House price dynamics Amsterdam considered. How does
district attractiveness affect the Amsterdam ripple effect?
Author:
Rob van Diepen
10538887
Supervisor:
J.B.S. Conijn
Master Thesis Finance and Real Estate Finance
University of Amsterdam
Faculty of Economics & Business
July, 2018
2
Statement of Originality
This document is written by Student Rob van Diepen who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Abstract
This thesis investigates the ripple effect within the city of Amsterdam based on attractiveness indicators of districts within Amsterdam. It tries to shed more light on the way in which the Amsterdam ripple effect develops itself. The used dataset contains data of housing transactions within Amsterdam over the period 1995-2017 in combination with attractiveness indicators. The ripple effect has been investigated by constructing quarterly returns based on Hedonic Price Indices. Vector Autoregression models in combination with Granger Causality tests show a pattern of housing price developments rippling out from the most attractive districts to the least attractive districts within Amsterdam. Impulse Response Functions are used to simulate price shocks within certain attractiveness parts in Amsterdam to show the effect of price movements on other parts in Amsterdam with a different level of attractiveness. The results indicate that the level of attractiveness is an indicator in which way the housing price developments ripple out over the city of Amsterdam.
4
Table of Content
1. Introduction ... 7
2. Literature review ... 10
2.1 The Dutch housing market ... 10
2.2 The Amsterdam housing market ... 14
2.3 The ripple effect ... 16
2.3.1 Country level ripple effects ... 17
2.3.2 City level ripple effects ... 21
2.4 Factors influencing house prices ... 23
2.5 Factors influencing attractiveness ... 26
3. Description Amsterdam districts ... 28
3.1.1 Centrum-West ... 28 3.1.2 Centrum-Oost ... 29 3.1.3 Westerpark... 29 3.1.4 Bos en Lommer ... 29 3.1.5 Oud-West/ De Baarsjes ... 29 3.1.6 Geuzenveld/Slotermeer ... 30 3.1.7 Osdorp ... 30
3.1.8 De Aker/ Nieuw Sloten ... 30
3.1.9 Slotervaart ... 30
3.1.10 Oud-Zuid... 31
3.1.11 Buitenveldert/Zuidas ... 31
3.1.12 De Pijp/ Rivierenbuurt ... 31
3.1.13 Oud-Oost ... 31
3.1.14 Indische Buurt/ Oostelijk Havengebied ... 32
3.1.15 Watergraafsmeer ... 32
3.1.16 IJburg/ Zeeburgereiland ... 32
5 3.1.18 Oud-Noord ... 33 3.1.19 Noord-Oost ... 33 3.1.20 Bijlmer-Centrum ... 33 3.1.21 Bijlmer-Oost ... 34 3.1.22 Gaasperdam/ Driemond ... 34 4. Hypotheses ... 35 5. Data ... 36 6. Methodology ... 42
6.1 Hedonic Price Models ... 42
6.2 VAR model ... 43
6.3 Impulse Response Functions ... 44
7. Results ... 44
7.1 Results Hedonic Price Indices ... 45
7.2 Results ripple effect ... 49
7.3 Impulse Response Functions ... 52
7.4 Summary of Empirical Results ... 55
8. Robustness checks ... 57
8.1 Results Hedonic Price Indices ... 58
8.2 Results ripple effect ... 58
9. Conclusion ... 61
9.1 Results ... 61
9.2 Suggestions for further research ... 62
References ... 64
Appendices ... 72
Appendix 1 – T-test on comparison of mean returns ... 72
Appendix 2 – SD-test on comparison of Standard Deviation of returns... 73
Appendix 3 – Variables description Hedonic Price Indices ... 74
6
Appendix 5 – Granger Causality Wald test ... 75
Appendix 6 – Robustness Granger Causality Wald test ... 77
Appendix 7 – Impulse Response Functions table form ... 78
Appendix 8 – Unit-root test ... 80
Appendix 9 – Stability of the VAR-model ... 80 Appendix 10 – SD-test on comparison of Standard Deviation of returns for attractiveness groups 80
7
1. Introduction
Occasion
Since the big crisis of 2008, housing prices in the Netherlands have been recovering. However, different regions recovered at different speeds. The housing prices in Amsterdam recovered more rapidly than the rest of the Netherlands for instance. Between 2015 and 2017, the median house price in Amsterdam increased by 46 percent, while the median house price in the rest of the Netherlands increased with 5 percent on average according to Vastgoedactueel (2017). It is evident that some differences appear between regions, since supply and demand determine the price development in different housing markets. A so-called ripple effect can exist due to temporal differences in housing price developments between regions. Research by Teye et al. (2017a) showed that a ripple effect exists for the Netherlands, where the housing prices in the Netherlands are following the housing prices of Amsterdam. This means that there is a so-called lead-lag effect in housing prices between regions. This effect can be explained by factors that shift the housing demand from Amsterdam to surrounding areas. Such factors are according to Teye et al. (2017a) the lower affordability of housing in Amsterdam or spill-over effects from the expectations of housing price developments of home-owners. Housing price trends are rippling out over the Netherlands in this way.
This ripple effect is not only visible in the Netherlands in general, but even within Amsterdam differences in housing price recovery appeared. Some districts recovered more rapidly than others. Within a city ripple effects have been researched by Oikarinen (2005) and Liao et al. (2015) in foreign countries. Oikarinen (2005) found a ripple effect for the Helsinki metropolitan area in Finland and Liao et al. (2015) for Singapore. Later the same has been found for Amsterdam by Teye et al. (2017b) and Mathlener (2017). They found that some districts in the city of Amsterdam have a lead-lag effect on other adjacent districts.
That the ripple effect exists is clear, but very limited research has been conducted on the underlying economic mechanisms and possible explanations of the ripple effect. There is a missing gap in the literature about the underlying mechanisms and explanations of the ripple effect, where research could add to the current knowledge. Some research has been done by Liao et al. (2015) and Brady (2014) on factors as foreign investment, income, interest rates and unemployment. Teye et al. (2017b) and Meen (1999) suggested additional research on underlying factors of the ripple effect in their papers. The suggestion that they make is to include spatial interrelationships exist between districts. These spatial interrelationships might be caused by social-economic activities that include internal migration. The key factor behind internal migration is the utility maximizing principle for
8 inhabitants when they chose where they want to live. The attractiveness of different neighborhoods might be the missing link in the spatial interrelationship argument.
Research question and sub-questions
Dröes et al. (2017) claim that housing price developments can be explained by three categories, which are observable housing characteristics, macro-economic fundamentals and non-observable housing characteristics. Especially the last factor is less researched in combination with the ripple effect and can be used to research the underlying factors of the ripple effect in Amsterdam. The non-observable housing characteristics become more important since the housing price growth of the past years in Amsterdam cannot be explained by simple housing market fundamentals as GDP growth and the interest rates (AT5, 2017a). Earlier research by Liao et al. (2015) and Brady (2014) focused on the macro-economic fundamentals. The observable housing characteristics are normally included in the hedonic price indices when the ripple effect is researched. The added value of research to the ripple effect lies in the observable housing characteristics. These non-observable housing characteristics can be rephrased to the attractiveness of the district. Therefore, the research question that this thesis tries to answer is the following:
“What is the effect of the attractiveness of the different residential districts on the ripple effect in the city of Amsterdam?”
The aim of this thesis is to research the influence of attractiveness of different district on the ripple effect. At first, the ripple effect itself over Amsterdam will be researched. Secondly, the attractiveness of the different districts will be included to determine the effect of the attractiveness of the different districts on the ripple effect. The main research question will be divided into a number of sub-questions that are related to the main question. The sub-questions that this thesis will answer are the following:
Sub-question 1: “What determines the attractiveness of a district from a residential perspective? “ Sub-question 2: “Does the ripple effect in Amsterdam exist over attractiveness groups?”
Sub-question 3: “How is attractiveness linked to the Amsterdam ripple effect?”
To be able to answer those sub-questions and after that the main question, hypotheses will formulated. Only sub-question 1 will be answered directly by the literature review, but the other questions will be answered by statistical hypotheses. Those hypotheses will be based on the findings in the literature review and will be formulated in section 4.
9 Relevance
The importance of researching the ripple effect lies in the fact that residential real estate represents a key function in multiple markets. For inhabitants it is their home and affordability of houses is important social good. Starters and people who want to move from their first house have difficulties to get affordable houses (AD, 2018). This problem becomes more and more urgent and municipalities and policy makers would be interested to get insight into what determines the price developments of dwellings. This thesis will add insights into the housing price development of different districts.
On the other hand, Geltner et al. (2013) indicate that residential real estate is also an important investment class, where investors want to make use of the beneficial risk-return ratio. This thesis will give insights into the price development in this important asset class. At first, Amsterdam is considered as the most volatile residential real estate market of the Netherlands according to Vastgoedactueel (2017). Getting insight into the development of this volatile market is beneficial for investors, because it allows for a better judgment of investment and disinvestment opportunities. The part of the return that will be considered in this thesis is the capital return, which is based on the empty value of the dwelling. Since investors have the option to sell empty rental dwellings in the owner-occupier market, getting insight into the price developments of empty dwellings is beneficial. Secondly, investors normally consider real estate from an asset management perspective, but the residential real estate market is formed by a large amount through the user market. Considering investment opportunities with the angle of the inhabitants’ perspective, will give new insights that investors want to include in their investment decisions.
The third category of stakeholders that has an interest in the outcomes of this thesis is the lenders. Banks provide mortgages to residents to enable them to buy a house. The insights that this thesis will give about price developments in the Amsterdam housing market allow banks to get a better view of the risk-profile of the lenders and the underlying asset. From a risk management perspective, this enables banks to improve the judgment of mortgage requests, which is beneficial for the functioning of the housing market (NVB, 2014).
This thesis contributes to the existing literature in different ways. At first this thesis uses attractiveness of the districts as a leading indicator for the ripple effect rather than geographic indicators and contributes in this way to the knowledge gap of explaining the Amsterdam housing prices. Secondly, this thesis tries to provide an explanation for the way in which different areas have a lead-lag effect on each other. Thirdly, this thesis makes use of Impulse Response Functions to research the Amsterdam ripple effect and visualizes housing price relations within Amsterdam.
10 Structure
The remaining part of thesis is organized as follows: Section 2 elaborates on the existing literature about the ripple effect; Section 3 provides an overview of the districts in Amsterdam that will be included in this thesis; Section 4 states the hypotheses that will be tested in this thesis; Section 5 gives an overview about the dataset and data sources; Section 6 discusses the methodology that will be used.; Section 7 contains the results of the empirical analysis.; Section 8 provides the robustness checks, and Section 9 concludes.
2. Literature review
The literature review will elaborate on the existing literature on the ripple effect and house price dynamics in general. At first, the Dutch and in particular the Amsterdam housing market will be elaborated on. This provides the necessary background information about the regulation and the working mechanisms behind the Dutch housing market. Secondly, the ripple effect itself will be explained, followed by a discussion of the different results from research on the ripple effect in different countries and cities. Finally, factors that have an influence on house prices will be reviewed.
2.1 The Dutch housing market
The Dutch population is expected to growth from 17 million inhabitants in 2017, to 18 million inhabitants in 2031 (CBS, 2017a). However, big differences are expected for areas within the Netherlands. On top of that, the trend is that more and more people live alone instead of living together. This means that there is an increasing demand for single person’s dwellings. Both affordability and space can be an issue to fulfill that demand. Boelhouwer et al. (1996) already stated that demographic developments are an important explanation for future price movements. The ongoing trends are expected to continue in the coming decades as well, where both the number of households and the composition of the households are important characteristics for predicting price movements according to their findings.
Considering the geography of the Dutch housing market learns that most people live in the western part of the country, or what is called the Randstad. About half of the people in the Netherlands live in the provinces Zuid-Holland, Noord-Holland, Utrecht and Flevoland, which form the Randstad together (PBL, 2017). According to the prognoses of CBS (2017a) and PBL the population growth will be concentrated in the Randstad in the period 2016 – 2040. The eight regions with the strongest prognoses are all located in the Randstad. Other growth regions are some cities in the province of Noord-Brabant as Eindhoven, which is also a major economic area (CBS, 2017b).
11 The Randstad area currently faces insufficient supply to fulfill the current demand according to the NVM (2016), which is the Dutch agency of real estate brokers. Especially the case of Amsterdam is severe, but also cities as Utrecht, The Hague and Rotterdam show scarcity at the housing market. The housing shortage is estimated to reach 200,000 dwellings in 2020 according to ING (2018). It is expected that this will push the prices further in the coming years. However, ING states that the owner-occupied sector is still affordable, because of the fact that the average household only spends 22 percent of its disposable income to net housing payments and the real housing prices are still not at the same level as in 2008 in the Netherlands. On the other hand, the real housing prices in Amsterdam are above the level of 2008. This makes Amsterdam a different case from the rest of The Netherlands.
The Dutch housing market is according to the Dutch Banking Association (2014), NVB, dominated by social housing. Almost a third of the dwellings in the Netherlands are owned by a housing corporation. The owner-occupied sector in the Netherlands counts for 56 percent of the dwellings, which is below the average of 62.3 percent of the rest of Europe (NVB, 2014). The liberal rental sector is relatively small in the Netherlands compared to other countries. The Dutch government strongly regulates the social housing sector, to protect renters against sharp rent increases and tries to stimulate owner-occupied housing. Boelhouwer and Hoekstra (2009) state that the Dutch government is harming the housing market with their policies. First of all, there is support for homeowners by the tax deductibility of interest expenses, while the rental sector is regulated to limit rent increases. This widens the gap between the rental sector and owner-occupied sector and makes it more difficult for people to move from a social rental dwelling to an owner-occupied dwelling. Secondly, there is no coordination between housing supply regulation and housing demand regulation. Therefore, a mismatch between supply and demand is created at several places in the Netherlands with a quantitative housing shortage as a consequence. The demand is stimulated by fiscal treatment of owner-occupied dwellings, while tight spatial regulation is limiting new housing supply.
The limitation of new housing supply is part of the explanation for the housing shortage in the bigger cities. Boelhouwer (2005) indicates that there is mismatch between the rapid housing price growth and the stagnation of the construction activities. He claims that the liberalization of the housing supply in the Netherlands in the nineties in combination with the creation of an environment where supply and demand cannot react in the right way, are causing the mismatch. Before the nineties, the Dutch government strongly regulated the construction of new dwellings by planning and regulation. The current situation is a tight spatial regulation of municipalities, that is frustrating new constructions according to Boelhouwer. This causes that supply cannot react
12 properly to the rising demand. Over the period 1995 till 2003 housing prices increased by 59 percent on average, while the amount of new constructed homes decreased by 35 percent. This contradiction is a phenomenon that is only visible in the Dutch housing market.
The regulation of the social rental sector and the tight spatial planning is not only a problem for new constructions, but also frustrates the switch from rental sector to owner-occupied sector for people. Hoekstra and Boelhouwer (2014) claim that middle-income households will get difficulties to get an affordable home. The Dutch state had to reform the regulation regarding social housing. This resulted that the households with an income above a certain threshold no longer have access to social housing. The threshold for 2018 is a yearly income above 36,798 euros as stated by the Dutch government (2018). The people who earn slightly more than this threshold are not allowed to rent a regulated rental dwelling, but they also cannot afford the owner-occupier sector dwellings. These people will find difficulties to find a suitable dwelling in the near future.
Another group that has difficulties to buy an owner-occupied dwelling is the group of starters. Neuteboom and Brounen (2011) researched the accessibility of owner-occupied dwellings for first time buyers in the Netherlands. They claim that smaller parts of the housing market are suitable for young households due to affordability issues and competition. These issues occur due to a larger group of bidders for affordable houses where young households have to compete to. The competition is more often richer with a higher equity stake which they built up over the years. This problem for young households is a problem for the total housing market too. Since young households provide the liquidity in the market according to Neuteboom and Brounen.
Besides the liquidity in the market, the mobility of the residents in the Netherlands is also an important characteristic of the functioning of the housing market. De Jong and Brouwer (2012) researched the mobility of people in the Netherlands. They found that the group of people between the ages 25 and 34 shows the highest mobility and that people older than 65 are less mobile than people between the ages 55 and 64. This supports their expectation that elderly people are less mobile in terms of migration. They also found that people above 55 tend to move within the same municipality or to a smaller, less populated municipality. Younger people are more eager to move to a different municipality. The group of people of 65 and older is expected to be half as big as the group of people between 20 and 64 in 2040 (CBS, 2010), which indicates that this group will be an important determinant of the future mobility of residents in the Netherlands. De Jong and Brouwer researched their mobility based on their individual characteristics, but also on the characteristics of their current dwellings. They found that elderly people’s moving behavior depends on individual characteristics, of which the most important are health and life-events such as retirement. Housing characteristics also play an important role in the decision to move. A decline in the attractiveness of
13 the neighborhood can be an important motive for elderly to move.
A well-functioning housing market is important for the Dutch economy, since over half of the dwellings are an owner-occupied dwelling. A large share of the Dutch owner-occupied houses is financed with debt. According to the NVB (2014), the Dutch mortgages have on average the highest loan-to-value ratio and debt-to-income ratio of Europe after France. On the other hand, the Dutch mortgage market shows amongst the lowest default rates of Europe. New mortgage policies are introduced by the Dutch government in 2014 (NVB, 2014). This resulted in a limitation of the maximum tax-deductibility of the interest paid on the mortgage. The previous regulation was too encouraging towards borrowing, which resulted into new regulation on tax deductibility of mortgage interest payments. The maximum tax rate at which the interest can be deducted will decrease from 52 percent in 2013 to 38 percent in 2040. On top of that the maximum loan-to-value ratio for mortgages will be decreased from 110 percent in 2013 to 100 percent in 2018. The maximum loan-to-value will be lowered to decrease the risk of underwater mortgages. These are mortgages where the outstanding loan is bigger than the value of the dwelling.
Underwater mortgages are not necessarily a problem for banks or homeowners according to the NVB (2014), but it is a problem for the functioning of the housing market. Underwater mortgages can slowdown the housing market, because it can give difficulties for people to refinance a residual mortgage on the current dwelling to a new mortgage for the new dwelling. On top of that, behavioral economics come into play. Van Dijk (2015) argues that homeowners with an underwater mortgage do not take the market value as reference for their reservation price, but the mortgage loan. Sellers cannot afford to sell the house for less than the mortgage value. This results in homeowners that do not succeed to sell their houses.
Besides the risks in the Dutch housing market, there are also some risk-mitigating factors according to the NVB (2014). The first important factor in the Dutch housing market system is the high amount of pension savings, which provides a considerable buffer to enable elderly people to pay off their mortgages after retirement. Because of the elaborative pension system, the pension assets per capita ratio of the Netherlands is the highest in the world. The second risk-mitigating factor is the social security system in the Netherlands. The social system provides people who lost their job with social benefits. This enables them to pay their housing costs even when they do not have an income. The third factor is the strong legal protection of lenders in the Netherlands. When a borrower defaults, banks are entitled to the assets and income of the borrower. This provides a strong incentive for borrowers to meet their financial obligations to the bank. The fourth important factor is the central administration of consumer loans. This enables lenders to make a better risk perception of the potential borrower. The last risk-mitigating factor is the National Mortgage
14 Guarantee Scheme (Nationale Hypotheek Garantie). More than 25 percent of the new mortgages were closed under this scheme in the year 2013. The NHG is a fund backed by the Dutch government and provides the payment to banks in the case of a default from a borrower which had a mortgage with NHG. The bank bears the first ten percent loss on the mortgage, but the remaining part is backed by the government.
The opportunity to borrow to buy a house is important for the functioning of the Dutch housing market. Besides this, people also have their personal considerations when they decide to buy a house with a mortgage. Teye et al. (2017c) claim that homeowners in the owner-occupier sector consider two main risks by buying a house with a mortgage. The first one is default risk, while the second one is the property price risk. The default risk is investigated by Teye et al. (2015) and they found that family breakdowns are the most important factor for defaults on mortgages. The effect of unemployment is found to be small, which can be explained by the social benefits system in the Netherlands. The effect of the interest rate is not found significant by Teye et al. (2015), but has an effect on the total amount of mortgage debt in the Netherlands. The explanation for the small influence of the interest rate and unemployment rate of mortgage defaults are the strict conditions for people to get a mortgage. Banks include these variables in their risk assessment to provide their clients with mortgage that they can afford, even if some variables as the interest rate change. The property price risk relates to residual debt by selling of the property. When the prices of houses in the market are declining on a large scale, it may cause immobility or the loss of equity value for homeowners.
2.2 The Amsterdam housing market
Amsterdam is the capital city of the Netherlands and besides this it is also the most popular city to live according to JLL (2017). The city is an important economic part of the country with among others the business district the Zuidas and the main airport Schiphol close by. This makes the city an important market for real estate investors. The amount of money invested in real estate in the Netherlands was more than 19.5 billion euros in 2017 according to CBRE (2018), where Amsterdam is considered as the most important real estate market within the Netherlands. This indicates that Amsterdam is seen as a major investment market worldwide.
Besides attracting investments from all over the world, the city of Amsterdam has a system of ground leases which is not common in the Netherlands. The system of ground leases means that homeowners pay a certain amount to the municipality each year. The system has been invented in 1896 (Parool, 2017) and around 80 percent of the ground in Amsterdam is owned by the municipality. As a consequence dwellings in the districts Centrum-West and Centrum-Oost do not have ground leases that often. However, for dwellings built after 1896 ground leases are common.
15 From 2017 onwards, the municipality of Amsterdam gives homeowners the option to buy off the ground leases forever.
In 2017, Amsterdam had a housing stock of 428,035 dwellings. Around 30 percent was owner-occupied. 42 percent was owned by housing corporations and 28 percent owned by other landlords (CBS, 2017c). The percentage of owner-occupied dwellings is remarkably lower than the rest of the Netherlands. The rental sector in Amsterdam also contains a lot of social housing dwellings. Those houses are intended to enable people to get an affordable home. To qualify for social housing, people should not earn more than 36,798 euros in 2018 as stated by the Dutch government (2018).
Janssen-Jansen and Schilder (2015) indicate that accessibility of the housing market in Greater Amsterdam strongly decreases when households come above the maximum income for social housing. Their income is too high to be eligible for social housing, but the debt capacity is also too low to get a mortgage. They claim that the housing market for the Greater Amsterdam area is balanced for social housing since there is enough quality housing to provide access to low-income people, but as a whole the market is the most unbalanced in the Netherlands in terms of accessibility. This is considered as a side effect of the big social housing stock in Amsterdam. Also the high-demand in the private rental sector causes a bigger gap between the social rental sector and liberal rental sector, which harms the transition of people from the social sector to the private rental sector.
Research on the housing market of Amsterdam itself is conducted by Teye et al. (2017b) and Mathlener (2017), but they both used a different division of Amsterdam to conduct their research on. Teye et al. (2017b) used the division of Amsterdam till 2010. In 2010 the division of the city district was reduced from fourteen to seven districts (Parool, 2012). Mathlener (2017) used a division of eleven districts, but this is not an officially used division. The reclassification in 2010 was used to bring back the number of districts to seven with each approximately 100,000 inhabitants (Parool, 2012). This reclassification was an obligation for the municipality of Amsterdam from the Minister of Home Affairs to abolish the old structure with part municipalities. The new structure with just seven districts will not be used in this thesis as a base to research the ripple effect. The reason for this is that the districts are too big to determine the attractiveness and too much information on housing price developments will be lost. Instead of the seven districts, the 22 areas used by the municipality of Amsterdam (AT5, 2017b) will be the base for the division of the different parts of Amsterdam. The 22 areas allow more deviation in attractiveness factors and incorporate more different housing price patterns. Using the 99 sub-districts or 476 neighborhoods is not possible due to too few observations in transactions each quarter.
16
2.3 The ripple effect
Liao et al. (2015) claim that an investigation of the ripple effect should contain two parts. At first there is a significant pronounced shock in the economy that affects the region or city. Secondly, a price diffusion mechanism exists, whereby the results of the economic shock or house price changes in one particular market also affect the house prices in other markets or areas.
One of the first studies to the ripple effect is conducted by Holmans (1990) and MacDonald and Taylor (1993). Both studies are focused on the United Kingdom. Holmans’ research focused on the pattern in house price development between the South-east of the UK and the North of England. Later MacDonald and Taylor contributed to the knowledge about the ripple effect by focusing on the difference between run housing prices and long-run housing prices. They found that the short-term housing price dynamics show weak evidence for a ripple effect from London to the rest of the UK.
Over the years, two major streams of research on the ripple effect can be distinguished. The first category assumes a long-run common trend for house prices among districts with short-term deviations that can differ among districts. This view is mostly researched by unit root tests, because two housing price trends need to be stationary in their first differences and cointegration tests. The second view assumes different trends among different regions or districts. The differences in housing price levels can be explained by the own lagged price levels or lagged price levels in other districts. The second view is mostly researched with VAR-models and Granger Causality tests. When the ripple effect will be researched, it is important to make use of models that allow for different regional housing price dynamics to draw the right conclusion and to keep in mind that there might be a difference in short-term and long-term trends according to Oikarinen and Engblom (2016). Teye and Ahelegbey (2017) consider the ripple effect as short-term dynamics and convergence as long-term dynamics.
The second part of the ripple effect as stated by Liao et al. (2015) is the diffusion mechanisms where underlying factors can actually explain the ripple effect. This area of research is less explored. Underlying factors can be macro-economic factors as foreign investment (Liao et al. 2015) or income, interest rates or the consumer price index as researched by Brady (2014). Liao et al. found that the foreign capital investment in the center of Singapore formed an underlying shock in that housing market, which caused a ripple effect to the other areas. On the other hand, foreign investments in other areas did not have a significant influence on the housing prices in the city center. The research of Liao et al. provides an example of a macro-economic explanation for the ripple effect, but the topic of unobservable housing characteristics in combination with the ripple effect has been less researched. Meen (1999) proposed spatial migration as an explanatory variable
17 for the ripple effect, but the characteristics of why people would move have not been considered in combination with the ripple effect.
2.3.1 Country level ripple effects
The ripple effect itself has been researched in many countries across the globe. However, most studies are concentrated on the United Kingdom and the United States. The first evidence of the ripple effect has been published by Holmans (1990). Later on MacDonald and Taylor (1993), Meen (1999), Tsai (2014) and Cook and Watson (2016) extended the research to the ripple effect in the UK. They found different evidence and results for the UK.
Holmans (1990) was one of the first researchers who observed house price patterns in the UK. His data period was from 1960 till 1990 and he observed that the south-east of the UK tended to lead the real estate cycle. He was also one of the first that recognizes that the pattern differs in upturn and downturn markets. After that, MacDonald and Taylor (1993) presented their research on the UK by first focusing on the long-term housing price relationships and then working backward to the short-term housing price dynamics. They assumed a long-term trend, with short-term deviations from that trend. The short-term deviations became known as the ripple effect. The long-term trend was tested with cointegration tests. The short-term dynamics were tested with Impulse Response Functions. They found long-term cointegration trends for the different regions in the UK with Greater London. However, they found weak evidence for short-term housing price dynamics by Impulse Response Functions between areas. The paper is a first attempt to describe the housing price pattern in the UK based on data. The first short-coming of this research is that the methodology did not include an explanation and is solely data driven. A second short-coming is that they did not allow for differences in the quality of the dwellings. This would require hedonic price indices to be constructed on the level of the different regions. At that time, these regional price indices were not available.
One of the first persons who tried to provide an explanation for the ripple effect was Meen (1999) with research on the UK housing market. Meen suggested four possible explanations for the ripple effect. The first explanation is migration. The motivation behind migration is that households want to take advantage of price differences between regions. When the prices are high in London and lower in the North, households might want to move to the North to take advantage of this price differences. This would lead to the equalization in house prices between London and the North. The second explanation suggested by Meen is equity transfer, which has to do with the repeated sales of houses. When house prices are going up in London, repeated sales cause an equity growth for homeowners in London. As a result, people will have more buying power and this will drive the prices up in the other regions. The third explanation is spatial arbitrage, where differences between
18 house prices would be eliminated by spatial arbitrage. The last explanation that is suggested by Meen is spatial patterns in the determinants of house prices. This means that there is no link between regions, but the underlying factors or regressors for the different regions are the same. As these underlying factors follow similar patterns, the house prices in different regions would also follow similar patterns. Meen found evidence that the ripple effect is caused by changes within a region rather than changes between regions. The explanation for this is that structural differences between different regional housing markets exist. The paper of Meen contributes to the literature by proving that these regional differences show a distinct spatial pattern.
Tsai (2014) argues that the long-term trend requires constant ratios between transaction prices and constant ratios for transaction volumes of different regions. The argument for this is that regional house price corrections to the overall market occur through both the house price and transaction volumes. With unit root tests he found evidence for this long-term trend in housing prices in the United Kingdom. He considered ten different regions within the UK and found that both housing prices and transaction volume had a significant long-term trend based on panel unit-root tests.
Cook and Watson (2016) extended the work of Tsai (2014) by testing the long-term convergence based on changes in house prices rather than their levels. The second distinction that they made is the distinction between the recovery phase of the market and the recessionary phase of the market. They found that regions that are closer to London show a bigger co-movement with London and secondly the co-movement is bigger during the recovery phase of the economy. The bigger co-movement during the recovery phase is in line with other research and can be explained by the fact that London corrects more sharply during downturns than other regions, which results in less co-movement during recessionary markets. The last result of the study of Cook and Watson is that the convergence is found most present during the full sample period.
The results and claims from the work on the ripple effect in the UK have also been tested in other countries. Chien (2010) used housing prices in Taiwan from different cities to research the ripple effect. His results were different from the main consensus in the UK, which is that London leads the housing prices of the rest of the country. The capital of Taiwan, Taipei, is a regional global city where the house prices are significantly higher than the rest of the country. Chien states that the housing prices in Taipei are on average 50% higher than in other cities. The ratio of house price to income is 35% higher in Taipei than in the rest of the country. This is not uncommon since the housing prices in Amsterdam were in the first quarter of 2018 49% more expensive than the rest of the Netherlands based on data of the NVM (2018). Per square meter the difference is even bigger with 103%. Teye et al. (2017a) found a significant ripple effect from Amsterdam to the rest of the
19 Netherland despite the big housing price differences. Chien found that the housing prices in the two other big cities in Taiwan, Kaohsiung and Taichung, have a significant ripple effect to the rest of the country based on endogenous two-break LM unit root tests.
The second country with a big contribution to the research of the ripple effect is the United States. Among others Brady (2014) and Cohen et al. (2016) investigated housing price dynamics in the US from different angles. Besides proving the existence of the ripple effect, Brady (2014) and Cohen et al. (2016) also tried to include economic explanations for the ripple effect in their research. Brady used Impulse Response Functions based on a single equation Spatial Autoregressive model to investigate the spatial diffusion pattern of housing price developments in the United States on state level. He investigated the effect of shock in a certain variable in a surrounding state on the housing prices. He found that housing prices increase after a shock in real income by 1.5 percent and an increase of four percent when unemployment decreases by one percent. The response to the number of housing starts is also positive, but is rather small with less than 0.1 percent. Housing prices tend to not response to shocks in inflation since Brady did not found significant evidence for this. The response to an increase in population is rather large with three percent and is found to be significant for up to sixteen quarters. Brady found a rather large response of a decrease of twenty percent in housing prices to a one percent increase in interest rate. This can be explained by the fact that Brady used the effective interest rate, rather than the simple interest rate. All in all, it can be concluded that significant ripple effects exists with a certain economic shocks in surrounding states. Cohen et al. (2016) extended the work of Brady (2014) by including the distance between Metropolitan Statistical Areas (MSA) to investigate the spatial effects in combination with housing price dynamics. Where Brady used quarterly data, Cohen et al. used yearly data in combination with Spatial Vector autoregressive models. They incorporated geography by including distance variables for the different Metropolitan Statistical Areas. They found evidence for a spatial diffusion pattern, whereby information on lagged house price in neighboring areas helps to explain the current house price changes in the area. This result was found after controlling for the lagged house price developments in the worn area. The latest key finding is that the lagged effect of neighboring MSA’s increased after the crisis period in 2007. This is evidence of more spatial dependency of housing prices after the big burst in 2007.
Oikarinen and Engblom (2016) compared a panel data model that does not allow for regional variations in housing price dynamics with a model that allows for regional differences in housing price dynamics. Their approach is different than the general approach of using VAR models. They argue that due to the lack of data on housing prices and market fundamentals, housing price dynamics are studied by the use of panel-data models. The potential problem with this type of
20 models is that ordinary fixed-effects are used. The underlying assumption of this is that housing price dynamics are the same for every region that is included in the analysis. According to Oikarinen and Engblom this can be corrected by using panel-data models that allow for regional differences. By using these models they found small differences between the different regions in Finland in the short-term housing price dynamics. However, the long-run elasticity of housing prices with income differs significantly across different cities. This indicates that different cities react differently to fundamentals, which is in line with the findings of Meen (1999). The paper tried to add to the literature by including an economic explanation for housing price dynamics and thereby allows for regional differences in both the short-term and long-term.
Another approach to the research the ripple effect is suggested by Yang, Yu and Deng (2018). They used a complete new approach by using a high-dimensional generalized VAR to distinguish between the effects of spillover determinants as city GDP, population and secondary education. With this approach they provided a solution to calculate spillover effects from a high-dimensional VAR with limited data. The first step that they take is to construct pairwise net spillover effects between two regions in China based on forecast error variance decompositions. The average spillover effect is 43 percent, which indicates that 43 percent of the housing price variation in a region can be explained by price developments in other regions. Yang et al. found that cities in China with a higher administrative status, city GDP, population and secondary education have a larger spillover effect than other cities. Their approach is a solution for the lack of transaction data and can be a good solution to research the ripple effect in emerging markets.
Comparing the findings in other countries to the Netherlands results in more or less the same findings as found for the UK. The consensus for the UK is that housing price developments start in the South-east or more specific London and ripple out to the rest of the country. Teye et al. (2017a) found that Amsterdam is the leading housing market in the Netherlands where the developments in housing prices ripple out to all other regions except the province of Zeeland. They used Granger-Causality test with lead-lag relationship to research the short-term dynamics and they used Autoregressive distributed lags (ARDL)-Bound cointegration techniques to test the long-term relationship. The result is that Amsterdam is found to be cointegrated with the provinces of Friesland, Groningen, Limburg, Overijssel, Utrecht and Zuid-Holland. Especially the cointegration with Friesland and Groningen is remarkable, since those provinces are situated with a large distance from Amsterdam and have the lowest housing prices. Teye et al. suggest further research with more advanced econometric models to investigate the ripple effect in the Netherlands. Another direction for further investigation suggested in the paper is the extent to which housing affordability motivates house movers and internal-migrants from Amsterdam.
21 Teye and Ahelegbey (2017) tried to detect the spatial house price diffusion pattern in the Netherlands. They used Bayesian network approach, which is a data-driven technique to produce dependency patterns. In their study they modeled the housing prices for the 12 provinces in the Netherlands to research the spatial interactions between house prices. They found temporal dependence and house price diffusion patterns, which is also called the ripple effect, for different provinces. Noord-Holland was leading from 1995-2005 and the province Drenthe in the period 2005-2016. They used a data driven method in the form of the Katz measure to see where the housing price diffusion would start. This is a different approach than simply assuming that the central district or biggest city is the starting point of housing price development.
So far the ripple effect has been investigated by two streams of research. The first category of researchers considers the ripple effect as a long-term trend with short-term deviations. They consider unit-root tests for stationary house prices as evidence for a long-term trend where short time dynamics are evidence for the ripple effect. The second category considers the ripple effect as a spillover effect. Within this category, the spatial diffusion pattern has been the most present explanation. The spillover phenomena can be tested by granger causality tests and VAR models. Examples of the first category are Oikarinen and Engblom (2016) in Finland and Teye and Ahelegbey (2017) in the Netherlands and Teye et al. (2017a) also in the Netherlands. The second category of research is among others represented by Yang et al. (2018) in China and Oikarinen (2005) in Finland. Some studies combine both methods as Teye et al. (2017a) did for the Netherlands.
2.3.2 City level ripple effects
Besides the ripple effect at country or regional level, another stream of researchers has focused on the ripple effect within a city. Examples are Ho et al. (2008), Liao et al. (2015), Oikarinen (2005), Teye et al (2017b), Holmes et al. (2015) and Mathlener (2017). The ripple effect within cities has been researched less extensively, but from a migration point of view the ripple effect should be more persistent within a city. By comparing the migration numbers of Amsterdam of 2015, it can be concluded that almost half of the movements in Amsterdam came from people that moved within Amsterdam (CBS, 2015).
The ripple effect within Amsterdam has been researched by Teye et al. (2017b). Their research was focused on the risk interrelationships of Amsterdam sub-districts. They found a lead-lag relationship form the city center to the more peripheral sub-districts. This lead-lead-lag relationship has been tested by pairwise Granger causality tests. The drawback of using pairwise Granger-Causality test is that is data-driven and does not try to include an explanation for the phenomena. Besides the lead-lag relationship, they found a decreasing variation between sub-district in house price growth rates over time. This decreasing variation has been tested by the interdistrict deviation
22 and is intended to shed more light on the growth rate of housing price in certain sub-district compared to other sub-districts. The result is that more peripheral sub-districts show a larger deviation from the city trend in house price growth. These peripheral sub-districts are sub-districts with lower house prices than other districts. In general, the house prices grow faster in the central sub-districts and are also found to be riskier in the central sub-districts. Teye et al. suggested for further research is to include internal migration, which was one of Meen’s (1999) explanations, explicitly in an econometric model for further research.
The second study that has been conducted on the ripple effect in Amsterdam is the work of Mathlener (2017). He found a lead-lag relationship for adjacent sub-district in Amsterdam. The result is in line with Teye et al. (2017b) with a ripple effect from the city center to the peripheral sub-districts. The key difference is that Mathlener finds evidence for housing price developments that ripple out to the adjacent sub-district, while Teye et al. also found that the city center has a lead-lag relationship to non-adjacent sub-districts.
In contrast to the findings of Teye et al. and Mathlener, where house prices tend to ripple out from the center sub-districts to the peripheral sub-districts, are the findings of Ho et al. (2008) in Hong Kong. The paper of Ho et al. uses a model in the Hong Kong housing market to investigate the effect of wealth shocks in the housing market. They defined their sub-districts or submarkets in quality tiers, where the quality tiers were based on the size of the dwellings. Two policy changes in the Hong Kong housing market have been investigated. The first one was a policy change that caused an increase in the flow of people moving from the public housing rental sector to the private owner-occupier market. The second policy change caused this flow to decrease. Both policy changes resulted in a lead-lag effect from the lower-quality tier to higher-quality tiers. This result is contradicting to other findings. Other studies indicate that the city center leads other areas. The city center normally contains higher quality housing and the peripheral areas normally contain lower quality housing.
Liao et al. (2015) investigated the effect of a policy change for foreign investors in the Singaporean housing market. At first they proved that a ripple effect exists within the Singaporean housing market by using hedonic price indices and conducted a structural break-point analysis afterwards. The result is a ripple effect from the city center to the suburbs. Liao et al. found evidence that an economic shock in the city center ripples out to the suburbs by use of an Impulse Response Function based on a Structural VAR model. The case study that they used is the legislation for foreign investors to invest in certain parts of the city center. As a consequence more foreign capital flowed into the Singaporean housing market, but the house prices in the suburbs also increased which was found to be caused by this capital inflow. On the other hand, foreign investments in sub-districts
23 have a minor influence on house prices in the central district.
Hong Kong and Singapore are so called city states which are not entirely comparable with a city as Amsterdam. The reason for this is that the probability of a spill-over effect to other cities close-by is limited, since other cities would be across the border. This implies that people would have to emigrate. The case of London or Paris should be more comparable to the city of Amsterdam from a migration perspective. Holmes et al. (2017) shed more light on the long-term convergence of different districts within Paris. They used a two-step approach to test the probability of long-run convergence between two city districts. The probability of stationary house price differentials between two districts is negatively affected by a differences in unemployment, differences in demographics and differences in the characteristics of the housing supply-side. Furthermore, they found that a smaller distance between districts is associated with a faster adjustment to the long-run equilibrium. This indicates that closer located districts show a higher housing price relationship than districts that are further located from each other. This is in line with the findings of Teye et al. (2017b) and Mathlener (2017) in Amsterdam.
Oikarinen (2005) found a lead-lag relationship national wide from the Helsinki Metropolitan Area to other regions, but within the Metropolitan Area, the housing prices in the suburbs Granger-cause the housing prices in the city center. This evidence is not in line with the research of Teye et al. (2017b), Mathlener (2017) and Liao et al. (2015), but can be explained by the fact that the Helsinki Metropolitan Area is rather small compared to the broader economy according to Oikarinen (2005). She suggests that employment growth and migration are important factors for the within city ripple effect, but the in-depth explanation is left for further research. The house prices in Finland lead GDP changes, which imply that informational factors play an important role in explaining the ripple effect in Finland.
2.4 Factors influencing house prices
The ripple effect is a consequence of house price movements and is more focused on the interrelationships of different housing markets. House price developments itself are also studied extensively. According to Dröes et al. (2017) house price developments can first be explained by observable dwelling characteristics as square meters, type of dwelling etcetera. Secondly, the developments in house prices can be explained by macro-economic factors as inflation, interest rates and GDP growth. Thirdly, the price developments can be explained by non-observable dwelling characteristics as facilities in the neighborhood. The unexplainable part is what is left out. The third category is an important determinant to explain the Amsterdam house price development, but is less researched than the other two categories. The attractiveness of the district can be linked to the non-observable dwelling characteristics.
24 Dröes et al. (2017) used the three factors combined to explain the Amsterdam housing price development for the period 1990 till 2017. However, the factors are normally used separately in combination with the ripple effect. The observable housing characteristics are incorporated by constructing the hedonic price indices. In this way, a correction can be made for the differences between dwellings. Examples of this in combination with the ripple effect are Teye et al. (2017b) and Oikarinen and Engblom (2016) with Granger Causality tests to research housing price dynamics. The macro-economic factors can be used to explain the different breaking-points in the ripple effect. Examples of this research are Brady (2014) and Liao et al. (2015) that used the factors in combination with an Impulse Response Function to explain the different breaking points in house price developments that cause the ripple effect. The non-observable housing characteristics have not been taken into account together with the ripple effect according to the author’s knowledge, but this thesis tries to use these characteristics to explain the way in which the housing price developments ripple out within the city of Amsterdam.
The observable housing characteristics can be used to explain the price differences between two products according to Rosen (1974). The work of Rosen is widely used in real estate hedonic price indices and provides a way to correct for differences in housing characteristics. A theory about the exact characteristics is lacking, but Francke (2017) names several observable factors that can be used to explain differences in price between dwellings. Among others the property type might give an explanation. The property type is commonly dividend in single-family houses and multifamily houses. The property subtype indicates whether a dwelling is a maisonnette or detached house for example. Other factors that explain differences are the construction year, lot size, property size and the level of maintenance.
Macro-economic factors are incorporated in the ripple effect more often than non-observable housing characteristics. As stated earlier, Brady (2014) and Liao et al. (2015) included these factors as an economic explanation for the ripple effect. Other studies applied macro-economic factors directly to housing prices. Donald Jud and Winkler (2002) used the growth of the population, real changes in income, construction costs and interest rates together with housing prices in 130 metropolitan areas in the United States. They separately modeled demand and supply to construct a price change as a function of different variables. Real housing prices are found to be strongly influenced by growth in population, change in real income, constructions costs and the interest rate. On top of that Donald Jud and Winkler claim that the stock market has an effect on housing prices via a certain wealth effect. The stock market both affects the housing market at the current phase as well with a lagged relationship. Housing price appreciation varies across areas due to location fixed-effects according to Donald Jud and Winkler. These location fixed effects for specific
25 areas are correlated with restrictions in land availability and other policies.
Dröes and Van de Minne (2015) used a different approach by incorporating the transaction data of 200 years from the Herengracht in Amsterdam. They found that the macro-economic factors that influence housing price differ over time. They found that construction costs, housing stock and population growth were the most important determinants in the 19th century. The most important factors at the end of the 20th century were GDP and the interest rate. The results for the 20th century are more or less in line with Brady (2014) and Donald Jud and Winkler (2002). These macro-economic factors are indirectly linked to housing price in this thesis by the time dummies in the hedonic price indices. Macro-economic factors have an influence on the housing market, but these developments happen over time. These developments over time are therefore incorporated in the time dummies.
The non-observable housing characteristics can be included in the housing price developments, but this is not applied very often and the used approach differs between studies. Dröes et al. (2017) used the asking price as a reference for non-observable housing characteristics as energy-efficiency or local facilities. The theory behind this is that homeowners will ask a higher price when the non-observable characteristics are more favorable. Dröes et al. incorporated this by subtracting the asking price from the transaction price. The drawback of this method is that a higher asking price can be a strategic decision of the seller. A different approach has been used by Visser and Van Dam (2006). They used four dimensions into the determination of the price of a dwelling. The dimensions that they used are physical dwelling characteristics, physical environmental characteristics, social environmental characteristics and functional environmental characteristics. The last three dimensions can be included into non-observable housing characteristics. The physical environmental characteristics cover the surrounding characteristics as parks and water, building density and the quality of the built surroundings. The social environmental characteristics include the composition of the population, unemployment rates and the average income in the neighborhood. The functional environmental characteristics include the distance to certain facilities and infrastructure.
They linked those characteristics to price developments of dwellings directly by the use of hedonic price indices. They found that physical environmental characteristics have a minor influence on the price. However, they found that the parks or forest in the near neighborhood have a positive influence on the price per m² of the dwelling. Visser and Van Dam found that social environmental characteristics as social reputation of the neighborhood and the share of non-western immigrant in the neighborhood have a negative impact on the price per square meter. The results for functional environmental characteristics are different for cities and rural areas. For cities it is found that the
26 presence of a highway in the near distance has a negative impact on the price due to noise and odor nuisance. For rural areas, the presence of a highway in the near distance is beneficial, because it increases the accessibility. This has a positive impact on the price per square meter. The result for traffic noise in cities is also found by Theebe (2004). He found that noise levels above 65 decibel have a negative impact up to twelve percent on the price of dwellings.
Most of the studies on housing price dynamics assume certain homogeneity over different districts or housing markets. Ferreira and Gyourko (2011, 2012) tried to validate this assumption by investigating difference between housing price booms between metropolitan areas in the United States. Ferreira and Gyourko (2011) found differences in the United States for the starts of housing booms. The time differences of housing booms between metropolitan areas were up to eleven years. They (2012) found that geographical closeness and economic similarity are important factors. On top of that, they found that a higher amount of college graduates and a higher median income in a neighborhood are related to an earlier housing boom than for other district with a lower amount of college graduates and lower median income. The importance of their work is that differences in housing booms can even exist over neighborhoods and can be explained by economic and geographical factors.
2.5 Factors influencing attractiveness
Macroeconomic factors as GDP growth, inflation rate and interest rate are important determinants for housing prices. Therefore, these factors are important considerations when house price dynamics are researched. However, from an inhabitant perspective the attractiveness of the neighborhood and district will be important considerations when they want to buy a house in that particular area. The attractiveness of a district can be considered from different angles. JLL (2017) researched with the Residential Ranking the attractiveness of different Dutch cities with more than 50,000 inhabitants. They did this in a quantitative way with eight main themes with in total 37 indicators. Not only housing market fundamentals are included, but also economic and demographic developments and the availability of different types of local amenities are included. Unfortunately, JLL is not fully transparent in which themes and which indicators it used to construct the Residential Ranking.
The Leefbaarometer (2016) measures the livability of different areas in the Netherlands and uses five main themes, which are dwellings (18%), inhabitants (15%), facilities in the city (25%), safety (24%) and physical surroundings (18%). The measure is based on data availability and is a uniform measure for all areas in the Netherlands according the Leefbaarometer 2.0 instrumentontwikkeling (2015). The drawback of the method is that the score is expressed as a dimension score. The reason for this is that the score of the model from the Leefbaarometer is a
27 relative score compared to the average of the Netherlands.
The underlying dimensions in the Leefbaarometer are dwellings, inhabitants, facilities in the city, safety and physical surroundings. Each of those dimensions has different underlying variables that are taken into consideration. Those underlying factors depend among others on the explainability of the variable for the Leefbaarometer, the data for that variable has to be available for every place in the Netherlands and has to be available at the six digits zip code number. Besides that, the data has to be available in time series and they have to be reliable.
The variable dwellings is focused on the structure of the housing stock. It takes into account the surface of the dwellings, but also the construction period and dwelling type. Secondly the ownership ratio and vacancy are taken into account. Inhabitants of the neighborhood have an impact on the livability of the neighborhood. Factors of the inhabitants that are taken into account in the Leefbaarometer that are inhabitants specific are the income of the people, the level of education and the social benefits. Variables that are included, but which have more to do with the population composition are developments in age, number of households and the mutation grade.
The facilities in the neighborhood are an important characteristic of the livability. The variable accounts for 25 percent and includes the distance to healthcare, retail shops, leisure, education, public transport and hospitality. The indicator safety in the neighborhood consists of two main themes. The first one is nuisance from people behavior. The second one is criminality. The nuisance from people behavior comes from nuisance of drug use, youth, local residents, vandalism, litter and daubing. The criminality numbers are based on among others violent crimes, robberies and burglaries. The physical surroundings variable includes the distance of dwellings to nuisance related surroundings as railroads, wind turbines and highways. Besides the nuisance related surroundings, also pleasure related surroundings are included. Examples of these are the distance to parks and water.
The Leefbaarometer gives a score based on the five factors as mentioned. The score is an indication score and is applicable
Figure 1: the attractiveness heat map of Amsterdam. Light blue indicates very attractive, red indicates very unattractive.
28 at different levels. The lowest level is the grid level, which is an area of 100 by 100 meters. The second lowest level is the neighborhood, followed by sub-districts, zip code areas and municipalities. The different levels allow for different dimensions within a level. A high level grid can exists in a low level neighborhood for this reason. The score from the Leefbaarometer is an identification score ranging from very inadequate to excellent. The attractiveness heat map for Amsterdam can be found in figure 1.
3. Description Amsterdam districts
This section will provide more background information about the different districts of Amsterdam that will be included in this thesis. The different districts that will be elaborated on are defined in table 1. Each district is part of a certain city part as defined by the municipality of Amsterdam (2017). City part Districts
Centrum Centrum-West, Centrum-Oost
West Westpark, Bos en Lommer, Oud-West/De Baarsjes
Nieuw-West Geuzenveld-Slotermeer, Sloterdijk, Osdorp, De Aker, Sloten en Nieuw-Sloten, Slotervaart
Zuid Oud-Zuid, Buitenveldert/Zuidas, De Pijp/Rivierenbuurt
Oost Oud-Oost, Indische Buurt/Oostelijk Havengebied, Watergraafsmeer, IJburg/Eiland Zeeburg
Noord West, Oud-Noord, Oost
Zuidoost Bijlmer-Centrum, Bijlmer-Oost, Gaasperdam/Driemond
Table 1: division of the districts of Amsterdam over the seven city parts.
3.1.1 Centrum-West
The district Centrum-West is one of the oldest parts of Amsterdam and nowadays the busiest district of the city according to the municipality of Amsterdam (2017-1). The 17th century constructed Grachtengordel-West is one of the neighborhoods in the districts and is UNESCO world heritage. The inhabitants of the district are typically from Dutch (60%) or Western (17%) origin which is above average for the city of Amsterdam. The average income is above the average of the rest of Amsterdam and the unemployment is low with four percent compared to the average of seven percent in the city of Amsterdam. The safety in the district is the worst in Amsterdam according to the safety-index score of the municipality. However, the inhabitants are satisfied with their district and rank it above average. They grade their district with an 8.0, which makes it the fourth best district. The district consists of approximately 28 percent owner-occupied dwellings and had 43,228 inhabitants in 2017.
