deviation [%]

50 20 10 5 2.5 1 0 −1 −2.5 −5 −10 −20 −50 −95

Data: c© 2013 swisstopo (JD100042)

Table 5.6: Scenario effects in supply by development purpose 2015

Indicator [%] Abs. Deviation

Projects built -34.15 -10887 Projects for lease 6.18 427 Projects for sale 22.81 4112 Projects for own-use -222.21 -15426 Living units built -0.01 -48 Lease 2.26 1009 Sale 30.27 33629 Own-use -348.71 -34686

Dynamics The dynamics of real estate production by purpose is shown using living units as an example (Fig. 5.12). The knees of the curves are in the same year of simulation for all purposes, which indicates that some parcels are not available regardless of purpose. Divergence between the project segments steadily increases as a consequence of proportional sampling from the project pool.

Spatial variation The maps in Fig. 5.13(a) show the effects of the increased share of sale projects spatially. Only two municipalities show

200

5.2. Scenario

Figure 5.12: Developed living units by purpose

0

50000

100000

150000

2000 2010 2020 2030 Year

N e w

ly b

u ilt

li vi

n g

u n

its [ −

]

Purpose

Lease

Sale

Own−use

Run

Baseline

Scenario

more projects for own-use in the scenario. All other municipalities are found to have less development for own-use. For commercial projects, it is the other way around. Most municipalities have more development with the purposes lease or sale.

It is expected that the projects for commercial purpose get built in municipalities with high rent prices since a positive parameter is estimated for the variable (Table 4.37). The opposite is expected for projects built for own-use. However, comparison with the spatial pattern of rent price levels in municipalities (Fig. 5.13(b)) does not allow that conclusion. While this shows that lower prices in the city of Zurich in the scenario deters commercial developments, it still increases in most other municipalities when the rent level also falls. An explanation could be that projects unable to locate in the city of Zurich are ’distributed’ to agglomeration municipalities.

The descriptive statistics in Table 5.7 show positively skewed dis- tributions for purpose specific numbers of projects. This means that a few municipalities get a lot more construction at the expense of other municipalities. Most (interquartile) municipalities experience variation between -74% and 84%. Comparing the standard deviation, this shows that deviations for projects to be sold are the most unevenly distributed over municipalities.

201

Chapter 5. Simulating Zurich’s land development

Figure 5.13: Deviation in the number of projects and rent price levels in 2015

(a) Projects

Projects for lease Projects for sale Projects for own−use

Deviation [%]

50 20 10 5 2.5 1 0 −1 −2.5 −5 −10 −20 −50 −87.5

(b) Rent price level

Deviation [%]

8.2 5 2.5 1 0 −1 −2.5 −5

Data: c© 2013 swisstopo (JD100042)

202

5.2. Scenario

Table 5.7: Descriptive statistics of deviations [%] of projects by purpose over municipalities 2015

Indicator Min Q1 Mean Median Q3 Max Sd

Projects for lease -66.7 -1.3 28.8 14.8 42.1 350.0 58.2 Projects for sale -66.7 17.2 60.6 43.8 83.7 500.0 75.2 Projects for own-use -87.5 -74.1 -65.7 -69.0 -62.3 200.0 24.3

Centrality of developments One hypothesis is that more commercial development happens at central locations. In this analysis, Centrality is measured as parcels with higher than average accessibility. This means that there are two centrality structures related to the considered modes (car, PT). Also, centrality is defined in relation to job locations, i.e. large replacements of jobs would have an effect on the centrality of locations.

The deviations in Table 5.8 show that the number of projects located on parcels with high accessibility is considerably lower (25%) in the scenario. The results by purpose segments suggest that this effect originates from the reduction of projects for own-use. The opposite effect is found for commercial development projects. One can also hypothesise that some of the effect is the consequence of the different preferences since accessibility is positively related to the rent price and rent price is again positively related to commercial development. The effects are stronger regarding car accessibility. The results are plausible, but cannot indicate whether more built space is provided, because it is not accounted for in the size of projects. This can be accounted for by analysing living units instead. The hypothesised effect is only visible for PT accessibility where the results show more living units developed on parcels with high accessibility. The contrary is visible with respect to car accessibility. A possible reason is that PT accessibility generally rises more in the scenario (Fig. 5.18). At this point, it is not possible to clarify the effect due to the redistribution of jobs and network travel times.

Dynamics The increase of the indicator is regular between years of travel simulation4. Discontinuity can be seen after years of travel simulation, which shows that the indicator also depends on variations of the accessibility pattern.

4The travel model is run every five years.

203

Chapter 5. Simulating Zurich’s land development

Table 5.8: Deviations in the number of developments 2015

Indicator [%] Abs. Deviation

High car accessibility Living units -2.2 -7630 Projects -25.0 -3497

Commercial 8.6 675 Own-use -68.4 -4172

High PT accessibility Living units 0.4 1094 Projects -21.3 -1736

Commercial 5.8 294 Own-use -66.2 -2030

Figure 5.14: Time series of projects on parcels with high accessibility by purpose

0

2500

5000

7500

2000 2005 2010 2015 Year

N u

m b

e r

o f p

ro je

ct s

[− ]

Purpose (Mode for acc.)

Commercial (car)

Own−use (car)

Commercial (PT)

Own−use (PT)

Run

Baseline

Scenario

204

5.2. Scenario

Table 5.9: Effects on land consumption in 2015

Indicator [%] Abs. Deviation

Built area -1.36 -840890 Floor capacity -0.39 -358134

Spatial variation According to the definition, parcels with high accessibility are only found in a subset of municipalities, which is why most municipalities have no projects at such locations and consequently zero deviation (Fig. 5.15). The maps visualise where high accessibility can be found and show the importance of the two main cities of Zurich and Winterthur. The trend is not the same across the municipalities with highly accessible parcels.

Land consumption The number of parcels for residential development is calculated for 2015 because of the unplaced projects in later simulation years (Fig. 5.16). At this point, the built area is 1.36 percent smaller in the scenario. This would be evidence for a more resource friendly development with respect to land. In the simulation, the effect is a result of the project characteristics in the development pool. The indicator regarding capacity of allowed floor space also indicates more efficient use with similar reasons.

Dynamics The steepness of the curves in Fig. 5.16 indicates a much higher demand for residential parcels. As a result, suitable parcels for development can no longer be found, which results in zero consumption and unplaced projects before the end of the simulation period. Earlier, it is more the case for the reference than for the scenario. In the reference, more projects are needed to meet demand and thus more parcels are consumed. The remaining parcels are not used due to a incorrectly set filter that excludes residential parcels. Extrapolation of the trend suggests that, without the mistake, parcels would have run out around 2020. This artefact of the simulation has to be corrected. Non-residential development does not show this problem. Plenty of non-residential land is available in the study area for the given demand.

Spatial variation Changed development activity also leads to a different pattern in remaining floor capacities, see Fig. 5.17. Consequently,

205

Chapter 5. Simulating Zurich’s land development

Figure 5.15: Projects with high accessibility by purpose

Commercial (car) Own−use (car)

Commercial (PT) Own−use (PT)

Deviation [%]

100 50 20 10 5 2.5 1 0 −1 −2.5 −5 −10 −20 −50 −100

Data: c© 2013 swisstopo (JD100042)

206

5.2. Scenario

Figure 5.16: Parcels for development over time according to main cate- gories of use

0

10000

20000

30000

2000 2010 2020 2030 Year

[− ]

Indicator

Remaining residential parcels

Remaining non−residential parcels

Unplaced projects

Run

Baseline

Scenario

there is no visible pattern in the spatial distribution of deviations for residential or non-residential zones.

Compactness The compactness of simulated spatial development is measured with travel-related indicators. The simulation shows less com- pactness with 1.86% higher travelled distance (Table 5.10). The travel time indicator has a higher increase: 2.69%. Analysis per mode shows more PT use in the scenario, most clearly in the respective mode shares. The effect in travel time and travelled distance thus originate to some extent in the agent’s mode choice. The absolute deviation is the same, but the car mode is dominant with a share of more than 95%. The transport model is thus badly calibrated towards mode share. The current model split reported in the micro-census is a car share of 58% and a PT share of 32% (Hofer, 2012). Both car and PT accessibility are higher in the scenario, 1.45 percent and 0.08 percent respectively.

Dynamics The time series of the overall accessibility index shows steps because the travel simulation is run only every five years (Fig. 5.18). After each travel model run, accessibility increases, which is a result of having more jobs in the simulated area. The effect is stronger for PT,

207

Chapter 5. Simulating Zurich’s land development

Figure 5.17: Floor capacities by zoning 2015

Total Residential Non−residential Deviation [%]

15.2 10 5 2.5 1 0 −1 −2.5 −5 −10

Data: c© 2013 swisstopo (JD100042)

Table 5.10: Effects on settlement compactness in 2015

Indicator [%] Abs. Deviation

Person meter travelled 1.86 2562174 Person meter travelled by car 1.58 2104570 Person meter travelled by PT 10.03 457604 Person minutes travelled 2.69 4001 Person minutes travelled by car 2.16 2973 Person minutes travelled by PT 9.36 1028 Mode share car -0.26 -0.0025 Mode share PT 8.15 0.0025 Car accessibility 0.08 0.0009 PT accessibility 1.45 0.0439

208

5.2. Scenario

Figure 5.18: Mode specific accessibility indexes over time

1.0

1.5

2.0

2.5

3.0

2000 2005 2010 2015 Year

A cc

e ss

ib ili

ty [ −

]

Run

Baseline

Scenario

Mode

Car

Public transport

which means that public transport users would generally profit more.

Spatial variation Accessibility indexes change for each municipal- ity (Fig. 5.19). The municipalities around Winterthur show higher gains than those around Zurich in regard to both modes (car and PT). The two major cities also profit in the scenario. It is again difficult to find the reasons for the difference.

Rent prices The only simulated price at this point is rent price. The scenario shows a lower mean for rent price, mainly due to lower maximum prices (Table 5.11). For the maximum price in the canton, the relative deviation is almost 18 percent. The minimum price is higher by 7.4 percent. The smaller standard deviation shows reduced price spread in the scenario.

Dynamics The divergence of the maximum price seems to be related to the accessibility update in 2010 (Fig. 5.20). While the mean rent price of approximately 1000 Swiss Francs is reasonable, a minimum rent price about 16 Swiss Francs is not. A maximum of around 13,000 Swiss Francs is plausible as well.

209

Chapter 5. Simulating Zurich’s land development

Figure 5.19: Accessibility deviations by mode

Car accessibility PT accessibility

Deviation [%]

50 20 10 5 2.5 1 0 −1 −2.5 −5 −10 −20 −50 −99.4

Data: c© 2013 swisstopo (JD100042)

Table 5.11: Deviations in rent price statistics 2015

Indicator [%] Abs. Deviation

Minimum 7.4 1.3 Mean -0.9 -9.6 Maximum -17.8 -2277.5 Standard deviation -5.8 -26.3

Spatial variation Simulation results show rent price increases in few peripheral municipalities (Fig. 5.13(b)). In most municipalities, the rent price level decreases. Model estimates suggest that the effects are most likely related to a different distribution of jobs in gastronomy, the size of created living units and population density (Table A.9). The scenario assumptions modify the pool of living units from which is sampled. In the scenario pool, the average size of a living unit is smaller, which is a reason for lower rent price estimates.

210

5.2. Scenario

Figure 5.20: Rent price statistics over time

0

5000

10000

2000 2005 2010 2015 Year

R e

n t p

ri ce

[ C

H F

]

Statistic

Minimum

Mean

Maximum

Standard deviation

Run

Baseline

Scenario

5.2.3 Conclusions from simulation

The simulation experiment tried to reveal the effects of a consolidated real estate industry, i.e. an industry with more commercial developers. It is shown that questions regarding the development of a real estate industry can be investigated with microsimulation models of transport and land use. The simulation is, however, on a proof-of-concept level. The strength lies in the richness of information produced. The effects can be analysed in their spatio-temporal dynamics on various geographical units of analysis. Visualisation is essential to get an overview. This work uses maps and plots of time series. Distribution statistics support the visual presentation. Identifying causalities requires detailed analysis and is time-consuming.

The simulation is flawed in many aspects, which is why it is not possible to draw conclusions regarding the macro-level effects from the simulation. Insufficiencies include: • The implemented real estate development model only represents

green field development. Events such as replacement of buildings, renovations and demolitions should also be represented.

• Land use regulations are only represented on a draft level. While zoning type and allowed densities are important, in reality there are additional regulations.

211

Chapter 5. Simulating Zurich’s land development

• The project structure needs to be implemented correctly, i.e. the project entity should be related to buildings, which are again related to units of use. This would also allow accommodating mixed-use buildings. Indicators such as the number of buildings are biased in the current structure.

• Calibration should be improved. Following calibration approaches for large-scale dynamic transport microsimulation models (Flöt- teröd et al., 2012) are a possible way forward.

A model that implements an appraisal norm could improve the simula- tion, as suggested by (Foti and Waddell, 2014, p. 5). The appraisal-based approach would allow modelling provided quantities endogenously, based on market prices. Such an approach allows a valuation of built space quality on parcels considered for development. More information would be needed since the appraisal norm is more detailed than the assessment presented here. Notably, it would allow the consideration of signals from capital markets, which would add an important influence that has been neglected so far.

Availability of parcels is problematic when locating projects in the simulation since they need larger parcels and thus possible options are used up sooner. A model that covers land development by allowing new parcel layouts could be a solution.

212

Chapter 6

Conclusion The conclusions presented in this chapter first address the hypotheses presented in the introduction and then discusses their outcome. Conclu- sions regarding the methodology are drawn next. Part 3 lays out possible research opportunities for the future.

6.1 Verification of hypotheses and findings The main hypotheses presented at the beginning of this research were that the characteristics of the real estate developers themselves would influence their decisions and, consequently, spatial development. The following set of hypotheses, set out at the beginning, is discussed in the following.

1. Micro level (a) There are behavioural differences among real estate develop-

ers. (b) The choice of a real estate developer for a development site de-

pends on the characteristics of the developer. The developer’s resources, such as property, knowledge and money, influences his valuation and thus the choice for a development option.

(c) Heterogeneity in developers’ decision-making can be mea- sured by estimating location choice models for specific devel- oper types.

(d) Specialised professional developers build in central (highly accessible) places.

2. Macro level (a) To simulate the development process more accurately, differ-

ent developer types need to be considered. (b) The consolidation of a real estate industry (having more profes-

sional developers) leads to more efficient spatial development,

Chapter 6. Conclusion

e.g. less land consumption or less energy use in the transport sector.

This dissertation analysed location choices of real estate developers by conducting in-depth personal interviews and discrete location choice model estimation (micro level) for verification of these hypotheses. The model estimations were then used to simulate a developer type specific scenario with a land use transport interaction (LUTI) model of the Canton of Zurich, Switzerland to assess the effects on spatial development (macro level).

It is found to be meaningful to define the developer as main decision maker. Either he himself is the owner of a property or he is the owners’ representative. With this definition, a variety of developers contributes to the building stock evolution.

Evidence could be found from both the qualitative and quantitative methods for the first two hypotheses on the micro level (1a, 1b). The qual- itative analysis found development purpose and level of professionalism as two attributes that show differences in developers’ decision-making processes. For the development purpose, different decision-making crite- ria have been reported among developers. Developers for self-use projects are more concerned with the long-term development of the location and land prices, whereas sales-oriented developers are more interested in short-term price trends and absorption rates in real estate markets. The use of different evaluation methods is found to vary according to the level of professionalism. While professional developers do advanced analysis, e.g. with geographic information systems (GISs), non-professional de- velopers rely on ad hoc approaches and make more use of local market knowledge. This is related to the situation that local market knowledge is harder to maintain for professional developers because their activity sphere is reported to be larger. Also, the search space for acquisition is found to be larger for professional developers. In addition, professional developers who are well known can profit from offers, i.e. owners of property approach the development experts to sell them a lot or to buy development services.

An analysis of tasks fulfilled during the development process showed instances of development service providers and promoters within the sub-population of selling developers. Promoters take the full risk of development by buying the property, which they improve to sell later. Ownership of land is reported to be of major importance since it is the pri- mary resource for a developer. In some cases, a land bank is available that predetermines location choice to a large extent and helps get favourable loans.

214

6.1. Verification of hypotheses and findings

The estimation of location choice models with deterministic basic interactions showed significant differences in parameter estimates across developer types, thus giving evidence that developer characteristics influ- ence development events. However, significant differences can only be found in the four attributes among those tested. These attributes are ac- cessibility car, new neighbouring buildings, accessibility public transport (PT) and the share of recreation area in traffic analysis zone (TAZ).

Hypothesis 1c is verified on the first level. The estimated parameters are developer-type-specific and the difference can be tested for signifi- cance. Deterministic basic interaction is applicable to the sample, while advanced model forms, such as latent class models or mixed logit models did not have enough data to be supported. These are still favourable to apply if better data on developers is at hand.

The insights into how much the evaluation of considered attributes varies, have to be read with care due to endogeneity. A concluding statement for hypothesis 1d is therefore not possible. It is problematic to drop the price variable, since it is central from a theoretical point of view. Incorrect sign and insignificance hint at omitted variables and inaccurate price measurement.

Further model estimations confirm that purpose is a meaningful at- tribute for developer discrimination. When the developers are categorised according to the purposes of sale, lease and own-use, the estimation re- sults showed expected signs. Differentiation according to purpose seems to separate the developers with commercial interest more clearly from the more consumer-like developers (self-providers). In connection to location choice, it is found that self-providing developers are discouraged by higher rent prices, whereas commercial developers are attracted by them. Higher taxes (Eigenmietwert) due to higher rent price levels at a location can be a reason for self-providers to dislike high rent price levels. Another explanation can be that self-providing developers are outbid by commercial developers and thus end up with parcels in low price areas.

A good fit of the project to zoning constraints is the second vari- able with considerable impact on choice probability. The projects of all developer types exploit allowed volumes on the parcels. Estimation results further suggest that location factors add some explanation for self-providing developers, whereas in the case of commercial developers, the reflection of location factors in the price is more relevant. Location factors, such as lake view, sunshine exposure and recreation area are appreciated, but less relevant compared to price and zoning constraints. An interpretation is that commercial developers are more profit oriented. Knowing the developer’s purpose allows a more specific calculation of

215

Chapter 6. Conclusion

the property. The idea of heterogeneous developers can thus be integrated in appraisal-based approaches.

In the current state of the simulation, the consideration of developer types is not found to improve simulation results. This can be partly con- cluded from the estimation results, which are not achieving a better fit to the data than models without developer type consideration. The influ- ence of developer heterogeneity is hard to isolate from other influences with the available data and is probably less influential than project type and market prices. Confirmation of hypothesis 2a is not possible in that general formulation. It depends on the problem to be investigated in the future whether developer heterogeneity can be ignored or should be taken seriously. Similarly, it might be advisable for practical reasons to work on the basis of single buildings as observations. These observations of single buildings are theoretically less adequate than entire projects, but more likely to be available and would probably involve less preparation effort.

While the simulation could prove the concept of use in spatial policy assessments, it is not mature enough to actually quantify the expected effects of a consolidated real estate industry (2b). Scenario effects, such as reduced land consumption or increased travel distances, are dominated by artefacts, such as vanishing development opportunities or a badly calibrated mode choice model.

The location choice models show negative parameters for accessibility to employment, which is unexpected. One possible reason is that accessi- bility has been defined using inadequate points of interest. More specific and detailed accessibility would probably give better results.

6.2 Discussion of approach

There is something inherently contradictory in the approach that tries to simplify something that needs to be complex. To find the right amount of complexity is difficult, as is tracking down causalities within the simula- tion. It should be carefully evaluated if a certain problem actually needs to be targeted with such methods if the methods are resource intensive.

The qualitative information might be most useful to inform data col- lection for model estimation. If the data is given, as in this case, modelling is limited to what the data allows, e.g. if real estate developers point at parcel availability as being highly relevant for their location choice, then researchers should ideally know where these parcels are and be able to survey them.

There is a lot of competition in the real estate market in the Canton

216

6.2. Discussion of approach

of Zurich, and it is questionable if the real estate developers really have a choice regarding sites for development. In the interviews as well, state- ments were made that ’the options available are just taken’. UrbanSim also has a model that implements developer assessments of parcels with a pro forma type calculation (Foti and Waddell, 2014, p. 5). This allows evaluating newly generated development templates on a parcel considered. Replacement of buildings can be handled with such an approach and the connection to financial markets can be integrated by considering current and estimated interest rates. The model was not chosen for this study be- cause it does not allow direct investigation of location choice preferences. To advance the implementation of the land use transport interaction model of the Canton of Zurich, implementing Swiss appraisal norms following the example of the pro forma based approach is suggested.

The availability of data determines the quality of model operationalisa- tion to a large extent. Missing data cannot always be collected because of privacy issues or budget constraints. Therefore, it is even more important to fully exploit existing data. Some potential lies in the combination or integration of different data sources to reveal new connections. The com- bination of datasets also allows for cross-checks and thus the assessment of data quality.

Calibration methods for LUTI models were found to be quite ad hoc. The manual adjustment of parameters is a substantial task given computation times of several days for one run. Thus, a better method for that task is needed. Adopting the method presented by Flötteröd et al. (2012) to LUTI models can be a way forward.

The data produced with simulation is large. Appropriate methods and tools are necessary for its assessment. Visualisation certainly is a way forward, but appropriate statistics and indicators would be helpful to ’connect the dots’ to form a meaningful picture. This becomes especially challenging for holistic concepts, such as sustainability, since a lot of aspects need to be controlled. A set of appropriate indicators is still needed in the context of LUTI simulations.

What Lowry (1964) found for gravity models still holds for state-of- the-art LUTI models.

´´The statistical regularities on which a gravity model is based represent the outcome of myriad forces operating in an un- specified technological and institutional environment. We can be fairly sure that both the environment and the forces at play are subject to change, but the gravity principle offers few clues as to the impact of a specific change on the param- eters of the model. In other words, if fitted to the current

217

Chapter 6. Conclusion

environment, the model is subject to obsolescence at an un- known rate; and if used to test the impact of radically new public policies (or major changes in transport technologies, or changing standards of living), the parameters fitted from current data may be quite irrelevant.”(Lowry, 1964, p. 22)

The shift of preferences over time may be analysed by deploying appro- priate choice models and panel data. Repeated observation of the same choices over time in a monitoring process would then make preference shifts detectable.

6.3 Suggested further research

More information on developers The population of real estate devel- opers remains unknown to a large extent. More data on these actors is needed. One way forward is to enrich observations on development projects with information in a central firm register (central firm index (ZEFIX)). However, more important are better land price models on the basis of transaction data to make evaluations. This data exists but is diffi- cult to obtain due to privacy concerns. It would be especially interesting to relate the transactions with development activity to assess pre- and post-development ownership durations.

The influence of the ownership structure is interesting and relevant from a theoretical point of view as argued by (McNamara, 1983). While models of portfolio decision-making were envisaged, ownership could not be investigated with the data available. Land register data could be used to research this aspect. Since ever more land registers are run on digital systems, opportunities to do so get better from year to year. Further, one could test the available categorisations of developer types in the federal building and housing register (GWR) project data once enough observations are available. These categories name the legal status of the developers, which could, for example, be used to estimate a model for cooperatives.

Construction model In the simulation, a decision of a developer leads to immediate availability of the respective projects. This is unrealistic since the projects need time to be constructed. A regression model of construction time could be used to improve that point. Information in the DOCUMEDIA dataset on approval and construction period could be used.

218

6.3. Suggested further research

Simulation with sampling The transport model needs refinements. As evaluation has shown, simulating a 10% sample is not enough to get sufficient coverage in space (subsection 5.2.2). Either a 100% scenario is simulated or a regression model assigns travel indicators. Simulation of 100% populations is still too expensive (computation time and adequate machines). One could regress the travel indicators to available information on the land use side using the 10% sample.

Activities Currently, the only trip purpose considered is work. This should be improved by adding further activities such as leisure travel, de- liveries and shopping, which account for 42%, 16% and 12%, respectively, of daily travel distance (Hofer, 2012). Work currently accounts for 30%, which means that it is not even the highest share. Also, subcategories of these traditional trip purposes could be investigated, e.g. travel during lunch period (Pendyala et al., 1991, p. 403). Destination choice should also be included in the transport model (Horni, 2013). Destination choice comprises the ’within the day’ location choices that are frequent (such as shopping, eating or leisure). This more detailed destination choice could feed back to real estate models, allowing to consider the needs for restructuring of the building stock more accurately. The description of units could be analogous to living units, but would further comprise work units, restaurant units or leisure units. An example for shopping units is given by Horni (2013, p. 71).

Integration of travel model To be able to learn more about the speed of changes, it is important to update transport system conditions all year. Even this may not be useful and approaches of continuous simulation, such as presented by Märki (2014), might be a way forward.

Modelling complex systems Constructing complex phenomena by putting together understandable pieces is a tempting thought in favour of mi- crosimulation. A difficulty in practical application is to bring all models to a comparable performance level. When is the transport model sufficiently calibrated in regard to the rent price model? Which of the sub-models contributes the most uncertainty in the results? How do expenses balance with benefits and insight? Such questions remain unanswered to a large extent, in the realm of transport modelling (Hartgen, 2013) and LUTI modelling as well. First attempts have been shown by Ševčíková et al. (2007) and have also been applied to practice (Puget Sound Regional Council, 2013). More research is still needed in this direction.

219

Chapter 6. Conclusion

Usability of LUTI UrbanSim has recently been reimplemented as doc- umented by Foti and Waddell (2014). The authors explain that the new implementation is done with a minimum of self-written code replacing functionality with popular python libraries. Highlighted libraries include the new statistical library pandas, JSON and StatsModels. Pandas pro- vides most tools available in standard statistical software such as R or SPSS. Configuration is now done via JSON, which replaces the XML structure. A mechanism to handle very large choice sets, such as in ur- ban location choice models, has been added to the StatsModels package functionality. The integration of a transport model is on the development agenda. Otherwise, StatsModels provides the necessary tools. Together with the visualisation and spatial analysis functionality, it is a powerful toolbox for urban analysis and simulation. As a next step, it is suggested that the existing simulation in this thesis should be migrated to this new version of UrbanSim.

The dissertation revealed the major potential of LUTI models for understanding complex problems and allowing better-informed decisions. However, there remains a lot to be done to exploit this potential in practice. The effort is substantial, but feasible (Hurtubia et al., forthcoming). One remaining element is to find beneficial use cases in an urban management process. An important prerequisite is improved data availability.

220

Bibliography Abelson, R. P. (1968) Simulation of social behavior, in Handbook of

social psychology, vol. 2, 274–356, Addison-Wesley, Reading.

Abraham, J. E. and J. D. Hunt (2007) Random utility location, produc- tion, and exchange choice; additive logit model; and spatial choice microsimulations, Transportation Research Record, 2003, 1–6, January 2007.

Abraham, J. E. and J. D. Ortuzar (1999) Policy analysis using the sacra- mento MEPLAN land use-transportation interaction model, Transporta- tion Research Record, 1685, 199–208, January 1999.

Abraham, J. E., T. Weidner, J. Gliebe, C. Willison and J. D. Hunt (2005) Three methods for synthesizing base-year built form for integrated land use-transport models, Transportation Research Record, 1902, 114–123, January 2005.

Alda, W. and J. Hirschner (2011) Projektentwicklung in der Immobilien- wirtschaft : Grundlagen für die Praxis, Vieweg + Teubner, Wiesbaden.

Alonso, W. (1964) Location and Land Use: Toward a General Theory of Land Rent, Harvard University Press, Cambridge.

Anas, A. and R. J. Arnott (1993) Technological progress in a model of the housing – land cycle, Journal of Urban Economics, 34 (2) 186–206, September 1993.

Anas, A. and Y. Liu (2007) A regional economy, land use, and transporta- tion model (relu-tran c©): Formulation, algorithm design, and testing, Journal of Regional Science, 47 (3) 415–455.

Anderstig, C. and L.-G. Mattsson (1991) An integrated model of resi- dential and employment location in a metropolitan region, Papers in Regional Science, 70 (2) 167–184.

Bibliography

Arentze, T. A. and H. J. P. Timmermans (2007) A multi-agent activity- based model of facility location choice and use, disP, 43 (170) 33–44.

Ashworth, A. (2008) Pre-contract studies: development, economics, ten- dering and estimating, Blackwell, Oxford.

Axhausen, K. W. (2008) Accessibility: long-term perspectives, Journal of Transport and Land Use, 1 (2) 5–22.

Axhausen, K. W. and L. Hurni (eds.) (2005) Zeitkarten Schweiz 1950 - 2000, IVT and IKA, ETH Zürich, Zürich.

Ball, M. (1983) Housing policy and economic power : the political economy of owner occupation, Methuen, London.

Ball, M. (1985) Coming to terms with owner occupation, Capital & Class, 8 (3) 15–44, October 1985.

Ball, M. (1986a) The built environment and the urban question, Environ- ment and Planning D: Society and Space, 4 (4) 447 – 464.

Ball, M. (1986b) Housing analysis: Time for a theoretical refocus?, Hous- ing Studies, 1 (3) 147–166, Juli 1986.

Ball, M. (1998) Institutions in British property research: A review, Urban Studies, 35 (9) 1501–1517, August 1998.

Ball, M. (2002) Cultural explanation of regional property markets: A critique, Urban Studies, 39 (8) 1453–1469, Juli 2002.

Ball, M. (2003) Markets and the structure of the housebuilding industry: An international perspective, Urban Studies, 40 (5-6) 897 –916, May 2003.

Ball, M. (2006) Markets & Institutions in Real Estate & Construction, Wiley-Blackwell, Oxford.

Balmer, M. (2007) Travel demand modeling for multi-agent traffic simu- lations: Algorithms and systems, Ph.D. Thesis, ETH Zurich, Zurich, May 2007.

Barra, T. l. d., B. Pérez and N. Vera (1984) TRANUS-j: putting large models into small computers, Environment and Planning B: Planning and Design, 11 (1) 87 – 101.

222

Bibliography

Batty, M. (2007a) Cities and Complexity: Understanding Cities with Cellular Automata, Agent-Based Models, and Fractals, MIT Press, Cambridge.

Batty, M. (2007b) Complexity in city systems: Understanding, evolu- tion, and design, Technical Report, 117, Centre for Advanced Spatial Analysis (CASA), University College London (UCL), London.

Batty, M. (2009) Urban modeling, in R. Kitchin and N. Thrift (eds.) International Encyclopedia of Human Geography, 51–58, Elsevier, Oxford.

Batty, M. and P. Longley (1994) Fractal cities: a geometry of form and function, Academic Press Professional, Inc., San Diego.

Beckmann, K., U. Brüggemann, J. Gräfe, F. Huber, H. Meiners, P. Mieth, R. Moeckel, H. Mühlhans, H. Schaub, R. Schrader, C. Schürmann, B. Schwarze, K. Spiekermann, D. Strauch, M. Spahn, P. Wagner and M. Wegener (2007) ILUMASS: Integrated land-use modelling and transport system simulation, Technical Report, Deutsches Zentrum für Luft- und Raumfahrt (DLR), Berlin.

Ben-Akiva, M. E. (1973) Structure of passenger travel demand models, Ph.D. Thesis, Massachusetts Institute of Technology, Cambridge.

Benenson, I. and P. M. Torrens (2004) Geosimulation : Automata-based Modelling of Urban Phenomena, John Wiley & Sons, Hoboken.

Bernardin, V., F. Koppelman and D. Boyce (2009) Enhanced destination choice models incorporating agglomeration related to trip chaining while controlling for spatial competition, Transportation Research Record, 2132, 143–151, December 2009.

Berndt, E. K., B. H. Hall and R. E. Hall (1974) Estimation and inference in nonlinear structural models, in S. V. Berg (ed.) Annals of Economic and Social Measurement, vol. 3/4, 653–665, National Bureau of Economic Research, New York.

Bierlaire, M. (2012) Biogeme, http://biogeme.epfl.ch/.

Bierlaire, M. and M. Fetiarison (2003) BIOGEME: A free package for the estimation of discrete choice models, paper presented at the 3rd Swiss Transportation Research Conference, Ascona, March 2003.

223

Bibliography

Bodenmann, B. R. (2011) Location choice of firmswith special emphasis on spatial accessibility, Ph.D. Thesis, ETH Zurich, Zurich.

Borning, A., H. Ševčíková and P. Waddell (2008) A domain-specific language for urban simulation variables, in Proceedings of the 9th Inter- national Digital Government Research Conference, 207–215, Digital Government Society of North America, Montreal, May 2008.

Bosredon, M., A. Dobson, D. Simmonds, P. Minta, T. Simpson, K. An- drade, H. Gillies and K. Lumsden (2009) Transport/economic/land-use model of Scotland: Land-use modeling with DELTA, paper presented at the 11th International Conference on Computers in Urban Plannind and Urban Management (CUPUM), Hong Kong, June 2009.

Boyce, D. and H. Bar–Gera (2003) Validation of multiclass urban travel forecasting models combining origin–destination, mode, and route choices, Journal of Regional Science, 43 (3) 517–540.

Boyce, D. and Y.-F. Zhang (1997) Calibrating combined model of trip dis- tribution, modal split, and traffic assignment, Transportation Research Record, 1607, 1–5, January 1997.

Briassoulis, H. (2000) Analysis of land use change: Theoretical and mod- eling approaches, in S. Loveridge (ed.) The Web Book of Regional Science, Regional Research Institute, West Virginia University, Mor- gantown.

Bröhl, A., A. Lüscher, P. Meier, M. Regi, M. Salvi, J. Schellenberg and M. Waeber (2011) Immobilien aktuell, Technical Report, 1, Zürcher Kantonalbank, Zürich, May 2011.

Bundesversammlung der Schweizerischen Eidgenossenschaft (1980) Bundesgesetz über die Raumplanung, January 1980, http: //www.admin.ch/opc/de/classified-compilation/ 19790171/index.html.

Bundesversammlung der Schweizerischen Eidgenossenschaft (1983) Bundesgesetz über den Erwerb von Grundstücken durch Personen im Ausland, December 1983, http://www.admin.ch/opc/de/ classified-compilation/19830373/index.html.

Caindec, E. K. and P. Prastacos (1995) A description of POLIS. The projective optimization land use information system, Technical Report, 95-1, Association of Bay Area Governments, Oakland.

224

Bibliography

Cardell, N. S. and F. C. Dunbar (1980) Measuring the societal impacts of automobile downsizing, Transportation Research Part A: General, 14 (5–6) 423–434, October 1980.

Cascetta, E. and A. Papola (2001) Random utility models with implicit availability/perception of choice alternatives for the simulation of travel demand, Transportation Research Part C: Emerging Technologies, 9 (4) 249–263, August 2001.

Cascetta, E. and A. Papola (2009a) Dominance among alternatives in random utility models, Transportation Research Part A: Policy and Practice, 43 (2) 170–179, February 2009.

Cascetta, E. and A. Papola (2009b) Dominance among alternatives in random utility models, Transportation Research Part A: Policy and Practice, 43 (2) 170–179.

Chang, J. S. (2006) Models of the relationship between transport and land-use: A review, Transport Reviews, 26 (3) 325–350.

Chorus, C. G., T. A. Arentze and H. J. Timmermans (2008) A random regret-minimization model of travel choice, Transportation Research Part B: Methodological, 42 (1) 1–18, January 2008.

Christaller, W. (1933) Die zentralen Orte in Süddeutschland: eine ökonomisch-geographische Untersuchung über die Gesetzmässigkeit der Verbreitung und Entwicklung der Siedlungen mit städtischen Funk- tionen, Wissenschaftliche Buchgesellschaft, Jena.

Cifarelli, D. M. and E. Regazzini (1987) On a general definition of con- centration function, Sankhyā: The Indian Journal of Statistics, Series B (1960-2002), 49 (3) 307–319, December 1987.

Coiacetto, E. (2001) Diversity in real estate developer behaviour: A case for research, Urban Policy and Research, 19 (1) 43–59.

Coiacetto, E. (2009) Industry structure in real estate development: Is city building competitive?, Urban Policy and Research, 27 (2) 117 – 135.

Coles, L. (2012) Projektentwickler als Stadtgestalter : Typologisierung von Wertschöpfungsstrategien im Spannungsfeld von Stadtproduktion, Bauqualität und wirtschaftlichem Interesse, vol. 14, VDG, Weimar.

Cramer, D. (2008) Entwicklung einer theoretisch optimalen Strategie für die Bearbeitung von Immobilien-Projektetentwicklungen in der

225

Bibliography

deutschsprachigen Schweiz untersetzt mit einer Fallstudie, Master the- sis, Center for Urban & Real Estate Management, Zürich.

Cressie, N. a. C. and H. J. Whitford (1986) How to use the two sample t-test, Biometrical Journal, 28 (2) 131–148.

Daly, A. and S. Hess (2013) Simple approaches for random utility mod- elling with panel data, http://www.stephanehess.me.uk/ papers/Daly_Hess_simple_panel_final.pdf.

de la Barra, T. (1989) Integrated land use and transport modelling: deci- sion chains and hierarchies, Cambridge University Press, Cambridge.

de Palma, A., N. Picard and M. de Lapparent (forthcoming) Modeling real estate investment decisions in households, in R. Hurtubia, M. Bierlaire, P. Waddell and A. de Palma (eds.) Handbook on integrated transport and land use modeling for sustainable cities, EPFL Press, Lausanne.

Diappi, L. and P. Bolchi (2006) Gentrification waves in the inner-city of Milan, in J. P. Van Leeuwen and H. J. P. Timmermans (eds.) Innovations in Design & Decision Support Systems in Architecture and Urban Planning, 187–201, Springer Netherlands, Dordrecht.

Diaz, J. (1999) The first decade of behavioral research in the discipline of property, Journal of Property Investment & Finance, 17 (4) 326–332, October 1999.

DiPasquale, D. (1999) Why don’t we know more about housing supply?, The Journal of Real Estate Finance and Economics, 18 (1) 9–23.

Docu Media Schweiz GmbH (2013) Home | documedia, December 2013, http://www.docu.ch/.

Domencich, T. A. and D. McFadden (1975) Urban Travel Demand: A Behavioral Analysis, North-Holland, Amsterdam.

Dong, H. and J. Gliebe (2010) Exploring the taste heterogeneity in home developers’ location choice, paper presented at the Proceedings of the 57th Annual Northamerican Meetings of the Regional Science Association International, Denver, November 2010.

Dong, H. and J. Gliebe (2011) Forecasting location of new housing in integrated models of land use, Transportation Research Record, 2255, 79–90, December 2011.

226

Bibliography

Dubin, J. A. and D. L. McFadden (1984) An econometric analysis of residential electric appliance holdings and consumption, Econometrica, 52 (2) 345–362, March 1984.

Dugundji, E. and J. Walker (2005) Discrete choice with social and spatial network interdependencies: An empirical example using mixed general- ized extreme value models with field and panel effects, Transportation Research Record, 1921, 70–78, January 2005.

Echenique, M. H., A. D. J. Flowerdew, J. D. Hunt, T. R. Mayo, I. J. Skidmore and D. C. Simmonds (1990) The MEPLAN models of bilbao, leeds and dortmund, Transport Reviews, 10 (4) 309–322.

Ettema, D., K. Jong, H. J. P. Timmermans and A. Bakema (2007) Puma: Multi-agent modelling of urban systems, in E. Koomen, J. Stillwell, A. Bakema and H. J. Scholten (eds.) Modelling Land-Use Change, vol. 90, 237–258, Springer Netherlands, Dordrecht.

Farooq, B. (2010) Evolution of urban built space, Dissertation, University of Toronto, Toronto.

Farooq, B. and E. J. Miller (2012) Towards integrated land use and trans- portation: A dynamic disequilibrium based microsimulation framework for built space markets, Transportation Research Part A: Policy and Practice, 46 (7) 1030–1053, August 2012.

Fassbind, F. and Göhner AG (1960) Göhner 1890-1960, Göhner AG, Zürich.

Flötteröd, G. (2009) Cadyts – a free calibration tool for dynamic traffic simulations, paper presented at the 9th Swiss Transportation Research Conference, Ascona, September 2009.

Flötteröd, G., M. Bierlaire and K. Nagel (2011) Bayesian demand cali- bration for dynamic traffic simulations, Transportation Science, 45 (4) 541–561, November 2011.

Flötteröd, G., Y. Chen and K. Nagel (2012) Behavioral calibration and analysis of a large-scale travel microsimulation, Networks and Spatial Economics, 12 (4) 481–502, December 2012.

Form, W. H. (1954) The place of social structure in the determination of land use: Some implications for a theory of urban ecology, Social Forces, 32 (4) 317–323, May 1954.

227

Bibliography

Fotheringham, A. S. (1988) Consumer store choice and choice set defini- tion, Marketing Science, 7 (3) 299–310, June 1988.

Fotheringham, A. S. and A. Curtis (1992) Encoding spatial information: The evidence for hierarchical processing, in A. U. Frank, I. Campari and U. Formentini (eds.) Theories and Methods of Spatio-Temporal Reasoning in Geographic Space, 269–287, Springer, Berlin.

Foti, F. S. and P. Waddell (2014) UrbanSim2: Building a bridge from urban modeling to the open source GeoSpatial community, paper presented at the FOSS4G, Portland, September 2014.

Frei, A. (2012) Networks, geography and travel: Travel between infras- tructure and social structure, Ph.D. Thesis, ETH Zurich, Zurich.

Friedrich, S. (2004) Stadtumbau Wohnen, Vdf Hochschulverlag, Zurich.

Fröhlich, P. (2008) Änderungen der Intensitäten im Arbeitspendlerverkehr von 1970 bis 2000, Ph.D. Thesis, ETH Zurich, Zurich.

Geurs, K. T. and B. van Wee (2004) Accessibility evaluation of land-use and transport strategies: review and research directions, Journal of Transport Geography, 12 (2) 127–140, June 2004.

Gilbert, N. and P. Terna (2000) How to build and use agent-based models in social science, Mind & Society, 1 (1) 57–72, March 2000.

Gläser, J. and G. Laudel (2004) Experteninterviews und qualitative In- haltsanalyse als Instrumente rekonsturierender Untersuchungen, VS Verlag für Sozialwissenschaften, Wiesbaden.

Gmünder, M., N. Braun, V. Killer and B. Vitins (2010) Agglomerations- definition Schweiz, Schlussbericht, IVT, ETH Zurich and B, S, S. Volkswirtschaftliche Beratung, Basel / Zürich, February 2010.

Goodman, L. A. and W. H. Kruskal (1959) Measures of association for cross classifications. II: Further discussion and references, Journal of the American Statistical Association, 54 (285) 123–163, March 1959.

Gore, T. and D. Nicholson (1991) Models of the land-development pro- cess: a critical review, Environment and Planning A, 23 (5) 705 – 730.

Graaskamp, J. A. (1981) Fundamentals of Real Estate Development, Urban Land Institute, Washington D.C.

228

Bibliography

Gregor, B. (2007) Land use scenario DevelopeR: Practical land use model using a stochastic microsimulation framework, Transportation Re- search Record, 2003, 93–102.

Gruber, R., R. Zbinden and W. A. Schmid (2000) Räumliche Effekte von Swissmetro, Modellsimulation der Auswirkungen von Verkehrsinfras- trukturänderungen, Technical Report, F5b, ORL ETH Zurich, Bern.

Guevara, C. A. and M. Ben-Akiva (2006) Endogeneity in residential location choice models, Transportation Research Record, 1977, 60–66.

Guevara, C. A. and M. E. Ben-Akiva (2013) Sampling of alternatives in multivariate extreme value (MEV) models, Transportation Research Part B: Methodological, 48 (2013) 31–52, February 2013.

Guo, J. Y. and C. R. Bhat (2007) Operationalizing the concept of neigh- borhood: Application to residential location choice analysis, Journal of Transport Geography, 15 (1) 31–45, January 2007.

Guy, S. and J. Henneberry (2000) Understanding urban development processes: Integrating the economic and the social in property research, Urban Studies, 37 (13) 2399–2416, December 2000.

Guy, S. and J. Henneberry (2002) Bridging the divide? Complementary perspectives on property, Urban Studies, 39 (8) 1471–1478, Juli 2002.

Haag, G. (1990) Master equations, in C. S. Bertuglia, G. Leonardi and A. G. Wilson (eds.) Urban dynamics: Designing an integrated model, 69–83, Routledge, London.

Haase, R. (2011) Ertragspotenziale – Hedonische Mietpreismodellierun- gen am Beispiel von Büroimmobilien, Dissertation, ETH Zürich, Zürich.

Haider, M. and E. Miller (2004) Modeling location choices of housing builders in the greater toronto, canada, area, Transportation Research Record, 1898, 148–156, January 2004.

Hansen, W. G. (1959) How accessibility shapes land use, Journal of the American Planning Association, 25 (2) 73–76.

Hartgen, D. T. (2013) Hubris or humility? accuracy issues for the next 50 years of travel demand modeling, Transportation, 40 (6) 1133–1157, November 2013.

229

Bibliography

Hartmann, M. (2013) Landpreisentwicklung und Erreichbarkeit, Bachelor thesis, ETH Zürich, Institut für Verkehrsplanung und Transportsysteme, Zürich.

Harvey, D. (1985) The urbanization of capital: studies in the history and theory of capitalist urbanization, Johns Hopkins University Press, Oxford.

Hauri, E. and V. Steiner (2006) Siedlungswesen Schweiz - Raumentwick- lung und Wohnungswesen, Technical Report, Bundesamt für Woh- nungswesen, Bern.

Hausman, J. and D. McFadden (1984) Specification tests for the multino- mial logit model, Econometrica, 52 (5) 1219–1240, September 1984.

Hausman, J. A. and D. A. Wise (1978) A conditional probit model for qualitative choice: Discrete decisions recognizing interdependence and heterogeneous preferences, Econometrica, 46 (2) 403–426, March 1978.

Healey, P. (1991) Models of the development process: A review, Journal of Property Research, 8 (3) 219–238, December 1991.

Healey, P. (1992) An institutional model of the development process, Journal of Property Research, 9 (1) 33–44, March 1992.

Healey, P. (1994) Urban policy and property development: the institu- tional relations of real-estate development in an old industrial region, Environment and Planning A, 26 (2) 177–198.

Heckman, J. J. (1981) Dynamic discrete probability models, in C. F. Manski and D. McFadden (eds.) Structural analysis of discrete data with econometric applications, 114–197, MIT Press, Cambridge.

Heidemann, C. (1992) Methodologie der Regionalplanung: Die erste und einzige Bilderfibel der Regionalplanung, Technical Report, 16, Institut für Regionalwissenschaft, Universität Karlsruhe.

Hensher, D. A. and T. Ton (2002) TRESIS: A transportation, land use and environmental strategy impact simulator for urban areas, Transporta- tion, 29 (4) 439–457, November 2002.

Hess, S., M. E. Ben-Akiva, D. Gopinath and J. L. Walker (2011) Advan- tages of latent class over continuous mixture of logit models, Working

230

Bibliography

Paper, Institute for Transport Studies, Institute for Transport Studies, Leeds, May 2011.

Hess, S., A. Stathopoulos and A. Daly (2012) Allowing for heterogeneous decision rules in discrete choice models: an approach and four case studies, Transportation, 39 (3) 565–591.

Hofer, T. (2012) Mobilität im Kanton Zürich, Technical Report, 10/07, Statistisches Amt des Kantons Zürich, Zürich.

Honda, Y. (1985) Testing the error components model with non-normal disturbances, The Review of Economic Studies, 52 (4) 681–690, October 1985.

Horni, A. (2013) Destination choice modeling of discretionary activities in transport microsimulations, Ph.D. Thesis, ETH Zurich, Zurich.

Hunt, J. D. and J. E. Abraham (2003) Design and application of the PECAS land use modelling system, paper presented at the 8th Comput- ers in Urban Planning and Urban Management Conference, Sendai, May 2003.

Hunt, J. D. and J. E. Abraham (2005) Design and implementation of PECAS: A generalised system for allocating economic production, exchange and consumption quantities, in M. Lee-Gosselin and S. T. Doherty (eds.) Integrated Land-Use and Transportation Models: Be- havioural Foundations, 253–274, Elsevier, Amsterdam.

Hunt, J. D., D. S. Kriger and E. J. Miller (2005) Current operational urban land-use–transport modelling frameworks: A review, Transport Reviews, 25 (3) 329–376.

Hunt, L. M., B. Boots and P. S. Kanaroglou (2004) Spatial choice mod- elling: new opportunities to incorporate space into substitution patterns, Progress in Human Geography, 28 (6) 746–766, December 2004.

Hurtubia, R., M. Bierlaire and F. Martínez (2012) Dynamic microsimu- lation of location choices with a quasi-equilibrium auction approach, paper presented at the 12th Swiss Transport Research Conference, As- cona, May 2012.

Hurtubia, R., M. Bierlaire, P. Waddell and A. de Palma (eds.) (forth- coming) Handbook on integrated transport and land use modeling for sustainable cities, EPFL Press, Lausanne.

231

Bibliography

Iacono, M. and D. Levinson (2008) Review of methods for estimating the economic impact of transportation improvements, paper presented at the 88th Annual Meeting of the Transportation Research Board, Washington D.C., August 2008.

Iacono, M., D. Levinson and A. El-Geneidy (2008) Models of transporta- tion and land use change: A guide to the territory, Journal of Planning Literature, 22 (4) 323–341.

Institute of Transport and Logistics Studies (2009) TRESIS v1.10 - in- stitute of transport and logistics studies - the university of sydney, http://sydney.edu.au/business/itls/tresis.

Isard, W. (1951) Interregional and regional input-output analysis: A model of a space-economy, The Review of Economics and Statistics, 33 (4) 318–328, November 1951.

Isard, W. (1956) Location and space-economy, John Wiley & Sons Inc, Cambridge.

Jackson, M. (1991) Will You Be There, song text, Epic, Los Angeles, November 1991.

Johnson, D. E. (2008) Fundamentals of Land Development: A Real-World Guide to Profitable Large-Scale Development, Wiley, Hoboken.

Kahneman, D. and A. Tversky (1979) Prospect theory: An analysis of decision under risk, Econometrica, 47 (2) 263–291, March 1979.

Kain, J. F. (1985) Housing and Neighborhood Dynamics: A Simulation Study, Harvard University Press, Cambridge.

Kaiser, E. J. (1968) Locational decision factors in a producer model of residental development, Land Economics, 44 (3) 351–362, August 1968.

Kaiser, E. J., R. W. Massie, S. F. Weiss and J. E. Smith (1968) Predicting the behavior of predevelopment landowners on the urban fringe, Journal of the American Institute of Planners, 34 (5) 328–333.

Kaiser, E. J. and S. F. Weiss (1970) Public policy and the residential development process, Journal of the American Institute of Planners, 36 (1) 30–37.

232

Bibliography

Kelly, E. D. (1994) The transportation land-use link, Journal of Planning Literature, 9 (2) 128–145, November 1994.

Kemeny, J., J. Kersloot and P. Thalmann (2005) Non-profit housing influ- encing, leading and dominating the unitary rental market: Three case studies, Housing Studies, 20 (6) 855–872.

Killer, V. (2011) Understanding overlapping functional commuting re- gions with confidence ellipses and social network methods, Technical Report, 714, IVT, ETH Zurich, Zurich.

Killer, V., R. Fuhrer, D. Guth, C. Holz-Rau and K. W. Axhausen (2013) Road accessibility of Germany and Switzerland 1970-2007, Arbeits- berichte Verkehrs- und Raumplanung, 936, IVT, ETH Zurich, Zurich.

Kim, T. J., J. H. Rho and S. Suh (1989) Integrated urban systems modeling: theory and applications, Kluwer Dordrecht.

Kowald, M. (2013) Focussing on leisure travel: The link between spatial mobility, leisure acquaintances and social interactions, Dissertation, ETH Zurich, Zurich.

Kubli, U., A. Lüscher, M. Salvi, P. Schellenbauer, J. Schellenberg, P. Moser, U. Rey and C. Bischoff (2008) Wertvoller Boden - Die Funk- tionsweise des Bodenmarktes im Kanton Zürich, Technical Report, Zürcher Kantonalbank, Statistisches Amt des Kantons Zürich, Zürich, May 2008.

Kuster-Langford, J. (1989) Wohnbaulandpreise im Umland von Zürich: Eine empirische Untersuchung der räumlichen Preisunterschiede unter besonderer Berücksichtigung der Auswirkungen der kommunalen Nutzungsplanung und Baulandpolitik, Geographisches Institut, Abt. Anthropogeographie, Universität Zürich, Zürich.

Landis, J. and M. Zhang (1998a) The second generation of the california urban futures model. part 1: Model logic and theory, Environment and Planning B: Planning and Design, 25 (5) 657–666.

Landis, J. and M. Zhang (1998b) The second generation of the california urban futures model. part 2: Specification and calibration results of the land-use change submodel, Environment and Planning B: Planning and Design, 25 (6) 795–824.

233

Bibliography

Landis, J. D. (1994) The california urban futures model: a new genera- tion of metropolitan simulation models, Environment and Planning B: Planning and Design, 21, 399–420.

Leishman, C. (2009) Spatial change and the structure of urban housing sub-markets, Housing Studies, 24 (5) 563–585.

Lendi, M. (1996) Grundriss einer Theorie der Raumplanung: Einleitung in die raumplanerische Problematik, vdf Hochschulverlag AG, Zurich.

Leontief, W. W. (1944) Output, employment, consumption, and invest- ment, The Quarterly Journal of Economics, 58 (2) 290–314, February 1944.

Löchl, M. (2006) Real estate and land price models for UrbanSim’s greater zurich application, Technical Report, Netzwerk Stadt Landschaft (NSL), ETH Zurich, Zurich.

Löchl, M. (2010) Application of spatial analysis methods for understand- ing geographic variation of prices, Dissertation, ETH Zurich, Zurich.

Louviere, J. J., D. A. Hensher and J. D. Swait (2000) Stated Choice Methods: Anaylsis and Application, Cambridge University Press, Cam- bridge.

Lowry, I. S. (1964) A model of metropolis, Technical Report, RM-4035- RC, RAND Corporation, Santa Monica, August 1964.

Luce, R. D. (1959) Individual Choice Behavior, John Wiley and sons, New York.

Mackett, R. L. (1991) LILT and MEPLAN: a comparative analysis of land-use and transport policies for leeds, Transport Reviews, 11 (2) 131–154.

Malik, K. (2013) Human Development Report 2013. The rise of the South: Human progress in a diverse world., United Nations Development Programme (UNDP), New York.

Manski, C. F. and D. McFadden (1981) Structural analysis of discrete data with econometric applications, MIT Press, Cambridge.

Märki, F. (2014) An Agent-Based Model for Continuous Activity Planning of Multi-Week Scenarios, Ph.D. Thesis, ETH Zurich, Zurich.

234

Bibliography

Marschak, J. (1960) Binary-choice constraints and random utility indica- tors, in Mathematical Methods in the Social Sciences, 1959: Proceed- ings of the First Stanford Symposium, 312–329, Stanford University Press.

Martínez, F. J. (1992) The bid - choice land-use model: an integrated economic framework, Environment and Planning A, 24 (6) 871 – 885.

Martínez, F. J. (1996) MUSSA: Land use model for santiago city, Trans- portation Research Record, 1552, 126–134, June 1996.

Martínez, F. J. (2000) Towards a land use and transport interaction frame- work, in D. A. Hensher and K. Button (eds.) Handbook of Transport Modelling, vol. 1 of Handbooks in Transport, 127–164, Elsevier, Den Haag.

Martínez, F. J. and P. Donoso (2010) The MUSSA II land use auction equilibrium model, in F. Pagliara, J. Preston and D. Simmonds (eds.) Residential Location Choice, 99–113, Springer, Berlin.

Mayring, P. (2002) Einführung in die qualitative Sozialforschung: Eine Anleitung zu qualitativem Denken, Beltz, Weinheim.

McFadden, D. (1974) Conditional logit analysis of qualitative choice behaviour, in P. Zarembka (ed.) Frontiers in Economics, 105–142, Academic Press, New York.

McFadden, D. (1976) The revealed preferences of a government bureau- cracy: Empirical evidence, The Bell Journal of Economics, 7 (1) 55–72, April 1976.

McFadden, D. (1977) Modelling the choice of residential location, Cowles Foundation for Research in Economics, New Haven.

McFadden, D. (1978) Modeling the choice of residential location, in A. Karlqvist, L. Lundqvist, F. Snickars and J. Weibull (eds.) Spatial Interaction Theory and Planning Models, 75–96, North Holland, Ams- terdam.

McFadden, D. (1981) Econometric models of probabilistic choice, in C. Manski and D. McFadden (eds.) Structural Analysis of Discrete Data: With Econometric Applications, 198–272, MIT Press, Cam- bridge.

235

Bibliography

McFadden, D. (2000) Economic choices, December 2000, http://www. nobelprize.org/nobel_prizes/economic-sciences/ laureates/2000/mcfadden-lecture.pdf.

McFadden, D. and K. E. Train (2000) Mixed MNL models for discrete response, Journal of Applied Econometrics, 15 (5) 447–470.

McKeever, J. R. (ed.) (1973) The Community Builders Handbook, Urban Land Institute, Washington D.C.

McNamara, P. F. (1983) Towards a classification of land developers, Urban Law and Policy, 79 (6) 87–94.

Metro Regional Government (2010) Metro: Publications and tech- nical papers, http://www.oregonmetro.gov/index.cfm/ go/by.web/id=24906.

Miller, E. J., J. Douglas Hunt, J. E. Abraham and P. A. Salvini (2004) Microsimulating urban systems, Computers, Environment and Urban Systems, 28 (1–2) 9–44, January 2004.

Miller, H. J. (1999) Measuring space-time accessibility benefits within transportation networks: Basic theory and computational procedures, Geographical Analysis, 31 (1) 1–26.

Miller, J. H. and S. E. Page (2007) Complex Adaptive Systems: An Intro- duction to Computational Models of Social Life, Princeton University Press.

Mills, E. S. (1967) An aggregative model of resource allocation in a metropolitan area, The American Economic Review, 57 (2) 197–210, May 1967.

Miyamoto, K., R. Udomsri, S. Sathyaprasad and F. Ren (1996) A decision support system for integrating land use, transport and environmental planning in developing metropolises, Computers, Environment and Urban Systems, 20 (4-5) 327–338, Juli 1996.

MODELISTICA (2013) TRANUS, http://www.tranus.com/.

Muth, R. F. (1969) Cities and Housing, University of Chicago Press, Chicago.

Nicolai, T. W. and K. Nagel (forthcoming) Integration of agent-based transport and land use models, in R. Hurtubia, M. Bierlaire, P. Waddell

236

Bibliography

and A. de Palma (eds.) Handbook on integrated transport and land use modeling for sustainable cities, EPFL Press, Lausanne.

Orcutt, G. H. (1960) Simulation of economic systems, The American Economic Review, 50 (5) 894–907, December 1960.

Ostrom, T. M. (1988) Computer simulation: The third symbol system, Journal of Experimental Social Psychology, 24 (5) 381–392, September 1988.

Patton, M. Q. (1990) Qualitative evaluation and research methods, Sage Publications, Inc, Thousand Oaks.

Pendyala, R. M., K. G. Goulias and R. Kitamura (1991) Impact of telecom- muting on spatial and temporal patterns of household travel, Trans- portation, 18 (4) 383–409, December 1991.

Picard, N. and C. Antoniou (2011) Econometric guidance, SustainCity Deliverable, 5.1, THEMA, Paris.

Portugali, J. (2000) Self-Organization and the City, Springer, Berlin.

Puget Sound Regional Council (2013) 2013 land use forecast validation report, Technical Report, Puget Sound Regional Council, Seattle, Juli 2013.

Pumain, D. and G. Haag (1991) Urban and regional dynamics – towards an integrated approach, Environment and Planning A, 23 (9) 1301–1313.

Putman, S. (1996) Extending DRAM model: Theory-practice nexus, Transportation Research Record, 1552, 112–119, January 1996.

Putman, S. H. (1974) Preliminary results from an integrated transportation and land use models package, Transportation, 3 (3) 193–224, October 1974.

Putman, S. H. (1975) Urban land use and transportation models: A state- of-the-art summary, Transportation Research, 9 (2-3) 187–202, Juli 1975.

Ratcliffe, J., M. Stubbs and M. Keeping (2004) Urban Planning and Real Estate Development, Taylor & Francis, New York.

Revelt, D. and K. Train (1998) Mixed logit with repeated choices: House- holds’ choices of appliance efficiency level, Review of Economics and Statistics, 80 (4) 647–657, November 1998.

237

Bibliography

Rey, U. (2009) Die Krise schlägt nicht überall durch, Technical Report, 09/09, Statistisches Amt des Kantons Zürich, Zürich.

Rey, U. (2010) Dynamischer Zürcher Wohnungsmarkt, Wohnbautätigkeit und Leerstand 2009/2010, Technical Report, 10/07, Statistisches Amt des Kantons Zürich, Zürich.

Rey, U. (2011) Zentrale Lagen begehrt, Technical Report, 11/12, Statistis- ches Amt des Kantons Zürich, Zürich.

Rey, U. (2013) Neubauten wie vor 40 Jahren, Technical Report, 13/02, Statistisches Amt des Kantons Zürich, Zürich.

Rho, J. H. and T. J. Kim (1989) Solving a three-dimensional urban ac- tivity model of land use intensity and transport congestion, Journal of Regional Science, 29 (4) 595–613, November 1989.

Ricardo, D. (1817) On the Principles of Political Economy, and Taxation, J. Murray, London.

Rosen, S. (1974) Hedonic prices and implicit markets: Product differenti- ation in pure competition, Journal of Political Economy, 82 (1) 34–55, January 1974.

Royal Institute of British Architects (2013) RIBA plan of work 2013, Technical Report, Royal Institute of British Architects, London.

Ruming, K. (2010) Developer typologies in urban renewal in sydney: Recognising the role of informal associations between developers and local government, Urban Policy and Research, 28 (1) 65–83, March 2010.

Rybczynski, W. (2007) Last Harvest: How a Cornfield Became New Daleville, Scribner, New York.

Salvi, M., P. Schellenbauer and H. Schmidt (2004) Preise, Mieten und Renditen, Technical Report, Zürcher Kantonalbank, Zürich.

Salvini, P. and E. J. Miller (2005) ILUTE: An operational prototype of a comprehensive microsimulation model of urban systems, Networks and Spatial Economics, 5 (2) 217–234.

Schalcher, H.-R., Kammer Unabhängiger Bauherrenberater (Schweiz) and Eidgenössische Technische Hochschule (Zürich) (eds.) (2009) Im- mobilienmanagement : Handbuch für Immobilienentwicklung, Bauher- renberatung, Immobilienbewirtschaftung, Zürich : Schulthess, Zürich.

238

Bibliography

Schilling, H. (1973) Kalibrierung von Widerstandsfunktionen, Techni- cal Report, 73/1, Lehrstuhl für Verkehrsingenieurwesen, ETH Zurich, Zurich.

Schirmer, P., C. Zöllig, K. Müller, B. R. Bodenmann and K. W. Axhausen (2011) The Zurich case study of UrbanSim, paper presented at the 51th Congress of the European Regional Science Association, Barcelona, September 2011.

Schirmer, P., C. Zöllig Renner, K. Müller, B. R. Bodenmann and K. W. Axhausen (forthcoming) Transport and land use microssimulation on the canton of zurich using UrbanSim, in R. Hurtubia, M. Bierlaire, P. Waddell and A. de Palma (eds.) Handbook on integrated transport and land use modeling for sustainable cities, EPFL Press, Lausanne.

Schirmer, P. M., M. A. B. van Eggermond and K. W. Axhausen (2013) Measuring location in residential location choice: An empirical study on the canton of zurich, paper presented at the 13th International Conference on Computers in Urban Planning and Urban Management (CUPUM), Utrecht, July 2013.

Schmid, P., L. Trüb, A. Sonderegger and H. C. Daeniker (2007) Mehr als wohnen - Argumente und Fakten, Technical Report, Schweizerischer Verband für Wohnungswesen, Sektion Zürich und Finanzdepartement der Stadt Zürich, Zürich, March 2007.

Schmitz, A. (2004) Residential Development Handbook, Urban Land Institute, Washington, D.C.

Schüssler, R. and P. Thalmann (2005) Was treibt und hemmt den Woh- nungsbau? Ergebnisse einer Bauträger- und Investorenbefragung, Tech- nical Report, 76, Bundesamt für Wohnungswesen (BWO), Grenchen.

Ševčíková, H., A. E. Rafteryb and P. A. Waddell (2007) Assessing un- certainty in urban simulations using bayesian melding, Transportation Research Part B: Methodology, 41 (6) 652–659.

SIA (2001) Leistungsmodell, Standard, SN 508 112, SIA, Zürich.

Simmonds, D. C. (1999) The design of the DELTA land-use modelling package, Environment and planning A, 26 (5) 665–684.

Simmonds, D. C. and O. Feldman (2005) Land-use modelling with DELTA: Update and experience, paper presented at the 9th Computers

239

Bibliography

in Urban Planning and Urban Management Conference, London, June 2005.

Southworth, F. (1995) A technical review of urban land use–transportation models as tools for evaluating vehicle travel reduction strategies, Tech- nical Report, 6881, Oak Ridge National Laboratory, Center for Trans- portation Analysis Energy Division, Oak Ridge, Tennessee, Juli 1995.

Stadt Zürich, Stadtentwicklung (2008) Vermietungspraxis, Mi- etermix und Nachbarschaft bei Stadtzürcher Genossenschaften, August 2008, https://www.stadt-zuerich.ch/ content/dam/stzh/prd/Deutsch/Stadtentwicklung/ Publikationen_und_Broschueren/Stadt-_und_ Quartierentwicklung/Wohnen/Befragung_Genoss_ Vermietpraxis_02_Praesentation_Genoss.pdf.

STASA (2013) STASA - publications, http://www.stasa.de/en/ Veroeffentlichungen.shtml.

Statistics Canada (2009) Guide to the modgen visual interface, Technical Report, v10.1.0, Statistics Canada, Ottawa, May 2009.

Statistics Canada (2011) Modgen (model generator), September 2011, http://www.statcan.gc.ca/microsimulation/ modgen/modgen-eng.htm.

Stover, V. G. and F. J. Koepke (1988) Transportation and land develop- ment, Prentice Hall, Englewood Cliffs.

Strauch, D., R. Moeckel, M. Wegener, J. Gräfe, H. Mühlhans, G. Rind- fuesser and K.-J. Beckmann (2005) Linking transport and land use planning: The microscopic dynamic simulation model ILUMASS, in P. M. Atkinson, G. M. Foody, S. E. Darby and F. Wu (eds.) Geodynam- ics, 295–311, CRC Press, Boca Raton.

Strohm, D. (2012) Entwickeln mit System, Neue Zürcher Zeitung Sonder- beilage Immobilien, June 2012.

Synthicity team (2014a) Synthicity, http://www.synthicity. com/.

Synthicity team (2014b) UrbanSim — urbansim 0.2dev documen- tation, September 2014, http://synthicity.github.io/ urbansim/index.html.

240

Bibliography

Teye-Ali, C., P. Davidson and R. Culley (2013) The comparative efficacy of combining latent class and nested logit (LCNL) to simultaneously account for inter-alternative correlation and taste heterogeneity so as to improve toll route choice forecasts, paper presented at the International Choice Modelling Conference, Sydney, Juli 2013.

Thalmann, P. (2009) Hoch hinaus oder in die Breite?, paper presented at the ETH Forum Wohnungsbau 2009, 33–39, Zürich, April 2009.

Thalmann, P. (2012) Housing market equilibrium (almost) without vacan- cies, Urban Studies, 49 (8) 1643–1658, June 2012.

Thurstone, L. (1927) A law of comparative judgment, Psychological Review, 34 (4) 273–286.

Timmermans, H. J. P. (2007) The saga of integrated land use-transport modeling: How many dreams before we wage up, in K. W. Axhausen (ed.) Moving Through Nets, 219–248, Elsevier, Oxford.

Train, K. E. (2009) Discrete Choice Methods with Simulation, Cambridge University Press, Cambridge.

Tschopp, M., P. Fröhlich and K. W. Axhausen (2005) Accessibility and spatial development in switzerland during the last 50 years, in D. M. Levinson and K. J. Krizek (eds.) Access to Destinations, 361–376, Elsevier, Oxford, December 2005.

Turci, L., S. Pennec, L. Toulemon, A. Bringé, E. Morand and R. Baggio (2012) Demographic model user guide, Technical Report, 4.4, INED, Paris, April 2012.

UrbanSim Developers (2014) UrbanSim WebHome, http://www. urbansim.org/Main/WebHome.

UrbanSimProjekt (2011) The open platform for urban simulation and UrbanSim version 4.3, Technical Report, 4.3, University of California Berkeley and University of Washington, January 2011.

van Wezemael, J. E. (2005) Investieren im Bestand, Ostschweizerische Geographische Gesellschaft, St.Gallen.

Vanegas, C. A., D. G. Aliaga, P. Wonka, P. Müller, P. Waddell and B. Wat- son (2010) Modelling the appearance and behaviour of urban spaces, Computer Graphics Forum, 29 (1) 25–42.

241

Bibliography

Verburg, P. H., P. P. Schot, M. J. Dijst and A. Veldkamp (2004) Land use change modelling: current practice and research priorities, GeoJournal, 61 (4) 309–324, December 2004.

von Thünen, J. H. (1826) Der Isolierte Staat, Verlag von Gustav Fischer, Jena.

Vrtic, M., P. Fröhlich, N. Schüssler, P. Kern, F. Perret, S. Pfisterer, C. Schulze, A. Zimmermann, U. Heidl and K. W. Axhausen (2005) Verkehrsmodell für den öffentlichen Verkehr des Kantons Zürich, Tech- nical Report, Amt für Verkehr des Kantons Zürich, IVT, ETH Zürich, Ernst Basler + Partner AG, Planung Transport Verkehr (PTV), Zürich.

Waddell, P. (2000) A behavioral simulation model for metropolitan policy analysis and planning: residential location and housing market com- ponents of UrbanSim, Environment and Planning B: Planning and Design, 27 (2) 247–264.

Waddell, P. (2002) UrbanSim: Modeling urban development for land use transportation and environmental planning, Journal of the American Planning Association, 68 (3) 297–314.

Waddell, P. (2005) Towards an open platform for urban simulation, May 2005.

Waddell, P. (2011a) Integrated land use and transportation planning and modelling: addressing challenges in research and practice, Transport Reviews, 31 (2) 209–229.

Waddell, P. (2011b) Modelling real estate supply, http://www. urbansim.org/Documentation/Classroom/WebHome.

Waddell, P., H. Ševcıková, D. Socha, E. Miller and K. Nagel (2005) Opus: An open platform for urban simulation, paper presented at the Proceedings of the 9th Computers in urban planning and urban management (CUPUM) conference, University College London, UK.

Waddell, P. and G. F. Ulfarsson (2003) Dynamic simulation of real estate development and land prices within an integrated land use and trans- portation model system, paper presented at the 82nd Annual Meeting of the Transportation Research Board, Washington, Juli 2003.

Waddell, P. and G. F. Ulfarsson (2004) Introduction to urban simula- tion models: Design and development of operational models, in D. A.

242

Bibliography

Hensher, K. L. Button and P. Stopher (eds.) Handbook of Transport Geography and Spatial System, vol. 5, 203–236, Elsevier.

Wagner, P. and M. Wegener (2007) Urban land use, transport and environ- ment models, disP, 43 (3) 45–56.

Walker, J. and M. Ben-Akiva (2002) Generalized random utility model, Mathematical Social Sciences, 43 (3) 303–343, Juli 2002.

Walker, J. L. (2001) Extended discrete choice models: Integrated frame- work, flexible error structures, and latent variables, Ph.D. Thesis, Mas- sachusetts Institute of Technology, Cambridge.

Wallbaum, H., S. Kytzia and S. Kellenberger (2011) Nachhaltig bauen : Lebenszyklus, Systeme, Szenarien, Verantwortung, vdf, Zürich.

Wang, L. (2009) Advances in integrated urban modeling: Microsimulation models of the housing market, real estate development, and workplace choice, Dissertation, University of Washington, Seattle.

Wang, L. and P. Waddell (2013) Disaggregated real estate demand model with price formation for integrated land use and transportation modeling, paper presented at the 92nd Annual Meeting of the Transportation Research Board, January 2013.

Watkins, C. A. (2001) The definition and identification of housing sub- markets, Environment and Planning A, 33 (12) 2235–2254.

Wegener, M. (1995) Current and future land use models, paper presented at the 1st Land Use Model Conference, Dallas, February 1995.

Wegener, M. (2004) Overview of land use transport models, in D. A. Hensher, K. J. Button, K. E. Haynes and P. Stopher (eds.) Handbook of Transport Geography and Spatial System, vol. 5 of Handbooks in Transport, 127–146, Elsevier, Oxford.

Wegener, M. and F. Fürst (1999) Land-use transport interaction: State of the art, Technical Report, 46, Institut für Raumplanung, Universität Dortmund, Dortmund.

Weidner, T., R. Donnelly, J. Freedman, J. E. Abraham and J. D. Hunt (2007) A summary of the Oregon TLUMIP model microsimulation modules, paper presented at the 86th Annual Meeting of the Transporta- tion Research Board, Washington, D.C., January 2007.

243

Bibliography

Weiss, M. A. (1987) The rise of the community builders : the American real estate industry and urban land planning, Columbia University Press, New York.

Weiss, S. F. (1966) Residential developer decisions ; a focused view of the urban growth process, Center for Urban and Regional Studies, Institute for Research in Social Science, University of North Carolina, Chapel Hill.

Widler, K. (2013) Mobilisierung von Nutzungsreserven - Eigentümer als Schlüsselakteure der Innenentwicklung, Master thesis, ETH Zürich, Zürich, August 2013.

Wilson, A. G. (1998) Land-use/transport interaction models: Past and future, Journal of Transport Economics and Policy, 32 (1) 3–26.

Würth, N. and P. Meier (2014) Immobilienbarometer 4. quartal 2013, Technical Report, 12, Zürcher Kantonalbank, Zürich, January 2014.

Ye, X., K. Konduri, R. M. Pendyala, B. Sana and P. Waddell (2009) A methodology to match distributions of both household and person attributes in the generation of synthetic populations, paper presented at the 88th Annual Meeting of the Transportation Research Board, Washington, D.C., January 2009.

Zaborowski, C., H. Froidevaux, W. Ott and M. Arend (2001) Wirkungen und Nutzen der Wohnbauförderung im Kanton Zürich, Technical Re- port, 13/02, Statistisches Amt des Kantons Zürich, Wüest & Partner, econcept AG, Zürich.

Zeebroeck, N. v., B. V. Pottelsberghe and W. Han (2005) Issues in measur- ing the degree of technological specialisation with patent data, Working Papers CEB, 05-016.RS, ULB – Universite Libre de Bruxelles, Brus- sels.

Zhou, B. and K. Kockelman (2008) Microsimulation of residential land development and household location choices: Bidding for land in austin, texas, Transportation Research Record, 2077, 106–112, December 2008.

Zhou, B. B. and K. M. Kockelman (2011) Change in land use through microsimulation of market dynamics, Transportation Research Record, 2255, 125–136, December 2011.

244

Bibliography

Zöllig, C. and K. W. Axhausen (2011) A conceptual, agent-based model of land development for UrbanSim, paper presented at the Proceedings of the 51th ERSA Conference, Barcelona, September 2011.

Zöllig, C. and K. W. Axhausen (2012) Heterogeneity of real estate de- velopers in canton zurich, paper presented at the 12th Swiss Transport Research Conference, Ascona, May 2012.

Zöllig, C., R. Hilber and K. W. Axhausen (2011) Konzeptstudie Flächen- nutzungsmodellierung, Technical Report, IVT, ETH Zürich und Map- puls AG, Zürich.

Zöllig Renner, C. and K. W. Axhausen (2013) Comparing estimation results of land use development models using different data bases available in switzerland, paper presented at the 92nd Annual Meeting of the Transportation Research Board, Washington, D.C., January 2013.

Zöllig Renner, C. and K. W. Axhausen (forthcoming) A real estate devel- opment model with heterogeneous agents, in R. Hurtubia, M. Bierlaire, P. Waddell and A. de Palma (eds.) Handbook on integrated transport and land use modeling for sustainable cities, EPFL Press, Lausanne.

Zöllig Renner, C., T. W. Nicolai and K. Nagel (forthcoming) Agent-based land use transport interaction modeling: state of the art, in R. Hurtubia, M. Bierlaire, P. Waddell and A. de Palma (eds.) Handbook on integrated transport and land use modeling for sustainable cities, EPFL Press, Lausanne.

Zondag, B. (2007) Joint modelling of land-use, transport and economy, Dissertation, Technical University Delft, Delft.

Zürcher Kantonalbank (2011) ZKB - Wohneigentumsindex (ZWEX), August 2011, http://www.zkb.ch/de/center_worlds/ eigenheimcenter/marktinfos/wohneigentumsindex_ zwex_/index.html.

245

Glossary

accessibility "...accessibility is defined as the potential of opportunities for interaction." (Hansen, 1959, p. 71). 34

agent-based modelling Modelling of complex systems via modelling of system components as agents which interact. 21, 90

agglomeration An agglomeration is a connected area of multiple munic- ipalities with at least 20000 inhabitants. Definition by BfS. 39

builder The builder executes the development decision of a developer. These are usually construction firms. 53

calibration Process of adapting a general model to specific circum- stances. 5, 188, 217

complex system Complex systems are characterised by non-ergodicity (long term), phase transition, emergence and universality (Batty, 2007a). 28

development consortium The development consortium is a union of contractors for the realisation of a construction project. 52

endogeneity Endogeniety is the fact that an observed explanatory vari- able is correlated with an unobserved variable. The correlated observed variable picks up the effect of the unobserved variable. The estimated parameter is consequently biased. 104

heteroscedasticity Heteroscedasticity is the characteristic of a set of ran- dom variables that we find different dispersions in sub-populations. 104

impact The impact of a variable is a statistic which describes the in- fluence of the variable on the utility. It is calculated as follows:

I = median( xchoice ) ∗ βx (6.1)

Glossary

Where:

βx : estimated coefficient of variable x x : variable

. 160 infrastructure Infrastructures comprises all installations which facilitate

or protect certain activities. 29

land development Land development denotes actions which alter the landscape. These actions can be physical such as terrain movements or facility building, or legal such as to parcel up an area in lots. 51, 52

land use Land use is defined by the activity the land is used for. Activities and therefore land uses vary from minutes to decades. Here more persistent land uses are of interest. 27, 28, 30

land use development Land use development is defined as the process of land uses rearranging themselves in space. This process is guided by the provision of locations especially prepared for certain activities. e.g. soccer is play on a soccer ground. 52

planning Planning is the creation of plans for action to modify an object to meet defined goals. (Heidemann, 1992). 21

real estate developer Decision maker in respect of one or more develop- ment projects. 52

real estate development Real estate development is the process of creat- ing and modifying real estates. It can be done by adding, altering or replacing built space. 52, 53

sustainability Not using more resources than the eco-socio-economical system can produce. 29

urban modelling The process of identifying appropriate theory, translat- ing this into a mathematical or formal model, developing relevant computer programs, and then confronting the model with data so that it might be calibrated, validated, and verified prior to its use in prediction. (Batty, 2009, p. 51). 40, 188

urban system An urban system is a complex of interacting subsystems of which a major part is created by human decisions. Urban systems are embedded in an ecosystem. 28–30

validation Process to prove that something is correct or works correctly. 189

248

Acronyms

ABM agent-based model. 41, 90 ABS agent-based simulation. 25, 89, 90 ARE ZH Cantonal Office for Spatial Development 1. 19, 152, 182 AvCR average concentration ratio over the considered dimensions of

specialisation.. 131, 132

BfS Swiss Federal Statistical Office 2. 19 BLCM building location choice model. 10, 147, 150, 154, 156, 158 BLP Berry, Levinsohn and Pakes. 104 BTM building transition model. 184, 187

CA cellular automata. 90 CBD central business district. 184 CNL cross nested logit. 94 Cns construction. 231 CR concentration ratio. 130, 131 CRB Swiss centre for construction rationalisation. 10, 117, 222 CUFM California urban futures model. 87 CV coefficient of variation. 190, 191

DCA discrete choice analysis. 24, 25, 89–91, 101, 109, 120, 122, 146 DCM discrete choice modelling. 90, 91, 93, 109 DCT discrete choice theory. 46, 91, 105

egid Federal building identifier 3. 182 ELCM employment location choice model. 11, 49, 184, 185, 187, 231 ERM employment relocation model. 185–187 ETM employment transition model. 49, 185, 187

FACS free agents on a cellular space. 90

GAS geographic automata systems. 90

1Amt für Raumentwicklung 2Bundesamt für Statistik 3Eidgenössischer Gebäude Identifikator

Acronyms

GEV generelised extreme value. 92, 94, 97 GIS geographic information system. 149, 214 GPL general public license. 50 GUA geographical units of analysis. 36, 49–51, 179, 180 GUI graphical user interface. 50 GVZ Building Insurance of the Canton of Zurich 4. 19, 117, 152, 173,

182 GWR federal building and housing register 5. 116, 117, 147–149, 152,

171, 182, 192, 193, 218

HEV house owners association 6. 168–170 HLCM household location choice model. 11, 49, 186–188, 233 Hlt health. 185, 231 HR hotels and restaurants. 185, 231 HRM household relocation model. 185, 187, 188 HTM household transition model. 49 HUDS harvard urban development simulation. 39, 40

IAPRU implicit availability/perception random utility. 103 IIA independence from irrelevant alternatives. 91, 94, 95, 105 IID independent and identically distributed. 93, 96, 99 IIN independence from irrelevant nests. 95

LC latent class. 82, 96, 97, 146, 176, 177 LCC land cover change. 36 LUC land use change. 36 LUDM land use development model. 35 LUSDR land use scenario developer. 40 LUTI land use transport interaction. 3, 7, 19, 22, 23, 25, 35–37, 40, 53,

84, 88, 179, 180, 183, 187, 188, 214, 217, 219, 220

MAS multi agent systems. 41, 47, 90 MATSim Multi-Agent Transport Simulation. 149, 152, 183, 186–188 Mfg manufacturing. 231 MFH multi family housing. 82, 122, 131, 151, 154–158, 162–165 ML mixed logit. 96, 97, 146 MMNL mixed multinomial logit. 92, 96, 102 MNL multinomial logit. 4, 49, 82, 86, 91–96, 99, 103, 105, 146, 150,

164, 186, 187 MNP multinomial probit. 4, 92, 97

4Gebäudeversicherung Kanton Zürich 5Gebäude- und Wohnungsregister 6Hauseigentümerverband

250

Acronyms

NA not available. 118 NL nested logit. 91, 94, 95, 99

O1 developer with one project for own-use. 128, 163, 165 OGEV ordered general extreme value. 94 Om developer with multiple projects for own-use. 128, 131, 145, 163 OPUS open platform for urban simulation. 48, 50, 51

PCL pairwise cross nested logit. 94 PLCM project location choice model. 10, 147, 150, 154, 156, 158, 164,

167, 173, 176, 184, 187 PT public transport. 72, 138, 149, 152–156, 163, 184, 186, 203, 207,

209, 215, 231

RCL random coefficient logit. 96 REDM real estate development model. 10, 49, 152, 180 REPM real estate price model. 11, 49, 50, 185, 187, 230 RIBA Royal Institute of British Architects. 66 RMSE root mean square error. 175 ROI return on investment. 49, 50, 85 RPL random parameter logit. 82 Rtl retail. 231 RUM random utility maximisation. 46, 91–93, 97, 104

SAKZ Zurich Cantonal Statistical Office 7. 19, 115, 152, 168–171, 189, 190, 192, 194, 249

SCGE spatial computational general equilibrium. 39, 47 SD standard deviation. 191 SFH single family housing. 82, 122, 131, 151, 154–158, 161, 162, 164,

165 Smc developer with one or multiple projects developed for sale. 128, 145 SME Small and medium enterprises. 134 SNF Swiss National Science Foundation. 19 SP stated preference. 92 SQL structured query language. 181 Srv service. 185, 231 SVW Swiss Housing Association 8. 19

TAZ traffic analysis zone. 86, 160, 180, 190, 215 TLUMIP transportation and land use model integration project. 40, 48 Trd trade. 231

7Statistisches Amt Kanton Zürich 8Schweizerischer Verband für Wohnungswesen

251

Acronyms

Trn transport. 231

UGM urban growth models. 36

WLCM workplace location choice model. 11, 185, 187, 232

ZEFIX central firm index 9. 218

9Zentraler Firmen Index

252

Appendix A

Appendix

Appendix A. Appendix

A.1 Descriptive statistics of DOCUMEDIA data

A.1.1 Dimensions and format Entities: Construction projects Number of records: 59073 Number of variables: 93 ID variable: objektnr Variable groups: • Address details of developer, planer and engineer • Localisation of construction site • Details on constructed structure

A.1.2 Levels of categorical variables

Table A.1: Building types (01) according to Swiss centre for construction rationalisation (CRB) classification

Level Count Proportion

Behelfswohnungen 39 0.066 Mehrfamilienhäuser 10126 17.161 Terrassenhäuser 232 0.393 Einfamilienhäuser / Villen 25471 43.168 Alterswohnungen und Alterssiedlungen 154 0.261 Alterswohnheime 117 0.198 Kinder- und Jugendheime 30 0.051 Studenten- und Lehrlingswohnheime 8 0.014 Reiheneinfamilienhäuser 245 0.415 Kinderhorte und Kindergärten 316 0.536 Primar- und Sekundarschulen 460 0.780 Berufs- und höhere Fachschulen 83 0.141 Mittelschulen und Gymnasien 9 0.015 Heilpädagogische und Sonderschulen 36 0.061 Hochschulen und Universitäten 49 0.083 Bibliotheken und Staatsarchive 49 0.083 Forschungsinstitute 43 0.073 Industrie und Gewerbe 1 0.002 Lagerhallen 1129 1.913 Mehrgeschossige Lagerbauten 9 0.015 Mechanisierte Lager und Kühllager 31 0.053 Silobauten und Behälter 322 0.546 Verteilzentralen 16 0.027

Continued on next page

254

A.1. Descriptive statistics of DOCUMEDIA data

Level Count Proportion

Industriehallen 15 0.025 Industrielle Produktionsbauten 312 0.529 Betriebs- und Gewerbebauten 1251 2.120 Werkstatt / Atelier / Lager privat 373 0.632 Remise 2199 3.727 Futterlagerräume und Treibhäuser 24 0.041 Stallungen und landwirt. Prod.anlagen 1443 2.446 Tierheime und Veterinärstationen 12 0.020 Tierspitäler 4 0.007 Schlachthöfe 10 0.017 Jauchegrube 61 0.103 Heiz-, Fernwärme- und Kraftwerkbauten 44 0.075 Wasseraufbereitungsanlagen 141 0.239 Kehrichtverb.- und Wiederaufb.anlagen 18 0.031 Tankstellen (auch Gas) 266 0.451 Kunstbauten, Masten und Türme 12 0.020 Wärme- und Kälteverteilanlagen 51 0.086 Elektrische Verteilanlagen 23 0.039 Autogewerbe 179 0.303 Handel und Verwaltung 1 0.002 Ladenbauten 915 1.551 Warenhäuser und Einkaufszentren 63 0.107 Bürobauten, einfache Anforderungen 156 0.264 Bürobauten, erhöhte Anforderungen 1209 2.049 Verwaltungsgeb., Banken, Rechenzentren 92 0.156 Gemeindehäuser 47 0.080 Rathäuser und Regierungsgebäude 2 0.003 Gerichtsgebäude 1 0.002 Polizeigebäude,Untersuchungsgefängnisse 12 0.020 Strafvollzugsanstalten 5 0.008 Arztpraxen und Aerztehäuser 156 0.264 Krankenhäuser 73 0.124 Universitätskliniken 7 0.012 Pflegeheim, Sanatorium, Rehabilitation 107 0.181 Heilbäder und Spezialinstitute 2 0.003 Tagesheime und geschützte Werkstätten 38 0.064 Restaurationsbetriebe 661 1.120 Hotel- und Motelbauten 102 0.173 Kantinen 33 0.056

Continued on next page

255

Appendix A. Appendix

Level Count Proportion

Herbergen und Massenunterkünfte 10 0.017 Raststätten, Cafeterias, Tea-Rooms 101 0.171 Klubhütten 141 0.239 Campinganlagen 3 0.005 Sport-, Turn- und Mehrzweckanlagen 344 0.583 Tribünenbauten und Garderobengebäude 85 0.144 Frei- und Hallenbäder 231 0.391 Reithallen 20 0.034 Bootshäuser 18 0.031 Freizeitzentren und Jugendhäuser 35 0.059 Aussen-, Kinder- und Parkanlagen 413 0.700 Zoolog.-, botan. Gärten, Gewächshäuser 74 0.125 Fitnesscenter/-raum 30 0.051 Einstellgaragen und Parkhäuser 252 0.427 Strassenverkehrsgebäude 2 0.003 Werkhöfe 62 0.105 Busbahnhöfe, Zollanlagen, Wartehallen 20 0.034 Bahnbetriebsbauten, Seilbahnstationen 24 0.041 Flughafenbauten 1 0.002 Postgebäude und Fernmeldegebäude 23 0.039 Garagen/Fertiggaragen 1979 3.354 Zeughäuser 2 0.003 Oeffentliche Zivilschutzanlagen 4 0.007 Zivilschutz-Ausbildungszentren 4 0.007 Feuerwehrgebäude 14 0.024 Militäranlagen, milit. Schutzanlagen 4 0.007 Kirchen und Kapellen 90 0.153 Kirchgemeindehäuser 76 0.129 Friedhofanlagen 33 0.056 Abdankungshallen 6 0.010 Krematorien 1 0.002 Klöster 2 0.003 Ausstellungsbauten 7 0.012 Museen und Kunstgalerien 29 0.049 Wohlfahrts-, Klubhäuser, Kulturzentren 18 0.031 Konzertbauten und Theaterbauten 11 0.019 Musikpavillons 2 0.003 Kino-, Diskothek- und Saalbauten 58 0.098 Kongresshäuser und Festhallen 2 0.003

Continued on next page

256

A.1. Descriptive statistics of DOCUMEDIA data

Table A.2: Levels of offer type offertcode

Level Count Proportion

1 keine Angabe 3791 6.4 2 Offerte an Bauherr 21210 35.9 3 Offerte an Planer 33414 56.6 4 Offerte an Ing. 658 1.1

Data: DOCUMEDIA

Level Count Proportion

Radio-, Fernseh- und Filmstudios 10 0.017 Wintergarten 2591 4.391 Gartenhäuser / Pavillons 591 1.002 Swimmingpools, Jacuzzi 303 0.514 Carport 1547 2.622 Parkplätze 465 0.788 öffentliche WC-Anlagen 31 0.053 Personenunterstände 46 0.078 Strassen 3 0.005 Eisenbahnen 1 0.002 Kanalisationen ausserorts 1 0.002 Bauwerke für Entwässerungen 1 0.002 Lärmschutzmassnahmen 11 0.019 Fussgänger- und Radfahrerbrücken 1 0.002 Stützmauern 59 0.100 Ufer- und Sohlensicherungen 1 0.002 Fischteiche 2 0.003 Mechanische Abwasserreinigungsanlagen 2 0.003 Biologische und chemische Abwasserreinig 1 0.002 Antennanlagen 9 0.015 Hochspannungs-Trafostationen 2 0.003 Silos 1 0.002

Data: DOCUMEDIA

257

Appendix A. Appendix

Table A.3: Levels of construction stage baustadiumcode

Level Count Proportion

1 Projekt 2795 4.7 2 Gesuch 4204 7.1 3 Bewilligt 44061 74.6 4 5 undef 374 0.6 5 Bau beendet 7466 12.6 6 8 undef 167 0.3 7 9 undef 5 0.0 8 11 undef 1 0.0

Data: DOCUMEDIA

Table A.4: Levels of project type bauartcode

Level Count Proportion

1 Neubau 21384 36.2 2 Anbau 14430 24.4 3 Umbau 18095 30.6 4 Renovation 789 1.3 5 Abbruch 4375 7.4

Data: DOCUMEDIA

Table A.5: Levels of purpose verwendungszweck

Level Count [%]

1 0 undef 12994 22.0 2 Vermietung 5903 10.0 3 Verkauf 4275 7.2 4 Eigenbedarf 35900 60.8 5 9 undef 1 0.0

Data: DOCUMEDIA

258

A.1. Descriptive statistics of DOCUMEDIA data

A.1.3 Descriptives

259

A ppendix

A .

A ppendix

Table A.6: Descriptives of cleaned DOCUMEDIA dataset

Variable Type Class Nb_NA NA_share Nb_Null Nb_zeros Zeros_share Min Q25 Mean Median Q75 Max

Object_nb integer integer 0 0.0 0 0 0.0 4985.0 99449684.0 76972737.7 99622579.0 99811230.0 100002008.0 Developer_nb integer integer 0 0.0 0 11 0.0 0.0 2490302.0 3509869.8 2620729.0 2797836.0 100419631.0 Offer_type integer factor 0 0.0 0 NA NA NA NA NA NA NA NA Object_name integer factor 16001 27.1 0 NA NA NA NA NA NA NA NA Site_street character character 55 0.1 0 NA NA NA NA NA NA NA NA Site_zip_code integer integer 0 0.0 0 0 0.0 6280.0 8155.0 8430.7 8414.0 8636.0 9922.0 Site_place character character 0 0.0 0 NA NA NA NA NA NA NA NA Site_language character character 0 0.0 0 NA NA NA NA NA NA NA NA Site_canton character character 0 0.0 0 NA NA NA NA NA NA NA NA Site_district_code character character 0 0.0 0 NA NA NA NA NA NA NA NA Site_district_name character character 0 0.0 0 NA NA NA NA NA NA NA NA Reference_nb character character 404 0.7 0 NA NA NA NA NA NA NA NA Description character character 0 0.0 0 NA NA NA NA NA NA NA NA Construction_stage integer factor 547 0.9 0 NA NA NA NA NA NA NA NA Project_type integer factor 0 0.0 0 NA NA NA NA NA NA NA NA Year_built integer integer 50425 85.4 0 0 0.0 1259.0 2000.0 1997.1 2002.0 2006.0 2014.0 Application_date double Date 900 1.5 0 NA NA 1996-07-26 NA 2005-01-02 2005-02-09 NA 2010-11-12 Approval_date double Date 14775 25.0 0 NA NA 1999-01-05 NA 2004-09-21 2004-08-04 NA 2010-12-04 Application_duration double numeric 15003 25.4 0 170 0.3 0.0 59.0 101.1 79.0 113.0 2712.0 Construction_start double Date 33803 57.2 0 NA NA 1998-10-01 NA 2004-08-06 2004-05-01 NA 2012-09-01 Construction_end double Date 45373 76.8 0 NA NA 1999-01-01 NA 2004-12-06 2004-10-01 NA 2015-01-01 Construction_duration double numeric 45446 76.9 0 0 0.0 1.0 3.0 7.3 6.0 9.0 1512.0 Construction_cost_min double numeric 49 0.1 0 0 0.0 0.0 0.1 1.8 0.2 0.9 2200.0 Construction_cost_max double numeric 49 0.1 0 0 0.0 0.0 0.1 1.9 0.2 0.9 2200.0 Nb_buildings integer integer 990 1.7 0 2904 4.9 0.0 1.0 1.3 1.0 1.0 79.0 Nb_dwellings integer integer 8435 14.3 0 20432 34.6 0.0 0.0 3.4 1.0 1.0 620.0 Nb_floors integer integer 9808 16.6 0 22017 37.3 0.0 0.0 1.1 1.0 2.0 33.0 Nb_basements integer integer 13159 22.3 0 32507 55.0 0.0 0.0 0.3 0.0 1.0 14.0 Nb_parking_lots integer integer 14998 25.4 0 40263 68.2 0.0 0.0 1.1 0.0 0.0 510.0 GFA integer integer 15721 26.6 0 42994 72.8 0.0 0.0 23.1 0.0 0.0 200000.0 Footprint integer integer 15615 26.4 0 39862 67.5 0.0 0.0 69.1 0.0 0.0 670000.0 Parcel_area integer integer 59016 99.9 0 0 0.0 1.0 70.0 14964.2 3500.0 16000.0 200000.0 Volume integer integer 13648 23.1 0 34833 59.0 0.0 0.0 1816.0 0.0 0.0 630793.0 Purpose integer factor 12995 22.0 0 NA NA NA NA NA NA NA NA Building_type_code integer factor 68 0.1 0 NA NA NA NA NA NA NA NA Building_type_name character character 68 0.1 0 NA NA NA NA NA NA NA NA Construction_cost double numeric 49 0.1 0 0 0.0 0.0 0.1 1.8 0.2 0.9 2200.0

260

A.2. Additional models in simulation

A.2 Additional models in simulation

Table A.7: Coefficients of the income regression model to update house- holds from demographic simulation

Variable Estimate SD t-Value

Constant 3755.2 11.95 314.23 Level of education 1653.0 6.20 266.51 Cars in household 2.4 6.41 0.37 Persons in household 1748.9 4.48 390.74

Source: Schirmer et al. (forthcoming)

Table A.8: Coefficients of the car availability model to update households from demographic simulation

Variable Car onwernership Estimate SD t-Value

Constant 0 0.9257 0.045 20.6 Level of education 0 -0.1932 0.020 -9.7 Distance to CBD Zurich 1 0.0001 0.000 22.9 Household income 1 0.0000 0.000 -0.5 Persons in household 1 0.8334 0.022 37.7

261

Appendix A. Appendix

Table A.9: Coefficients of the real estate price model (REPM) defining rent prices of living units

Variable Estimate SD t-Value

Constant 2.9256 0.08 34.89 Accessibility_Car 0.1084 0.01 13.37 Accessibillity_PT 0.0218 0.00 13.02 Built in 1921 to 1930 0.1261 0.02 7.18 Built in 1981 to 1990 0.0070 0.01 1.02 Built in 1991 to now 0.0708 0.01 10.96 Built pre 1921 0.0762 0.01 8.79 Distance to station 0.0000 0.00 -4.12 Foreigners within 300m 0.0000 0.00 2.60 Proximity to highway (<100m) -0.0532 0.02 -3.42 Is a single family house 0.0719 0.02 4.78 Jobs in Hotels and Gastro 0.0001 0.00 25.68 View of Lake (ha) 0.0000 0.00 17.86 Population density (ln) -0.0574 0.00 -16.39 Size in sqm (ln) 0.7887 0.01 123.80 Slope of terrain 0.0401 0.00 11.61 Sunshine index (evening) 0.0074 0.00 8.55

Adj. Likelihood ratio index: 0.782 Number of observations: 6497

Source: Schirmer et al. (forthcoming)

262

A .2.

A dditionalm

odels in

sim ulation

Table A.10: Coefficinets of employment location choice model (ELCM) according to distinguished sectors

Mfg Cns Trd Rtl HR Trn Srv Hlt

Average zonal income - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** Accessibilty car + 1.36 ** + 1.24 ** + 1.65 ** + 2.42 ** + 1.67 ** + 1.86 ** + 1.85 ** + 0.63 ** Accessibilty PT + 0.04 ** + 0.04 ** + 0.11 ** + 0.06 ** + 0.01 + 0.00 + 0.05 ** + 0.01 Distance to highway access - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 * + 0.00 - 0.00 ** + 0.00 ** + 0.00 ** Distance to station - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** Distance to Zurich CBD + 0.00 ** + 0.00 + 0.00 ** + 0.00 ** + 0.00 + 0.00 ** - 0.00 - 0.00 ** Household density (km2) - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** - 0.00 ** Job density (km2) + 0.00 ** + 0.00 ** + 0.00 ** + 0.00 ** + 0.00 ** + 0.00 ** + 0.00 ** + 0.00 ** Share of same jobs (zone) + 5.82 ** + 5.26 ** + 8.63 ** + 8.19 ** + 7.37 ** + 6.13 ** + 3.85 ** + 4.85 **

Adj. likelihood ratio index: 0.17 0.11 0.23 0.18 0.13 0.26 0.21 0.17 Number of observations: 15714 9187 11895 10143 7038 14390 33170 12382 ∗∗∗ p < 0.005, ∗∗ p < 0.01, ∗ p < 0.05

Source: Schirmer et al. (forthcoming)

263

Appendix A. Appendix

Table A.11: Coefficinets of workplace location choice model (WLCM)

Variable Estimate t-Value

βdist Xηdist distance to job (βdist ) 4.9 27.97 distance to job (ηdist ) -0.106 -9.75

Adj. ρ2 0.18 Number of observations 5291

Source: Schirmer et al. (forthcoming)

264

A.2. Additional models in simulation

Table A.12: Coefficients household location choice model (HLCM)

Variable Estimate Impact

Household Distance to previous location -7.070 ** Distance to workplace -3.220 * η for distance to previous location 0.163 ** η for distance to workplace 0.203 ** (interaction distance to previous location) -26.892 (interaction distance to workplace) -16.627

Accessibility Car accessibility (car) -0.302 ** -2.758 PT accessibility (no car) 0.541 ** 5.877

Built environment Close to network? -0.304 * 0.075

Points of interest Distance to CBD Zurich (young HH)1 0.067 ** 0.760 Distance to highway access (car)1 -0.092 ** -0.159 Distance to station (no car)1 -0.230 -0.171 Retail density -0.003 ** 0.000 Distance to school1 0.298 ** 0.107 Service density -0.001 ** 0.000

Socioeconomic structure Households of same age 0.684 ** 0.271 Residential unit Building age 0.360 ** 1.248 New building? 0.578 ** 0.578 Rent-income ratio -3.400 ** -16.609 Rooms per person -0.677 ** -1.124 Room size 0.000 ** 0.000

LL(0) -2679.736 LL(conv.) -1261.079 ρ2 0.529 Adj. ρ2 0.522 ∗∗ p < 0.05, ∗ p < 0.1

Source: Schirmer et al. (forthcoming)

265

A.3. Interview guidelines

A.3 Interview guidelines

267

SNF Projekt Urban Transformation

Fragenkatalog für Bauherrschaft

C. Zöllig, IVT, ETH Zürich

K.W. Axhausen, IVT, ETH Zürich

November 2011

Fragenkatalog für Bauherrschaft____________________________________________________4. November 2011

1 Einleitung Das Institut für Verkehrsplanung und Transportsysteme der ETH untersucht im Rahmen eines

Forschungsprojektes des Schweizerischen Nationalfonds das Investitionsverhalten von

Immobilienentwicklern. Dazu gehören auch Sie als Privatperson mit ihren Bauprojekten. Uns

geht darum das unterschiedliche Verhalten von Bauherren und dessen Auswirkungen auf die

räumliche Entwicklung besser zu verstehen. Schlussendlich soll unter Berücksichtigung von

Bauherrentypen erklärt werden, wo welche Gebäude erstellt werden. Für die Verkehrsplanung

ist dies von Interesse, da Gebäude oft den Ort für eine bestimmte Aktivität (z.B. arbeiten)

vorgeben. Bei der langfristigen Planung ist es deshalb wichtig zu wissen wo, welche Gebäude

entstehen.

Informationsziel

Mit den Interviews sollen mögliche unterschiedliche Verhaltensweisen von Bauherren

herausgefunden werden. Im Vordergrund stehen dabei Strategien, Motivationen und

Entscheidungsverhalten bei der Realisierung von Bauprojekten.

Der Fragebogen soll:

1. Den befragten Bauherrn einem Typen1 zuordnen können.

2. Die Unterschiede im Entscheidungsverhalten zeigen, bezüglich:

1. Entscheidungskriterien

2. Zugrunde liegender Informationen

3. Betrachteter Möglichkeiten

Im Folgenden finden Sie 30 Fragen in vier Themenbereiche aufgeteilt. Die Fragen werden vor

dem Interview zur Verfügung gestellt, um eine Vorbereitung zu ermöglichen. Das Interview

wird nach Möglichkeit im direkten Gespräch durchgeführt und dauert ca. 40 bis 50 Minuten.

Zur nachträglichen Transkription wird das Gespräch nach Einwilligung des Befragten

aufgezeichnet. Die Transkription wird den Befragten zur Überprüfung und Korrektur

zugestellt. Die Auswertungen und Resultate werden anonymisiert.

1 Wir unterscheiden primär zwischen Promotoren und Entwicklern mit Nutzungs- /Bewirtschaftungsabsicht. Entwickler mit Nutzungsabsicht werden zudem als mit oder ohne Portfoliostrategie arbeitend erfasst.

1

Fragenkatalog für Bauherrschaft____________________________________________________4. November 2011

2 Fragen

2.1 Informationen zum Bauherr 1. Welches war die Hauptmotivation zur Realisierung ihres letzten Bauprojektes? (Verkauf

des Projektes, Erstellung zur eigenen Nutzung)

2. Welche Funktionen übernahmen Sie im Bauprozess? (Finanzierung, Nutzungskonzept, Entwurf, Planung, Bau)

3. Wie viele externe Partner waren am Projekt beteiligt?

4. Welche Leistungen wurden von externen Projektpartnern abgedeckt? (Planung, Nutzungskonzept, Entwurf, Bau, Finanzierung)

5. Wie viele Bauprojekte haben Sie bereits realisiert?

6. Welche Gesamtgrösse hatten diese? (CHF, m2, Anzahl Wohnungen etc.)

7. Für welche Nutzungen haben Sie gebaut?

8. Wie hoch war die Eigenkapitalquote an den Projekten?

2.2 Entscheidungsprozess 9. Wie viele Immobilien besitzen Sie?

10. Wie sind ihre Immobilien strukturiert? (Nach Nutzungen/Gebäudetypen)

11. Welche Strategie verfolgen Sie bezüglich Ihrer Immobilien?

12. Welche Diversifizierungsziele verfolgen Sie bei Ihren Immobilieninvestitionen?

13. Was wäre zur Zeit eine vorteilhafte Erweiterung ihres Immobilienbesitzes?

14. Nach welchem Hauptentscheidungskriterium entscheiden Sie sich für oder gegen ein Bauprojekt?

2

Fragenkatalog für Bauherrschaft____________________________________________________4. November 2011

15. Welcher Anteil des Einkommens wenden Sie für Hypotheken auf? (%)

16. Nach wie vielen Jahren soll die Investition amortisiert sein?

17. Welche weiteren Kriterien berücksichtigen Sie?

18. Welches sind die wichtigsten Informationsquellen, die der Investitionsentscheidung zu Grunde liegen?

19. Wer entscheidet über die Realisierung eines Bauprojektes?

2.3 Art des Bauprojektes 20. Welche Arten von Bauprojekt habe Sie realisiert? (Im Bezug auf Nutzungen)

21. Beinhaltet Ihr Bauprojekt Nutzungsmischungen?

22. Gibt es Projekttypen, die Sie favorisieren?

2.4 Standortwahl für das Bauprojekt 23. Wie gingen Sie vor bei der Standortsuche?

24. Welches Gebiet betrachteten Sie für ihr Bauprojekt? (International, National, Regional, Lokal)

25. Wie haben Sie die Attraktivität des Bauplatzes abgeschätzt?

26. Welches sind aus Ihre Sicht die wichtigsten fünf Eigenschaften der Parzelle?

27. Welches sind aus Ihrer Sicht die wichtigsten fünf Eigenschaften der Gemeinde?

28. Haben Sie die Bautätigkeit anderer in der Umgebung eines möglichen Standortes berücksichtigt?

29. Wie kamen Sie zum Bauland?

30. Wie lange vor Einreichung der Baubewilligung haben Sie das Bauland beschafft?

3

SNF Projekt Urban Transformation

Fragenkatalog für Immobilienentwickler

C. Zöllig, IVT, ETH Zürich

K.W. Axhausen, IVT, ETH Zürich

November 2011

Fragenkatalog für Immobilienentwickler_______________________________________________3/ November 2011

1 Einleitung Das Institut für Verkehrsplanung und Transportsysteme der ETH untersucht im Rahmen eines

Forschungsprojektes des Schweizerischen Nationalfonds das Investitionsverhalten von

Immobilienentwicklern. Es geht in erster Linie darum das unterschiedliche Verhalten von

Entwicklern und dessen Auswirkungen auf die räumliche Entwicklung zu beleuchten.

Schlussendlich soll unter Berücksichtigung von Immobilienentwicklertypen erklärt werden,

wo welche Immobilien erstellt werden. Für die Verkehrsplanung ist dies von Interesse, da

Immobilien die Lokalitäten für verschiedene Aktivitäten darstellen. Bei der langfristigen

Planung ist deshalb der Veränderung des Immobilienangebotes Rechnung zu tragen.

Informationsziel

Mit den Interviews sollen mögliche unterschiedliche Verhaltensweisen von

Immobilienentwicklern in Erfahrung gebracht werden. Im Vordergrund stehen dabei

Strategien, Motivationen und Entscheidungsverhalten bei der Realisierung von Bauprojekten.

Der Fragebogen soll:

1. Die befragte Unternehmung Entwicklertypen1 zuordnen können.

2. Die Unterschiede im Entscheidungsverhalten zeigen, bezüglich:

1. Entscheidungskriterien

2. Zugrunde liegender Informationen (insbesondere über die Akteure)

3. Betrachteter Alternativenmenge

Im Folgenden finden Sie die entsprechenden Fragen in fünf Themenbereiche aufgeteilt. Die

Fragen werden vor dem Interview zur Verfügung gestellt, um die Vorbereitung zu

ermöglichen. Das Interview wird nach Möglichkeit im direkten Gespräch durchgeführt und

dauert ca. 45 bis 60 Minuten. Zur nachträglichen Transkription wird das Gespräch nach

Einwilligung des Befragten aufgezeichnet. Die Transkription wird den Befragten zur

Überprüfung und Korrektur zugestellt. Die Auswertungen und Resultate werden

anonymisiert.

1 Wir unterscheiden primär zwischen Promotoren und Entwicklern mit Nutzungs- /Bewirtschaftungsabsicht. Entwickler mit Nutzungsabsicht werden zudem als mit oder ohne Portfoliostrategie arbeitend erfasst.

1

Fragenkatalog für Immobilienentwickler_______________________________________________3/ November 2011

2 Fragen

2.1 Informationen zur Unternehmung 1. Welche Rechtsform hat Ihre Unternehmung? (Privatperson, Einzelfirma, GmbH, AG,

Genossenschaft, Verein, Stiftung)

2. Welches ist die Hauptmotivation zur Realisierung von Immobilienprojekten? (Verkauf des Projektes, Erstellung zur eigenen Bewirtschaftung)

3. Welche Funktionen übernimmt die Unternehmung im Bauprozess? (Finanzierung, Nutzungskonzept, Entwurf, Planung, Bau)

4. Wie viele externe Partner sind in der Regel am Projekt beteiligt?

5. Welche Leistungen werden von externen Projektpartnern abgedeckt? (Planung, Nutzungskonzept, Entwurf, Bau, Finanzierung)

6. Wie viele Projekte hat Ihre Unternehmung in den letzten drei Jahren realisiert?

7. Können Sie einen Bereich von Projektgrössen angeben, die sie realisiert haben? (CHF, m2, Anzahl Wohnungen etc.)

8. Auf welche Nutzungen ist die Unternehmung spezialisiert? (Bei Produktion bzw. Betrieb)

9. Wie hoch war die durchschnittliche Eigenkapitalquote an diesen Projekten?

10. Wie viele Mitarbeiter hat die Unternehmung?

11. Welchen Umsatz erzielte die Unternehmung im letzten Jahr? (in Mio.)

2.2 Entscheidungsprozess 12. Hält die Unternehmung ein Immobilienportfolio?

13. Wie viele Immobilien umfasst das Immobilienportfolio?

14. Wie ist das Immobilienportfolio strukturiert? (Nach Nutzungen/Gebäudetypen)

2

Fragenkatalog für Immobilienentwickler_______________________________________________3/ November 2011

15. Welche Strategie wird bezüglich des Immobilienportfolios verfolgt?

16. Welche Diversifizierungsziele verfolgt die Unternehmung bei Immobilieninvestitionen?

17. Was wäre zur Zeit eine vorteilhafte Veränderung ihres Immobilienportfolios?

18. Nach welchem Hauptentscheidungskriterium wird für oder gegen ein Projekt entschieden?

19. Welche Rendite wird angestrebt? (%/Jahr)

20. In welchem Zeitraum soll die Investition amortisiert sein?

21. Welcher Anteil des Projektes muss ab Plan verkauft sein?

22. Welche weiteren Kriterien werden berücksichtigt?

23. Welche sind die wichtigsten Informationsquellen, die der Investitionsentscheidung zu Grunde liegen?

24. Wer entscheidet über die Realisierung eines Bauvorhabens?

2.3 Art der Bauprojekte 25. Welche Art von Bauprojekten realisiert die Unternehmung? (Im Bezug auf Nutzungen)

26. Welche Nutzungsmischungen treten häufig auf?

27. Wie viele Gebäude werden bei den Bauprojekten üblicherweise realisiert?

28. Gibt es Entwicklungstypen, welche favorisiert werden? Werden Standardprojekte realisiert?

2.4 Standortwahl für die Bauprojekte 29. Wie gehen Sie vor bei der Standortsuche?

30. Welchen Markt betrachten Sie für neue Investitionen? (International, National, Regional, Lokal)

3

Fragenkatalog für Immobilienentwickler_______________________________________________3/ November 2011

31. Wie wird die Attraktivität eines Ortes abgeschätzt?

32. Welches sind die wichtigsten fünf Eigenschaften der Mikrolage?

33. Welches sind die wichtigsten fünf Eigenschaften der Makrolage?

34. Inwiefern wird auf die räumlich Verteilung der Projekte geachtet? (Diversifizierung, Konzentration)

35. Inwiefern wird die Bautätigkeit anderer in der Umgebung eines möglichen Standortes berücksichtigt?

2.5 Einschätzung des Umfeldes 36. Welche Märkte identifizieren Sie innerhalb der Immobilienproduktion?

37. Welche Märkte bearbeitet Ihre Unternehmung?

38. Wie kommen Sie in der Regel zum Bauland?

39. Wie lange vor Einreichung der Baubewilligung wird in der Regel das Bauland beschafft?

4

A.4. Land price model

A.4 Land price model

Table A.13: Fixed effects per municipality of panel linear model

Nb. municipality Estimate Std. Error t-value Pr(>|t|)

1 21.95 291.11 0.08 0.94 2 147.67 319.21 0.46 0.64 3 106.40 300.63 0.35 0.72 4 23.89 304.23 0.08 0.94 5 52.08 314.59 0.17 0.87 6 -148.58 307.92 -0.48 0.63 7 140.62 300.15 0.47 0.64 8 -126.53 341.52 -0.37 0.71 9 106.49 281.49 0.38 0.71 10 -11.29 307.30 -0.04 0.97 11 -81.92 309.00 -0.27 0.79 12 24.59 291.87 0.08 0.93 13 -139.18 310.45 -0.45 0.65 14 -38.24 264.46 -0.14 0.89 21 -270.42 383.45 -0.71 0.48 22 -209.70 319.08 -0.66 0.51 23 -266.29 329.64 -0.81 0.42 24 -60.70 306.50 -0.20 0.84 25 -184.00 304.64 -0.60 0.55 26 -123.44 329.65 -0.37 0.71 27 -176.40 310.23 -0.57 0.57 28 -176.81 299.08 -0.59 0.55 29 -186.84 321.66 -0.58 0.56 30 -179.31 302.98 -0.59 0.55 31 -41.72 298.18 -0.14 0.89 32 -244.22 337.02 -0.72 0.47 33 -225.99 304.72 -0.74 0.46 34 -124.24 291.00 -0.43 0.67 35 -192.88 310.66 -0.62 0.53 36 -209.93 320.78 -0.65 0.51 37 -229.95 326.13 -0.71 0.48 38 -225.71 324.00 -0.70 0.49 39 -279.91 315.32 -0.89 0.37 40 -193.21 310.12 -0.62 0.53

Continued on next page

277

Appendix A. Appendix

Municipality Estimate Std. Error t-value Pr(>|t|)

41 -361.61 317.64 -1.14 0.25 42 -127.36 301.10 -0.42 0.67 43 -175.55 316.21 -0.56 0.58 44 -113.32 296.58 -0.38 0.70 51 -16.72 270.24 -0.06 0.95 52 -14.35 290.31 -0.05 0.96 53 122.87 309.21 0.40 0.69 54 84.32 295.36 0.29 0.78 55 6.82 297.09 0.02 0.98 56 -44.94 306.84 -0.15 0.88 57 -21.59 301.25 -0.07 0.94 58 -155.17 294.96 -0.53 0.60 59 -139.07 307.04 -0.45 0.65 60 -46.10 290.63 -0.16 0.87 61 -42.30 283.01 -0.15 0.88 62 -84.98 293.35 -0.29 0.77 63 72.26 263.17 0.27 0.78 64 -150.60 288.11 -0.52 0.60 65 -57.53 346.22 -0.17 0.87 66 73.26 271.94 0.27 0.79 67 -13.54 295.12 -0.05 0.96 68 -191.74 301.89 -0.64 0.53 69 90.29 304.54 0.30 0.77 70 -240.55 340.82 -0.71 0.48 71 -123.70 312.24 -0.40 0.69 72 -102.07 253.27 -0.40 0.69 81 -70.05 339.63 -0.21 0.84 82 -38.34 281.92 -0.14 0.89 83 138.78 291.10 0.48 0.63 84 -71.24 302.27 -0.24 0.81 85 -18.24 313.24 -0.06 0.95 86 55.17 298.72 0.18 0.85 87 86.72 290.92 0.30 0.77 88 -160.06 265.10 -0.60 0.55 89 12.57 284.94 0.04 0.96 90 88.14 300.09 0.29 0.77 91 41.43 286.47 0.14 0.88 92 36.20 304.44 0.12 0.91 93 7.69 298.39 0.03 0.98

Continued on next page

278

A.4. Land price model

Municipality Estimate Std. Error t-value Pr(>|t|)

94 87.58 289.89 0.30 0.76 96 90.50 292.76 0.31 0.76 97 2.78 285.11 0.01 0.99 98 -80.39 317.85 -0.25 0.80 99 79.82 311.34 0.26 0.80 100 -91.59 292.60 -0.31 0.75 101 9.11 300.76 0.03 0.98 102 -280.14 251.67 -1.11 0.27 111 -68.91 309.19 -0.22 0.82 112 32.70 289.79 0.11 0.91 113 -187.43 315.88 -0.59 0.55 114 -197.72 280.53 -0.70 0.48 115 13.88 322.34 0.04 0.97 116 -53.36 316.50 -0.17 0.87 117 -19.68 299.12 -0.07 0.95 118 -96.92 319.98 -0.30 0.76 119 54.78 341.60 0.16 0.87 120 -174.53 308.98 -0.56 0.57 121 31.31 308.09 0.10 0.92 131 1.95 315.79 0.01 1.00 132 157.13 306.29 0.51 0.61 133 101.00 289.42 0.35 0.73 134 -11.03 329.16 -0.03 0.97 135 854.98 284.91 3.00 0.00 ** 136 24.77 296.54 0.08 0.93 137 316.52 283.19 1.12 0.26 138 93.85 293.28 0.32 0.75 139 567.29 278.43 2.04 0.04 * 140 -18.15 322.03 -0.06 0.96 141 159.52 288.09 0.55 0.58 142 339.73 307.79 1.10 0.27 151 768.45 271.97 2.83 0.00 ** 152 560.27 266.10 2.11 0.04 * 153 55.51 302.54 0.18 0.85 154 716.62 277.87 2.58 0.01 ** 155 312.84 289.66 1.08 0.28 156 847.10 277.96 3.05 0.00 ** 157 214.74 314.55 0.68 0.49 158 95.24 277.54 0.34 0.73

Continued on next page

279

Appendix A. Appendix

Municipality Estimate Std. Error t-value Pr(>|t|)

159 186.97 288.29 0.65 0.52 160 492.83 267.13 1.84 0.07 . 161 770.75 288.76 2.67 0.01 ** 171 -87.53 300.75 -0.29 0.77 172 55.87 301.70 0.19 0.85 173 -85.87 322.12 -0.27 0.79 174 -89.82 315.97 -0.28 0.78 175 -186.20 362.35 -0.51 0.61 176 -144.12 290.98 -0.50 0.62 177 126.90 306.10 0.41 0.68 178 9.72 306.68 0.03 0.97 179 -354.34 379.21 -0.93 0.35 180 -127.33 319.97 -0.40 0.69 181 -197.87 324.67 -0.61 0.54 182 -133.40 316.75 -0.42 0.67 191 127.86 302.15 0.42 0.67 192 63.77 296.45 0.22 0.83 193 -155.95 289.64 -0.54 0.59 194 182.66 321.82 0.57 0.57 195 -63.38 270.45 -0.23 0.81 196 99.71 305.86 0.33 0.74 197 4.47 305.44 0.01 0.99 198 132.39 312.17 0.42 0.67 199 41.74 282.73 0.15 0.88 200 112.14 298.42 0.38 0.71 211 -291.79 322.21 -0.91 0.37 212 -249.30 307.09 -0.81 0.42 213 -36.85 305.51 -0.12 0.90 214 -152.56 329.17 -0.46 0.64 215 34.88 276.33 0.13 0.90 216 -236.49 325.44 -0.73 0.47 217 -26.86 311.98 -0.09 0.93 218 -216.72 306.17 -0.71 0.48 219 -110.74 308.37 -0.36 0.72 220 -178.05 354.01 -0.50 0.61 221 -165.56 292.24 -0.57 0.57 222 -318.25 326.49 -0.97 0.33 223 -137.79 309.12 -0.45 0.66 224 1.12 280.69 0.00 1.00

Continued on next page

280

A.4. Land price model

Municipality Estimate Std. Error t-value Pr(>|t|)

225 -108.73 319.38 -0.34 0.73 226 -255.90 323.64 -0.79 0.43 227 -1.96 306.86 -0.01 0.99 228 -155.15 315.71 -0.49 0.62 229 -91.58 294.56 -0.31 0.76 230 85.96 323.95 0.27 0.79 231 -180.58 313.04 -0.58 0.56 241 139.70 261.90 0.53 0.59 242 43.92 302.05 0.15 0.88 243 102.61 327.29 0.31 0.75 244 -35.55 301.82 -0.12 0.91 245 317.61 313.70 1.01 0.31 246 18.74 311.08 0.06 0.95 247 237.20 310.27 0.76 0.44 248 378.60 264.79 1.43 0.15 249 165.98 286.02 0.58 0.56 250 39.18 323.69 0.12 0.90 251 47.38 288.15 0.16 0.87 261 633.38 332.23 1.91 0.06 .

*** p < 0.001, ** p < 0.01, * p < 0.05, . p < 0.1 Data: Zurich Cantonal Statistical Office (SAKZ)

281

Cirriculum Vitae

Msc Christof Zöllig Renner

PERSONAL DETAILS

Date of birth January 6th, 1981 Marital status Married

Gender Male Address Regensdorferstr. 190 8049 Zürich

Citizenship Swiss E-Mail [email protected]

HIGHER EDUCATION

2008 – Ph.D. in Spatial Planning, Prof. K.W. Axhausen

ETH Zurich, Institute for Transport Planning and Systems

2005 – 2008 M.A. in Education, Emphasis: Geomatics

ETH Zurich, Switzerland

2001 – 2007 Msc in Geomatics Engineering and Planning

ETH Zurich, Switzerland

08/2004 – 02/2005 Semester abroad

Universidad Politécnica de Valencia (UPV), Spanien

EMPLOYMENT EXPERIENCE

Analyst and Simulation Engineer

10/2014 - ASE (Analysis Simulation Engineering) GmbH, Switzerland

Research Assistant

2008 – 05/2014 Institute for Transport Planning and Systems, ETH Zurich, Switzerland

2007 – 2008 Institute for Spatial and Landscape Planning, ETH Zurich, Switzerland

Visiting Student Researcher

02/2014 – 03/2014 Urban Analytics Lab, Institute of urban and regional development, University of California, Berkeley

Internship

09/2005 – 10/2005 Scheifele, Zurich, Switzerland

03/2005 – 06/2005 ERR-Raumplaner FSU SIA, St. Gallen, Switzerland

08/2004 Herzog Ingenieure ETH / SIA, Davos, Switzerland

TEACHING

Assisting and lecture

2008 – Lecture Planning concepts, Msc, ETH Zurich

2013 Module Valuation methods, DAS transportation engineering, ETH Zurich

Lecture in

2007 Environmental planning, Msc, ETH Zurich

2007 Economic geography, Bsc, Zürcher Fachhochschule Winterthur, Winterthur

2007 Geography, secondary education, Aargauische Maturitätschule für Erwachsene

PROJECTS

2010 – 2013 SustainCity, Micro-simulation for the prospective of sustainable cities in Europe

14.11.2014 1/3

2010 – 2012 Urban transformation, Agent-based simulation of real estate development

2010 Study of concepts for a national land use model in Switzerland, sponsor: Federal Office for Spatial Development ARE

PROFESSIONAL ACTIVITIES

Editor

2008 – 2013 NSL-Newsletter

Conference organizing committee

2011 Infrastructure research of NSL graduate students

2010 Meeting of NSL graduate students

2010 English seminar

Reviewing for

Cities, disP, JTLU, Transportation

SKILLS

Languages German Mother tongue

English Effective Operational

French Threshold or intermediate

Spanish Threshold or intermediate

Computer Operating System Windows, Linux

Office MS Office, OpenOffice, Latex

Geographic Informations Systems ArcGIS, QGIS, PostgreSQL

Statistical software SPSS, R

Programming languages Python, SQL, bash

Graphical programs Illustrator, gimp

Others SVN, eclipse

PUBLICATIONS

Zöllig, C. and K.W. Axhausen (2013) Modelling real estate development with heterogeneous agents, in 13th Swiss Transport Research Conference, paper presented at 13th Swiss Transport Research Conference, Ascona, April 2013.

Zöllig Renner, C. and K.W. Axhausen (2013) Comparing Estimation Results of Land Use Development Models Using Different Data Bases Available in Switzerland, paper presented at Transportation Research Board 92nd Annual Meeting, Washington, D.C., January 2013.

Zöllig, C. and K.W. Axhausen (2012a) Uncovering the heterogeneity of real estate developers in the canton Zurich, SustainCity Working Paper, 3.6, IVT, ETH Zürich, Zurich.

Zöllig Renner, C. and K.W. Axhausen (2012) Comparing Estimation Results of Land Use Development Models Using Different Data Bases Available in Switzerland, Arbeitsberichte Verkehrs- und Raumplanung, 796, IVT, ETH Zurich, Zurich.

Zöllig, C. and K.W. Axhausen (2012b) Assessment of infrastructural investment using agent based accessibility, in K. T. Geurs, K. J. Krizek and A. Reggiani (eds.), Accessibility Analysis and Transport Planning, 54–70, Edward Elgar Publishing, Cheltenham, UK.

Zöllig, C. and K.W. Axhausen (2012c) Heterogeneity of real estate developers in Canton Zurich, paper presented at 12th Swiss Transport Research Conference, Ascona, May 2012.

Zöllig, C., R. Hilber and K.W. Axhausen (2011a) Konzeptstudie Flächennutzungsmodellierung, Report for the

Cirriculum Vitae 14.11.2014 2/3

ARE, IVT, ETH Zürich und Mappuls AG, Zürich.

Zöllig, C. and K.W. Axhausen (2011a) How to model the gains from infrastructure investment?, Arbeitsberichte Verkehrs- und Raumplanung, 673, IVT, ETH Zürich, Zurich.

Zöllig, C. and K.W. Axhausen (2011b) A conceptual, agent-based model of land development for UrbanSim, paper presented at 51th ERSA Conference, Barcelona, September 2011.

Sutter, J.A. and C. Zöllig (2011) Einführung in Geographische Informationssysteme, EducETH.

Zöllig, C. and K.W. Axhausen (2011c) Integrierte Flächennutzungs- und Transportmodelle verbessern die Beurteilung von Verkehrsinfrastrukturprojekten, Forum Raumentwicklung, 11 (3) 17–19.

Koll-Schretzenmayr, M. and C. Zöllig (2010a) Transport and Land Use Models, disP, 46 (3) 70–75.

Zöllig, C. and K.W. Axhausen (2010) Calculating benefits of infrastructural investment, Arbeitsberichte Verkehrs- und Raumplanung, 612, IVT, ETH Zürich, Zurich.

Koll-Schretzenmayr, M. and C. Zöllig (2010b) Innenentwicklung akteuresbezogen, in M. Klemme and K. Selle (eds.), Siedlungsflächen entwickeln, 214–227, Verlag Dorothea Rohn, Detmold.

Zöllig, C. (2007) Stadtentwicklung zwischen Koordination und Selbstorganisation, Analysen, 3 (006) 28.

Zöllig Renner, C., T.W. Nicolai and K. Nagel (forthcoming) Agent-based land use transport interaction modeling: state of the art, in R. Hurtubia, M. Bierlaire, P. Waddell and A. de Palma (eds.), Handbook on integrated transport and land use modeling for sustainable cities, xx–xx, EPFL Press, Lausanne.

Zöllig Renner, C. and K.W. Axhausen (forthcoming) Real estate development models with heterogeneous agents, in R. Hurtubia, M. Bierlaire, P. Waddell and A. de Palma (eds.), SustainCity, xx–xx, EPFL Press, Lausanne.

Schirmer, P., C. Zöllig Renner, K. Müller, B.R. Bodenmann and K.W. Axhausen (forthcoming) Case study Zürich, in R. Hurtubia, M. Bierlaire, P. Waddell and A. de Palma (eds.), SustainCity, xx–xx, EPFL Press, Lausanne.

Cirriculum Vitae 14.11.2014 3/3