Master Thesis
Effect of shopping centers on the value of surrounding residential properties in the Netherlands
This thesis investigates the effect of proximity of small neighborhood shopping centers on the value of surrounding residential properties in the Netherlands. Housing transaction data of the Dutch Association of Realtors, neighborhood data of the CBS and shopping center data of Locatus for the years 2016 and 2017 are used. Results of the hedonic regression model show that the proximity of shopping centers negatively influences house prices in the near vicinity by 1.6 percent. As distance increases, the effect becomes positive up to a maximum point after which house prices decline. Furthermore, this thesis does not provide enough evidence for a significant effect of the size of a shopping center on the value of surrounding residential properties.
July 1, 2018
Rosanne Steur, 10753230
MSc Finance: Real Estate Finance & Finance Thesis supervisor: dhr. dr. M.A.J. Theebe
1
Statement of originality
This document is written by student Rosanne Steur who declares to take full responsibility for the contents 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.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
2 Table of Contents
Statement of originality 1
1. Introduction 3
2. Literature review 6
2.1 Relationship between shopping centers and the value of surrounding houses 6
2.2 Different externalities 7
2.3 Non-monotonic relationship 9
2.4 Theory of Christaller 10
2.5 Difference between shopping centers in the Netherlands and U.S or Canada 11
3. Data 12
3.1 Descriptive statistics housing transaction data 12
3.2 Descriptive statistics neighborhood data 15
3.3 Descriptive statistics shopping center data 17
3.4 Distance between houses and nearest shopping center 18
4. Methodology 19
4.1 Hedonic regression model 19
4.2 Alternative regression specification 21
4.3 Regression diagnostics 22
4.4 Expectation of control variables 23
4.5 Endogeneity 24
5. Results 26
5.1 Proximity of shopping centers 26
5.2 House characteristics 33
5.3 Neighborhood control variables 34
6. Robustness check 35 7. Conclusion 38 8. References 40 Appendix 43 Appendix 1 43 Appendix 2 43 Appendix 3 44
3
1. Introduction
A recently published study of ‘Pararius’, an online platform for owner-occupied and rental houses, argues that people attach great importance to the location of a house. Pararius reveals that the majority of the people is willing to give up living area or garden surface, to live in a desired location (Van Leeuwen, 2018). Especially, the social and functional factors, like the quality of the neighborhood and the proximity of employment, are important. Therefore, the price of a house is determined by the characteristics of the house such as the size and the age, as well as by the environmental factors of the location (Song and Knaap, 2004). Supported by Visser and Van Dam (2006), the presence of a highway road, public transport, schools and shopping centers could therefore influence the value of a residential property, as part of the environmental amenities.
This thesis focusses on the relationship between small neighborhood shopping centers and the value of residential properties. Comparable to other forms of externalities, shopping centers create positive as well as negative effects. This translates into attraction and aversion effects that influence buyer’s willingness to pay for a house. Hence, house prices should incorporate both the positive and negative external effects of shopping centers (Des Rosiers, Lagana & Thériault, 2001). Living in the vicinity of a shopping center exerts positive externalities like the ease in shopping, amusement facilities and less travel time. However, living too close to a shopping center also creates negative externalities such as traffic noise, pollution and congestion (Sirpal, 1994). Wilhemsson (2000) and Theebe (2004) both provide evidence that traffic noise has a significantly negative impact on the value of surrounding houses. Furthermore, Wilhelmsson (2000) asserts that negative effects decline with distance.
During last years, several research is conducted regarding the effect of the size of a shopping center on the value of surrounding residential properties. These studies focus among other things, on the U.S and Canada. For example, Sirpal (1994), Des Rosiers, Lagana, Thériault & Beaudoin (1996) examine the difference in price of otherwise similar residential properties, located around different sized shopping centers in Gainesville and Canada, respectively. From these studies, it is found that the size of a shopping center has a positive and significant effect on the value of surrounding residential properties. In addition to Sirpal (1994), Des Rosiers et al. (1996) also focus on the proximity of shopping centers. To supplement existing research, this thesis will examine whether the effect of proximity of shopping centers on the value of residential properties holds in the Dutch market, in the time period 2016-2017 as well. The variable of interest is the distance of shopping centers. Hence, the main focus is on the effect distance might have regarding to the negative externalities of shopping centers. The negative externalities might exceed the
4 positive effects, the closer the residential properties are to the shopping center. Therefore, this thesis will focus on the following question:
‘Does the proximity of shopping centers affects the value of surrounding residential properties in the Netherlands?’ Furthermore, if there is a relationship between house prices and the proximity of shopping centers, the next question is whether this effect weakens or changes as distance increases. Hence, the sub-question is:
‘Does the effect of proximity of shopping centers on the value of surrounding residential properties diminish or becomes the opposite as distance increases?’
This thesis contributes to the current literature by focusing on small neighborhood centers in the Dutch market, because the effect on house prices in the Netherlands is examined on a small scale. Furthermore, European and U.S type of shopping centers differ in terms of function and size. According to the International Council of Shopping Centers (ICSC), U.S. and Canadian shopping centers are characterized by convenience-, neighborhood-, community- or regional centers. A convenience center is most comparable to the Dutch small shopping center in terms of typical type of anchor shops. The European kind of shopping centers are much smaller than the U.S. or Canadian types of shopping centers (ICSC). Apart from this difference, to examine if the results found by Des Rosiers et al. (1996) and Sirpal (1994) are also to be found in the Netherlands in the time period 2016-2017, data of 2016 and 2017 will be used.
For this thesis, the database of the Dutch Association of Realtors (NVM) is used for the transaction prices and house characteristics. In addition, data about the neighborhood characteristics is obtained from Statline, the database of the Central Bureau of Statistics. Furthermore, data about the size and location of shopping centers is gathered from Locatus. Results show that the proximity of shopping centers negatively influences house prices in the near vicinity by 1.6 percent. As distance increases, the effect becomes positive up to a maximum point after which house prices decline.
The outcome of this thesis can be of interest for municipalities, policy makers and developers when determining a zoning plan for new construction. They can consider the effect shopping centers might have on the valuation and affordability of houses. On top of that, it could be interesting for homeowners and potential buyers because the value of a house is part of household’s wealth and it determines their profit in case of sale (Hurst & Lusardi, 2004). Lastly,
5 the outcome is worth knowing to investors who invest in retail and the associated homes in the direct surrounding. Considering the effect of shopping centers, they can anticipate on the return of the investment.
The remainder of this thesis continuous as follows. Section 2 provides the literature review. Subsequently, section 3 evaluates the data used by presenting descriptive statistics. Thereafter, section 4 describes the methodology and formulates the hypothesis. Then, section 5 analyses the results, followed by the robustness check in section 6. Finally, section 7 provides a conclusion regarding this research including limitations and recommendations for further research.
6
2. Literature review
To start with, section 2.1 describes the effect of proximity of shopping centers on house prices by evaluating existing literature about the relationship between shopping centers and the value of surrounding residential properties. Section 2.2 reviews studies concerning different sources of externalities. Section 2.3 describes the non-monotonic price-distance relationship. Section 2.4 explains the theory that is the basis of the retail structure. Lastly, section 2.5 provides an overview of the difference between the U.S or Canadian and the Dutch shopping centers.
2.1 Relationship between shopping centers and the value of surrounding houses
Studies done by Sirpal (1994), Addae-Dapaah and Lan (2010) and Des Rosiers et al. (1996) focus on the effect of shopping centres. First of all, Sirpal (1994) examines the effect of the size of a shopping center on the value of surrounding residential properties. He provides evidence that the size of shopping centers has a significant positive effect on the value of residential properties in the surrounding. Sirpal (1994) conducts his research with data of residential properties around three shopping centers of various sizes, in a city in Florida in June 1990. The study is done by means of a multiple regression model, using various forms. The second study by Addae-Dapaah and Lan (2010) investigates the effect of shopping centers on the value of residential apartments in Singapore. Using a hedonic price model it is found that shopping centers positively contribute to the price of an apartment. However, this positive effect non-consistently declines as distance increases.
The most similar paper regarding to this thesis, is the paper from Des Rosiers et al. (1996). They not only investigate the effect of size of a shopping center on the value of surrounding residential properties, but also the effect of the proximity of a shopping center. They conduct their research in Canada between 1990 and 1991. In accordance with Sirpal (1994), they provide evidence of a positive relationship between the size of a shopping center and residential property values, confirmed using various functional forms. Moreover, they provide evidence that the value of residential properties rises the more proximate the property to the nearest shopping center, up to a maximum point within 200-300 meters, after which prices fall. From this, they conclude that the relationship between house prices and distance to shopping centers is non-monotonic.
In addition, research is conducted regarding the effect of shopping centers on residential properties in the Netherlands. First of all, two of these researches focus on the effect of restructuring of shopping centers. Using a repeat sales method and a difference-in-difference model, Voorn (2013) provided evidence of a positive effect of the restructuring of a shopping center on the value of surrounding residential properties in Zwolle and Den Haag, in the period
7 2000-2012. Secondly, De Vries (2015) confirms this effect with a hedonic price model. Apart from these studies, Visser and van Dam (2006) implicitly take into account the effect of shopping centers on value of surrounding residential properties. This study investigates the effect of characteristics of the residential environment on the value of dwellings in the Netherlands. So, existing literature about the effect of shopping centers on the value of surrounding residential properties, has been conducted in the Netherlands as well as abroad. However, the research conducted in the Netherlands is limited. Therefore, this thesis contributes to the existing literature by focusing on small established neighborhood shopping centers in the province Noord-Holland.
2.2 Different externalities
This section evaluates studies regarding the effect of different sources of externalities on the value of surrounding residential properties. This section contributes to this thesis, because there are similarities with respect to the methodology and control variables.
During the years, studies investigated the effect of different sources of externalities. To start with, evidence for the effect of non-residential land use is inconsistent. Grether and Mieszkowski (1978) look among other things at highways and industrial properties. They argue that non-residential land use do not affect the value of surrounding dwellings. While Kain and Quigley (1970) suggest a negative effect of non-residential land use on the value of adjacent residential properties. Furthermore, Dröes and Koster (2016) and Koster and Ommeren (2015) focus on visible externalities which negatively affect the value of surrounding houses. Besides the visible externalities, the effect of nuisance in terms of noise and crime can also have a negative impact on house values. Lynch and Rasmussen (2001), Theebe (2004), Wilhelmsson (2000) and Dekkers and Van der Straaten (2009) investigate the effect of different forms of nuisance. Besides the negative impact of externalities, there are also externalities that positively affect the value of residential properties. According to Luttik (2000), attractive environmental attributes can have substantial positive effects on house prices. Nature and landscape characteristics such as a water view and an attractive landscape can increase the value of a house by 8-10% and 5-12% respectively. On the other hand, traffic noise decreases the value of a house by 5%.
In accordance to this, Theebe (2004) and Wilhelmsson (2000) came to the same conclusion. They examine the impact of traffic noise on residential property values in the Netherlands and Sweden respectively. They both find evidence that traffic noise has a substantial negative impact on the value of surrounding houses. Each additional decibel above 55 dB, decreases on average the value of a house by 0.4% (Theebe, 2004). According to Wilhemsson (2000), properties located in a busy area with lot of noise are on average 30% worth less than houses in a more quite region.
8 Moreover, Collins and Evans (1994) provide evidence that the effect of noise differs among house types. Their results show that the prices of detached houses are more affected by aircraft noise compared to semi-detached or terraced houses. A deeper insight is given by Dekkers and van der Straaten (2009). They classify traffic noise into three sources; road, railway and aircraft noise. Their results suggest that aircraft noise causes the main price decreases, followed by train and road traffic. Besides the nuisance of noise, Lynch and Rasmussen (2001) examine the impact of crime on residential property prices. They provide evidence that costs associated with crime does not influence the value of residential properties. Though, houses located in areas with a substantial crime level, assigned less value relative to houses in a more safe area.
Besides the impact non-visible externalities like nuisance, the impact of visible externalities is investigated by Dröes and Koster (2016), and Koster and Ommeren (2015). Dröes and Koster (2016) measure the external effects of windmills on surrounding residential house prices. Their results show, on average, a discount of 1.4% for houses in a range of two kilometers from a windmill. Moreover, they provide evidence that expectations regarding the placement of a windmill put a downward pressure on house prices as well. The subsequent research regarding the visible externalities is done by Koster and Ommeren (2015). They examine the effect of earthquakes on residential house prices, especially earthquakes as a result of natural gas production. They provide evidence of a substantial negative effect of earthquakes in terms of a decrease in house prices of 1.9%.
Although, most of the sources of externalities are not corresponding with this thesis, these studies all have similarities with this research in terms of methodology and control variables. First of all, except the study of Kain and Quigley (1970), all studies use a hedonic price analysis with the transaction price of a house as dependent variable. Moreover, they all include house characteristics, such as the house type, number of rooms, year of construction and the presence of a garden and parking facilities as independent variables. Besides this, most studies include neighborhood control variables as well. Since this study uses data of the NVM, which includes all of these house characteristics and the database of the CBS for neighborhood control variables, the structure of the regression model is predicated upon the methodology of these literature. The study of Des Rosiers et al. (1996) forms the basis for the specification of the proximity of shopping centers, which is used as the main independent variable.
9
2.3 Non-monotonic relationship
To answer the second question, whether the effect of proximity of shopping centers on the value of surrounding residential properties diminishes or becomes the opposite as distance increases, this section discusses the literature about a non-monotonic price-distance relationship.
As mentioned in the introduction, house prices reflect both the characteristics of the house as well as the environmental externalities. Recent literature indicates that the proximity of schools and shopping areas positively affects house prices (Visser and Van Dam, 2006). However, these externalities also generate negative effects for houses in the close surrounding. This suggests that the relationship between house prices and the distance of external facilities is non-monotonic. For example, des Rosiers et al. (1996) provide evidence for the fact that house prices tend to rise op to a maximum point, and then decrease. To capture the effect of non-monotonicity, a quadratic function is used. A negative significant sign of the squared term indicates that the relationship between distance and house prices is non-monotonic (Des Rosiers et al., 1996). Another way in which Des Rosiers et al. (1996) evince this relationship between house prices and distance, is by means of dummy variables for distance. Their results show that house prices tend to rise up to a maximum point within 200-300 meters, after which they drop. This confirms the non-monotonic relationship between house prices and distance as well.
In accordance with Des Rosiers et al. (1996), the study of by Addae-Dapaah and Lan (2010) concludes that the price-distance relationship of shopping centers and the surrounding residential houses, in particular apartments, is non-monotonic. Because their results show that the premium to house prices non-consistently declines within the first 500 meters. Furthermore, Sirpal (1994) provides evidence that house prices rise as the distance to a shopping center increases, after which they fall after a maximum point. However, he cannot support the non-monotonic price-distance relationship. Research should indicate whether the price-distance relationship is non-monotonic in this thesis.
10
2.4 Theory of Christaller
The central place theory of Christaller was originally developed to explain the distribution of cities within a country. However, it could also serves as a basis for the spatial distribution of the retail locations (Forbes, 1972). This theory reveals a hierarchical structure between the population and the services of shopping centers. The dispersion of shopping centers looks like a hexagon, in which the biggest centers are located at the most central spots, whereas the smaller neighborhood centers are located at non-central places to serve a smaller service area or hinterland (Evers et al., 2011). The higher-order centers are located at greater distance from each other, while the lower-order centers are more closer. The main idea of the central place theory is that centers of identical size, rank and function are equally divided among cities and inhabitants (Geltner, Miller, Clayton & Eichholtz, 2013).
Christaller’s theory describes two main subjects. First of all, each center has its own threshold market. This refers to the population that is needed to remain profitable. Secondly, the consumers’ maximum willingness to travel, the range, depends on the type of product (Dennis, Marsland & Cockett, 2002). Based on this, it is suggested that the proximity of shopping centers affects residential properties in the direct surrounding. Consequently, this thesis only investigates houses within a range of two kilometers from the shopping center. A range of two kilometers is used since this thesis investigates the effect of small neighborhood shopping centers. Moreover, Dröes and Koster (2016) and Van Duijn et al. (2014), used a range of two kilometers in their research as well.
11
2.5 Difference between shopping centers in the Netherlands and U.S or Canada
To investigate whether the effect of proximity of shopping centers is considerable in the Dutch market as well, this thesis focusses on the small neighborhood shopping centers in the Dutch retail market. This section evaluates the difference in type of shopping centers in the U.S or Canada and the Netherlands. This could be an explanation for the different effect of proximity and size of shopping centers on surrounding residential house prices.
The shopping centers, as we are familiar with now, are derived from the American shopping centers. During the twentieth century, the phenomenon of shopping centers in the U.S, has been spread around the world (Jackson, 1996). Whereas the U.S retail structure is not impeded by spatial planning, the Dutch retail market is formed by strict spatial planning concerning the retail structure. As a result, the Dutch retail market is characterized by a ‘fine-grained’ retail structure (Brayé, 2010). This structure could be positively as well as negatively interpret by consumers. On the one hand, consumers derive certainty as well as confidence from it and most of the shops are accessible by foot or bike which enables consumers to do their daily grocery while spending little travel time (NRW, IVBM & INretail, 2016). But, on the other hand the standard supply creates monotonousness. The opposite is a ‘croase-grained’ retail landscape. This requires consumers to travel more to peripheral retail locations (Evers, Kooijman & Krabben, 2011). According to the International Council of Shopping Centers (ICSC), there are different types of shopping centers, classified on size, concept and number of anchors. European shopping centers can be classified into very large, large, medium and small types, whereas Canadian and U.S. shopping centers can be divided into super-regional malls, regional malls, community-, neighborhood- and convenience shopping centers. Furthermore, the U.S and Canadian shopping centers also differ in square meters. Where the U.S and Canadian shopping centers ranges from less than 9,000 m2 to more than 250,000 m2, the Dutch type of shopping centers ranges from 5,000 m2 to more than 80,000 m2. Moreover, neighborhood centers in the Dutch retail market focus on convenience shopping. These centers include among other things, daily shops, fashion retailers, toys and home furnishing products. On top of that, they are in general characterized by at least one anchor shop, for example a supermarket (ICSC). However, shopping centers in the U.S aimed at a greater service area with a wider range of offerings including leisure and catering facilities. Shopping centers in the U.S are seen as the main spots in the cities, while this is not the main function in the Netherlands (Evers et al., 2011). Hence, the Canadian and the U.S shopping centers are not directly comparable to the European type of centers regarding size and concept. As a result it is expected that the of proximity and size of shopping centers affects house prices more positively in the U.S and Canada than in the Netherlands.
12
3. Data
This section provides the descriptive statistics of the data used in this thesis. Section 3.1 describes the housing transaction data, gathered from the NVM. Thereafter, section 3.2 gives an overview of the neighborhood characteristics obtained from Statline. Section 3.3 discusses the data regarding shopping centers obtained from Locatus. Finally, section 3.4 reports an overview of the distance between houses and the nearest shopping center.
3.1 Descriptive statistics housing transaction data
This thesis focusses on the relation between small neighborhood shopping centers and the prices of surrounding residential properties in the province Noord-Holland. To be precise, ten municipalities in the province Noord-Holland are considered namely; Alkmaar, Amsterdam, Beverwijk, Diemen, Edam-Volendam, Heerhugowaard, Heiloo, Hoorn, Purmerend, and Zaanstad. These municipalities are randomly selected based on the fact that they are located in the province Noord-Holland. Data of transaction prices, house characteristics, such as number of rooms, year of construction, square meters, garden surface, garden position and address is collected from the Dutch Association of Realtors (NVM). House price data of 2016 and 2017 is used for this study. By using data of 2016 and 2017, the impact of the economic crisis does not influence the results. On top of that, this is in accordance with existing literature that uses house price data for a time period of two years as well. Table 1 shows the number of transactions per municipality in a radius of two kilometer of a shopping center. Referring to column 1, the majority of the properties have been sold in Amsterdam, whereas the fewest properties have been sold in Hoorn.
Table 1 : Housing transactions per municipality
Table 1 provides an overview of the number of transactions per municipality in a radius of two kilometers of a shopping center in the period 2016-2017.
Municipality Number of transactions
Alkmaar 1,458 Amsterdam 15,871 Beverwijk 931 Diemen 446 Edam-Volendam 422 Heerhugowaard 309 Heiloo 430 Hoorn 300 Purmerend 1,737 Zaanstad 2,651 Total 24,555
13 Table 2 provides summary statistics of housing transaction data including the different house characteristics, which are included as control variables. The dataset contains 24,555 observations. It can be concluded from this data that the transaction prices are divergent. With a mean transaction price of approximately €365,000, the highest and lowest transaction price ranges from €2,175,000 to €95,000 with a standard deviation of approximately €282,000. Subsequently, dummy variables are created for the different house types. The most represented property type in this dataset is with 38 percent the upstairs apartments, followed by the single-family houses which are approximately 26 percent of all houses. Whereas the categories canal houses, farmhouses, bungalows and country houses, are all only a small part of the sample. Furthermore, table 2 provides summary statistics about house characteristics. First of all, the garden surface and living area are expressed in square meters. Referring to column 1, the average garden surface is approximately 25 meters and the mean of the living area is approximately 98 square meters. Moreover, the average number of rooms is 3.73, with a maximum of 13 rooms. In addition, the variables ‘Swimming pool’, ‘Elevator’ and ‘Monument’ are defined as dummy variables. The variable equals 1 if the dwelling offers an elevator, a swimming pool or if the dwelling has an monument status. Otherwise, the variable equals 0. Out of the 24,555 houses sold in 2016 and 2017, only 3 percent is defined as a monument and 18 percent has an elevator. On top of that, only 0.13 percent has a swimming pool. Subsequently, the average maintenance is defined as the average of the in- and outside maintenance to capture the problem of multicollinearity. Including both the in- and outside maintenance can lead to multicollinearity, because the value of these variables are generally the same. This is apparent from the high correlation of 0.70 too. The ‘Average maintenance’ is an estimation for the average maintenance and ranges from 1 to 9 where, the value of 1 refers to ‘worse maintenance’ and a value of 9 is ‘excellent maintenance’. The mean of the average maintenance is equal to 7.09.
Subsequently, dummies are created for the construction year of the house. These dummies represent the age of the dwelling. The variable takes 1 if the property is built in the defined period and 0 otherwise. The construction years are divided into nine categories, from 1500-1905, 1906-1930, 1931-1944, 1945-1959, 1960-1970, 1971-1980, 1981-1990, 1991-2000 and from 2001. Unfortunately, there are three observations of which the year of construction is unknown. Therefore, these three observations are dropped. Column 1 evaluates that most houses sold in 2016 and 2017, where built in the period 1906-1930, while the period 1945-1959 is the least common.
14 Table 2 : Summary Statistics housing transaction data
Table 2 reports the summary statistics of the housing transaction data obtained from the NVM. The table provides the mean, standard deviation, minimum and maximum of the house characteristics.
Variable Mean Std. dev. Minimum Maximum
Transaction price 365,019 281,676 95,000 2,175,000 House type Simple 0.01 0.12 0 1 Single-family 0.26 0.44 0 1 Canal house 0.00 0.07 0 1 Mansion 0.03 0.17 0 1 Farmhouse 0.00 0.03 0 1 Bungalow 0.00 0.07 0 1 Villa 0.01 0.10 0 1 Country house 0.00 0.02 0 1
Apartment ground floor 0.10 0.30 0 1
Apartment upstairs 0.38 0.49 0 1
Maisonette 0.03 0.17 0 1
Portico flat 0.08 0.27 0 1
Gallery flat 0.07 0.25 0 1
Apartment ground floor and
upstairs 0.01 0.11 0 1 House characteristics Garden (m2) 25.29 50.80 0.00 936.00 Living area (m2) 98.45 44.88 28.00 323.00 Number of rooms 3.73 1.49 1 13 Swimming pool 0.00 0.04 0 1 Elevator 0.18 0.38 0 1 Monument 0.03 0.17 0 1 Average maintenance 7.09 0.86 1 9
Construction year periods
1500 - 1905 0.13 0.34 0 1 1906 - 1930 0.19 0.40 0 1 1931 - 1944 0.09 0.28 0 1 1945 - 1959 0.05 0.21 0 1 1960 - 1970 0.13 0.33 0 1 1971 - 1980 0.08 0.27 0 1 1981 - 1990 0.10 0.30 0 1 1991 - 2000 0.10 0.29 0 1 After 2001 0.14 0.34 0 1 Parking facilities No parking 0.82 0.39 0 1 Parking spot 0.05 0.22 0 1
Carport without garage 0.05 0.22 0 1
Carport with garage 0.06 0.25 0 1
Garage without carport 0.01 0.09 0 1
Multiple parking 0.01 0.11 0 1 Garden location No garden 0.63 0.48 0 1 Garden North 0.02 0.15 0 1 Garden North-East 0.04 0.19 0 1 Garden East 0.03 0.18 0 1 Garden South-East 0.05 0.22 0 1 Garden South 0.07 0.25 0 1 Garden South-West 0.07 0.25 0 1 Garden West 0.05 0.21 0 1 Garden North-West 0.04 0.20 0 1
15 Garden maintenance Not finished 0.01 0.10 0 1 Neglected 0.00 0.05 0 1 Normal 0.80 0.40 0 1 Good 0.07 0.26 0 1 Well-maintained 0.12 0.32 0 1 Number of observations 24,555
Furthermore, dummies are created for additional characteristics such as parking facilities and the position of the garden. The average values of the dummy variables regarding the parking facilities show that the majority of the sold houses did not have access to a parking facility. Lastly, the descriptive statistics of the dummies for the sun position of the garden reveal that 63 percent of the houses sold in these municipalities did not have a garden. The remaining 37 percent of the houses did have a garden with different positions in the North, East, South and West. Moreover, 80 percent of the gardens are in ‘normal’ condition.
3.2 Descriptive statistics neighborhood data
To control for differences between municipalities, data about neighborhood characteristics is included. These data is obtained via Statline, the database of the Central Bureau of Statistics. The data contains among other information, population statistics, households characteristics and ethnic diversity data of immigrants, based on 4-digit postcodes. These data is available for both 2016 as well as 2017. However, data about the average disposable income is only available for 2015. Hence, the average disposable income is set at the level of 2015 because, multiplying all data with the same percental increase for 2016 and 2017 is neglectable. A limitation of this dataset is that it is linked to the most frequent 4-digit postcode. To obtain general statistics per specific postcode, relative numbers with respect to total inhabitants, total households and total housing supply are calculated. These relative numbers are representative for the characteristics of the specific neighborhoods based on 4-digit postcodes. Unfortunately, data about the district ‘Zuidschermer’ with postcode 1847 in Alkmaar and ‘Petroleumhaven’ with postcode 1041 in Amsterdam is not available. Therefore, the three observations within these districts are deleted.
16 Table 3 : Summary statistics neighborhood data
Table 3 reports the summary statistics of the neighborhood data obtained from Statline, the database of the CBS. The table provides the mean, standard deviation, minimum and maximum of the average disposable income, the inhabitants in the age category of 0-15 years, inhabitants of 65 years and older, Western immigrants, Non-Western immigrants and the share of
single- family houses.
Variable Mean Std. dev. Minimum Maximum
Age 0-15 years (%) 14.96 4.20 0.00 32.46
Age 65 and older (%) 14.83 6.25 2.36 31.90
Total Western immigrants (%) 15.68 6.36 0.00 31.81 Total non-Western immigrants (%) 25.34 16.02 0.00 72.31
Single-family house (%) 28.76 29.54 0.00 98.00
Average disposable income (in thousands) 27.37 8.49 16.40 62.29
In total there are six neighborhood control variables; • Relative number of children
• Relative number of elderly
• Relative number of Western immigrants • Relative number of non-Western immigrants • Relative amount of single-family houses
• The logarithm of the average disposable income per inhabitant
Referring to table 3, the relative number of children is presented by the variable ‘Age 0-15 years’, whereas the relative number of elderly is shown by the variable ‘Age 65 and older’. Within the municipalities, the relative amount of children and elderly is almost equal. Both categories are approximately 15 percent of the total inhabitants. Moreover, the relative amount of Western immigrants is 15.68 percent whereas the relative amount of non-Western immigrants is 25.34 percent. Subsequently, the variable ‘Single-family houses’ presents the relative amount of single-family houses in terms of total housing supply. According to column 1, on average 28.76 percent of the total housing supply are single-family houses. Furthermore, column 2 shows that the standard deviation of the relative amount of single-family houses is 29.54. This implies variation among municipalities concerning the relative number of single-family houses. Lastly, the average disposable income ranges from a minimum of €16,400 in the neighborhood ‘Poelenburg’ in Zaandstad to a maximum of approximately €62,000 in the neighborhood ‘Museumkwartier’ in Amsterdam, with a mean of approximately €27,000 in the total sample.
17
3.3 Descriptive statistics shopping center data
Data of shopping centers in the province Noord-Holland is obtained from Locatus. Locatus collects actual data about stores, shopping centers, service providers as well as restaurants in the Benelux. The data contains information about the type of shopping centers, the number of shopping centers per type and municipality, the address, the zip-codes and the long- and latitude coordinates of the shopping centers. To control for the effect of size of shopping centers, only the neighborhood centers, classified by Locatus as ‘Small neighborhood shopping centers’, are investigated. However, to capture the effect of size between the small neighborhood centers, a control variable for the size of the neighborhood shopping centers is included. Due to the lack of information about the Gross Leasable floor Area of shopping centers, the number of shops is used as a measure for the size of a shopping center. This is supported by Des Rosiers et al. (1996) since they found a strong correlation between the GLA and the number of shops. Table 4 shows an overview of the number of small neighborhood shopping centers and the average number of stores in a shopping center per municipality.
Table 4 : Number of neighborhood shopping centers per municipality
Table 4 reports the number of neighborhood shopping centers per municipality. In addition, column 2 provides an overview of the average number of stores in a small neighborhood shopping center.
The total sample includes 51 neighborhood shopping centers divided over the ten municipalities. Table 4 shows that half of the shopping centers is located in Amsterdam. The other half is divided among the other municipalities, whereas Diemen, Heerhugowaard, Heiloo and Hoorn only have one neighborhood shopping center. Column 2 shows that the size of a shopping center, measured in number of shops, differs among the municipalities. In contrast to
Amsterdam, which has on average a shopping with 31 stores, Heerhugowaard has on average a shopping center with only 8 stores.
Municipality Number of neighborhood shopping centers Average number of stores in a shopping center
Amsterdam 27 31 Alkmaar 4 23 Beverwijk 2 30 Edam-Volendam 2 23 Diemen 1 16 Heerhugowaard 1 8 Heiloo 1 11 Hoorn 1 17 Purmerend 6 29 Zaanstad 6 20 Total 51
18
3.4 Distance between houses and nearest shopping center
To measure the impact of the proximity of small neighborhood shopping centers on the value of surrounding residential properties, the Euclidean distance between the houses and the shopping centers in a radius of two kilometers is calculated. In addition, dummies are created for the different distance categories to investigate whether the effect of proximity differs among the distance categories. Subsequently, the number of shopping centers within two kilometers is used as a control variable. This variable expresses the shopping center density in the surrounding area.
Table 5 : Summary statistics about distance to nearest shopping center
Table 5 reports the summary statistics of the distance to nearest shopping. Only transaction in a radius of two kilometers are investigated. First of all, the distance to nearest shopping centers is presented in meters. Furthermore, the distance is divided in
six categories expressed by dummy variables. Lastly, the shopping center density of the area is expressed by the number of shopping centers within two kilometers.
VARIABLES Mean Std. dev Minimum Maximum
Distance to nearest shopping center in meters 944.72 511.98 7.61 1,999.79 Dummies for distance categories
Distance to nearest shopping center 250 m 0.07 0.25 0.00 1.00 Distance to nearest shopping center 250-500 m 0.18 0.38 0.00 1.00 Distance to nearest shopping center 500-750 m 0.18 0.38 0.00 1.00 Distance to nearest shopping center 750-1000 m 0.15 0.36 0.00 1.00 Distance to nearest shopping center 1000-1500 m 0.25 0.43 0.00 1.00 Distance to nearest shopping center 1500-2000 m 0.18 0.39 0.00 1.00 Shopping center density
Number of shopping centers within 2 km 2.12 1.14 1.00 8.00
Table 5 shows that the average distance to nearest shopping center is approximately 945 meters. The most represented category is the category of distance between 1000 and 1500 meters, where only 7 percent of the houses are located in a radius of 250 meter from the nearest shopping center. Moreover, the density of shopping centers is approximately 2 within two kilometers.
19
4. Methodology
This section evaluates the methodology used in this thesis. Section 4.1 explains the hedonic regression model with distance as a continues variable. Section 4.2 provides an alternative regression specification with distance as a categorial variable. Section 4.3 analyses the regression diagnostics. Then, section 4.4 describes the control variables and the expected sign. Lastly, section 4.5 evaluates the endogeneity problem.
4.1 Hedonic regression model
To answer the research question ‘Does the proximity of shopping centers affects the value of the surrounding residential properties in the Netherlands?’, an hedonic regression model is used. Research has shown that hedonic models are frequently applied in existing literature regarding residential topics. The hedonic approach adequately takes into account all the characteristics affecting house prices, such as house and neighborhood characteristics. This allows the marginal effect of the different characteristics to be determined separately (Des Rosiers et al., 1996). As specified by Des Rosiers et al. (1996), the dependent variable is the transaction price of a house, as a measure for the value of the residential property. More specifically, the logarithm of the transaction price is used. Using the logarithm, the residuals are closer to normality compared to the residuals of the transaction price (Sirpal, 1994). To answer the research question whether the proximity of shopping centers affects the value of surrounding residential properties and the sub-question; ‘Does the effect of proximity of shopping centers on the value of surrounding residential properties diminish or becomes the opposite as distance increases?’, the following hypothesis are formulated:
H0: “The distance between the shopping centre and the surrounding residential properties does not influence the transaction prices.”
H0: “The effect of proximity of shopping centers does not change as distance increases.”
In this thesis, two hedonic regression models are used to test the hypotheses. The hypotheses are tested for the total sample as well as for two sub-samples. The first sub-sample is the municipality Amsterdam and the other sub-sample contains the municipalities Alkmaar, Beverwijk, Edam-Volendam, Diemen, Heerhugowaard, Heiloo, Hoorn, Purmerend and Zaanstad. Amsterdam is chosen as a sub-sample because it is a capital city and it deviates in size from the other municipalities.
20 The first hedonic regression model controls for house characteristics as well as neighborhood characteristics. In addition, quarterly dummies are included to control for difference in time. Furthermore, this model controls for the distance between a house and the nearest shopping center, the shopping center density within two kilometers and the size of the shopping centers. After comparing a number of different regression specifications with respect to the goodness of fit, the first model is shown by the following formula:
Ln(Pijt) = α + β1Dij + β2Dij2 + β3LnSize
j + β4N2km + β5Xi + β6Zi + θt + εi (1)
Where, Ln(Pijt) is the logarithm of the transaction price of house i around shopping center j at time
t. Besides the fact that the residuals are closer to normality by means of a log transformation of the transaction price, the estimated coefficients can be interpret as percentage changes in the dependent variable resulting from a unit change in de independent variables (Thornton and Innes, 1989). Subsequently, α is the unknown constant followed by β1, β2, β3, β4, β5 and β6 that are the parameters to be estimated (Van Duijn et al., 2014). Furthermore, Dij is the distance between
shopping center j and house i, Dij2 is the quadratic term of distance between shopping center j and
house i, LnSizej is the logarithm of the size of shopping center j, N2km is the number of shopping
centers within two kilometers and Xi and Zi are vectors containing house and neighborhood
characteristics respectively. The variables are defined in more detail in the remainder of this section. Lastly, εit is the error term with conditional mean zero, that controls for unobserved variables
(Koster and Ommeren, 2015).
As mention above, Dij is the distance to the nearest shopping centers from house i to
shopping center j. This is the straight-line distance between two points in terms of longitude and latitude coordinates (Wang, Zhand & Draxler, 2009). Only houses in a radius of two kilometers from a shopping center are taken into control. The long- and latitude coordinates of the shopping centers are determined based on the average of the coordinates of the stores in the shopping center. This could result in some divergence from the actual distance. Although, this bias is assumed to remain insignificant. Next, the quadratic function of distance is included to investigate whether the price-distance relationship is non-monotonic. A significant coefficient of this squared term, indicates a non-monotonic relationship (Des Rosiers et al., 1996). Besides, to control for the effect of size between small neighborhood centers, the variable LnSizej is included. The size of a shopping
center is expressed in the number of shops. Subsequently, N2km is the number of shopping centers
21 Furthermore, Xi is the vector for house characteristics that contains living area and garden
surface in square meters. Besides it contains the number of rooms and dummies for different house types, construction years, parking facilities and the position and the maintenance of the garden. To avoid the multicollinearity problem, the reference categories are Single-family house, construction year 1906-1930, no parking, the northern position of the garden and a normal condition of the garden, respectively. On top of that, dummies are included indicating whether the house has an elevator or a swimming pool and whether the house is defined as a monument. To avoid the effect of outliers, extreme observations are winsorized. Using this option, the observations are not deleted, but corrected to other values in the dataset. The variables transaction price, number of rooms and living area are winsorized.
Besides, the vector Zi includes neighborhood characteristics with respect to the number of
Western and non-Western immigrants, the share of children and the relative amount of elderly. Besides this, this vector contains the average disposable income and the relative amount of single-family houses. Lastly, θt, is a vector for quarterly dummies to control for difference in time. The
default category for the quarterly dummies is the first quarter of 2016.
4.2 Alternative regression specification
According to existing literature, the relationship between house prices and distance can be non-monotonic. A second alternative to indicate whether this effect is present in this study, is by means of dummies for distance. By including dummies for distance, it could become clear whether house prices reach a maximum as a result of the effect of distance (Des Rosiers et al., 1996). Hence, to test the second hypothesis: “The effect of proximity of shopping centers does not change as distance increases.”, an alternative regression model is formulated and shown by the following formula:
Ln(Pijt) = α + β1Dij + β2LnSizej + β3N2km + β4Xi + β5Zi + θt + εi (2)
This model is an extension of the basic hedonic regression model. In this model, Dij is defined as
a categorial vector. Dummies are created in ranges of 250 and 500 meters. By means of comparing the signs of the estimated coefficients for the different distance categories, it can be concluded whether the price-distance relationship is non-monotonic and changes when distance increases. In this specification, six dummies are created. First of all, the dummy ‘Distance 250’ equals 1 if the distance of house i to the nearest shopping center j is less or equal to 250 meter and 0 otherwise. Secondly, the dummy ‘Distance 250-500’ equals 1 if the distance of house i to the nearest shopping j center is more than 250 meters and less or equal to 500 meters. The dummy equals 0 otherwise.
22 The same reasoning applies to the dummy variables ‘Distance 500-750’, ‘Distance 750-1,000’, ‘Distance 1,000-1,500’, ‘Distance 1,500-2,000’. The dummy variable ‘Distance 1,000-1,500’ is used as the default category. Furthermore, LnSizej and the vectors Xi, Zi and θt are the same as in model
(1).
Moreover, to investigate whether the results for the total sample, differ from the results per sub-sample, the regressions are applied on the two sub-samples. Again, only houses in a range of two kilometers from a shopping center are investigated. The regression specifications of model (1) and (2) are applied on the sub-sample that contains the municipality of Amsterdam. However, due to a lack of variety in the neighborhood data in the second sub-sample, the regression specifications contain neighborhood fixed effects instead of neighborhood control variables. Therefore, the regression specifications for the sub-sample, consisting of the municipalities Alkmaar, Beverwijk, Edam-Volendam, Diemen, Heerhugowaard, Heiloo, Hoorn, Purmerend and Zaanstad, are shown by the following formulas:
Ln(Pijt) = α + β1Dij + β2Dij2 + β3LnSize
j + β4N2km + β5Xi + θt + δm + εi (3)
Ln(Pijt) = α + β1Dij + β2LnSizej + β3N2km + β4Xi + θt + δm + εi (4)
Referring to model (3), δm is a neighborhood fixed effect to control for the difference between municipalities. Subsequently, Dij, Dij2, LnSizej, N2km and the vectors Xi and θt are the same as in model
(1). Similarly, according to model (4), δm is a neighborhood fixed effect as well. Furthermore, Dij,
LnSizej, N2km and the vectors Xi and θt are the same as in model (2).
4.3 Regression diagnostics
This section analyses the regression diagnostics. Firstly, to detect the problem of multicollinearity, the variance inflation factor of all variables is calculated. Several studies, among others the study of Mendenhall and Sincich (1996), assume that the problem of multicollinearity occurs when the variance inflation factor exceeds the value of 10. Hence, to avoid the problem of multicollinearity in this study, all variables in the model do have a variance inflation factor of less than 10. Additionally, the correlation matrix between the variables is determined. Appendix 1 exhibits the variance inflation factors and the correlation matrix of the neighborhood control variables. It is assumed that a correlation coefficient higher than (-)0.7 leads to multicollinearity (Mukaka, 2012). According to appendix 1, the correlation between the variables ‘Total Western immigrants’ and ‘Average disposable income’ is 0.73, which implies a high correlation. On top of that, the variables
23 ‘Single-family houses’ and ‘Total Western immigrants’ are also highly negatively correlated. However, results of the Walt-test indicate that these variables are relevant to the model. Besides, the variance inflation factors are less than 10. Therefore, the variables ‘Total Western immigrants’, ‘Average disposable income’ and ‘Single-family house’ are included as control variables. Subsequently, appendix 2 provides the variance inflation factors and the correlation matrix for house characteristics. The same reasoning holds for the variable ‘Number of rooms’ and ‘Living area’ as well.
Furthermore, the skewness of the distribution of the variables is analyzed. The skewness of ‘Living area’, ‘Number of rooms’ and the size of a shopping center, ‘Sizej’ deviates from symmetry.
Therefore, a log transformation for these variables is preferred in order to obtain a normal distribution. Moreover, it can be assumed that the residuals are normally distributed. This is supported by the graphs of the residuals in appendix 3. Though, the Breusch-Pagan test indicates that the residuals suffer from heteroskedasticity. To correct for this problem, standard errors are clustered at PC6 level.
4.4 Expectation of control variables
As mentioned in the previous sections, the regression models include control variables such as distance specifications, shopping center size, house characteristics and neighborhood characteristics. To start with the different distance bands. It is expected that shopping centers will have a positive effect on the price of surrounding residential properties, but living to close to the shopping center, will generate negative effects due noise and pollution. Therefore, the dummy for ‘Distance 250’ is expected to have a negative sign (Des Rosier et al., 1996). Moreover, according to Sirpal (1994), the size of a shopping center is expected to have a positive effect on the price of a house. Lastly, it is expected that the density of shopping centers negatively influences the price of a house because it re-enforces the negative externalities of shopping centers in terms of more nuisance like pollution and traffic noise.
Besides the distance specifications, house characteristics are also included as control variables. First of all, dummies are created for the different house types. According to Visser and Van Dam (2016), it is expected that the type of a house influences the price, because house types differ in price. Subsequently, the garden surface and the living area are both expressed in square meters. It is expected that they both have a positive effect on the transaction price of a house (Dröes and Koster, 2016). This effect is expected for the number of rooms too. The higher the number rooms, the higher the transaction price of a house (Sirpal, 1996) In addition, dummies are created for the presence of an elevator, a swimming pool and whether the house has a monument
24 status. The variables equal 1 if the property has one of these characteristics and it is expected that this will have a positive effect on the transaction price of a house (De Wit et al., 2013; Des Rosiers et al., 1996; Koster and Ommeren, 2015). Furthermore, the average maintenance is defined as the average of the in- and outside maintenance. The value of this variable is positively related to the average maintenance; the higher the value, the better maintenance. Hence, this will translate in a positive effect on the transaction price as well (Theebe, 2004). Additionally, dummies are created for the building periods of the dwelling. These dummies represent the age of the dwelling. However, age could influences the transaction price upwards, as well as downwards. On the one hand, age could have a negative effect on the price of a house as result of depreciation. On the other hand, age could have a positive sign due to of the vintage effect (Goodman and Thibodeau, 1995). Besides, dummies for parking facilities indicate the presence of a parking facility. It is expected that this will positively influence the price of a house (Van Duijn, Rouwendal & Boersema, 2014). In addition, dummies are created for the position and the maintenance of the garden. It is expected that a garden in the direction of the sun, is worth more. As a result, it is expected that the coefficients of ‘Garden South’ and ‘Garden South-West’ will have a positive sign (De Wit et al., 2013). Moreover, a good condition of the garden also adds value to a house and is therefore expected to have a positive sign as well (Van Duijn et al., 2014).
The last group of control variables are neighborhood characteristics. Firstly, the relative amount of children, represented by ‘Age 0-15 years’, and elderly, represented by ‘Age 65 and older’, are included. Previous research provides evidence that the share of elderly inhabitants has a positive effect on the transaction price. Hence, it is expected that the variable indicating the elderly inhabitants will have a positive sign (Des Rosiers et al., 1996). Furthermore, the total amount of Western and non-Western immigrants are added to the model. There is a negative relation between house prices and the amount of non-Western immigrants (Visser and Van Dam, 2006). Subsequently, the relative amount of single-family houses in terms of total housing supply is included. Lastly, the average disposable income per inhabitant is included. This variable is an indicator for welfare. It is expected that the transaction price and the welfare of a neighborhood are positively related (Andrews, Sánchez and Johansson, 2011).
4.5 Endogeneity
According to the central place theory of Christaller (1933) shopping centers are located on a central place in the neighborhood. Though, houses in the city center are in general more expensive than houses at peripheral locations. Hence, it is argued whether the rise in house prices resulted from the presence of a shopping center or from being located in the city center. To tackle this
25 endogeneity problem, only houses within two kilometers of a shopping center are investigated. By means of this, this study controls for the effect of a rise in house prices resulting from being located in the city center. Existing literature of Dröes and Koster (2016), Sirpal (1994) and Van Duijn et al. (2014) control for this endogeneity problem as well, by considering observations in a radius of two kilometers, 3,000 feet and two kilometers , respectively.
26
5. Results
This section provides the regression results for the total sample and the two sub-samples. Firstly, section 5.1 provides an interpretation of the regression results of model (1), (2), (3) and (4)
regarding the effect of proximity of shopping centers. Subsequently, section 5.2 gives an
interpretation of the results and control variables regarding the house characteristics. Lastly, section 5.3 discusses the results of the neighborhood control variables.
5.1 Proximity of shopping centers
According to column 1 of table 6, results show that house prices are related with the proximity of shopping centers. This relationship is supported by the significance of the variable ‘Distance to nearest shopping center’. This variable measures the proximity of shopping centers in kilometers. It turns out that increasing distance to the nearest shopping center positively effects house prices. Results show that if the distance increases with one kilometer, house prices increase with 23.4 percent. This effect is consistent with the results of Sirpal (1994). However, this is in contrast to the results of Des Rosiers et al. (1996), Addae-Dapaah and Lan (2010) who concluded that distance requires a premium to house prices. This can be explained by the fact that the investigated shopping centers in their research are of much larger scale. That kind of shopping centers offers more facilities and serves as the main places in the city. Next, to indicate a non-monotonic price-distance relationship, the quadratic function of distance is included. The quadratic term turns out to be significant. In accordance to the findings of Des Rosiers et al. (1996), Addae-Dapaah and Lan (2010), the non-monotonic price-distance relationship can be supported. The results of column 1 are graphically shown in figure 2.
Figure 2. Effect of proximity of shopping centers on house prices in the total sample.
0 0,01 0,02 0,03 0,04 0,05 0,06 0,07 0 0,5 1 1,5 2 2,5 Eff ec t o n tr an sac ti o n p ri ce Distance
Total sample
27 Figure 2 shows that living in the close surrounding of a shopping center, has a negative effect. However, as distance increases, the effect of proximity of shopping centers on house prices turns positive up to a maximum point, after which they decline. This is in line with results of Sirpal (1994. An explanation for this, is that the negative externalities, such as pollution and noise outweigh the positive externalities, as the ease of shopping and less travel time, in the very close surrounding of a shopping center (Des Rosiers et al., 1996). Moreover, figure 2 corresponds to the results in column 2. As shown in column 2 of table 6, distance to the nearest shopping center is expressed in the dummy variables ‘Distance 250’, ‘Distance 250-500’, ‘Distance 500-750’, ‘Distance 750-1000’, ‘Distance 1,000-1,500’ and ‘Distance 1,500-2,000’, whereas ‘Distance 1,000-1,500’ is the reference category. These dummies measure the effect of proximity for houses within 250, 250-500, 500-750, 750-1,000 or 1,500-2,000 meters of the shopping center, respectively. In line with figure 2, the results show that the proximity of shopping centers negatively effects house prices within a radius of 250 meter. Then, house prices increase within the distance band of 250 to 500 meters, relative to houses within 1,000-1,500 meters from the shopping center. The insignificance of the variable ‘Distance 500-750’ indicates that the effect of the proximity of shopping centers is zero from 750 meters. Therefore, the variable ‘Distance to nearest shopping center’, in column 1, is winsorized at 750 meters. Although, the variable ‘Distance 1,500-2,000’ turns out to be significantly positive. An explanation for this can be that another shopping centers positively influence the price of a house in that distance band. As a result, the first null hypothesis; “The distance between the shopping centre and the surrounding properties does not influence the residential transaction prices” as well as the second hypothesis “The effect of proximity of shopping centers does not change as distance increases.” can be rejected.
To investigate whether the effect is applicable to Amsterdam as well, the regression is applied to the sub-sample that contains the municipality of Amsterdam. The estimated coefficients of the dummy variables, presented in column 4, indicate that the proximity of shopping centers first have a negative effect, after which the effect turns positive. Although, the negative effect within the distance band of 250 meters, turns out not to be significant in Amsterdam.
28 Figure 3. Effect of proximity of shopping centers on house prices in Amsterdam.
Nevertheless, based on the results of column 3, figure 3 does provide evidence for a negative effect in the direct vicinity. In accordance with figure 2, the effect turns positive when distance increases. Hence, the results are in line with the results of the total sample. The estimated coefficients of the dummy variables suggest that the effect of proximity of shopping centers is zero from 1,500 meters. Therefore, the variable ‘Distance to nearest shopping center’, in column 3, is winsorized at 1,500 meters. Additionally, the regression is applied on a second sub-sample that consists of the municipalities Alkmaar, Beverwijk, Edam-Volendam, Diemen, Heerhugowaard, Heiloo, Hoorn, Purmerend and Zaanstad. According to column 6, the estimated coefficient for the first distance category within 250 meters, indicates that the proximity of shopping centers negatively affects house prices within 250 meters. This is in line with the results of the total sample. However, the remaining distance categories turn out to be insignificant. To investigate whether the effect of the total sample is applicable in this sub-sample too, figure 4 graphically shows the results of column 5. -0,005 0 0,005 0,01 0,015 0,02 0,025 0 0,5 1 1,5 2 2,5 Eff ec t o n tr an sac ti o n p ri ce Distance
Amsterdam
Distance winsorized at 1 km29 Figure 4. Effect of proximity of shopping centers on house prices in the sub-sample that contains Alkmaar, Beverwijk,
Edam-Volendam, Diemen, Heerhugowaard, Heiloo, Hoorn, Purmerend and Zaanstad.
Hence, despite the insignificance of the dummy variables for the distance categories, figure 4 shows that houses in the close vicinity are negatively affected by the proximity of a shopping center in this sub-sample as well. So, the results are in line with the effect of both the total sample and the sub-sample of Amsterdam.
Furthermore, in contrast to the results of Sirpal (1994), the variable ‘Ln Shopping center size’ appears to be insignificant in the total sample as well as in the sub-sample containing the nine municipalities. This variable measures the effect of size of a shopping center. Moreover, results in column 3 and 4 show that the size of shopping centers negatively influences the price of adjacent residential dwellings in Amsterdam. A one percent increase in the shopping center size, would result in a discount of approximately 0.02 percent on the transaction price of a house, keeping other variables constant.
Subsequently, the coefficient for shopping center density, measured by the number of shopping centers within two kilometers, is significantly negative at 1 percent level in the total sample as well as in Amsterdam. This implies that each additional shopping center within two kilometers from the house, negatively influences the house price by approximately 2 percent. This sign is as expected, because this can cause more nuisance.
To conclude, the results obtained with dummies for distance as well as with the quadratic distance term, provide evidence that house prices are affected by the proximity of shopping centers. Moreover, the regression results per sub-sample provide evidence that the effect of proximity is consistent in the two sub-samples as well. Furthermore, the results provide enough evidence to confirm that the proximity of shopping centers and the surrounding residential property prices have a non-monotonic relationship.
0 0,01 0,02 0,03 0,04 0,05 0,06 0 0,5 1 1,5 2 2,5 Eff ec t o n tr an sac ti o n p ri ce Distance
Sub-sample
Distance 2 km30 Table 6 : Regression results
Table 6 provides an overview of the regression results of model (1), (2), (3) and (4) to investigate the effect of proximity of shopping centers on the value of surrounding residential properties in a radius of two kilometers. The dependent variable is Log (Transaction price). Distance specifications, house characteristics and neighborhood characteristics are used as independent variables. The regression uses data of 2016 and 2017. Column (1) and (2) provide regression results for the total sample. Besides, column (3) and (4) provide the results for the municipality of Amsterdam. Then, column (5) and (6) provide the results for the municipalities Alkmaar, Beverwijk, Edam-Volendam, Diemen, Heerhugowaard, Heiloo, Hoorn, Purmerend and Zaanstad together. Robust standard errors are reported in parentheses. All standard errors are clustered at PC6 level. Significance at 1%, 5% and 10% level are indicated by ***, **, and *, respectively. Furthermore, the adjusted R-squared and AIC are reported as goodness of fit criteria. Moreover, for the overall regressions and the regressions for the sub-sample Amsterdam, the default dwelling is a Single-family house with a garden on North and normal maintenance, no parking facilities, constructed in the period 1906-1930 and located in the distance band of 1,000-1,500 meters. Furthermore, for the regression for the sub-sample in column (5) and (6), the default dwelling is a Single-family house with a garden on the South-West and normal maintenance, no parking facilities, constructed in the period 1960-1970 and located in the distance band of 1,000-1,500 meters. Furthermore, in column 2 and 4 the variable ‘Distance to nearest SC’ is winsorized at 750 and 1,500 meters respectively.
(1) (2) (3) (4) (5) (6)
VARIABLES Total Sample Total Sample Amsterdam Amsterdam Sub-sample Sub-sample
Distance to nearest shopping center 0.234*** (0.0673) 0.0537** (0.0242) 0.0785*** (0.0243) Distance to nearest shopping center2 -0.227*** (0.0675) -0.0360*** (0.0132) -0.0287** (0.0125) Distance 250 -0.0157* -0.00513 -0.0496*** (0.00883) (0.0105) (0.0123) Distance 250-500 0.0194*** 0.0183** -0.00468 (0.00679) (0.00757) (0.0119) Distance 500-750 0.00714 0.00861 -0.00805 (0.00632) (0.00738) (0.00903) Distance 750-1,000 0.00979 (0.00746) 0.0230*** (0.00753) 0.000474 (0.00872) Distance 1,500-2,000 0.0171*** (0.00605) -0.00239 (0.00671) 0.0176 (0.0122) Ln Shopping center size 0.00649 (0.00493) 0.00525 (0.00491) -0.0230*** (0.00623) -0.0235*** (0.00613) -0.00677 (0.0235) -0.00660 (0.0231) Number of SC within 2km -0.0217*** (0.00203) -0.0211*** (0.00206) -0.0158*** (0.00249) -0.0160*** (0.00250) 0.00697 (0.00527) 0.00685 (0.00533) Simple 0.00893 0.00892 0.0150 0.0153 -0.0315** -0.0305** (0.0181) (0.0181) (0.0316) (0.0316) (0.0134) (0.0134) Canal house 0.157*** 0.156*** 0.144*** 0.144*** 0.183*** 0.181*** (0.0269) (0.0267) (0.0220) (0.0220) (0.0463) (0.0467) Mansion 0.134*** 0.134*** 0.113*** 0.113*** 0.133*** 0.133*** (0.0115) (0.0115) (0.0147) (0.0147) (0.0132) (0.0131) Farmhouse 0.207** 0.208** -0.100*** -0.0870*** 0.386*** 0.388*** (0.0870) (0.0866) (0.0182) (0.0188) (0.0790) (0.0781) Bungalow 0.268*** 0.265*** 0.274*** 0.272*** 0.233*** 0.230*** (0.0254) (0.0255) (0.0482) (0.0488) (0.0225) (0.0225) Villa 0.260*** 0.261*** 0.240*** 0.243*** 0.294*** 0.293*** (0.0208) (0.0208) (0.0337) (0.0338) (0.0215) (0.0215) Country house 0.536*** 0.532*** 0.628*** 0.623*** 0.416*** 0.416*** (0.102) (0.103) (0.0281) (0.0284) (0.109) (0.111) Apartment ground floor 0.0523*** (0.00902) 0.0521*** (0.00899) -0.0484*** (0.00961) -0.0475*** (0.00958) -0.0679*** (0.0145) -0.0671*** (0.0145) Apartment upstairs 0.0432*** (0.00929) 0.0419*** (0.00924) -0.0380*** (0.0101) -0.0379*** (0.0101) -0.0761*** (0.0119) -0.0759*** (0.0119) Maisonette -0.0381*** -0.0395*** -0.0626*** -0.0634*** -0.102*** -0.102*** (0.0135) (0.0134) (0.0148) (0.0146) (0.0128) (0.0128) Portico flat -0.0169 -0.0186* -0.0559*** -0.0555*** -0.0700*** -0.0696*** (0.0109) (0.0109) (0.0111) (0.0111) (0.0129) (0.0130) Gallery flat -0.0494*** -0.0506*** -0.0813*** -0.0814*** -0.127*** -0.125*** (0.0119) (0.0118) (0.0126) (0.0126) (0.0127) (0.0128) Apartment ground 0.140*** 0.139*** 0.0338** 0.0345** -0.0880** -0.0879*
