i

MEASURING AND UNDERSTANDING THE HOUSE PRICE DYNAMICS

OF THE ABERDEEN HOUSING MARKET

ANTHONY OWUSU-ANSAH

MSc. Real Estate Management, KTH-The Royal Institute of Technology, Sweden.

BSc. (Hons) Land Economy, Kwame Nkrumah Univ. of Science and Technology, Ghana.

A thesis presented in fulfillment of the requirement for the degree of Doctor of Philosophy

in the Business School, University of Aberdeen.

August 2012

ABSTRACT

Housing constitutes a major part of the real wealth of many countries as well as households

overall wealth portfolio. As a result, there is the continuous need to measure house price dynamics

and understand the housing markets especially at the local levels as opposed to the national and

regional levels. This thesis focuses on the Aberdeen housing market in Scotland and considers

a number of issues in relation to measuring and understanding the price dynamics of the study

area. The thesis empirically highlights why house price indices should be confined to the local

markets, produces constant-quality house price indices for the study area, measures the accuracy

of five different index construction models, tests if temporal aggregation affects the construction

of house price indices, and applies the indices to examine the relationship between new residential

construction and various determinants and also estimates house price elasticity of supply for the

local housing market.

The thesis adopts a quantitative research approach. The dataset for the house price index

construction is obtained from the Aberdeen Solicitors’ Property Centre and spans from January

2000 to December 2010 with about 57,000 property transactions and includes both the house prices

and a number of physical and locational characteristics. The dataset for examining the determinants

of new residential construction is also obtained from a number of sources including new housing

starts from the Scottish Government Statistics and building warrants granted from the Aberdeen

City Council.

The thesis finds that the national and regional house price indices do not portray accurate

picture of house price trend in the local housing market and so it is important for price trend analysis

to be confined to the local level. The study also finds that in the construction of house price indices,

the lower level of temporal aggregation, that is, monthly and quarterly is to be preferred and at this

level of temporal aggregation, the explicit time variable hedonic and the Quigley’s hybrid models

perform better by producing the lowest mean squared errors. At the broader level of temporal

aggregation however, the strictly cross-sectional hedonic model does better. In general, the repeat-

sales method is found to be the least preferred method of constructing house price indices.

The thesis also finds that changes in house prices, time on the market, planning regulation,

lagged stock, and lagged and future housing starts are the main factors that influence new residential

construction in Aberdeen. These factors have the expected signs with the changes in house prices

having a large positive coefficient of about 3.7% in the current period. All these factors are measured

at the local housing market level. The national variables, material costs and interest rates, however do

not have any effect on new housing construction in Aberdeen. This suggests that when modeling the

local housing market, local variables are more useful than the national variables. These influential

local variables would not be measured properly or would be ignored entirely when the national or

regional housing markets are modelled and stresses the need to conduct housing studies at the local

level. The price elasticities of supply estimated are in the range of 2.0-3.2 for housing starts, and

0.01-0.02 for housing stock. The starts elasticities are higher than most of the local housing markets

elasticities in the UK. Thus, private developers in Aberdeen respond more to a change in house

prices by initiating new construction than most of the local authority districts in the UK.

These findings have a number of practical and policy implications and the methods could be

applied to estimate house price indices for the other local housing markets in the UK when the data

system in the UK is improved. More studies are needed at the local levels in the UK so that some of

the issues raised in the thesis can be explored further.

ii

ACKNOWLEDGEMENT

The successful completion of this PhD thesis would not have been possible without the as-

sistance and help received from a number of people. First and foremost, I would like to thank my

supervisors: Dr Rainer Schulz and Professor Deborah Roberts. Without their commitment and ded-

ication in providing me with sound advice and constructive criticisms, this final result would never

have been materialised. I also thank Professor Stanley McGreal of the University of Ulster and Pro-

fessor Bryan MacGregor of the University of Aberdeen who have acted as the External and Internal

examiners respectively.

I would also like to thank the staff within the Business School of the University of Aberdeen,

especially, Professor Norman Hutchison and other members of the Real Estate group, who have

helped me in many ways during the last three years. Beyond the University of Aberdeen, I also

thank Dr Raymond Abdulai of the Liverpool John Moores University, Dr Samuel Azasu of the

University of Wits in South Africa, Professor Paul Kojo Asabere of the Temple University Business

School in the USA, and Professor Mats Wilhelmsson and Dr Han-Suck Song of KTH-The Royal

Institute of Technology in Sweden for their ideas and encouragement.

My heartfelt thanks go to the Aberdeen Solicitors’ Property Centre for providing their hous-

ing transaction data for the thesis. I also thank Tom Snowling and Dawn Ramsay of the Aberdeen

City Council, Gavin Oag of the Grampian Assessor and Electoral Registration Office, and Bruce

Teubes of the Scottish Government Statistics Office for the various data they provided for the thesis.

Finally, and most importantly, I wish to thank my family and friends for their support through-

out my studies. Especially, I wish to thank my wife, Amma and my son, Francis; my parents, Mr.

C.K. Owusu-Ansah and Mrs. Susanna Owusu-Ansah; and Mr Francis Ohemen for their support,

prayers, understanding and encouragement. To them I dedicate this thesis.

iii

DECLARATION

I declare that this dissertation has been composed by myself, that it has not

been accepted in an previous application for a degree, that the work of which it

is a record has been done by myself and all quotations have been distinguished

appropriately and the source of information specifically acknowledge.

Signed

....................................................................................

ANTHONY OWUSU-ANSAH

iv

TABLE OF CONTENTS

1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1

1.1 PRELIMINARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 RESEARCH BACKGROUND . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.3 MOTIVATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.4 RESEARCH AIM AND OBJECTIVES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.5 RESEARCH METHODOLOGY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.6 CONTRIBUTIONS OF THIS THESIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.7 THESIS STRUCTURE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2 HOUSE PRICE INDEX CONSTRUCTION METHODS . . . . . 17

2.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.2 ATTRIBUTES WHICH DETERMINE THE QUALITY OF A

HOUSE PRICE INDEX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.3 THE THEORY OF HEDONIC PRICING MODELING . . . . . . . . . . . . . . . 22

2.3.1 An overview of the hedonic method . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.3.2 The hedonic method and housing characteristics . . . . . . . . . . . . . . . 24

2.3.3 Hedonic modeling and functional form selection . . . . . . . . . . . . . . . 38

2.3.4 Main sources of hedonic problems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

2.4 THE AVERAGE METHOD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

2.4.1 Advantages and disadvantages of the average method . . . . . . . . . . 49

v

2.5 THE HEDONIC REGRESSION METHOD . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2.5.1 Explicit time variable hedonic index (ETV) model . . . . . . . . . . . . . 50

2.5.2 Strictly cross-sectional hedonic index (SCS) model . . . . . . . . . . . . . 52

2.5.3 Advantages and disadvantages of the hedonic index method . . . . 53

2.6 THE REPEAT SALES METHOD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

2.6.1 The original repeat-sales (ORS) model. . . . . . . . . . . . . . . . . . . . . . . . . 55

2.6.2 The weighted repeat-sales (WRS) model . . . . . . . . . . . . . . . . . . . . . . . 56

2.6.3 Advantages and disadvantages of the repeat-sales index method .59

2.7 THE HYBRID METHOD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

2.7.1 The Case and Quigley (CQ-hybrid) model . . . . . . . . . . . . . . . . . . . . . 62

2.7.2 The Quigley (Q-hybrid) model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

2.7.3 The Hill, Knight and Sirmans (HKS-hybrid) model . . . . . . . . . . . . . 65

2.7.4 The Englund, Quigley and Redfearn (EQR-hybrid) model . . . . . . 68

2.7.5 Advantages and disadvantages of the hybrid method . . . . . . . . . . . . 70

2.8 COMPARISON OF THE INDEX METHODS BASED ON

PREVIOUS EMPIRICAL FINDINGS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

2.8.1 Measuring of index accuracies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

2.8.2 Temporal aggregation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

2.9 HOUSE PRICE INDICES IN SOME SELECTED MARKETS . . . . . . . . 79

2.10 EXISTING HOUSE PRICE INDICES IN THE UK . . . . . . . . . . . . . . . . . . . 82

2.10.1 Halifax Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

2.10.2 Nationwide Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

2.10.3 A comparison of the Halifax and Nationwide indices . . . . . . . . . . . 89

vi

2.11 CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

3 THE STUDY AREA AND THE DATA . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

3.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

3.2 THE STUDY AREA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

3.2.1 Selection and justification of the geographical areas . . . . . . . . . . . . 99

3.2.2 An overview of Aberdeen housing market . . . . . . . . . . . . . . . . . . . 104

3.3 TYPES OF ECONOMIC DATA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

3.3.1 Cross-sectional data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

3.3.2 Time series data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

3.3.3 Panel data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

3.4 DATA GATHERING . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

3.4.1 Data identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

3.4.2 Data sources and availability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

3.4.3 Data coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

3.5 DATA PREPARATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

3.5.1 Data cleaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

3.5.2 Identification of repeat sales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

3.6 VERIFICATION OF THE FINAL DATASET. . . . . . . . . . . . . . . . . . . . . . . 132

3.6.1 Distribution of observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

3.6.2 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

3.7 CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

vii

4 EMPIRICAL RESULTS AND COMPARISON OFHOUSE PRICE INDEX CONSTRUCTION METHODS . . . . 155

4.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

4.2 PRESENTATION AND ANALYSIS OF THE HEDONIC

CHARACTERISTICS RESULTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

4.2.1 Hedonic characteristics estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

4.2.2 A test of the constancy of the implicit prices over time. . . . . . . . 166

4.3 EMPIRICAL RESULTS AND ANALYSIS OF THE VARIOUS

INDICES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

4.3.1 Comparison of the various index models . . . . . . . . . . . . . . . . . . . . . 174

4.3.2 Overall accuracy and precision of the index methods . . . . . . . . . 178

4.3.3 Statistical significance of the differences between models . . . . . 184

4.4 TEST OF TEMPORAL AGGREGATION . . . . . . . . . . . . . . . . . . . . . . . . . . 187

4.5 THE ABERDEEN HOUSING MARKET REPORT . . . . . . . . . . . . . . . . . 195

4.6 CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196

5 DETERMINANTS OF NEW RESIDENTIALCONSTRUCTION AND PRICE ELASTICITY OFSUPPLY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

5.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199

5.2 MOTIVATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201

5.3 DETERMINANTS OF HOUSING CONSTRUCTION . . . . . . . . . . . . . . 203

5.3.1 House prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204

5.3.2 Cost of construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

5.3.3 Interest rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210

viii

5.3.4 Planning regulations and other governmental policies . . . . . . . . . 210

5.3.5 Time on the market . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212

5.3.6 Other factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213

5.4 REVIEW OF EMPIRICAL MODELING AND PRICE

ELASTICITY OF HOUSING SUPPLY . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213

5.4.1 Modeling and price elasticities from some international

markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214

5.4.2 Modeling and price elasticities from the UK . . . . . . . . . . . . . . . . . 223

5.4.3 Conclusions from the literature review . . . . . . . . . . . . . . . . . . . . . . . 228

5.5 ESTIMATING HOUSING SUPPLY FOR THE ABERDEEN

HOUSING MARKET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

5.5.1 The econometric model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

5.5.2 Stationarity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236

5.6 DATA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237

5.6.1 Definition and sources of data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237

5.6.2 Presentation of data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240

5.6.3 Pre-test (stationary test) of the data . . . . . . . . . . . . . . . . . . . . . . . . . . 247

5.7 ESTIMATION RESULTS AND DISCUSSIONS . . . . . . . . . . . . . . . . . . . 249

5.7.1 Instrumental variable regression results . . . . . . . . . . . . . . . . . . . . . . 249

5.7.2 Price elasticities of supply . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255

5.8 CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258

6 SUMMARY AND CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261

6.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261

ix

6.2 SUMMARY OF THE RESEARCH AND ITS FINDINGS . . . . . . . . . . 262

6.2.1 First objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262

6.2.2 Second objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265

6.2.3 Third objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267

6.2.4 Fourth objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269

6.3 CONTRIBUTION AND RESEARCH IMPLICATIONS . . . . . . . . . . . . 272

6.3.1 Research contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272

6.3.2 Practical and policy implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274

6.4 RESEARCH LIMITATIONS AND DIRECTION FOR FUTURE

RESEARCH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276

6.5 CLOSING STATEMENT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281

APPENDICES.............................................................................................298

Appendix 1. The SCS monthly hedonic results.................................................298

Appendix 2. The SCS quarterly hedonic results....................................................321

Appendix 3. The SCS semi-annual hedonic results................................................333

Appendix 4. The SCS yearly hedonic results..........................................................338

Appendix 5. The ETV hedonic results....................................................................341

Appendix 6. The Q-hybrid hedonic results.............................................................343

Appendix 7. The absolute location values of the yearly SCS hedonic model..........345

Appendix 8. The Aberdeen housing market report project......................................357

x

1.1 Why the house price index project?. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357

1.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358

1.3 Methodology used for the index construction . . . . . . . . . . . . . . . . . . . . . . . 360

1.4 Results and comparison with the other indices . . . . . . . . . . . . . . . . . . . . . . 368

1.5 The Aberdeen housing market report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372

1.5.1 Change in house prices in Aberdeen city and suburbs and the

surrounding areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372

1.5.2 Constant quality house prices in Aberdeen city and suburbs and the

surrounding areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374

1.5.3 Volume of transactions in the Aberdeen city and suburbs . . . . . 375

xi

LIST OF ABBREVIATIONS

ACC Aberdeen City Council

ADF Augmented Dickey-Fuller

AHMR Aberdeen Housing Market Report

ASC Aberdeenshire Council

ASPC Aberdeen Solicitors’ Property Centre

BCIS Building and Construction Information Service

BLUE best linear unbiased estimator

BWA building warrant applications

BWG building warrants granted

CBD central business district

CML Council of Mortgage Lenders

CRER Centre for Real Estate Research

ETV explicit time variable

GDP gross domestic product

xii

GLS generalised least squares

GVA gross value added

HMA housing market area

IV instrumental variable

KPSS Kwiatkowski-Phillips-Schmidt-Shin

MCI material cost index

MAE mean absolute error

MSE mean squared error

MSOA Medium Level Super Output Areas

OLS ordinary least squares

ONS Office for National Statistics

ORS ordinary repeat-sales

PES price elasticity of supply

PPI property price index

RICS Royal Institution of Chartered Surveyors

xiii

RPI retail price index

SCS strictly cross-sectional

SUR Seemingly Unrelated Regression

TOM time on the market

WLS weighted least squares

WRS weighted repeat-sales

WTP willingness-to-pay

xiv

LIST OF FIGURES

2.1 A Comparison of the Halifax and Nationwide Index Series

at the National Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

2.2 A Comparison of the Halifax and Nationwide Index Series

at the Scottish Regional Level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

3.3 Housing Market and Council Areas (Source: Aberdeen

City and Shire Strategic Development Planning Authority,

2011) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

3.4 A map of UK showing the location of Aberdeen (Source:

Tiesdell and Allmendinger, 2004) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

3.5 Population trend in Aberdeen from 2000 to 2010 (Source:

General Office for Scotland, 2011) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

3.6 Population trend in Aberdeen compared to Edinburgh and

Glasgow from 2000 to 2010 (Source: General Office for

Scotland, 2011) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

3.7 Economic activity rate - adults of working age in Aberdeen

compared to Edinburgh and Glasgow from 2004 to 2010

(Source: Annual Population Survey, via Nomis, 2011) . . . . . . . . . . . . 108

3.8 Unemployment rate among those of working age in

Aberdeen compared to Edinburgh and Glasgow from 2004

to 2010 (Source: Annual Population Survey, via Nomis, 2011) . . . . . . 108

3.9 Dwellings of the Population in Aberdeen . . . . . . . . . . . . . . . . . . . . . . . 110

3.10 Dwelling types in Aberdeen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

xv

3.11 Distribution of sales by year and dwelling type . . . . . . . . . . . . . . . . . . 134

3.12 Distribution of sale price by various transformations . . . . . . . . . . . . . . 135

3.13 Geographical distribution of the 57,150 sales by the various

ASPC areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

3.14 Geographical distribution of the 32,347 flat sales by

transaction prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

3.15 Geographical distribution of the 16,467 semi-detached

sales by transaction prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

3.16 Geographical distribution of the 8,336 detached sales by

transaction prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

3.17 Geographical distribution of the 14,303 repeat sales by

dwelling types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

3.18 Geographical distribution of the 14,303 repeat sales by

transaction prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

4.19 A 3-Dimensional plot showing the absolute locational

values estimated by the ETV hedonic yearly model . . . . . . . . . . . . . . . 164

4.20 A 3-Dimensional plot showing the absolute locational

values estimated by the SCS hedonic model for year 2005 . . . . . . . . . 165

4.21 A 3-Dimensional plot showing the absolute locational

values estimated by the Q-hybrid yearly model . . . . . . . . . . . . . . . . . . 165

4.22 A graph showing the annual indices of the various models . . . . . . . . . 175

4.23 A graph showing the semi-annual indices of the various models . . . . . 175

4.24 A graph showing the quarterly indices of the various models . . . . . . . . 176

xvi

4.25 A graph showing the monthly indices of the various models . . . . . . . . 176

5.26 A graphical presentation of the housing starts . . . . . . . . . . . . . . . . . . . . 242

5.27 A graphical presentation of property price index . . . . . . . . . . . . . . . . . 242

5.28 A graphical presentation of material cost index . . . . . . . . . . . . . . . . . . 243

5.29 A graphical presentation of interest rate . . . . . . . . . . . . . . . . . . . . . . . . 243

5.30 A graphical presentation of the percentage of building

warrants granted . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244

5.31 A graphical presentation of the number of housing stock . . . . . . . . . . . 244

5.32 A graphical presentation of the average number of days

properties stay on market . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245

5.33 A graphical presentation of the volume of properties sold . . . . . . . . . . 245

5.34 A graphical presentation of the growth in GDP . . . . . . . . . . . . . . . . . . 246

5.35 A graphical presentation of the growth in population . . . . . . . . . . . . . . 246

6.36 Geographical distribution of the 75,330 sales by the various

index areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359

6.37 The Aberdeen housing price index for the five areas . . . . . . . . . . . . . . 369

6.38 A comparison of the index series for Aberdeen (local),

Scotland (regional) and UK (national) . . . . . . . . . . . . . . . . . . . . . . . . . 371

6.39 A comparison of the quality-adjusted and the ASPC

average index series for the Aberdeen housing market . . . . . . . . . . . . . 371

6.40 Volume of property transactions in Aberdeen from 2007Q1

– 2012Q1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375

xvii

LIST OF TABLES

2.1 Property characteristics included in previous studies . . . . . . . . . . . . . . . 26

2.2 Neighbourhood and accessibility variables used in previous

studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

2.3 The various hedonic methods and models . . . . . . . . . . . . . . . . . . . . . . . . 48

2.4 Some recent studies on house price index construction methods . . . . . . 75

2.5 Definitions and Code Names of Variables Included in the

model for estimating the Halifax indices . . . . . . . . . . . . . . . . . . . . . . . . . 87

3.6 The ASPC housing market areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

3.7 Distribution of property status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

3.8 Missing values in the dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

3.9 Magnitude and significance of the dropped variables with

the ETV hedonic model with 5,956 observations . . . . . . . . . . . . . . . . . 127

3.10 Frequency distribution of newbuilding, hasgrage and

hasgarden dummy variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

3.11 Frequency distribution of the dwelling type dummy variable . . . . . . . . 128

3.12 Frequency distribution of the main variables . . . . . . . . . . . . . . . . . . . . 128

3.13 Number of dwellings and sales, 2000:I - 2010:IV . . . . . . . . . . . . . . . . . 132

3.14 Descriptive statistics of selling price over the years . . . . . . . . . . . . . . . 137

xviii

3.15 Descriptive statistics of selling price over the years . . . . . . . . . . . . . . . 138

3.16 List of the independent variables and their description . . . . . . . . . . . . . 139

3.17 Magnitude and significance of the neighbourhood and

accessibility variables sourced from the SNS . . . . . . . . . . . . . . . . . . . . 141

3.18 List of the independent variables and their description . . . . . . . . . . . . . 152

4.19 A comparison of the ETV, SCS and Q-hybrid hedonic

estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

4.20 A test of the assumption of constancy of implicit prices of

the housing attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171

4.21 Accuracy of the house price index construction models

using the MSE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

4.22 The Morgan-Granger-Newbold Correlation Tests . . . . . . . . . . . . . . . . . 186

4.23 Tests of disaggregation of price trends over time . . . . . . . . . . . . . . . . . 190

4.24 Estimates of house prices, returns and volatilities across

time aggregation and the various methods . . . . . . . . . . . . . . . . . . . . . . 194

5.25 Published studies on housing supply . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

5.26 Definition of data variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238

5.27 Summary statistics of data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241

5.28 Unit root tests for stationarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248

5.29 Empirical regression results using the IV approach . . . . . . . . . . . . . . . 250

5.30 Price elasticities of supply in Aberdeen . . . . . . . . . . . . . . . . . . . . . . . . 257

xix

6.31 List of the independent variables and their description . . . . . . . . . . . . . 360

6.32 The House Price Index Areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362

6.33 Rate of price changes in UK, Aberdeen and the surrounding

areas for 2012Q1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373

6.34 Constant-Quality House Prices in Aberdeen and the

surrounding areas for 2012Q1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374

6.35 Volume of sales by property in Aberdeen and surrounding

areas for 2012Q1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376

xx

Chapter 1

INTRODUCTION

1.1 PRELIMINARY

This chapter outlines the main elements of the thesis. It begins with the research back-

ground and leads to the research motivation, aim and objectives. The chapter will then

discuss the methodology adopted and the contributions of the study. At the end, the struc-

ture of the thesis will be outlined.

1.2 RESEARCH BACKGROUND

Real estate is so important a subject that it cannot be left out in any serious macroeconomic

deliberation and the collective quest for wealth creation and economic development. This

is true whether in the advanced or developing world. Land, for example, is a primary

commodity that provides space for human and economic activities. Thus, for many people

around the world it is a very strategic economic asset. It is, therefore, not surprising that in

many countries in the developing world, landed property accounts for about 50% to 75%

of the national wealth (Bell, 2006).

Real estate can be broadly grouped into residential real estate and non-residential real

estate (Brueggeman and Fisher, 2001). Residential real estate includes flats, single-family

1

houses and multi-family properties such as apartment blocks. This real estate provides

accommodation for households and is often called "housing". The non-residential real

estate includes commercial real estate such as hotels, factories, office, warehouses and

retail buildings, agricultural real estate, land and corporate real estate like hospitals and

universities. This study exclusively focuses on housing.

Housing forms a major component of the real estate market. In the UK, the most

valuable asset of the nation’s wealth is housing with a total value of £4,048 billion in 2010,

representing about 61% of the total wealth in the UK (ONS, 2010). Various studies in-

cluding Green (1997) and Coulson and Kim (2000) have established the effects of housing

investment on a country’s economic growth. These studies note that housing investment

may stimulate GDP growth more than other types of investment because a lot of jobs, both

direct and indirect, are created by the sector. Housing has played a major role in shap-

ing the business cycles of countries like USA, Britain, New Zealand, Australia and Canada

(Hale, 2008).

Furthermore, price of housing has impact on the net wealth of people who own it. In

Britain, home ownership has grown rapidly in the UK between 1971 and 2010 from about

49% to 68.5% (ONS, 2011). Since about 75% of house purchases in the UK are financed

with a mortgage loan, and the average mortgage repayment accounts for almost 19% of

the average household income in the early 2000s (ONS, 2004), a change in house price

can affect the value of home owner wealth and consumption expenditure. As Tsatsaronis

and Zhu (2004) observe, since the behaviour of house prices affects individual expenditure,

the aggregate expenditure is also affected. House price changes affect consumer spending

2

and business investment patterns through the wealth effect, which in turn affects the wider

macro economy and the entire business cycle dynamics. The importance of accurate in-

formation on house prices to households, real estate developers, banks and policy makers

therefore cannot be over-emphasised.

Measuring of house prices is however not easy. This is because housing has a set of

unique characteristics with regard to location, structural composition as well as neighbour-

hood and environmental quality. That is, unlike the stock market for instance, the property

market is highly heterogeneous in the sense that there are different market segments and

property types and so no two properties can be considered as being the same. Two residen-

tial properties can be similar with regard to some of the physical characteristics but they

can never be the same because no two properties can be located at the same place. Also,

there is no central location or trading place where properties are transacted, and there are

also high transaction costs associated with property. The heterogeneity that exists in the

property market, the lack of central trading place, and the high transaction costs have made

gathering of information in the property market difficult.

Even in situations, where raw data about property transactions exists, it is normally

difficult to understand and easily make use of such data because of the housing market

heterogeneity. One way in which raw property data can be meaningful and useful is to

transform the data so as to construct property price indices (showing price movements over

time). However, housing, like other capital assets is a composite asset and so it is sold

wholly as one unit. When the factors that cause the changes in house prices are uncovered,

it helps to understand the house price changes. The factors that may cause changes in house

3

prices may be due to the shifts in the relationships between demand and supply; and may

also be caused by differences in dwelling characteristics and/or differences in or quality

of locational and neighbourhood attributes of the properties sold. It is therefore necessary

to remove the effect of the different and mixed of housing and locational attributes on the

house prices and to efficiently measure the variation in house prices caused by inflation on

a standardised basis.

The analysis of housing markets can be at national, regional or local level depending

on the data required. If it is at the macro level, then crude data aggregate might be sufficient.

But an index aggregated from reliable local information would be better. Reliable local

information is necessary when the level of analysis is local (Hwang and Quigley, 2006;

Maclennan, 1977). In examining the British housing market, Meen (1999) notes that the

housing market in the UK may be best described as a series of different local markets that

are interlinked rather than characterising it as a single national market. One of the important

features of house price indices is that they should be local in terms of spatial coverage

(Costello and Watkins, 2002). Indeed, making assumptions based on the aggregate nature

and behaviour of markets when constructing house price indices at the neighbourhood level

may make the resulting index unreliable (Munro and Maclennan, 1986).

In the literature, constant-quality house prices indices have been applied in several

areas. These areas include asset pricing, empirical tests of housing market efficiency, hedg-

ing mechanism for house price volatility, estimating of real estate derivatives and home

equity insurance, estimating the relationship between house prices and housing demand,

as well as modeling the supply of housing. One of the areas that has received little atten-

4

tion and in which results are inconclusive is the modeling of housing supply. As Ball et al.

(2010) point out, even simple issues and questions like the price elasticity of housing sup-

ply can produce so many different and wide range of estimates. Studies of housing supply

include Lee (1999), Malpezzi (1999), Mayer and Somerville (2000), Malpezzi and Maclen-

nan (2001), Neto (2005), Wigren and Wilhelmsson (2007), Ball et al. (2010), etc.. Also,

most all the studies on the supply of housing are conducted at the national and regional lev-

els. At the local housing markets, housing supply has been examined using cross-sectional

dataset.

In terms of index construction, there are three main quality-controlled index con-

struction approaches, the hedonic, repeat-sales and the hybrid which is a combination of

the first two approaches. With the hedonic technique, the price of the property is regressed

on the characteristics of the property and may be applied on a period by period basis or esti-

mated on pooled transaction data with time dummies as additional regressors. In each case

however, objections have been raised as to the difficulties involved in identifying all the

relevant price influences and the correct functional form (Case and Quigley, 1991; Shiller,

1993).

Due to the difficulty involved in identifying and gathering all the property and loca-

tional characteristics, the repeat sales method standardises the characteristics of the prop-

erties with reference only to themselves, by confining the analysis to properties which have

been sold at least twice (Bailey, Muth and Nourse, 1963). The method is however criticised

by throwing away all the single sale transactions thereby wasting a lot of transaction data.

Thus, the fraction of properties which are repeat sales is likely to be small in any market

5

(Palmquist, 1982; Mark and Goldberg, 1984; Case, 1986; Case and Shiller, 1987, 1989;

etc.).

The hybrid method utilises the desirable features of both the hedonic and the repeat

sales techniques to the estimation of the house price indices. It uses all available infor-

mation on property sales, whether single or repeat transactions and also capitalises on the

added precision when multiple transactions exist, by comparing transaction prices for the

same properties. The method is therefore used to estimate the house prices for a standard-

ised unit by combining data from single transactions where one sale is observed; multiple

where the physical and locational characteristics are the same; and multiple transactions

where the physical and locational characteristics of the property have changed between

sales. The hybrid method has however been rarely used in practice.

The various methods have been compared with respect to accuracy in different situ-

ations (see for example, Case and Quigley, 1991; Quigley, 1995; Haurin and Hendershott,

1991; Thibodeau, 1997; etc.) but the results seem to be inconsistent. Case, Pollakowski

and Wachter (1991) for instance find the repeat sales indices to increase more slowly than

those constructed using other methods. They also do not find any clear efficiency gains

from using the hybrid method. This is in contrast with Case and Quigley (1991) who find

the hybrid method to be more reliable.

1.3 MOTIVATION

In the UK, the main sources of house price information are the Halifax, Nationwide, HM

Land Registry and the Council of Mortgage Lenders (CML). Each provides house price

6

indices of the private housing market at regional and national levels. The HM Land Registry

and the Council of Mortgage Lenders (CML) indices are based on the repeat-sales and

mean prices of the transactions respectively. Therefore, the CML price information is

likely to be influenced by the differences in property characteristics and locational attributes

and so they are not constant-quality. Both the Nationwide and the Halifax indices are

constructed using the hedonic regression technique but their indices do not exist at the

local levels but only at the national and the twelve regional levels including Scotland.

Aggregate indices however should be composed of local indices and so require un-

derstanding of local index construction. Also, for local market analysis, local price indices

should be relied on. However, such local price indices are not readily available in the UK

(Costello and Watkins, 2002). Costello and Watkins (2002) therefore recommends that

efforts should be made to construct house price indices that can consistently achieve the

fundamental aims of monitoring changes in price at the local level in the UK. This thesis

contributes to the development of such local information.

The preceding section shows that the literature on house price index construction is

voluminous and there are several methods to produce house price index numbers. However,

the literature has reached no firm conclusion as to which method performs best in terms

of accuracy. This may be due to the way in which the index accuracies are measured.

An average index is usually used as a benchmark index against which the index from the

methods are measured. This is clearly problematic since the average index does not control

for property heterogeneity. There is therefore the need for further empirical studies on the

examination of index accuracy.

7

Beyond the issues of index construction technique selection and the need to monitor

house prices at the local level is the pooling of data across time in analysing trends and

volatilities in prices. Most studies in the literature have arbitrary pooled data into broader

representations of time to estimate indices. The perception is that pooling data together

helps to overcome the problem of small sample size, a common problem encountered in

studies using real estate transaction data. In doing this however, they implicitly assume

that the pooled sample will produce index numbers that are statistically equivalent to those

that would have been obtained from their constituent sub-samples. Since factors such as

buyer preferences and supply conditions play a crucial role in determining house prices,

and these may vary across time, it is necessary to test if pooling of data alters the predictive

performance.

The demand side of the housing market is an area which is more researched than the

supply side (DiPasquale, 1999). DiPasquale also notes that the empirical evidence on the

demand side of the housing market is much more convincing than the small evidence we

have on the supply side. Simple questions about the price elasticity of housing supply can

even produce a wide range of estimates (Ball et al., 2010). Most of the housing supply

studies, particularly in the UK, are conducted at the national and in some limited cases

regional levels (see for example Malpezzi and Maclennan, 2001; Tsoukis and Westaway,

1994; Bartlett, 1988) as compared to local housing market studies. Using aggregated data

is a problem because it hides interesting variation in the timing of real cycles across regions

and also shrouds inter-metropolitan area movements in population (Goodman, 1998). This

8

study will provide further evidence about housing supply at the local level using the house

price index that will be constructed together with other local variables.

1.4 RESEARCH AIM AND OBJECTIVES

The aim of this research is to estimate house price trend and use it to understand the supply

side of the Aberdeen housing market. The realisation of the aim requires the pursuit of the

following objectives:

(i) To review the state of current knowledge on house price index construction methods

and identify areas where further knowledge or research is required.

(ii) To apply the different methods to the same dataset, to empirically examine the

accuracy of the various methods and to examine the effect of pooling observations on

house prices across time.

(iii) To apply the best index method according to objective (ii) to produce constant-

quality house price indices for the Aberdeen housing market from which Aberdeen

Housing Market Reports are published quarterly.

(iv) To use the house price index series to learn more about the supply side of the

Aberdeen housing market.

9

1.5 RESEARCH METHODOLOGY

Research methodology has been defined differently by different people. For example, John-

ston (1983:4) defines methodology as a set of rules and procedures that indicates how re-

search and arguments are to be conducted within a discipline, that is, how information

can be collected and organised”. Frankfort-Nachmias and Nachmias (1996:13) also define

methodology as a system of explicit rules and procedures upon which research is based

and against which claims for knowledge are evaluated”. Research methodology can, thus,

means a collection of rules and procedures that are used to collect, organise and analyse

data in order to provide new information or confirm or refute an existing theory in a partic-

ular field.

There are three methodologies used to conduct research in the social sciences. These

are the quantitative research methodology, qualitative research methodology and the mixed

methodology.

The quantitative methodology, according to Bryman (2001:20), entails a deductive

approach to the relationship between theory and research, in which the accent is placed

on the testing of hypotheses. Barnes (2000) and Creswell (1994:2) also define quantitative

research methodology to include concepts composed of variables, measured with numbers

and analysed with statistical procedures in order to determine whether the theory hold or

not. This methodology can provide wide coverage of the range of situations, can be fast and

economical, and may be of considerable relevance to policy decisions, particularly when

statistics are aggregated from large samples. On the other hand, the methodology may not

10

be very effective in understanding processes or the significance that people attach to actions

(Easterby-Smith et al., 1991).

Qualitative research methodology on the other hand, involves an inquiry process of

understanding a social or human problem, based on building a complex, holistic picture,

through the formation of words and reporting of detailed views of informants, conducted

in a natural setting (Creswell, 1994). The methodology has the ability to look at change

processes over time, to understand people’s meanings, to adjust new issues and ideas as they

emerge and to contribute to the evolution of new theories. However, data collection can take

time and resources, analysis and interpretation of data may be very difficult. Also, many

especially policy-makers may give low credibility to studies based on this methodology

since it is more subjective to the researcher’s judgement (Easterby-Smith et al., 1991).

It should however be noted that, the choice of a particular research methodology is not

contingent upon the advantages and disadvantages associated with them. The overarching

factor that determines the particular methodology to be adopted is the research problem(s),

question(s) or aim and objective(s) of the study (Mertens, 2003; Creswell, 2003, 2007).

Thus, no particular methodology is necessarily superior to the other (Punch, 1998). Given

the objectives of this research, the quantitative research methodology is ultimately adopted

for this study. The study employs a number of statistical techniques based on quantified

variables and hypothesis testing, as the main methodologies in addressing the research

objectives set out in the previous section. The research procedures are implemented in

three main stages.

11

The first stage consists of an extensive and critical review of the relevant literature

and the objectives are two-fold: (i) to explore studies that are related to the present study;

and (ii) to identify the theoretical framework relevant to the research objectives and to help

design the appropriate research strategy.

In the second stage, descriptive statistics are used to describe the main features of

the data and provide simple summaries about the sample and measures in relation to the

research objectives. The aim of this stage is to provide a foundation for empirical analysis.

In the third stage, which is the empirical stage, empirical models are developed, ap-

plied, compared and tested. Empirical findings are discussed and interpreted to see whether

these results are consistent with theoretical indications and previous empirical findings.

1.6 CONTRIBUTIONS OF THIS THESIS

One major significant and practical contribution of this thesis is the construction of house

price indices for the Aberdeen housing market. Due to the important role housing plays

in the economic development of a country, the availability of a robust and reliable house

price indices is crucial. However, in the whole of Scotland, the Nationwide and/or Halifax

Building Society Regional index for Scotland is usually employed. Local indices are vir-

tually absent in the UK. Constructing a local house price index for the Aberdeen housing

market would, therefore, provide market participants and policy makers with better infor-

mation about house price changes in the local housing market. This thesis contributes to

the construction of house price indices for the Aberdeen housing market. Based on this,

12

the Aberdeen Housing Market Report (AHMR) are published on quarterly basis which the

practitioners and academics can use for housing market analysis.

The thesis is one of the few studies that extensively compares the various house price

index construction methods by using UK data and at the local level. Furthermore, this

study examines a major issue that most index construction studies in the literature ignore

- that is, the issue of pooling data together across time. Most studies in the literature have

arbitrary pooled data together across time to estimate indices. The perception is that pool-

ing data together within a time period helps to overcome the problem of small sample size,

a common problem encountered in studies using real estate transaction data, and which

can cause variability of index numbers at high frequencies. In doing pooling data together

within time, however, the studies implicitly assume that the pooled sample will produce in-

dex numbers that are statistically equivalent to those that would have been obtained from

their constituent sub-samples. Because the implicit prices of the housing characteristics are

likely to change at the broader level of temporal aggregation, this assumption is question-

able. The choice of temporal aggregation should be seen as a trade-off between bias and

variance. This thesis moves a step further to test if temporal aggregation really matters and

so contributes to knowledge regarding house price index construction.

This thesis also examines the relationship between house prices and housing starts.

As noted above, almost all the studies about the supply of housing are conducted at the

national levels and in some limited cases at the regional and local levels. However, even

though disaggregated-level studies are more desirable, it seems very difficult for such local

level studies to be conducted using consistent time series data. This thesis contributes to

13

the already existing literature by employing time series data and at the local housing market

level. This is also one application that helps our understanding of the housing market.

1.7 THESIS STRUCTURE

Chapter 2 provides a literature review of the various house price index construction meth-

ods. The chapter begins with a historical overview of the development of hedonic pricing

model and the process of hedonic modelling. It also discusses the problems often en-

countered in hedonic modeling such as the selection of the appropriate functional form, the

problem of heteroskedasticity, omitted variable bias, multicollinearity, as well as discussion

on the property attributes that has been included in previous studies. The three main house

price index construction methods (the hedonic, repeat-sales and hybrid) are presented. The

relative advantages and disadvantages of the various methods are discussed. Issues often

encountered in house price index construction such as the issues of measuring the accu-

racy of various index methods and temporal aggregation are also discussed. The chapter

also discusses existing house price indices in UK and some selected countries. Notably, the

Halifax and Nationwide indices produced for the UK housing market are discussed as well

as the S&P/Case-Shiller indices in the U.S.

Chapter 3 introduces the different geographical areas, the Aberdeen housing market.

In this chapter also, the data used for the house price index construction is prepared and

described. The source of the data and the raw dataset and its limitations are all discussed.

The data cleaning exercise is described in this chapter. By doing this, some variables

14

are dropped and new variables are generated to aid the analyses, and the final dataset is

presented and described.

Chapter 4 continues with the empirical analyses. Five models from the three main

house price index construction methods are implemented at various levels of temporal ag-

gregation. The models are the explicit time variable (ETV) hedonic model, strictly cross-

sectional (SCS) hedonic model, ordinary repeat-sales (ORS) model, weighted repeat-sales

(WRS) model and the Quigley’s hybrid (Q-hybrid) model . The implicit prices of the hous-

ing and locational attributes are tested to find out if they are constant over time. In the

chapter also, the out of sample technique is used to measure the mean squared error (MSE)

of the various index construction models to examine the accuracy of the different index

models. The effect of temporal aggregation on house price indices is also examined in

this chapter. The chapter also highlights how the thesis has contributed to the Aberdeen

Housing Market Report (AHMR) with the detailed report presented at the Appendix.

Chapter 5, the penultimate chapter, focuses on an application of house price indices.

The chapter uses constant-quality house price indices for the Aberdeen City district to-

gether with other variables to examine the determinants of housing construction and to

estimate the price elasticity of supply for the Aberdeen local housing market using dif-

ferent model specifications. The chapter compares the price elasticities of supply for the

Aberdeen local market with the price elasticities of supply for other local areas in the UK

and also highlights the need for housing market analysis to be confined to the local or dis-

trict levels.

15

Chapter 6 summarises and concludes the thesis. It provides several conclusions that

are drawn from the research. The main findings are presented in relation to the research

objectives. This is followed by policy implications and contributions the research makes

to knowledge. Finally, limitations of the research as well as the potential areas for future

research are provided to conclude the thesis.

16

Chapter 2

HOUSE PRICE INDEX CONSTRUCTION

METHODS

2.1 INTRODUCTION

It was pointed out in Section 1.5 that establishing the relevant existing rules and knowl-

edge in the form of literature review is necessary to identify the theoretical framework

relevant to the research and to help design the appropriate research strategy. This chap-

ter establishes that by providing an overview of the various house price index construction

methods. These methods are the average method, the hedonic method, the repeat-sales

method and the hybrid method. The chapter begins by providing an overview of the at-

tributes which determine the quality of a house price index in Section 2.2. The purpose for

providing this overview is to help identify the desirable properties of an index computation

method against which to benchmark the different methods.

In section 2.3, the hedonic pricing theory is discussed. This is important because

most of the house price index construction methods directly or indirectly rely on regres-

sion analysis. The average index construction method is described in Section 2.4, with

the advantages and disadvantages inherent in using the method. In Section 2.5, the hedo-

nic method is described. An overview of the hedonic method, functional form selection

and housing characteristics used in hedonic modeling are all discussed. The two varia-

tions of the hedonic price index construction method, the explicit time variable and strictly

17

cross-sectional are also presented and discussed in this section. Finally, the advantages

and disadvantages of using the hedonic method to construct house price indices are also

discussed in Section 2.5.

Section 2.6 presents the repeat-sales method of house price index construction. The

unweighted or original repeat-sales (ORS) and the weighted repeat sales (WRS) models are

discussed in this section, as well as the advantages and disadvantages of using the repeat-

sales method to construct house price indices.

In Section 2.7, the hybrid methods of constructing house price indices are presented.

Notably, the methods proposed by Case and Quigley (1991), Quigley (1995), Hill, Knight

and Sirmans (1997), and Englund, Quigley and Redfearn (1998) are all discussed. The

relative advantages and disadvantages of using the hybrid method are also discussed in this

section.

Section 2.8 compares the various house index construction methods based on previ-

ous empirical findings. It identifies and summarises the findings from some previous house

price index construction studies and uses them to compare the various house price index

methods. The absence of the "true" house price trend and the issue of temporal aggregation

are also briefly discussed. These two issues are of particular importance since they may

influence or determine the accuracy of the various index methods.

Section 2.9 discusses the existing indices in some selected countries such as the USA,

Sweden, Germany and New Zealand. Among other indices, the S&P/Case-Shiller indices

in the U.S. is discussed. In Section 2.10 also, existing house price information in the

UK is discussed. Notably, the Halifax, the Nationwide and the Land Registry house price

18

indices are the main existing indices discussed here. The coverages and methodologies of

these indices as well as their limitations are all discussed. The aim is to show the gap that

exists in the market. Also, the EU directive on national house price indices and standard

methodologies are briefly mentioned. Section 2.11 summarises the chapter.

2.2 ATTRIBUTES WHICH DETERMINE THE QUALITY

OF A HOUSE PRICE INDEX

Index numbers are used to aggregate detailed information on prices and quantities into

scaler measures of price and quantity levels or their growth (Diewert, 2008). Presenting

these detailed information in the form of indices makes it easy to see how the prices and

quantities have changed over time and also to facilitate comparisons of series with different

units of measurement (Brooks and Tsolacos, 2010). Index numbers are widely used to

display series for gross domestic product (GDP), consumer prices, stock prices, exchange

rates, house prices, etc. It is very important to ensure that the index numbers relied upon

are accurate, timely and robust. Obviously, the methodology used to construct the indices

will determine how accurate the index numbers are.

The attributes of a good house price index method should include (Bourassa et al.,

2006):

• Data requirement

The first attribute considered here is that data is difficult to come by, and so an index

construction method should not require huge amounts of data. If the data requirement is

19

huge, then the application of that index method would become difficult. If the method

requires specific property characteristics and such variables are absent, it may cause bias in

the estimated index numbers. More so, the subset of properties that form the basis for the

construction of the index should be a representative of the inventory of properties. That is,

there should be no sample selection bias in the construction of price indices.

• Standardisation

Secondly, one of the most important features for any house price index is that it

should be able to track price changes for a house that has the same set of characteristics

over time. That is, they should be standardised to account for changes in the stock and

neighbourhood mix of transactions in different time periods. In this case, the price changes

would only result from changes in the implicit prices or valuations of characteristics, and

not from differences in the characteristics of properties that have transacted in the various

periods. The index method therefore should be able to produce constant-quality indices.

• Implementation

Another attribute considered is that an index method should be transparent and easy

to construct or implement and it should not require complex estimation techniques or tech-

nical skills. This is because when an estimation technique is very complex, it becomes

very difficult to identify when a mistake is made, and it will be difficult for people without

such technical knowledge to replicate and implement the model. Easy methods therefore

facilitate transparency and makes replication easily possible.

20

• Revision

The fourth attribute is the stability of the index numbers when they are revised. Index

numbers are typically revised periodically when either the method used to incorporate the

data or the data themselves are updated (Clapham et al., 2006). Information does not come

in all at once and so when new data arrives after an index has already been published,

the index has to be reestimated to incorporate the new information (Shiller, 1993). In

doing this, the extent to which the index numbers change as additional data are included

in the index estimation is a great concern. That is, the house price index construction

method should not change the historical index values when new transaction data arrives.

When historical index numbers change after the arrival of new information, it becomes

a big concern for market participants and policy makers when these numbers are relied

upon in policy formulation, investment decisions and economic modeling. For example,

if government agencies use the house price index to construct a cost of living index that

serves as a basis for wage and benefits adjustments, the accuracy of the previous wage and

benefits adjustments will be questioned when the index numbers are changed. Again, when

the price indices are used to hedge against property risk, then changes in the historical index

numbers will be a big problem for all the users. Therefore, the particular index construction

method should not change the historical or already published index numbers significantly.

Thus, the desirable properties of an index computation method are that the method

should (i) require less data in its implementation (ii) use data which is representative of

the inventory (iii) be standardised for quality (constant-quality) (iv) be easy to implement

21

(v) not change the historical numbers when revised. It is expected that at least one of the

methods to be reviewed fulfills the criteria best.

Most of the house price index construction methods to be reviewed are directly or

indirectly based on the hedonic regression analysis and so the hedonic pricing theory is

discussed in the next section.

2.3 THE THEORY OF HEDONIC PRICING MODELING

2.3.1 An overview of the hedonic method

The first study found1 to have applied the hedonic model is the study by Haas (1922).

According to Colwell and Dilmore (1999), Haas (1922) applied the hedonic model in the

area of land economics to value farm lands. Berndt (1991) also mentions Waugh (1928) as

one of the first researchers to apply the hedonic model in the area of agricultural economics

to find out that quality variation existed among lots of asparagus. Several studies also

mention Court (1939), an automobile industry analyst, as one of the early people to have

used the hedonic model (see Griliches, 1958; Goodman, 1998 and Malpezzi, 2003).

Lancaster (1966) laid the theoretical foundation of the hedonic modeling in his con-

sumer theory studies where he provides a micro economic foundation for estimating the

value of utility generating characteristics. This was followed by another study in 1974

by Rosen who focused on housing characteristics. Rosen (1974) used the hedonic pricing

model to estimate the implicit prices associated with housing characteristics. The study of

1 There might be earlier studies nobody has detected

22

Rosen, therefore, is often cited to have developed the theoretical framework of the hedonic

pricing model and provides the foundation for non-linear hedonic pricing models (Sirmans

et al., 2005).

Housing is bought as a bundle of characteristics, and each of these attributes con-

tributes to the price of the property. A property is therefore purchased not for the prop-

erty’s sake but for the attributes that make up the property. However, since the individual

attributes do not have explicit prices, the price one pays for a particular property is the

sum of the implicit prices the market gives to the various attributes that make up the bun-

dle (Rosen, 1974). Since most of these housing characteristics are not traded explicitly

and their prices could not be observed directly on the market, the hedonic pricing model is

applied to estimate the marginal contribution of each property and neighbourhood charac-

teristic to the house price. The hedonic price function theory therefore provides the frame-

work for the analysis of differentiated goods such as housing, whose individual features we

do not have observable market prices for (Pagourtzi et al., 2003).

The hedonic technique involves the estimation of some regression relationship be-

tween the transaction prices Pt, of properties at time t, and their physical and locational

characteristics, X:

Pt = f(X) (2.1)

Like this study, most of the studies using the hedonic pricing model use the para-

metric approach. By this approach, the regression curve, f(.), is assumed to have some

prespecified functional form that is fully described by a finite set of parameters, which are

23

usually coefficients of the independent variables (Wooldridge, 2009). Accurate interpreta-

tion of the relationship between house prices and the housing characteristics depends on

two basic things; the inclusion of the "correct set" of property characteristics, X , and the

correct functional form, f(.), for the hedonic regression (Case and Quigley, 1991).

2.3.2 The hedonic method and housing characteristics

There are two main variables included in hedonic modeling. These are the dependent and

the independent variables. The dependent variable is usually the price of the property.

However, depending on the availability of the transaction price information and the na-

ture of the studies undertaking, measures such as asking price, assessment value, or rent

can be used. The independent variables are numerous. Wilhelmsson (2000) identifies four

main factors that affect demand for properties and hence the price to include the property’s

structural attributes, its location or neighbourhood amenities, its environmental attributes

and macro attributes like gross domestic product (GDP) and interest rate. The macro at-

tributes like GDP and interest rate affect the entire economy and are discussed in Chapter 6.

Only the structural or physical property characteristics, and its location in terms of neigh-

bourhood characteristics and accessibility, and environmental attributes are reviewed in this

section. However, as will be discussed below, in the empirical hedonic modelling in this

thesis, only the absolute locational values in the form of the X and Y coordinates expanded

to a third-degree polynomial are used to control for location and accessibility.

24

Physical characteristics

The physical characteristics of a house influence the price to be paid for the house.

Table 2.1 shows the structural physical characteristics employed in previous studies. While

the studies and the variables in the table are not exhaustive, the table is very useful in that

it serves as a guide in selecting the physical housing characteristics for the study. These

physical characteristics can be grouped as; accommodation and size; structural improve-

ment and materials used; and age and condition of the structure.

• Accommodation and size

The level of accommodation provided by a house can influence the value or price to

pay for the house. Such factors include the number of bedrooms and other rooms, the num-

ber of floors, floor size, land area etc. Generally, individual buyers have their own needs,

tastes and preferences concerning the amount of accommodation. Such accommodation

needs, tastes and preferences are influenced by the size of the family, prestige and status of

the individual etc. They therefore restrict their enquiries to properties having the number

of rooms or size of property that they want. If individuals get the amount and size of ac-

commodation they want, they will be willing to pay higher value for it than they would pay

for property with more or less the amount of accommodation they require. The number of

rooms (bathrooms, public rooms and bedrooms) dominantly affects price in the positive di-

rection (Sirmans et al., 2005). This means that, all other things being equal, as the number

of rooms increase, the price of the property also increases.

25

Tab

le2.1

.P

roper

tych

arac

teri

stic

sin

cluded

inpre

vio

us

studie

s

Au

tho

urs

Lo

tF

loo

rL

ivin

gO

ther

Ag

eo

fD

wel

-P

res-

Po

ol

Hea

-C

on

di-

Do

ub

leN

o.

of

No

.o

fN

o.

of

Gar

age/

Fir

e-B

ase-

size

area

area

area

pro

p-

lin

gen

ceo

fti

ng

tio

no

fg

lazi

ng

bed

-b

ath

-o

ther

on

site

pla

cem

ents

erty

typ

egar

den

syst

emd

wel

lin

gro

om

sro

om

sro

om

sp

ark

ing

Pal

mq

uis

t

(19

80

)×

××

××

××

×C

ase

and

Qu

igle

y

(19

91

)×

××

×A

dai

ret

al.

(19

96

)×

××

××

××

××

An

gli

nan

d

Gen

çay

(19

96

)×

××

××

Hil

let

al.

(19

97

)×

××

Hen

neb

erry

(19

98

)×

××

××

En

glu

nd

etal

.

(19

98

)×

××

××

××

×

26

Au

tho

urs

Lo

tF

loo

rL

ivin

gO

ther

Ag

eo

fD

wel

-P

res-

Po

ol

Hea

-C

on

di-

Do

ub

leN

o.

of

No

.o

fN

o.

of

Gar

age/

Fir

e-B

ase-

size

area

area

area

pro

p-

lin

gen

ceo

fti

ng

tio

no

fg

lazi

ng

bed

-b

ath

-o

ther

on

site

pla

cem

ents

erty

typ

egar

den

syst

emd

wel

lin

gro

om

sro

om

sro

om

sp

ark

ing

Bo

ura

ssa

etal

.(1

99

9)

××

××

Des

Ro

sier

s

etal

.(2

00

0)

××

××

×F

letc

her

etal

.

(20

00

a)×

××

××

××

×F

letc

her

etal

.

(20

00

b)

××

××

××

××

×M

cClu

skey

etal

.(2

00

0)

××

××

××

×W

ilh

elm

sso

n

(20

00

)×

××

×S

teven

son

(20

04

)×

××

××

××

××

Bo

ura

ssa

etal

.(2

00

6)

××

××

Wil

hel

mss

on

(20

09

)×

××

××

27

Rationally, as the number of rooms increases, the utility of homebuyers are expected

to increase as well because they will be able to allocate new rooms for other uses and ease

the pressure on existing rooms. However, from theory, the utility should diminish beyond

a certain number of rooms (Wooldridge, 2009).

• Structural improvement and materials used

The materials that go into the construction of a property and the structural improve-

ment made to the property affect the price to pay for the property. Physical factors such as

the type, style and quality of floor finishes, roof, ceilings etc. will influence the amenity to

be derived from living in a particular property and hence the price to pay for that property.

Structural improvements like the availability of garage, swimming pool, gardens, fence

wall etc. all affect the value of residential accommodation. Usually, the availability of

improvements like swimming pool, garages and gardens in a property will make rational

buyers pay higher price for such a property than they will pay for similar property without

such improvements, all other things being equal (Sirmans et al., 2005).

• Age and condition of structure

The age and condition of a property will also influence the price to pay for the prop-

erty. In examining the factors that are mostly included in hedonic models to determine

house prices, Sirmans et al. (2005) find that the age of property influence the value of the

property mostly in the negative direction. This is not surprising because as the age of the

property increases, the economic value of the property decreases and hence the utility to be

28

derived from the property decreases. Furthermore, homebuyers would have to spend addi-

tional money on maintenance when properties are old. They are, therefore, willing to pay a

price lower than a new property of similar but new attributes. Apart from the age, the con-

dition of the structure also affects the value to pay for the property. If a property is old but

has seen a lot of refurbishment, it will demand higher price than a similar property that has

not been refurbished. Furthermore, the time the property was constructed also affect the

vintage or the quality and style that are embedded in the dwelling.

As shown in Table 2.1, lot size, floor area, age of property, dwelling type, heating

system, quality of dwelling, number of bedrooms, number of bathrooms, as well as the

availability of garage are among the physical property variables mostly included in hedonic

models.

Location

The importance of location in real estate is well known. There is a real estate adage

that states that the three most important factors which determine property values are (i) lo-

cation, (ii) location and (iii) location. Spatially, no two properties are the same and indeed,

there is a consensus among valuers that location is the most important factor in property

value determination (McCluskey et al., 2000). The importance of location is evident by the

fact that location physically fixes a property in space and thereby defines its distance from

features such as commercial, transportation and leisure activities. Again, in cases where

houses within a particular sub-market or neighbourhood are homogeneous, many of the

amenities that are common to those properties are best represented by location (Gelfand et

29

al., 1998). Location is an inherent attribute of a house which directly determines the quality

and hence the market value of the house.

The locational influences on the value of residential property may arise from a num-

ber of sources. These are grouped under neighbourhood quality and accessibility (Mc-

Cluskey et al., 2000). Table 2.2 presents some of the neighbourhood and accessibility

factors that have been used in previous studies.

• Neighbourhood quality

Can (1990) defines neighbourhood in an economic sense as an area within which

relatively the same prices prevail for properties that permit approximately the same types

of uses and socioeconomic status. In his extensive review of literature to determine the

existence of neighbourhoods or submarkets, Jones (2002) concludes that there is a little

consensus on how neighbourhoods or submarkets should be identified. Neighbourhood

quality can be argued as an unobservable variable (Dubin and Sung, 1987). However, sub-

jective assessments can be made about some of these perceived quality indicators. The

neighbourhood quality factors that influence residential property values include (a) expo-

sure to adverse environmental factors (b) neighbourhood amenities (c) perceived levels of

neighbourhood security etc. Depending on the presence or absence of these amenities,

residential properties may reduce or increase in value (Gallimore et al., 1996).

Adverse environmental factors are those factors that impact negatively to the health

and comfort of the inhabitants within the neighbourhood. Such adverse environmental fac-

tors include traffic noise, air pollution, vibration, noise level, aesthetic and barrier effects

30

Table 2.2. Neighbourhood and accessibility variables used in previous studies

Authours Study area Variables used

Palmquist (1980) Washington, USA Distance to nearest park, location to highway,

access to recreational facilities

Case and Quigley (1991) Hawaii, USA Distance to shore

Des Rosiers and Canada Distance to highway exists, major work places,

Theriault (1992) parks, primary schools, Regional and

neighbourhood shopping centres, welfare housing

Garrod and Willis Britain Approximate distance from nearest urban centre,

(1992) post office and schools. The presence of pub,

post office, river, canal etc.

Adair et al. (1996) Ireland Aggregate accessibility index

So et al. (1997) Hong-Kong Walking distance to bus stations, availability of

car parks, car sport facilities, shopping centre, and

swimming pool

McCluskey et al. (2000) Northern Ireland X and Y coordinates

Bateman et al. (2001) Glasgow Neighbourhood variables like people who own

cars, unemployment rate, number of young

families, people who do not own property, etc.

Frew and Wilson (2002) Portland, USA Distance to city centre, highway and highway

intersection

Bourassa et al. (2003) New Zealand Distance to CBD and neighbourhood variables

like population densities, % of unemployed, rooms

per house, ethnic composition, % of people who

receive support, home ownership rate, etc.

Day (2003) Glasgow Straight line distance to city centre, car travel time,

walking distance

Wilhelmsson (2004) Stockholm, Sweden Distance from CBD

Rodriguez and Targa Colombia Shortest walking time to nearest BRT station,

(2004) travel time to CBD, and other railway stations

Gibbons and Machin London, UK Straight line distance to nearest railway station

(2005)

Bourassa et al. (2006) New Zealand Distance from CBD and distance from subcentre

Dehring and Dunse (2006) Aberdeen, Scotland Proximity to urban parks

31

(Wilhelmsson, 2002). These factors have negative relationship with residential property

values. That is, because such factors directly affect the health of people, if such factors

are high within the neighbourhood, the health and comfort of the people will be greatly at

risk and so people will not be willing to live in such places. They will only be willing to

live there when they are compensated by paying less to live there. For example, Wilhelms-

son (2002) empirically examines the effect that traffic noise has on single-family houses

in the Stockholm Municipality and finds that a single-family house that values at SEK

975,000 would be sold for SEK 650,000 if the property is located on or near a highly noise

road. Traffic noise in this case reduces single-family property values by as much as 30%.

Other empirical studies that have examined the effects of such environmental attributes on

house prices and estimated the willingness-to-pay (WTP) for negative externalities include

Hughes and Sirmans (1993) and Palmquist (1992); etc. Similarly, positive environmental

factors like clear air, good weather conditions etc. also affect property prices positively.