Master Thesis
MSc Business economics - Finance and Real Estate Finance
Differences in Dutch Regional House Price Development and their Determinants
Student name: Lars Kulk Student number: 10275568
Master: Business Economics - Finance & Real Estate Finance (dual track) Thesis supervisor: Dhr. Prof. Dr. J.B.S. Conijn
Abstract
House prices are once again increasing after a downturn in the Dutch economy. Though the house prices in the Netherlands are increasing, it is at different speeds in different parts of the country. In this thesis, the regional house price differences between four selected regions will be examined using a difference-in-difference model. Secondly, using panel regressions, possible determinants will be linked to these regional price differences in an effort to explain differences in regional price development in the Netherlands. The determinants used in the analysis are demographics, income levels, employment, housing stock and distance to
facilities. The difference-in-difference model shows that regional house price differences exist in the Netherlands. From the possible determinants population growth, household income and distance to facilities provide an explanation for the existence and development of regional house price differences in the Netherlands.
Statement of Originality
This document is written by Lars Kulk, who declares to take full responsibility for the content of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
Table of Contents
1. Introduction 4
2. Literature Review 7
2.1 Regional House Price Development 7
2.2 House Price Determinants 10
3. Hypotheses 14
3.1 Hypothesis 1 14
3.2 Hypothesis 2 14
4. Data and Descriptive Statistics 14
4.1 Dataset Construction 15 4.2 Data Overview 16 4.3 Descriptive Statistics 17 5. Methodology 19 5.1 Difference-in-difference model 20 5.2 Panel regressions 21 6. Results 23 6.1 Price Indices 23 6.2 DID 23 6.3 Panel regressions 25 7. Robustness Checks 29
8. Conclusion and Discussion 32
References 35 Appendices 37 Appendix A 37 Appendix B 38 Appendix C 39 Appendix D 39 Appendix E 44 Appendix F 46
1. Introduction
House prices are once again increasing after a downturn in the Dutch economy. The housing market experienced several difficult years during the most recent financial crisis. However, house prices in the Netherlands are rising at different speeds in different parts of the country. Particularly in Amsterdam, the capital of the Netherlands, house prices are rising more rapidly than in the other parts of the Netherlands (Graafland, 2015). Furthermore, the Dutch central bureau of statistics reported that regional house prices differences have risen in 2016 like in previous years (CBS, 2017). At the end of the 20th century house prices rose rapidly in the Netherlands comparable to the current development. After the rapid rise the Dutch housing market stagnated and a discussion arose in which the Dutch housing market was considered one market. Brounen and Huij (2004) disagree with this view (p. 126). The authors state that regional diversity in price development exists in the Netherlands. In different regions house price development is substantially different (Brounen & Huij, 2004, p. 127). This thesis examines the current differences in regional house price development. First, this thesis will examine whether these differences in house prices are significant. Second, this thesis will examine these differences using multiple determinants of regional house prices to see if these determinants can explain the differences in regional house price development. This study could provide insight in what determinants are important for the development of regional house prices in the Netherlands.
The research questions of this thesis will be as follows: Are there significant differences in regional house price development across the Netherlands? This question is followed by the question: How can these differences be explained?
The aim of this thesis is to provide an explanation for the differences in regional house price development. This research could be practically useful to real estate investors and consumers buying homes: buying a house in high growth areas would produce higher returns, though it is also possible that consumers are currently overpaying in high growth areas such as Amsterdam, and would be better off buying a house in a low growth area. More
specifically, knowing which regional house price determinants are important can increase accuracy in predicting future house price developments. Diversification is important for investors to reduce risks. Hartzell, Shulman and Wurtzebach (1987) show that diversification in regional housing markets is beneficial. The results further suggest that location matters (p. 94). Diversification becomes more efficient if accuracy in predicting future house price development increases due to more knowledge on regional house price determinants. This is beneficial for investors. Other benefits from better understanding regional house prices can be
found in the areas of policy and regulation. Knowing which determinants play an important role can make it simpler for policymakers to influence a regional housing market. The housing market is an important part of the economy; if it becomes easier to influence the housing market then, indirectly, the economy and especially the regional economy would also be more controllable. This study is a step in the right direction.
Though much research has been produced on house prices and its determinants, there
is little research that specifically focuses on regional house prices. Reichert (1990) shows that house prices react the same to national macroeconomic factors but react differently to local factors. Different regions may have different determinants of house prices (p. 388). Brady (2011) examined how house prices in one region affect house prices in other regions (p. 213). The results show that house price development diffuses to other regions over a period of time (Brady, 2011, p. 220). The main goal of Baffoe-Bonnie’s study (1998) was to look at the effects of monetary policy and macroeconomic aggregates on the housing market on a national and regional level (p. 180). Baffoe-Bonnie (1998) stated that macroeconomic fundamentals such as mortgage rates influence national and regional housing markets. However, regions respond different to monetary policy and the macroeconomics aggregates (p.193). Ohtake and Shintani (1996) looked at the relationship between demographics and the Japanese housing market (p. 190). They found a relationship between house prices and demographics in the short run due to the fact that the housing supply is price inelastic (p. 200). Abraham and Hendershott (1992) examined the patterns and determinants of
metropolitan house prices. The results show that important determinants such as employment vary in their explanatory power per region (p. 20). A paper by Archer, Gatzlaff and Ling (1996) examined the variation in the rates of house price appreciation within a metropolitan market in Miami (p. 335). Their study shows that house prices vary spatially. Van Dijk et al. (2011) and Brounen & Huij (2004) show the presence of regional house price differences in the Netherlands for different time periods than this thesis examines. This thesis differs from currently existing literature because the focus of this thesis is primarily the Dutch regional housing market and the determinants. Little research has been conducted on Dutch regional house prices and their determinants. The aforementioned articles all give their view for regional house prices. However, it will be interesting to see which specific factors play an important role in the Dutch regional housing market.
Four corop regions were selected for the analysis based on the state of their housing markets1. The four regions are Amsterdam, North-Drenthe, Zeeuwsch-Vlaanderen and Agglomeration ‘s-Gravenhage. Regarding house price development, Amsterdam is an
overheated housing market, North-Drenthe and Zeeuwsch-Vlaanderen are unstressed housing markets, and Agglomeration ‘s-Gravenhage is an average housing market. For these four regions a dataset is provided by the Dutch Association of Realtors (NVM) is provided. Data on the possible determinants of regional house prices is retrieved from the Dutch Central Bureau of Statistics (CBS). Data is collected for the time period 2000-20162.
The thesis will first examine whether there are significant differences in regional house prices between these regions. This will be done using a difference-in-difference model. First, price indices for each region used over the period 2000-2016 were created using a hedonic model. Four separate hedonic regressions are estimated to create the four prices indices of the selected regions. Quarterly prices indices are created with the use of quarterly time dummies. Afterwards, the price indices will be compared using the
difference-in-difference model for the entire period. This empirical set up will show the causal effects of an event over time (Stock & Watson, 2012, p. 533). The significant results from the average treatment effect can be interpreted as differences in house price development over time. The cut-off point used for the analysis is Q2 2013. The model will show if regional price
development differs before and after the cut-off point between the selected regions. The difference-in-difference model will be estimated using an OLS regression with dummy variables for whether the observation has been exposed to the treatment, a dummy to indicate the post treatment period and an interaction term. The interaction term resembles the average treatment effect. When there are significant differences in regional house price development, this thesis will look at how these differences can be explained. For the second part of the methodology, multiple panel regressions will be performed. The log of price indices will be included as the dependent variable in the regressions because this is the development this thesis is interested in. Multiple determinants will be added to the panel regressions in an effort to explain the regional house price development. The determinants that will be used are based on existing literature. These are: demographics, income levels, employment, housing stock and distance to facilities. Separate panel regressions will be performed with different configurations. The model will be estimated for the time period 2008-2015 instead of 2000-2016 due to limited data availability on the possible determinants. Magnitude and direction of
1 Amsterdam consists of only the municipality Amsterdam due the size and the number of transactions 2 Time period per variable varies due to data availability
the coefficient of the significant variables produced in the panel regression in collaboration with the changes in levels of the chosen determinants before and after the cut-off point for each region can give insights in important determinants. In the results section, each significant variable is discussed separately.
In the second part of this thesis an overview of related literature will be given in the literature review. The end of the second part will elaborate on the own contribution of the thesis to existing literature. Thereafter, the hypotheses are discussed. The fourth part will elaborate on the data and methodology. In the fifth part the results and robustness checks are discussed. In the last part, a conclusion and discussion is given in which the hypotheses and the results are discussed. Furthermore, the conclusion will elaborate on limitations and directions for future research.
2. Literature Review
This section will give an overview of the existing literature regarding the topic of this thesis. The literature review is divided into two parts. The first part provides an overview of
literature on regional house price development. The second part gives an overview on
literature on house price determinants. Together these two parts cover all existing literature on the topic of this thesis.
2.1 Regional House Price Development
Previous research focused on either the national housing market macroeconomic relationships or microeconomic relationships for specific urban areas. Reichert (1990) aimed at analyzing macroeconomic and microeconomic factors in the determining of new house prices at a regional level (p. 373). He used quarterly data over a 12-year period from 1975.2 until 1987.2. The financial variables that are used are in real terms. For all estimation models a double log model was used, and the coefficient can then be interpreted as elasticities (Reichert, 1990, p. 381). Reichert (1990) showed that house prices react the same to national economic factors but react differently to local factors. Different regions can have different determinants of house prices (p. 388). First, Reichert (1990) looked at national house prices in the U.S. and which determinants play an important role in them. This was done to set the stage for regional house price results (p. 381). The model for the entire U.S. explained 94% of the variation in U.S. real house prices, and the results showed that the important determinants are
interest rates. Determinants such as employment play a less important, though still significant, role in explaining real U.S. house prices (Reichert, 1990, p. 383). Secondly, Reichert (1990) looked at the regional real house prices for four regions in the U.S., namely the North-Central, Northeast, South and the West. The results from these four regions show that local factors indeed play an important role in explaining regional house prices. The results showed differences in the size and significance of the coefficients between regions. The results also showed homogeneity within broad regions. Furthermore, the results rejected the presence of a national housing market (Reichert, 1990, p. 387). These results suggest that local
determinants may play a more important role in determining the differences in house price development across the Netherlands.
Lamont and Stein (1999) examined how house price dynamics differ across cities with different levels of leverage. The authors look at how cities with different levels of leverage responded to an economic shock, in particular, an income per capita shock (p. 498). Data on borrowing patterns was used from 44 metropolitan areas from 1984 until 1994. The paper finds that house prices react differently to income shocks in high and low leveraged cities. A high-leveraged city has a large portion of homeowners with high loan-to-value ratios. Vice versa, a low-leveraged city has a large portion of homeowners with low loan-to-value ratios. Lamont and Stein (1999) use a cutoff of 80% LTV as a standard benchmark for high LTV’s. This means that LTV’s are considered high when they exceed 80% (p. 501). In Cities with high leverage house prices tend to react quickly to an income shock. In contrast, house prices reacted more gradually to an income shock in low-leveraged cities. The house prices in high-leveraged cities reacted quickly but also tended to overshoot the new long-run level. This overshooting reaches its peak in the fourth year (Lamont & Stein, 1999, p. 505). The price gap after an income shock widens in the first few years and narrows again in the following years, due to the different reaction of high and low-leveraged cities to an income shock (Lamont & Stein, 1999, p. 513). This could be a determinant that helps explain the difference in Dutch regional house prices. This could also be an explanation for the booming Amsterdam housing market.
Brady (2011) gives a possible reason for the existence of regional house price differences. In his paper he looks at how house prices in one region affect house prices in other regions, which is a question of interest for policy makers and economists (p. 213). The econometric model uses average house prices per county. In the sample there are 31 counties from California, with a time period from 1995 until 2002 (Brady, 2011, p. 214). An
that shows that house price development diffuses to other regions over a period of time. Spatial correlation is identified by multiple studies. Spatial correlation can occur in the case of shared market features between regions, or by the feedback effect. The positive-feedback effect occurs when the growth of the housing market positively influences a neighboring region through the attitude and behavior of market participants in surrounding regions (Brady, 2011, p. 220). This spatial correlation between regions, when controlling for conditions like construction, usually lasts about two years. The results of Brady’s study are not only useful for understanding the transmission of shocks within regions but also for other regional phenomena (Brady, 2011, p. 229). Currently in the Netherlands, house prices in Amsterdam are growing at a very high pace, while other parts of the Netherlands show a lack of growth in house prices. It could be the case that Brady’s theory on spatial correlation is applicable to the situation in the Netherlands.
The article by Meen (1999) discusses the ripple effect in the United Kingdom. The ripple effect is known as a phenomenon, in which a rise in house prices in a certain region gradually spread out to the rest of the country over time. In the UK the house price rise originated in the South-East and then gradually spread out to the rest of the country over time (p. 733). Multiple articles statistically explain the ripple effect but the economic reasoning is less straightforward. This article connects the regional housing market and the ripple effect. More specific the article suggests that structural differences between regional housing markets play an important role in the ripple effect (Meen, 1999, p. 733-4). House price data from the government and two major mortgage lenders in combination with a theoretic model is used to explain the ripple effect (Meen, 1999, p. 734-6). The results show that differences in coefficients between regions show distinct spatial patterns, which can be associated with the ripple effect. Furthermore the results show that the ripple effect is caused by adjustments within regions instead of between regions (Meen, 1999, p. 752). The ripple can be a possible explanation for regional price dynamics in the Netherlands.
The article by van Dijk, Franses, Paap & van Dijk (2011) looks at regional house price developments in the Netherlands. The data is provided by the Dutch Association of Realtors (NVM). The data consists of quarterly data on house prices for 76 regions. The dataset covers the period 1985 Q1 until 2005 Q4. In this dataset the cross-section dimension and time
dimension are large. Macroeconomic data is provided by the Dutch Central Bank (DNB) and the Dutch Central Bureau of Statistics (CBS) (van Dijk et al, 2011, p. 2098). A panel model is developed and used for estimation. The authors want to answer two questions with the model.
regions. The second question is whether the house prices in regions follow the trend in real GDP (van Dijk et al, 2011, p. 2101). The 76 regions are clustered in classes on similarities on house price dynamics. The results show that two classes are sufficient. The first class is classified as mainly rural regions close to large cities. The second class is classified as both the larger cities and some more remote rural regions. The house prices in regions in the first class show slightly higher average growth rates and stronger and faster reactions to changes in GDP than the second class (van Dijk et al, 2011, p. 2108-9). These results illustrate the
presence of differences in house price development across the Netherlands in the sample period. These findings are relevant for this study because they show the existence of differences in regional house prices development however, in a different time period. Furthermore, the findings show the presence of differences between more rural and urban areas in price development. The selected regions in this thesis also contain more rural and urban areas.
2.2 House Price Determinants
From a macroeconomic standpoint, the housing market is considered an important sector of the national economy. The national economy is influenced by monetary policy. Baffoe-Bonnie (1998) looked at the relationship between monetary policy and the housing market. Their main goal was to look at the effects of monetary policy and macroeconomic aggregates such as inflation and employment growth on the housing market on a national and a regional level. The focus is on the short and long run effect as well as anticipated and unanticipated monetary policies (p. 180). They used a VAR model and quarterly data from 1973.1 until 1994.4 to examine these effects. The data consists of variables including house prices, money supply, mortgage rates and employment growth (Bonnie, 1998, p. 184). Baffoe-Bonnie (1998) stated that macroeconomic fundamentals such as mortgage rates influence national and regional housing markets. Monetary policy has a moderate effect on the housing market but also indirectly influences the housing market through the mortgage rates, which are influenced by monetary policy, on a national and a regional level. Inflation does not have a significant influence on the housing market on either a national or a regional level, but employment growth does. Changes in employment create changes in income, which influence the demand for housing. The macroeconomic fundamentals can only partially explain the differences between regional house prices. Regions respond differently to monetary policy and the macroeconomics aggregates (p.193). These results from Baffoe-Bonnie (1998)
suggest that other factors, like local factors, could play a role in determining regional house prices.
Brounen and Huij (2004) studied the Dutch housing market at the beginning of the 21th century. At the end of the 20th century house prices rose rapidly in the Netherlands. After the rapid rise the Dutch housing market stagnated and a discussion arose if the Dutch housing market was to be considered one market. Brounen and Huij (2004) disagree with this view because a study by Capozza, Hendershott, Mack and Mayer (2002) examined the
determinants of real house price dynamics. This study showed that regional housing markets can react differently on macro-economic shocks. In the article by Brounen and Huij (2004) it is argued that the Dutch housing market is not a homogeneous market (p. 126). The Dutch housing market is analyzed by using an extensive dataset provided by the Dutch Association of Realtors (NVM) over the time period 1985 until 2004. In the dataset house prices are specified geographically and on house type. The authors examine the relationship between house price development and macro-economic development (Brounen & Huij, 2004, p. 127). The results show that differences exist in regional price differences between the “Randstad” region and the rest of the Netherlands but also between provinces. These differences arise due to the difference in sensitivity to macro-economic development per region. The results show that the “Randstad” region was less sensitive for unemployment than the rest of the
Netherlands (Brounen & Huij, 2004, p. 127-8). This article gives a possible determinant, which can explain the possible current differences in regional price development.
The role of demographics in the housing market was first brought to light by Mankiw and Weil (1989), who showed that the generation of baby boomers was a major cause of the growth in house prices in the United States (Ohtake & Shintani, 1996, p. 189). The paper by Ohtake and Shintani (1996) examines the relationship between demographics and the Japanese housing market from Japan’s own generation of baby boomers. An index of demographic factors from Mankiw and Weil is used as well as the Mankiw and Weil model for the analysis. Ohtake and Shintani distinguish themselves from previous research by looking at the short and the long-term effects. Secondly, they also take cohort effect into account (p. 190). Ohtake and Shintani (1996) find a relationship between house prices and demographics in the short run because the housing supply is price inelastic. Housing supply is elastic in the long run and therefore demographics influence housing stock, but not prices (p. 200). This suggests that demographics could play a role in explaining differences in the development of regional house prices, due to regional demographic differences.
Abraham and Hendershott (1992) studied the patterns and determinants of
metropolitan house prices using data from the Freddie Mac repeat sale database. U.S. house price data from 1977 until 1991 was used. In the dataset there are 30 U.S. cities, all with significant swings in house prices (Abraham & Hendershott, 1992, p. 1-2). The results showed that there were regional house price differences in the period from 1977 until 1991. Important determinants such as employment vary in their explanatory power per region (Abraham & Hendershott, 1992, p. 20). Abraham and Hendershott (1992) stated that swings in regional house prices clearly mimic regional economic cycles (p. 1). These economic cycles could possibly be an explanation for regional house price differences in the Netherlands, as these economic cycles are also present in the Netherlands.
The article by Harter-Dreiman (2004) and Gallin (2006) both studied the relationship between house prices and income. Harter-Dreiman (2004) used OFHEO weighted repeat sales index from Fannie Mae/Freddie Mac loan level data. For income data aggregated personal income from Woods and Pool is used. The dataset consists of 76 MSAs. Furthermore, all data is available for the sample period 1980-1998 (p. 323-4). A three-equation VEC model is used for the estimations. The results show a positive relationship between house prices and
personal income. Furthermore the results show an elastic supply function and the adjustment of the house prices to a demand shock is slow. 70 percent of the adjustment occurs within the first 5 years and 90 percent within 10 years after the shock. The results also imply that lagged price and income affect current prices and influence the path to equilibrium (Harter-Dreiman, 2004, p. 335). These results suggest income could be a determinant for regional house price development. As earlier mentioned, Gallin (2006) also looked at the relationship between income and house prices. This article is critical on existing literature covering income and house prices. Data is retrieved from the OFHEO. For house prices quarterly repeat-sales price index. Total personal income is used at a city level is available only at an annual basis (Gallin, 2006, p. 427). Gallin (2006) finds that fundamentals, like income, can influence house prices. However, the house prices do not appear to have a stable long- run equilibrium relationship with the fundamentals (p. 433-4).
There has been research at a national and a regional level on house price indices, but little is known on how house prices vary within a city or metropolitan area (Archer & Gatzlaff & Ling, 1996, p. 334). Archer, Gatzlaff and Ling (1996) examined the variation in the rates of house price appreciation within a metropolitan market. The metropolitan market used in the paper is that of Dade County (Miami) Florida, from 1971 until 1992 (p. 335). The results showed that over one-half of the 79 census tract groups studied experienced rates of
appreciation that were significantly different from the overall Miami metropolitan market. This shows that house prices vary spatially. The differences can partially be explained by variables such as the location (for instance, which municipality the property is in or its
distance to CBD) or local changes including population and ethnic mix (Archer et al., 1996, p. 351). However, it should be noted that the explanatory power of these variables is small. When controlling for the census tract group location of each home, the explanatory power is merely 12% (Archer et al., 1996, p. 335). Despite the fact that the explanatory power is only 12%, the results indicate that local factors do play a role in explaining differences in regional house prices.
Green, Malpezzi and Mayo (2005) examine price elasticities of housing supply. The article examines the price elasticities of 45 U.S. MSAs over the period 1979-1996. House price data is retrieved from the Fannie Mae repeat-sales index. The price elasticities are estimated using supply-elasticity equations (p. 335). The results show that in 23 MSAs the supply elasticity is significantly greater than zero. The supply elasticities vary substantially between MSAs (Green et al. 2005, p. 336-7). This article shows the housing supply
elasticities vary between regions. This can possible also be the case in the Netherlands suggesting that housing supply can be a possible determinant of regional house prices.
The difference between this thesis and existing literature is that this thesis focuses specifically on the Dutch housing market. Little research has been done on Dutch regional house prices and especially possible determinants. Furthermore, as Reichert (1990) shows, house prices react the same to national economic factor but react differently to local factors. Van Dijk et al. (2011) and Brounen & Huij (2004) show the presence of regional house price differences in the Netherlands, however in another time period than this thesis is examining. The articles previously discussed show evidence for regional house price differences in the past and that there can be multiple reasons for regional house price differences. It can thus be expected that in the case of regional house price differences in the Netherlands, one or more of the above-described determinants could explain the Dutch regional house price differences. In the following section the hypotheses are formulated, among others, based on the above-described literature.
3. Hypotheses 3.1 Hypothesis 1
The first part of the methodology is intended to clearly show whether, and how strongly, differences in regional house prices between the four selected regions exist. I expect there to be significant differences in house price development between the four regions over the period 2000-2016. Furthermore, I expect these differences to have increased even more after the cut-off point of Q2 2013. In this period the differences became stronger than before the cut-off point because the recovery of the Dutch economy and the housing market started. A price index retrieved from the Dutch Central Bureau of Statistics (CBS) shows that house prices in the Netherlands started growing in this period. The price index for the period 2000-2016 of the Netherlands can be found in Appendix A. After the financial crisis the
Amsterdam housing market became overheated. The recovery in the other regions lagged; therefore, differences in regional house price development should have increased after the cut-off point.
3.2 Hypothesis 2
The second part of the methodology looks at which role the determinants in the differences between regions play. The determinants: demographics, income levels, employment housing stock and distance to facilities will be used to explain the differences. I expect that the determinants employment and income will have a significant effect on the regional house price development in the Netherlands. I expect that specifically these two determinants will play an important role because these are most affected the downturn and recovery of the Dutch economy. Furthermore, multiple studies, like Reichert (1990), Abraham and Hendershott (1992) and Baffoe-Bonnie (1998), show a significant relationship between income and employment with house prices.
4. Data and Descriptive Statistics
Four regions are selected for the analyses based on the state of housing market. The four areas are Amsterdam, North-Drenthe, Zeeuwsch-Vlaanderen and Agglomeration ‘s-Gravenhage. Amsterdam is an overheated housing market. North-Drenthe and Zeeuwsch-Vlaanderen are unstressed housing markets. Agglomeration ‘s-Gravenhage is an average housing market.
All data is retrieved from two sources: the Dutch Association of Realtors (NVM) and the Dutch Central Bureau of Statistics (CBS). Data of the local determinants are available at CBS. NVM data will also be used, as their dataset contains transaction prices and multiple housing characteristics, such as the house size in square meters, over the period from 2000 to 2016. Regional data on house prices as well as house price determinants will be used over a sample period if available. The availability is given below per variable.
First, an overview is given in which all variables are separately described. The second part will describe the merging of the different datasets and construction of the final dataset. Finally, I will provide and elaborate upon descriptive statistics.
4.1 Dataset Construction
Multiple datasets from different sources are used to create the dataset used in the analyses. Firstly, the data from the NVM, containing residential transaction data such as transaction prices, housing characteristics, and variables including transaction dates and a municipality identification variable (Gm2017) is used to create the price indices. However, first the identification variable for municipality (Gm2017) had to be linked to the four corop regions. This is done manually. Outliers were present in the NVM dataset and had to be removed. The paper by Dröes and Koster (2016) use a similar dataset, which is also retrieved from the NVM. Based on this paper, house size larger than 250 m2 and smaller than 25 m2, and transaction prices higher than 1,000,000 or below 25,000 euro, are removed. Furthermore, observations are dropped when the construction year is unknown, below 1700, or above 2016. Observations are dropped in case the lot size exceeded 3000 m2, as well as for house volume exceeding 4000 m3. The results from the hedonic model are exported to Excel. From all time dummies the exponent is taken and multiplied by 100 to create each price index. The Excel sheet containing the price indices is imported into STATA as a separate dataset. Secondly, this created dataset is merged on corop region and year with the first dataset of the CBS. This dataset from CBS contains the explanatory variables like population growth and the elderly dependency ratio. Furthermore, this dataset contains identification variables for each region (Corop). This dataset is the final dataset, which is used for both parts of the methodology. The following section will elaborate on all variables present in all datasets used.
4.2 Data overview
Below are listed all the variables that are used in the analyses per part of the methodology. The sources of the data for each variable are also given.
The following variables are used in the hedonic model. All data used in the hedonic model is retrieved from the database of the Dutch association of realtors (NVM) and are available for every transaction over the period from 2000 to 2016. Lntrans is the log of transaction prices. Transaction prices for the period from 2000 to 2016 are available for all municipalities that belong to the four selected regions. Lnwoon is the log of house size in square meters. Perceel is the size of the lot in square meters. Maintenance_good is a dummy variable indicating whether maintenance is satisfactory. This variable is created using two other variables, inbu and onbu, which show, respectively, inside and outside maintenance on a scale of -1 to 9. Inbu and onbu are considered satisfactory when bigger than 6. If both inbu and onbu are bigger than 6, Maintenance_good will be equal to 1. Parking is another dummy variable, which is equal to 1 if a parking space is present at the property. Nkamers is a discrete variable indicating the number of rooms in the property. Garden is a dummy variable equal to 1 when the property in question has a garden. Monument is a dummy variable equal to 1 when the property has a monument. Monumental is a dummy variable equal to 1 when the property is monumental. Furthermore, dummy variables are included for house type. Wtype1 is no house, wtype2 is a row house, wtype3 is a semi-detached row house, wtype4 is a corner house,
wtype5 is a semi-detached house and wtype6 is a detached house. The construction year is
used for construction year dummies, and transaction year is used to create time dummies. The following variables are used in the panel regressions. All data described below is retrieved from the Dutch central bureau of statistics (CBS). Elderly dependency ratio is a ratio showing population aged 65 and over as a proportion of the population aged 20-64. The ratio is given in a percentage. In the dataset there is a ratio for each year per region for the period 2000-2016. Population growth is the relative year-on-year population growth per region for the period from 2000 to 2015. Housing stock growth shows the year-on-year increase in housing stock for each region per year as a percentage. This variable is available for the period from 2000 to 2015. Employment is the average number of jobs in December of each year at firms and institutions per region. The variable has to be multiplied by a 1000 to obtain the number of jobs. The variable is available for the period from 2008 to 2015. Household
income is the average disposable income per household per region for each year. The variable
needs to be multiplied by 1000 to obtain the income level in euros. Data is available for the period from 2006 to 2015. Distance is the average distance of to facilities in the vicinity. The
average is calculated out of distances to four different facilities. These four facilities are: a hospital, primary school, train station and retail shops. The distances are shown in kilometers. The data is shown per region per year. Data on distances is available for the period from 2006 to 2015.
Two identification variables are present in the dataset. Gm2017 is an identification variable containing a unique code corresponding to a municipality in the Netherlands. The dataset consist of 15 municipalities, belonging to the four selected regions. This variable is retrieved from the Dutch Central Bureau of Statistics (CBS). Corop is another identification variable used for the four regions. This discrete variable is equal to 1 for Amsterdam, 2 for North-Drenthe, 3 for Zeeuwsch-Vlaanderen and 4 for Agglomeration ‘s-Gravenhage. This variable is retrieved from CBS. The following section will give descriptive statistics on the final dataset.
4.3 Descriptive Statistics
The following tables show the descriptive statistics per region. The descriptive statistics of the residential transactions can be found in appendix C.
Table 1
Descriptive statistics: Amsterdam
Variable Obs Mean Std. Dev. Min Max
Price index 68 130.43 16.09 100.00 177.45
Elderly dependency ratio 68 16.99 0.57 16.20 18.20
Housing stock growth 64 6.89 3.61 3.30 17.80
Employment 32 545.39 24.53 514.40 591.50 Population growth 64 8.23 5.90 -0.30 16.10 Household income 40 30.23 1.98 26.10 33.20 Distance 40 1.40 0.32 0.50 1.73 Corop 68 1 1 1 Year 68 2008 2000 2016
Table 1 shows the descriptive statistics for Amsterdam. The mean price index is 130.43. The mean elderly dependency ratio is 16.99%. Housing stock grows with a mean of 6.89% per year. The mean of employment is 545,390 jobs. The relative population growth is, on
average, 8.23% per year. The mean household income is 30,230 euro. The average distance to facilities in the vicinity is 1.40 kilometer.
Table 2
Descriptive statistics: North-Drenthe
Variable Obs Mean Std. Dev. Min Max
Price index 68 135.53 12.45 100.00 151.64
Elderly dependency ratio 68 30.47 4.38 25.60 39.10
Housing stock growth 64 7.06 6.32 -10.10 15.70
Employment 32 72.50 1.03 70.40 74.30 Population growth 64 4.00 3.93 -0.90 12.00 Household income 40 33.54 1.79 29.70 36.10 Distance 40 4.55 1.32 0.80 6.03 Corop 68 2 2 2 Year 68 2008 2000 2016
Table 2 shows the descriptive statistics for North-Drenthe. The mean price index for this region is 135.53. The mean elderly dependency ratio is 30.47%. Housing stock grows with a mean of 7.06% per year. The mean of employment is 72,500 jobs. The relative population growth is, on average, 4.00% per year. The mean household income is 33,540 euro. The average distance to facilities in the vicinity is 4.55 kilometer.
Table 3
Descriptive statistics: Zeeuwsch-Vlaanderen
Variable Obs Mean Std. Dev. Min Max
Price index 68 155.45 20.50 100.00 183.18
Elderly dependency ratio 68 34.15 4.70 29.20 43.10
Housing stock growth 64 2.35 10.65 -36.70 11.70
Employment 32 40.09 1.08 40.60 43.50 Population growth 64 -0.84 2.43 -3.40 5.10 Household income 40 32.43 1.76 28.70 34.90 Distance 40 12.05 4.03 0.70 16.93 Corop 68 3 3 3 Year 68 2008 2000 2016
Table 3 shows the descriptive statistics for Zeeuwsch-Vlaanderen. The mean price index is 155.45. The mean elderly dependency ratio is 34.15%. Housing stock grows with a mean of 2.35% per year. The mean of employment is 40,090 jobs. The relative population growth is,
on average, -0.84% per year. The mean household income is 32,430 euro. The average distance to facilities in the vicinity is 12.05 kilometer.
Table 4
Descriptive statistics: Agglomeration's-Gravenhage
Variable Obs Mean Std. Dev. Min Max
Price index 68 134.33 10.91 100.00 150.45
Elderly dependency ratio 68 24.00 1.12 22.90 26.30
Housing stock growth 64 10.01 3.71 4.40 18.00
Employment 32 392.74 9.45 378.40 406.00 Population growth 64 8.42 2.90 0.00 12.20 Household income 40 33.37 1.73 29.50 35.70 Distance 40 1.64 0.41 0.50 2.13 Corop 68 4 4 4 Year 68 2008 2000 2016
Table 4 shows the descriptive statistics for the Agglomeration ‘s-Gravenhage. The mean price index for Agglomeration ‘s-Gravenhage is 134.33. The mean elderly dependency ratio is 24.00%. Housing stock grows with a mean of 10.01% per year. The mean of employment is 392,740 jobs. The relative population growth is, on average, 8.42% per year. The mean household income is 33,370 euro. The average distance to facilities in the vicinity is 1.64 kilometer. The following section will elaborate on the methodology used to answer the hypotheses.
5. Methodology
The methodology consists of two parts. First, multiple difference-in-difference regressions will be performed to check for regional price differences. The second part of the methodology will consist of panel regressions. These will show which determinants play a role in the regional house price differences in the Netherlands. The entire methodology will be
performed for different sample periods depending on data availability. The following sections will elaborate on the difference-in-difference model and the panel regressions used for the analyses.
5.1 Difference-in-difference model
The econometric model uses transaction prices, provided by NVM. All transaction prices of the selected regions over the time period 2000-2016 are used in the model. A price index for each region is created using the transactions prices provided by NVM. These price indices are created used a hedonic regression in which the created price indices control for housing characteristics. The price index is created using quarterly time dummies. The base year is the first quarter of 2000.
This part of the methodology is important for determining the validity of the first hypothesis. The first step is to determine whether there are significant differences in house price development between regions. This will be accomplished with a difference-in-difference (DID) model. This empirical set up will show the causal effects over time of an event (Stock & Watson, 2012, p. 533). This model will compare the price development over time of the selected regions. In the DID model, all four regions will be compared. This results in six different regressions. The difference-in-difference model will be estimated using an OLS regression with dummy variables showing an average treatment effect, which resembles the regional house prices price difference between the treatment group and the control group. In each regression one of the regions is the treatment group and the other region the control group. The following equation is the standard OLS equation for estimating a DID model:
Yit = α + β*Treati + γ*Postt + δ*Treati X Postt + εit
Yit is the outcome variable. This is the price index of the selected region i at time t. Treati is a
dummy variable that indicates whether i belongs to the treatment group (=1 for the treated region even before treatment). Postt is another dummy variable that indicates whether an
observation belongs to the post-treatment period (=1 even for non-treated regions).
β is the difference of mean outcome of the treatment group vs. the control group before the
treatment. γ is the difference of the mean outcome of the control group after the time of treatment relative to the control group before. δ is the difference-in-difference, the shift in mean outcome of treatment group from pre- to post-treatment relative to the mean shift in outcome for the control group. In other words, δ measures the change of the outcome in the treatment group from pre- to post-treatment relative to the control group (Peters, 2016, p. 12-3). The coefficient δ resembles the average treatment effect. The post-treatment period in the DID model is after the second quarter of 2013. This year is chosen because this was the end of financial crisis and when the recovery of the Dutch housing market began. A price index
retrieved from the Dutch Central Bureau of Statistics (CBS) shows that house prices in the Netherlands started growing in this period. The price index for the Netherlands can be found in Appendix A. As stated in the first hypothesis, this is the period in which it is expected that the regional price differences increased.
The significance of δ is very important for determining the validity of the first hypothesis. Whether the significance is positive or negative is not of importance for
answering the first hypothesis, as regional house price differences can be negative or positive. Regional house price differences are present in the case of a significant coefficient δ.
5.2 Panel regressions
Step two is determining the important determinants that explain the significant differences in regional house price development. The determinants that will be used are: demographics, household income, housing stock growth, employment and distance to facilities. The
determinants that will be used are based on previous reviewed literature. Reichert (1990) and Harter-Dreiman (2004) found a significant relationship between house prices and income levels. Ohtake and Shintani (1996) and Archer et al. (1996) find a relationship between house prices and demographics. Abraham and Hendershott (1992) and Baffoe-Bonnie (1998) find a significant relationship between house prices and employment. Archer et al. (1996) finds that house price differences can partially be explained by variables such as the location (for instance, which municipality the property is in or its distance to CBD). Green et al. (2005) show that housing stock can be a possible determinant for regional house price differences.
Multiple panel regressions will be used to find significant determinants for the regional house price developments in the Netherlands. Multiple determinants, such as those described above, will be added to the regression as independent variables.
In all panel regression the constructed price indices of the four selected regions is the dependent variable. The log of the created price indices will be included as the dependent variable in the regression because this is the main relationship that this thesis is interested in. In total four regressions will be performed with different configurations. The first regression will have no fixed effects. In the second panel regression time fixed effects will be taken into account. The third panel regression will take region fixed effects into account. The last panel regression will take both time and region fixed effects into account. Due to limited data availability, the panel regressions are performed for the period from 2008 to 2015. In this period data for all selected independent variables are available. All panel regressions are
robust in order to control for heteroskedasticity. The following regression equation could be used to find significant determinants for each region:
Ln Yit= β0 + β1* Elderly dependency ratioit + β3* Population growthit + β4* Housing stock
growthit + β5* Employmentit + β6*Household incomeit + β7* Distanceit + Φt + δi + εit
Ln Yit is the log of the price indices of each region i at time t. Each independent variable is
mentioned in the equation above for region i and at time t. Φt captures the time fixed effects and δi captures the region fixed in the panel regression.
The effects are shown by the coefficients of determinants such as income and employment, in the case when the coefficients are significant. The coefficient of each significant determinant will show in what direction and how strongly the determinant influences regional house prices. The coefficient of income and employment should be positive to have economic meaning; Reichert (1990) found a positive and significant relationship between house prices and income levels (p. 381-2). Abraham and Hendershott (1992) and Baffoe-Bonnie (1998) find similar results for employment levels. Increases in employment and income levels positively influence demand for housing, which results in price increases. Ohtake and Shintani (1996) and the Mankiw-Weil model with demographic factors shows a sharp decline in house prices after the age of 60 in Japan (p. 192). Therefore, a negative coefficient is expected for the variable elderly dependency ratio. Population growth should positively influence house price because a higher demand or housing is created.
Housing stock should negatively influence house prices because an increase in housing stock increases supply. For distance a negative coefficient is expected, as certain facilities being distant from the property would be inconvenient and should negatively influence house prices. Furthermore, Archer, Gatzlaff and Ling (1996) also find a negative relationship between house price appreciation rates and distance to CBD (p. 348).
The significance of the coefficient plays an important role in determining the validity of the second hypothesis. Magnitude and direction of the coefficient of the significant variables produced in the panel regression in collaboration with the changes in levels of the chosen determinants before and after the cut-off point for each region can give an explanation on regional house price differences in the Netherlands. In the results section, each significant variable is discussed separately. However, the direction and magnitude of the coefficient should make economic sense. The following section will discuss the results estimated by the difference-in-difference model and the panel regressions.
6. Results
In the following section the results from the difference-in-difference model and panel
regression will be discussed. First, the results from the difference-in-difference model will be shown and discussed. Secondly, the results from the panel regressions will be shown and discussed.
6.1 Price Indices
To perform the difference-in-difference analysis, price indices per region are needed. As mentioned above, these are created using the dataset from the NVM in a hedonic model. The regression output of the hedonic model can be found in appendix D. An overview of the price indices per region can be found in appendix E. The Dutch central bureau of statistics provides a price index for Amsterdam. This price index is compared to the constructed price index of Amsterdam to check for fidelity. Both indices show a similar pattern of price development in Amsterdam over the period 2000-2016. This comparison is illustrated in appendix B.
6.2 DID
In the classic difference-in-difference model the coefficient from treatpost shows the difference-in-difference effect (average treatment effect). This effect is the difference in regional house prices in the model between the four selected regions used in the analyses. Table 5 shows the results in a matrix format from the difference-in-difference model, in which price indices of the selected region are the dependent variable. Regression output from the difference-in-difference model can be found in appendix F.
Table 5
Difference-in-difference model matrix
Average treatment effect Amsterdam North-Drenthe Vlaanderen Zeeuwsch- Agglomeration ‘s-Gravenhage
Amsterdam x 16.4374*** -4.1668 13.5012***
North-Drenthe x x -20.6042*** -2.9362
Zeeuwsch-Vlaanderen x x x 17.6680***
Agglomeration ‘s-Gravenhage x x x x
Notes: Treatpost shows the average treatment effect. Significance is given by *** p<0.01, ** p<0.05, * p<0.1. The model shows that when Amsterdam is compared to North-Drenthe, with
treatment effect is significant at 1%. This result means that house prices grew about 16.44% after the second quarter of 2013 in Amsterdam than in the corop area North-Drenthe. This result shows that regional house price differences are present between these two areas. The R-squared of the model is 0.1755, which means the explanatory power of the model is roughly 17.55%.
The model shows that when Amsterdam is compared to Zeeuwsch-Vlaanderen, in which Amsterdam is the treatment group, the average treatment effect is -4.1668. The results imply that house prices in Amsterdam grew 4.17% les than house prices in
Zeeuwsch-Vlaanderen after the second quarter of 2013. However, the average treatment effect is not significant, so no conclusion can be drawn on regional house price differences between these two regions. Furthermore, this result is not in line with expectations. It should be noted that very few transactions took place in the corop area Zeeuwsch-Vlaanderen, especially in comparison with Amsterdam, which could have had an influence on this result. The R-squared of the model is 0.4763, which means the explanatory power is roughly 47.63%.
The model shows that when Amsterdam is compared to Agglomeration ‘s-Gravenhage, in which Amsterdam is the treatment group, the average treatment effect is 13.5012. The average treatment effect is significant at 1%. This shows that house prices grew about 13.50% more after the second quarter of 2013 in Amsterdam, than in the corop area Agglomeration ‘s-Gravenhage. This result shows that regional house price differences are present between these two areas. The R-squared is 0.1871. This difference-in-difference model has an explanatory power of 18.71%.
The model shows that when North-Drenthe is compared to Zeeuwsch-Vlaanderen, in which North-Drenthe is the treatment group, the average treatment effect is -20.6042. The average treatment effect is significant at 1%. This result shows that there are regional house Price differences between these two corop areas. The corop area North-Drenthe grew 20.60% less than the area Zeeuwsch-Vlaanderen did after the second quarter of 2013. Again it should to be noted that very few transactions took place in the corop area Zeeuwsch-Vlaanderen, which could have had an influence on this result because the result seems quite large. The R-squared of this difference-in-difference model is 0.3824. The explanatory power of the model is roughly 38.24%.
The model shows that when the two corop areas North-Drenthe and Agglomeration ‘s-Gravenhage are compared, with North-Drenthe as the treatment group, the average treatment effect is -2.9362. The average treatment effect is not significant. The result implies that house prices grew 2.94% less in the corop area North-Drenthe than in the corop area Agglomeration
‘s-Gravenhage after the second quarter of 2013. Due to the insignificant results no conclusion can be drawn regarding the presence of regional house price differences. The R-squared of the model in which these areas are used is 0.0130, which means the explanatory power of this model is 1.30%. This is quite low compared to the other difference-in-difference models used for the other comparisons.
The model further shows that when Zeeuwsch-Vlaanderen is compared to Agglomeration ‘s-Gravenhage, with Zeeuwsch-Vlaanderen as the treatment group, the average treatment effect is 17.6680. The average treatment effect is significant at 1%. The result shows that the corop area Zeeuwsch-Vlaanderen grew 17.67% more compared to the Agglomeration ‘s-Gravenhage after the second quarter of 2013. This result is larger than expected. Again it should to be noted that very few transactions took place in the corop area Zeeuwsch-Vlaanderen, which could have had an influence on this result. However, the
average treatment effect is significant, which means there are regional house price differences between the areas. The R-squared of this model is 0.4239, which means the explanatory power is 42.39%.
The above-described results from the difference-in-difference model show that there are regional house price differences present in the Netherlands between the selected regions. The CBS, as well, reported that regional house prices have risen in 2016 just like in previous years (CBS, 2017). The above-described results and the findings from the CBS present a solid basis to perform the second part of the methodology. The following section will show results from the second part of the methodology, in which these regional house price differences will be explained.
6.3 Panel regressions
Table 8 shows the results from the model in which the regional price differences found in the previous section of the results are explained. In this model the log of price indices of the selected regions is the dependent variable. The period used in table 8 is from 2008 to 2015. A conclusion on which determinants played a role in the regional house prices in the
Netherlands can be drawn in collaboration with the data from tables 6 and 7. These tables show the levels of the chosen determinants before and after the cut-off point for each selected region.
Table 6
2008 Q1- 2013 Q2
Amsterdam North-Drenthe Zeeuwsch-Vlaanderen
Agglomeration 's-Gravenhage
Elderly dependency ratio 16.44 31.34 35.21 23.25
Population growth 13.67 1.62 -2.25 9.79
Housing stock growth 7.26 3.65 -2.66 10.31
Employment 531.83 72.76 42.70 397.24
Household income 30.55 33.91 32.76 33.83
Distance 1.46 4.81 12.93 1.71
Table 7
2013 Q3 – 2015 Q4
Amsterdam North-Drenthe Zeeuwsch-Vlaanderen
Agglomeration 's-Gravenhage
Elderly dependency ratio 17.32 36.90 41.06 25.36
Population growth 13.96 0.08 -1.36 8.54
Housing stock growth 11.54 2.30 7.02 8.80
Employment 575.22 71.92 40.74 382.84
Household income 32.28 35.24 34.16 34.86
Table 8
Panel regression
(1) (2) (3) (4)
Ln(Price index) Ln(Price index) Ln(Price index) Ln(Price index)
Elderly dependency ratio -0.0072*** 0.0010 -0.0060** -0.0030
(0.0023) (0.0029) (0.0024) (0.0024)
Population growth 0.0122*** 0.0071*** 0.0120*** 0.0040*
(0.0018) (0.0022) (0.0015) (0.0020)
Housing stock growth 0.0003 0.0004 -0.0001 -0.0003
(0.0003) (0.0005) (0.0003) (0.0005) Employment -0.0003*** 0.0001 0.0004 0.0001 (0.0000) (0.0001) (0.0003) (0.0004) Household income 0.0155*** 0.0147*** 0.0037 0.0604** (0.0058) (0.0043) (0.0074) (0.0279) Distance 0.0312*** 0.0274*** -0.0326* -0.0808*** (0.0023) (0.0023) (0.0170) (0.0188) Constant 4.5137*** 4.2605*** 5.0355*** 3.1301*** (0.1529) (0.1006) (0.2003) (0.6744)
Time fixed effects No Yes No Yes
Region fixed effects No No Yes Yes
Observations 128 128 128 128
R-squared 0.8502 0.9185 0.8731 0.9482
Notes: Dependent variable: log of price indices. Standard errors are in parentheses. Significance is given by ***
p<0.01, ** p<0.05, * p<0.1.
The R-squared for regression (1) is 0.8502, which means the explanatory power is 85.02%. No fixed effects are used in the first regression. The second regression, with time fixed effects, has an explanatory power of 91.85% resembling a squared of 0.9185. The R-squared for regression (3) is 0.8731, which means the explanatory power is 87.31%. In regression (3) regional fixed effect are taken into account. This R-squared for the last
regression 0.9482, in which both time and region fixed account are taken into account. In all regressions the number of observation is 128 due to some missing values. The fourth
regression is most suited to analyze regional house price differences because this model has the highest explanatory power. Furthermore, this model controls for time and regional fixed effects. The model shows that population growth, household income and distance are possible determinants for regional house price differences in the Netherlands. In the following
paragraphs these determinants will one-by-one be discussed in combination with the data from tables 6 and 7.
significant at 10% and the coefficient is positive. When population growth increases with one unit change this will positively influence regional house prices by 0.40%. This coefficient makes economic sense. Growth in population creates a higher demand for housing. The higher demand for housing increases prices. When looking at tables 6 and 7 it is evident that only in Amsterdam population growth accelerated after the cut-off point. In North-Drenthe and Agglomeration ‘s-Gravenhage population grew but at a lesser rate than before the cut-off point. In Zeeuwsch-Vlaanderen population shrinkage decreased but still shrinks after the cut-off point. The fact that population growth accelerated in Amsterdam and decreased North-Drenthe and Agglomeration ‘s-Gravenhage can explain the significant regional price differences between Amsterdam in comparison with the regions North-Drenthe and
Agglomeration ‘s-Gravenhage. House prices grew faster in Amsterdam after the cut-off point than in North-Drenthe and Agglomeration’s-Gravenhage and population growth can be considered on of the determinants. The same could be argued for Zeeuwsch-Vlaanderen. However, no significant regional price differences were found between these two regions.
Income is another variable mentioned by Reichert (1990), which could play a role in explaining regional house prices. In this analysis, household income is used. When household income increases with one unit change this will influence the price indices of each region with 6.04%. The coefficient is significant at a 5% level. Table 6 and 7 show that Amsterdam has seen the biggest increase in household income after the cut-off point. The other regions have also seen an increase but a slightly lower one. The previous part showed that house price grew faster in Amsterdam than in North-Drenthe and Agglomeration ‘s-Gravenhage. Together with bigger increase in household income and the positive significant coefficient it can be
concluded that household income partially explains these regional house price differences between Amsterdam and these two regions. Again, the same could be argued for Zeeuwsch-Vlaanderen, however as in the previous part, no significant regional price differences were found between Amsterdam and Zeeuwsch-Vlaanderen. The positive link between house prices and income is in line with literature; Reichert (1990) also finds a positive and
significant link on a national and a regional level (p. 382). The results make economic sense because higher income creates a higher demand for housing, which causes prices to rise.
Distance was the last variable examined. When examining the possible determinant distance it is noted that the coefficient is negative and significant at a 1% level. When the variable distance increases with one unit change this will influence the price indices of the regions with -8.08%. This coefficient makes economic sense because houses get less attractive, when they are located further away from facilities like schools or a hospital. The
results are in line with existing literature and make economic sense as mentioned earlier. The paper by Archer, Gatzlaff and Ling (1996) find a negative relationship between house price appreciation rates and distance to CBD (p. 348). When examining tables 6 and 7 it is clear that for Zeeuwsch-Vlaanderen distance decreased after the cut-off point and for the other regions distance increased. Between Amsterdam and Zeeuwsch-Vlaanderen no significant regional price differences could be found. However, house prices grew faster after the cut-off point in Zeeuwsch-Vlaanderen than in North-Drenthe and Agglomeration ‘s-Gravenhage. The fact that distance increased in Zeeuwsch-Vlaanderen and decreased in the other two regions can partially explain the regional house price differences. Therefore, the variable distance can be considered one of the determinants of regional house prices in the Netherlands.
The variables elderly dependency ratio, housing stock growth and employment, based on related literature, are insignificant in the model. Therefore, no further no conclusions can be drawn on their influence on regional house price differences and cannot be considered possible determinants. The following section will perform robustness checks on the above analyses.
7. Robustness Checks
Table 9 and 10 shows the results from an adjusted analysis similar to the second part of the methodology to check the robustness. In table 9 a panel regression with a log-log model configuration is used for the variables household income, employment and distance. In this model the log of price indices is still the dependent variable. The period used in table 9 is from 2008 to 2015.
Table 9
Panel regression: log-log model
(1) (2) (3) (4)
Ln(Price index) Ln(Price index) Ln(Price index) Ln(Price index) Elderly dependency ratio -0.0045** 0.0022 -0.0032 -0.0075***
(0.0023) (0.0026) (0.0028) (0.0024)
Population growth 0.0103*** 0.0084*** 0.0112*** 0.0030
(0.0019) (0.0022) (0.0015) (0.0021)
Housing stock growth 0.0003 0.0004 0.0002 -0.0001
(0.0003) (0.0005) (0.0003) (0.0005)
Employment (log) 0.0811*** 0.1429*** 0.3329** -0.4437**
(0.0125) (0.0147) (0.1272) (0.1919)
Income household (log) 0.4032* 0.3434* 0.0152 3.8927***
(0.2074) (0.1829) (0.2927) (0.9080)
Distance (log) 0.2869*** 0.2931*** -0.3041* -0.4598**
(0.0220) (0.0203) (0.1820) (0.2238)
Constant 2.8646*** 2.5762*** 3.6508*** -5.3918**
(0.6930) (0.5796) (1.0213) (2.2435)
Time fixed effects No Yes No No
Region fixed effects No No Yes Yes
Observations 128 128 128 128
R-squared 0.8335 0.9233 0.8746 0.9443
Notes: Dependent variable: log of price indices. Standard errors are in parentheses. Significance is given by
*** p<0.01, ** p<0.05, * p<0.1.
The independent variables of household income, employment and distance can be interpreted as elasticities in this model. If the results of column 4 from table 8 and table 9 are compared, it is noticeable that variables changed regarding the significance. In table 8, population growth, household income and distance were significant. In table 9, elderly dependency ratio, employment, household income and distance are significant. Furthermore, the directions of the coefficients are the same for the coefficients, which are significant in both tables. When looking at the coefficients of population growth, household income, and distance it becomes clear that the magnitude of the coefficients changed. In table 8, the effect of population growth on the log of price indices is not comparable due to insignificance in table 9. In table 8, the effect of household income on the log of price index is 6.04%, and in table 9 the effect is 3.89%. In table 8 the effect of distance on the log of price index is -8.08% and in table 9 the effect is –0.46%. This robustness check shows that direction of the determinants is the same but differences in significance and magnitude of the determinants. Therefore, the estimation results are not robust. Furthermore, the explanatory power of the estimated models is similar tot those in table 8.
Table 10 shows the same analysis as in in table 8 with a different dependent variable. In this model the dependent variable is the normal price indices instead of the logarithm of these price indices. The period used in table 10 is the same as in table 8 and is from 2008 to 2015.
Table 10
Panel regression
(1) (2) (3) (4)
Price index Price index Price index Price index
Elderly dependency ratio -0.9993*** 0.1239 -0.8188** -0.5089
(0.3453) (0.4192) (0.3607) (0.3645)
Population growth 1.7165*** 1.0240*** 1.6882*** 0.5909*
(0.2566) (0.3322) (0.2237) (0.3033)
Housing stock growth 0.0342 0.0502 -0.0152 -0.0378
(0.0531) (0.0719) (0.0509) (0.0691) Employment -0.0380*** 0.0149 0.0533 -0.0095 (0.0070) (0.0122) (0.0406) (0.0649) Household income 2.1934*** 2.1882*** 0.4795 9.9089** (0.8307) (0.6252) (1.0619) (4.3731) Distance 4.6768*** 4.1832*** -4.5284* -10.9410*** (0.3444) (0.3326) (2.6862) (2.8619) Constant 77.4549*** 38.9598*** 153.8069*** -147.9170 (21.5345) (14.5224) (29.0465) (105.5107)
Time fixed effects No Yes No No
Region fixed effects No No Yes Yes
Observations 128 128 128 128
R-squared 0.8609 0.9234 0.8821 0.9494
Notes: Dependent variable: Price indices. Standard errors are in parentheses. Significance is given by ***
p<0.01, ** p<0.05, * p<0.1.
Comparing the results of column 4 from tables 8 and 10, it is noticeable that nothing happens with the significance of the variables. All three determinants are significant at the same level in both tables. All the directions of the coefficients are the same in both tables. The coefficient is larger due to the fact that the dependent variable is now a price index and not a log of price index; therefore, the coefficient should be roughly 100 times as large. While the coefficients are larger, there is still variation in the effect of the independent variables in regional house price development. For example, in table 8 the coefficient is 0.0040 and in table 10 the coefficient is 0.5909. This is more than 100 times larger. In table 8 the coefficient household income is 0.0604. In table 10 the coefficient is 9.9089. Again, this is more than 100 times
