The effect of hazardous sites on property values: A case study of the firework disaster in Enschede
Véronique Meertens Student number: 11152672 Supervisor: Martijn Dröes
Second Supervisor: Marcel Theebe
MSc Business Economics, Finance and Real Estate Finance University of Amsterdam
2
Abstract
This study investigates the contagion effect of the explosion of the firework storage on May 13, 2000, in Enschede on property values surrounding other firework storages. A Difference-in-Differences model is used to compare the property values before and after the event. The sample consists of properties in 15 municipalities in the Netherlands and is obtained from the Dutch Broker Association. The results imply that properties within a radius of 300 meters of a firework storage experienced a discount of approximately 3% on the transaction price. This discount only applies during the first two years after the explosion. These results seem to hold under a variety of robustness checks.
Statement of originality
This document is written by Student Véronique Meertens, 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.
3
Inhoudsopgave
1. Introduction ...4
2. Literature Review ...6
2.1 Externalities and Property Values ...6
2.2 Explosions at Undesirable Facilities and Property Values ...7
2.3 Risk Perception ...8
2.4 Efficient Real Estate Market ...9
2.5 Contagion Effect ...9
2.6 Homeownership and the Dutch Housing Market ...10
2.7 Hedonic Pricing Model...10
3. Methodology ...11
4. Data and Descriptive Statistics ...13
4.1 Dataset Dutch Broker Association ...13
4.2 Dataset Firework Storages ...16
4.3 Correlation Matrix ...18 5. Results ...18 5.1 Difference-in-Difference model...18 5.2 Geographic Scope ...21 5.3 Dynamic Response ...22 6. Robustness Checks ...23 7. Conclusion ...25 8. Reference List ...28
4
1. Introduction
On Saturday, May 13, 2000, a big explosion took place at the storage unit S.E. Fireworks in Enschede. A fire, that initially was assumed to be mastered, caused a chain reaction of explosions in several storage bunkers. In total 23 people were killed, 950 people were injured and an area of 42 hectares was ruined. Since the storage unit was positioned in the middle of a residential area, 400 homes were completely destroyed. Research after the disaster showed that the regulations were breeched and that there were more bunkers present than allowed. Instead of the legal amount of 158,500 kg, there was 177,000 kg of fireworks present at the warehouse.
The ravage, caused by the explosion, ensured a thorough investigation to explore the cause of this accident. Although the way the fire started is still disputable, it was clear that the extent of the damage was much greater due to the breeching of regulations. Interviews with the surrounding citizens stated that they were not aware of the risk of living close to the firework warehouse and some even did not know that there was a storage unit. Therefore, this explosion can be used as a natural event to investigate the risk perception of citizens living close to a firework warehouse and the indirect impact it has on surrounding property values.
Already several studies investigated the link between undesirable facilities and the value of housing (also see the literature review), but none of these has focused on the warehouses of fireworks (Farber, 1998). The explosion showed that the firework is highly flammable and thus can be called a hazardous site or an undesirable facility. A considerable part of the researches focus on Hazardous Waste Sites (HWS). Since firework is no official waste, it cannot be called an HWS. Besides, since people are more aware of the existence and the risk of an HWS than of a firework warehouse, it is interesting to investigate the impact a firework storage has on property values separately. Moreover, the firework disaster in Enschede provides us with an ideal quasi-natural experiment to measure the impact.
Fisher et al. (1991) conducted a research about risk perception and the willingness to pay for risk reduction. They find that people are more willing to pay for risk-reduction when the risk is a direct threat to themselves, instead of a threat for the entire population. The disaster in Enschede has made the direct threat of living close to a firework warehouse more visible.
5 Consequently, a change in transaction prices of properties surrounding other firework storages is expected. Therefore, this research will investigate the following question:
Do firework warehouses in the Netherlands have an impact on property values in the surrounding areas?
To examine the effect of a firework warehouse on property values this study will look into 15 municipalities that accommodated a warehouse at the moment of the explosion. The data consists of transactions prices between 1990 and 2010 of those municipalities and are obtained from the Dutch Broker Association (NVM). A Difference-in-Differences model will be used to examine the effect on the treatment group, which consists of the transactions that occurred within a certain distance of the warehouse and after the explosion. By including a hedonic pricing model there can be controlled for housing and neighborhood characteristics. Hypothesized is a negative effect within the treatment group, but this effect diminishes over time and distance. This diminishing effect over distance is assumed since people might feel less threatened further away from the warehouse and over time since people might tend to forget the explosion after a while and thus perceive less risk. Consequently, multiple stages of risk perception will be investigated.
This research will also test how efficient the housing market is and how rational the housing market participants truly are. If the housing market is efficient and the participants are truly rational, the housing prices should not have dropped after the explosion, since the buyers have already taken the extra risk of living close to a firework warehouse into account. In an efficient housing market, the housing prices should reflect all the available information. In this case, the explosion might have provided extra information about the riskiness of the storaging of fireworks and thus have affected housing prices.
The results of this study show a negative effect on transaction prices of properties located within a radius of 300 meter of a firework storage. This effect is only perceptible in the first two years after the explosion. This could imply that the government should set a new minimum distance to a firework storage. From an investor’s perspective this could be an arbitrage opportunity since the transaction prices drop after the explosion, but restore to their original value within two to three years.
6 This study will discuss the relevant related literature in section 2. Section 3 and 4 cover respectively the methodology and data used for this research. Section 5 discusses the empirical results of the tests and section 6 introduces the robustness checks. The conclusion and limitations are given in section 7.
2. Literature Review
In order to articulate the right hypothesis, this section will give an overview of the main theories and studies of the effect of externalities and undesirable facilities on property values. Thereafter risk perception will be discussed, followed by the efficiency of the housing market and the contagion effect theory. The last part of the literature review introduces the theory of a hedonic price model and the related literature.
2.1. Externalities and property values
In the past decades multiple studies have been conducted to reveal the determinants of house prices. A recent study of Dröes and van de Minne (2016) showed that these determinants vary over time. Wilkinson (1973) stated that house prices are dependent on both internal and external abilities and these should be measurable. This study will focus on the effect of an external ability on house prices. These externalities can have a positive or negative effect on the value of the properties. For instance, externalities like a transit line (McMillen & McDonald, 2004; Grass, 1992), a specific view (Bond, Seiler & Seiler, 2002) or high quality primary schools (Gibbons & Machin, 2003) have a significant positive effect on house prices. While environmental externalities often have a negative impact on value, like an industry site (Li & James Brown, 1980), traffic intensity (Hughes & Sirmans, 1992) or airport noise (Espey & Lopez, 2000).
Especially Hazardous Waste Sites (HWS) cover a sizeable part of the studies regarding the negative effect of environmental externalities. McCluskey and Rausser (2003) showed with a research conducted in Texas that house appreciation rates were lower when situated close to an HWS. Even when the site was cleaned-up, the time to restore the equilibrium was quite long. The high perceived risk near the waste sites cause the prices to drop, although, this effect diminishes while distance to the site increases (McClelland, Schulze & Hurd, 1990; Lober & Green, 1994). However, stigmatization, in addition to this perceived risk, about these
7 HWS does not only affect property values, but other aspects of the economy, like business development and touristry (Slovic et al., 1991). Although a firework storage is not an official HWS since it does not store waste, it remains highly flammable and thus a real hazard. Therefore, this study expects a firework warehouse to be an externality that has a negative impact on house prices.
2.2. Explosions at undesirable facilities and property values
Some studies already investigated the effect of an actual explosion of such an HWS directly on perceived risk and indirectly on property values. Carroll, Clauretie, Jensen and Waddoups (1996) studied several stages of an explosion at a chemical plant in Nevada, USA, in 1988. They compared data from before the explosion, after the explosion, the announcement of relocation and lastly the announcement of the rebuilding 100 miles further away. They used a hedonic price model with multiple distance dummies. They stated that properties within 2 miles of the explosion show lower values both before and after the explosion, but restored values after the announcement of relocation. However, the study reveals that after the explosion the house prices became more sensitive to the distance from the incident. They suggest that higher perceived risk might be an explanation for this change in value. They conclude that these results support the informational efficiency of real estate markets.
Hansen, Benson and Hagen (2006) did further research to investigate the effect of perceived risk after an explosion on house prices. Therefore, they examined house prices around two pipelines; only one of them has exploded in the past. Results showed that before the accident there was no significant price difference for houses around both pipelines. This suggests a lack of perceived risk. After a deadly explosion, the prices around that specific pipeline decreased significantly. However, the prices around the accident-free pipeline did not experience a significant change. They also investigated the influence of distance to the exploded pipeline on the property values. As expected, the distance is negatively correlated to the price effect. Lastly, their results showed that the effect of the explosion diminishes over time. These results contradict the informational efficiency of the real estate market.
The media in the Netherlands stated that citizens of Enschede were ignorant of the firework warehouse in the middle of the living area. Consequently, it is plausible that the house prices around the firework warehouse did not reflect the actual risk.
8 2.3. Risk perception
Every person perceives risk in a different way. There are multiple factors that influence a person’s perception. Naturally a person’s character is one of the explanatory factors, like its attitude or risk sensitivity. But also someone’s cultural background can have an impact on the perception (Sjöberg, 2000). Since this research will examine if the risk perception of the citizens changed after the explosion, it is important to understand how they perceive the risk of an explosion and how willing they are to avoid it.
Fischer et al. (1991) studied these two questions. They investigated which form of perceived risk created the most concerns among citizens. The participants were students and had to hand in three questionnaires; one filled in by themselves, one by a grandparent and one by a parent. This ensured a diverse sample. They were asked to name which risk they perceived the highest. Their research shows that if people have to nominate their own concerns, risks regarding health, safety and the environment are often not mentioned. However, when asking the participants directly which risk in these three topics causes the most concern, 3.3% nominates a fire and/or explosion. In addition, the research shows that the older generations (parents/grandparents) are more concerned about their safety than the student generation. Fischer et al. also looked into the willingness to pay for risk-reduction. They found that respondents were willing to pay a little bit more than a day’s pay to reduce their safety risk. Also half of the respondents had the opinion that it was their own and the government’s responsibility.
Rogers (1997) researched how an event affected the risk perception. He used the natural event of a facility that caught fire. He compared surveys prior and post the fire. To ensure the results were not biased he controlled for the natural change in risk perception over years. He found no significant evidence that people adjusted their risk perception after the event. However, the damage of the explosion in the firework storage was more extensive than the event used in the study. Another research that examined the perception of environmental health risk before and after a refinery explosion, causing 17 casualties, did find that the results are different among subgroups in society. Age, gender and income are significant determinants in risk perception (Cutchin, Martin, Owen & Goodwin, 2008). Consequently, it is important to control for neighbourhood characteristics in this research.
9 2.4. Efficient real estate market
As stated in the introduction, this study can be used to test the efficiency of the real estate market. F. Fama is seen as the father of the Efficient Market Hypothesis (EMH). In 1965 he found evidence for both implications of the random walk model; successive independent price changes and prices moving conform a probability distribution. The impossibility for an investor to make profit from it was one of the economic implications. Multiple researches have been conducted to test this hypothesis on the Real Estate Market. Darrat and Glascock (1989) found that the real estate returns are affected by the lagged values of base money and market returns. This could argue the efficient market theory. In addition, Gatzlaff and Tirturoglu (1995) recapitulate multiple reviews concerning this subject. They conclude that every submarket experiences a different form of efficiency. For instance, long-run returns on the housing market tend to be less efficient than short-run returns.
The inefficiency of the real estate market can also be explained by the smoothing and lagging principles. Confident information and the deficiency of trading result in the unavailability of high quality price information. Appraisals are based on previous and current transactions, causing the smoothing and lagging effect (Geltner, MacGregor & Schwann, 2003). In addition, the presence of transaction costs disrupts the system of supply and demand.
2.5. The contagion effect
The effect this study is examining is also called the contagion effect. Contagion occurs when an event causes a change or disturbance in a different market. The contagion effect in economic context is usually known for the global spread of financial crisis. The interdependence of international economies causes a domino effect during times of distress. Aloui, Aïssa and Nguyen (2011) found proof for this contagion effect between the United States and the BRIC countries during the recent crisis. Moreover, they found that this interdependence applies to both good and bad times.
In the case of this study, the firework explosion probably increased the awareness of the risk of living close to a firework storage, contaminating the values of properties surrounding other storages. A particular contagion effect in the real estate market that has been much debated is the effect of foreclosed properties. Harding, Rosenblatt and Yao (2009) found a discount on
10 property values nearby a foreclosed property of 1%. They do find a diminishing discount over both time and distance. This pattern is also expected for this study.
2.6. Homeownership and the Dutch housing market
People owning a home tend to be more satisfied with their life and have a higher self-esteem (Rossi & Weber, 1996). In the United Stated they even call it part of ‘the American Dream’ to have your own home. This ownership also has positive external effects like higher participation in group or neighborhood activities and a safer neighborhood (Brounen, Cox & Neuteboom, 2012; Rohe & Bassolo, 1997). In 2014, 4.2 million houses in the Netherlands were occupied, which is 56% of the total housing stock. Each year more owner-occupied houses are sold. In the succeeding year this already increased with 16.1% (CBS, 2016).
Prior to the financial crisis in 2008, property values increased tremendously. Therefore, most people assumed buying a house was a risk-free investment. Unfortunately, the subprime mortgage crisis revealed that the opposite was true and that housing prices are also dependent on external (market) factors that we cannot control for (Theebe, 2002). To reduce the risk of investing in a house, it is important to determine as much factors as possible. This research will focus on an undesirable site as one of those factors.
2.7. Hedonic pricing model
As stated earlier, to answer the research question a difference-in-difference model is used. However, to control for omitted variable bias a hedonic pricing model is used. The price of a house is dependent on the buyer’s valuation of its internal and external characteristics (Freeman, 1979). Consequently, not only the attributes of the house itself, but also of the neighbourhood are taken into account. The word hedonic originates from the Greek ‘hedonikos’, and in economic terms this captures “the utility or satisfaction one derives from the consumption of goods and services” (Chin & Chau, 2003).
Two main issues occur when using a hedonic price model, namely misspecification and market segmentation (Chin & Chau, 2003). This misspecification occurs when important attributes are left out, or irrelevant attributes are put in. Attributes that have high production costs and ensure high utility are the most important ones (Butler, 1982). Besides, Mok et al.
11 (1995) state that the bias of too little attributes is in most cases small and neglectable. Therefore, this study examines which variables are truly relevant for the pricing model. Since the tests are concentrated on 15 separate municipalities, the study controls for market segmentation. In addition, the standard errors are clustered on the level of the four-digit zip code.
3. Methodology
To test the hypothesis a Difference-in-Differences (DiD) model will be used. This model compares the average change over a specific time for a control group and a treatment group. In this case the treatment group will contain the properties within 300 meters of firework warehouses (the data and result section will further explain this specific number). The main sample will consist of transactions between January 1, 1990 and December 31, 2015. The specific event time for this research will be May 13, 2000 (the day of the explosion). The basic regression will look like:
ln 𝑃𝑖𝑡 = 𝛼𝐹𝑊𝑖+ 𝛿𝑃𝑂𝑆𝑇𝑡+ 𝛾𝐹𝑊𝑃𝑂𝑆𝑇𝑖𝑡+ 𝜖𝑖𝑡 (1)
Here ln Pit reflects the transaction price of property i in time t. FWi is a dummy variable that equals one if the property is within 300 meters of a firework warehouse. It is expected that these firework storages are located at more economical or industrial neighbourhoods. However, the variable FWi captures this so called selection effect. POSTt is a dummy variable
that equals one if the observation is from after the explosion. FWPOSTit reflects the
interaction variable that equals one if the observation is within 300 meters of the warehouse and is observed after the explosion and last 𝜖𝑖𝑡 is an independently and identically distributed error term. In this regression 𝛼 will reflect the impact of living close to a hazardous waste site. The main parameter of interest is 𝛾 and captures the treatment effect or, in other words, the price development relative to the control group after the explosion.
Since the price of a property is dependent on its internal and external characteristics, a hedonic price model is added to the model. Besides, the amount of firework stored in the warehouse is added. Resulting in the following model:
12 ln 𝑃𝑖𝑡 = 𝛼𝐹𝑊𝑖 + 𝛿𝑃𝑂𝑆𝑇𝑡+ 𝛾𝐹𝑊𝑃𝑂𝑆𝑇𝑖𝑡+ 𝛽𝑋𝑖𝑡+ 𝜖𝑖𝑡 (2)
Where 𝛽𝑋𝑖𝑡 captures the characteristics of the property in a hedonic price regression. 𝑋𝑖𝑡 contains characteristics like size of the property or the house, number of floors and rooms or the condition of the property. This model controls for systematic differences in housing characteristics between the control and treatment group.
Since property values tend to change over time, it is important to control for time fixed effects. Besides, as stated in the literature review, the value is determined by neighbourhood characteristics. Therefore the model should also control for (four digit) zip code fixed effects. The term 𝛼𝐹𝑊𝑖 is captured in the zip code fixed effects and 𝛿𝑃𝑂𝑆𝑇𝑡 in the time fixed effects. Therefore, the new model is:
ln 𝑃𝑖𝑡 = 𝛾𝐹𝑊𝑃𝑂𝑆𝑇𝑖𝑡 + 𝜈𝑛 + 𝜃𝑡+ 𝛽𝑋𝑖𝑡+ 𝜖𝑖𝑡 (3)
Where 𝜈𝑛 reflects the zip code fixed effects measured in neighbourhood n. This approach captures, for example, differences in amenities across neighbourhoods. The four digit zip code level is a relatively large region, so it might unfortunately not control for all location attributes. 𝜃𝑡 captures the fixed effects in year t.
As stated earlier, it is expected that the effect diminishes over time and distance. People living further away might feel less threatened since the explosion might nog reach them. To investigate this hypothesis the model will be extended with distance dummies, resulting in the following model:
ln 𝑃𝑖𝑡 = ∑ 𝛾𝑑𝐹𝑊𝑃𝑂𝑆𝑇𝑖𝑡𝑑 𝑑̅
𝑑
+ 𝜈𝑛 + 𝜃𝑡+ 𝛽𝑋𝑖𝑡+ 𝜖𝑖𝑡 (4)
Where d denotes different distances to the warehouse, each with a longitude of 75 metres. 𝑑̅ is set on 450 meters, ensuring to capture all statistical significant distance frames. Again 𝛾𝑑 is the parameter of interest, reflecting the treatment effect in distance frame d.
13 Nowadays, the media ensures that almost everyone notices a deadly explosion. Nonetheless, it might be possible that people forget the impact of such an explosion after a certain amount of time. Moreover, it may take time for housing markets to adjust. Therefore, equation 3 will also be extended with time dummies:
ln 𝑃𝑖𝑡 = ∑ 𝛾𝑡𝐹𝑊𝑃𝑂𝑆𝑇𝑖𝑡𝑑 𝑡̅
𝑡
+ 𝜈𝑛 + 𝜃𝑡+ 𝛽𝑋𝑖𝑡+ 𝜖𝑖𝑡 (5)
The time dummies will capture time frames of one year. The first time dummy 𝑡 will start on May 13, 2000. Where 𝑡̅ is set on 4 years after the explosion to examine the dynamic response. Again 𝛾𝑡 is the parameter of interest, reflecting the treatment effect in distance frame t.
4. Data and descriptive statistics
In order to answer the main question of this research, two datasets are used. First the data from the Dutch Broker Association will be discussed. Second the data of the firework storages will be introduced and last these datasets will be combined in a correlation matrix.
4.1 Dataset Dutch Broker Association (NVM)
For the price and characteristics of the houses the database of the Dutch Broker Association (NVM) is used. This panel dataset contains the transaction and asking price of the houses sold within the 15 municipalities of the firework warehouses between 1990 and 2010. The individual dimension of the panel data is the (four digit) zip code. In addition, characteristics like size, number of bed- and bathrooms and the presence of a garden are provided. In total the dataset contains 102,078 observations. The average transaction price was €207,739 with an average asking price of €220,009. The spread of the transaction price is from €35,849 to €715,000. These variables are winsorized at a 1% -level to control for biasedness. Table 1 shows the descriptive statistics of the final dataset.
To calculate the distance between the observation and a firework storage, the Pythagoras’ theorem is used:
14 By inserting the difference between the XY coordinates of the observation and the nearest firework storage in the equation, the sky width between the points can be calculated:
The descriptive statistics show that the houses sold had an average distance of 5 kilometers to a firework storage, with a minimum of 20 meters. After exhaustive testing, the radius of 300 meters around a firework storage seems to be a turning point in both sign and significance. Therefore, this radius is used for the treatment group in equation 1-3 and 5. Table 1 shows that 0,13% of the observations lie within this radius of 300 meters. 0,01% of the selling’s took place after the explosion on May 13, 2000 and were located within this radius. Note that the treatment group contains only a small set of observations. Hence the test of the statistical significance of the treatment effect will be conservative.
mean st.dev. min max
Transaction price (€) 207,739 126,333 35,849 715,000 Asking Price (€) 220,009 136,581 38,571 775,000 Selling year 2002 5.142 1990 2010 Distance FW 5091 4307 20.5 28,731 FW <300m 0.0013 0.0362 FW <300m, within 2 years 0.0001 0.0108 Construction year <1945 0.2705 0.4442 Construction year 1945-1959 0.0543 0.2266 Construction year 1960-1974 0.3441 0.4751 Construction year 1975-1989 0.2654 0.4415 Construction year 1990-2000 0.0658 0.2479 Parcel (m2) 248.11 283.34 25 2,784 House size (m2) 133.52 44.092 45 1,128 Number of floors 2.7959 0.5274 1 8 Number of rooms 4.8071 1.3014 1 58 Parking space 0.4019 0.4903
Quality inside - Good 0.8897 0.3133
Quality outside - Good 0.9038 0.2948
Detached 0.1081 0.3105
Simple house 0.0420 0.2007
Single-Famile 0.7160 0.4509
Farmhouse 0.0039 0.0619
Observations 102,078
Table 1. Descriptive statistics housing dataset
𝑑 = √((𝑥2− 𝑥1)2+ (𝑦
15 Figure 1 – Locations of firework storages in the Netherlands on May 13, 2000
16 Zooming in on the characteristics of the properties sold, the analytics show that the average parcel and house size are respectively 248 and 134 square meters. Since these two variables showed outliers they are winsorized at a 1%-level prior the analysis. The number of floors range from 1 to 8, while the number of rooms has a maximum of 58.
The construction dates of the properties are widespread. Approximately a quarter was constructed before 1945, while only 5% originates from the period 1945-1959. 34% is built between 1960 and 1975, 27% between 1975 and 1989 and only 7% between 1990 and 2000. Both nature and quality of the property are determinants of the transaction price. Approximately 90% of the properties were sold in good state for both in- and outside. Only 10% of the houses were detached, while 70% of the properties were single-family houses. This is probably due to the fact that most of the municipalities used for this study are located in a rural area. 40% of the observations included a parking space.
4.2 Dataset firework storages
After the explosion the Dutch newspaper ‘NRC’ created a dossier for all the news concerning the accident. From this dossier the locations of the other firework warehouses stated in the Netherlands at that time are obtained. At the time of the explosion there were 15 other (registered) firework storages. See Figure 1 for the exact locations. The firework warehouses are reasonably spread over the Netherlands, providing a divers sample. The amount stored varies from 50 to 1,800 tonne with an average of 600 tonne.
Each country has its own regulations regarding firework storage. In the Netherlands they distinguish warehouses that store less or more than 10,000 kg of fireworks. A mention to the municipality is sufficient when the storage is less than 10,000 kg. When this number is exceeded an official permit, provided by the province, is required. Another important distinction is the kind of firework. If it is solely consumer used, the clearance of each repository must at least be between 20 and 48 meters (depending on the amount stored). However, if the repository contains firework for theatre purposes, the minimum clearances increases to 400-800 meters (“Vuurwerkbesluit”, n.d.). In the case of Enschede, the houses were approximately 40 meters away.
17 Table 2 – Correlation Matrix
Transaction price (€) Asking Price (€) Selling Year Distance FW Constructio n year Parcel (m2) House size (m2) Number of floors Number of rooms Parking Space Quality Inside Quality Outside Property type Transactionprice (€) 1 Asking Price (€) 0,9929 1 Selling Year 0,4322 0,4272 1 Distance FW 0,0698 0,0666 0,0773 1 Construction year 0,0060 0,0006 0,0620 0,1146 1 Parcel (m2) 0,2634 0,2740 -0,0127 0,0954 -0,0113 1 House size (m2) 0,6765 0,6798 0,0511 0,0617 0,0123 0,2972 1 Number of floors 0,1180 0,1109 0,0268 -0,0047 0,0267 -0,1313 0,2152 1 Number of rooms 0,4759 0,4749 0,0965 0,0063 -0,0092 0,1311 0,5934 0,3339 1 Parking Space 0,2410 0,2466 -0,0331 0,1300 0,0897 0,2819 0,3300 -0,0242 0,1394 1 Quality Inside 0,1544 0,1455 0,0058 0,0575 0,1244 0,0102 0,1207 0,0476 0,0078 0,1176 1 Quality Outside 0,1530 0,1444 0,0084 0,0552 0,1330 0,0086 0,1262 0,0535 0,0263 0,1159 0,8131 1 Property type 0,4999 0,5014 -0,0029 0,1128 0,0705 0,3178 0,5452 -0,0398 0,3017 0,3604 0,2063 0,2109 1
18 4.3 Correlation matrix
Table 2 shows the correlation of the different variables used for this study. As expected, the transaction price and the asking price are highly correlated. They do not have a perfectly linear relationship since bargaining often ensures a difference between the two prices. The selling year is moderately correlated to the transaction price (0,43). DiPasquale and Wheaton (1994) found a visible housing market dynamic, explained by estimates of income and price elasticities. This dynamic can explain the correlation between the selling year and the transaction price. Next to the selling year, the house size, the number of bedrooms and the property type have a moderate positive correlation with the transaction price. Property type is in this case a variable that captures a ranking where the height of the number reflects the luxury of the property. All these correlations are as expected. The other highly correlated variables are the quality of the property in- and outside (0,81). Apparently it is unusual that the maintenance of the interior and the exterior of the property is not the same. Obviously are the variables house size and number of bedrooms and floors positively correlated. All the other variables have a weak or almost no linear relationship.
5. Results
This section discusses the results obtained by the five equations stated in the methodology section. First the results of equations (1)-(3) will be discussed since they form the complete Difference-in-Differences model. In the subsequent subsections the geographic scope and dynamic response, examined by respectively equations (4) and (5), are discussed.
5.1. Difference-in-Differences model
Table 3 shows the results of the equations (1) to (5). For all equations the results are controlled for time fixed effects. In order to calculate standard errors that are robust for serial correlation and heteroscedasticity, they are clustered at the (four-digit) zip code level. As stated in the methodology section, firstly a classical DiD model is executed. Since a diminishing negative effect is expected for this study, the short-term effect is of most relevant. Therefore, the interaction variable is set on observations that are both located in the 300 meter radius and originate between May 13, 2000, and May 13, 2002 (the treatment effect). The possible negative effect might be lifted when looking at the more long-term (dynamic response function). The basic DiD model (1) shows a negative but insignificant coefficient for the interaction variable. For the observations that meet both conditions stated above, it is
19 estimated that they experience a decrease in transaction price of 12%. This model has an R-squared of 0.408. Next, this model is extended with more control variables. In the second model the DiD is expanded with 15 housing characteristics (2). The big increase of the R-squared to 0.786 shows that the model is more explanatory. However, the coefficient of the interaction variable is still insignificant and decreased to a negative effect of 4%.
Since certain areas in a municipality are more valuable than others, it is important to not only control for time fixed effects, but also zip code fixed effects (3). As stated earlier, this way there can be controlled for different attributes across locations. The results show that on average the treatment group experienced a discount of 3% on their selling price. This indicates that homebuyers were more aware of the risks of living close to a firework storage. The perception of risk translates into a lower demand for these particular properties, resulting in a lower house price. This model, controlling for both zip code and time fixed effects, shows an R-squared of 0.947, indicating that it is a good linear fit.
The control variables contain multiple housing characteristics. These variables have signs as expected and are all significant at the 1%-level. The size of the parcel and house are both set in logarithm since it is not expected that these have a linear relationship with the dependent variable transaction price. Consequently, if the parcel and house size increase with 1%, the transaction price increases with respectively 0,23% and 0,45%. This indicates that homeowners attach more value to the house size than parcel size. Also as expected, if the property is detached, has a parking space and is in good condition the price increases. In addition, the number of floors and rooms are positively correlated to the selling price. For every extra floor or room in the property, the price increases with 1%. In comparison to an estate, a simple-, single-family- or farm house is lower in value. The construction date is equivalent to the age of the property. Aside from the vintage effect, where the particular age of the property results in a higher value, the age usually has a negative correlation with the price. The coefficients of the construction variables are relative to a property built between 2000 and 2010. The results show a negative correlation between age of the property and the transaction price. The properties built before 1945 seem to experience a lower discount than properties built between 1945 and 1960. This might be due to the vintage effect.
20 Table 3. Results average effect
(1) (2) (3) (4) (5)
Independent Variables DiD
Hedonic Entity
Fixed Geographic Dynamic Price
Controls Effects Scope Response
FW storage ≤300m 0.056 0.179***
(0.053) (0.058)
FW storage ≤300m, ≤2 yrs after explosion -0.120 -0.041 -0.027*** See figure 2 See figure 3 (0.067) (0.031) (0.003)
Constructed before 1945 -0.132*** -0.185*** -0.185*** -0.185***
(0.042) (0.016) (0.016) (0.016)
Constructed between 1945 and 1959 -0.140*** -0.187*** -0.187*** -0.187***
(0.046) (0.014) (0.014) (0.014)
Constructed between 1960 and 1984 -0.165*** -0.159*** -0.159*** -0.159***
(0.032) (0.011) (0.011) (0.011)
Constructed between 1985 and 1999 -0.046 -0.050*** -0.050*** -0.050***
(0.025) (0.011) (0.011) (0.011)
Parcel in m2 (log) 0.086*** 0.229*** 0.229*** 0.229***
(0.023) (0.007) (0.007) (0.007)
House size in m2 (log) 0.660*** 0.445*** 0.445*** 0.445***
(0.026) (0.009) (0.009) (0.009) # Floors -0.009 0.013*** 0.013*** 0.013*** (0.009) (0.003) (0.003) (0.003) # Rooms 0.027*** 0.010*** 0.010*** 0.010*** (0.003) (0.001) (0.001) (0.001) Parking place 0.008 0.062*** 0.062*** 0.062*** (0.012) (0.004) (0.004) (0.004)
Quality inside - good 0.063*** 0.092*** 0.092*** 0.092***
(0.010) (0.003) (0.003) (0.003)
Quality outside - good 0.069*** 0.055*** 0.055*** 0.055***
(0.010) (0.004) (0.004) (0.004) Detached 0.030 0.110*** 0.110*** 0.110*** (0.021) (0.007) (0.007) (0.007) Simple house -0.340*** -0.218*** -0.218*** -0.218*** (0.030) (0.011) (0.011) (0.011) Single-family -0.298*** -0.143*** -0.143*** -0.143*** (0.021) (0.006) (0.006) (0.006) Farmhouse -0.118** -0.080*** -0.080*** -0.080*** (0.049) (0.023) (0.023) (0.023) Constant 11.128*** 7.606*** 8.706*** 8.706*** 8.706*** (0.045) (0.166) (0.061) (0.061) (0.061)
Year fixed effects Yes Yes Yes Yes Yes
Entity fixed effects No No Yes Yes Yes
Clustered SE on PC4 Yes Yes Yes Yes Yes
Observations 102,078 102,078 102,078 102,078 102,078
R-squared 0.408 0.786 0.947 0.947 0.947
Note: The indicator 'FW storage ≤300m, ≤2 yrs after explosion' is one if the observation is between May 13, 2000 and May 13, 2002 and within 300 meters of a firework storage. Clustered standard errors (PC4) in parentheses,
21 5.2. Geographic scope
One of the expectations of this study is that the effect after the explosion diminishes over distance. People living further away from a firework storage will perceive less risk. This can be explained by the possible unawareness of the location of the firework storage or the resident’s assumption that an explosion would not reach their house.
To examine this diminishing effect, the distance to the nearest firework storage is divided into steps of 75 meters. Since the amount of observations within a radius of 150 meters is not enough to draw a proper conclusion, the tested distances range from 150 to 450 meters. The results show that properties sold within 150 to 225 meters from a firework storage experience an estimated discount of 3.7% (Figure 2). The discount already decreases to 3.2% when the radius shifts from 225 to 300 meters. The effect within 300 and 450 meters is not significant, justifying the cut-off value of 300 meters used in equations (1)-(3). The upward trend line confirms the expectations that the treatment effect is negatively correlated to distance to a firework storage. The positive estimation of the treatment effect between 375-450 can possibly be explained by a selection effect. If homebuyers have a strong preference for a certain neighborhood, but do not want to live close to the storage, they might be willing to pay extra for the properties located further away.
Figure 2: Geographic Scope
-15,0% -10,0% -5,0% 0,0% 5,0% 10,0% 15,0% 20,0% 150-225 225-300 300-375 375-450 Ef fe ct on tra nsac ti on pr ice of prop erty
Distance to a firework storage (m)
Significant at 0.01-level Not significant Trendline
22 5.3. Dynamic response
Besides the geographic effect, a time effect is presumed. During the year following the explosion the media kept reminding the Dutch citizens of the tragic event in Enschede. However, when the lawsuit was closed, the attention around this incident reduced. This might have resulted in a drop in risk perception around firework storages. If so, the negative effect should reduce over time. Figure 3 shows the treatment effect over the four years succeeding the explosion. The results show an estimated negative effect of -1.9% in the first year after the explosion and this even falls to -2.8% in the following year. This extra drop in the second year might be explained by the fact that the lawsuit took place in the second year after the explosion, causing extra attention to the incident. Alternatively, it takes time for housing markets to adjust given the nature of the housing markets. Multiple factors like transaction costs, construction costs, bargaining time and competitive supply cause a low adjustment rate (Wheaton, 1990). The third and fourth year following the explosion show positive treatment effects. This indicates a lack of risk perception or could reflect the stricter regulations regarding firework storages. One concern is that there is no certainty that all firework storages were still in place during the succeeding years. This could create biased estimations since the estimated effect is not caused by the risk perception anymore. However, a closure or moving of a storage could also be an explanation of the sudden increase in property values from three years on after the explosion. The lack of a firework storage could have increased the popularity of the neighborhood and thus the property values. Overall the time scope shows an upward trendline as expected.
Figure 3: Dynamic Response
-10,0% -5,0% 0,0% 5,0% 10,0% 15,0% 0-1 1-2 2-3 3-4 Eff ec t on tr ans ac ti on pr ice of pr oper ty
Years post the explosion of May 13, 2000
Significant at 0.01-level Significant at 0.05-level Not significant Trendline
23
6. Robustness Checks
In this section the results of several robustness checks are discussed (Table 4). All regressions are extensions or adjustments of the third equation (3) in the methodology section. For each of them the standard errors are clustered at the zip code level. The first equation (1) examines an interaction effect between the amount of firework stored in the warehouse and the treatment effect. The second equation (2) reduces the control group to observations within 1 kilometer of a storage. In the third equation (3) the dependent variable is set on asking price instead of transaction price. And lastly, the dependent variable is set on the difference between the transaction and asking price (4).
The amount of firework stored in these storages might influence the knowledge of the existence. Besides, it might affect the risk perception since a higher amount of firework can cause a bigger explosion. Therefore, the regression (3) in Table 2 is expanded with a variable that controls for the amount of firework located at these storages and an interaction variable of the amount of firework with the treatment dummy. The estimation of the treatment effect interacting with the stock is -0.3% and is not statistically significant (Table 4, equation 1). This considerable difference with the baseline estimation could indicate that the baseline estimation is overestimated. The R-squared remains 0.95
The current control group has a distance radius of 300 to 29,000 meters to a storage. However, there still might be unobserved local effects that can be controlled for by reducing the control group. Therefore, the maximum distance of the control group is set on 1 kilometer. Consequently, the number of observations reduced to 4,270. In comparison with the baseline estimation, the estimation of the treatment effect increased to -3.1% and is still significant at the 1% level. The housing characteristic control variables encounter a small change. However, no coefficient changed in sign or significance. The R-squared increased from 0.947 to 0.957, meaning the linear fit is higher when the control group is reduced to 1 kilometer.
As shown in the data section, the asking price and transaction price are highly correlated. However, the asking price reflects the reaction on the risk perception of the homeowners, while the transaction price reflects the homebuyers. Therefore, it is interesting to also investigate the asking prices. Since homeowners might want to sell sooner because of the
24 Table 4 – Robustness Checks Results
Dependent Variable Logarithm of Transaction
Price Logarithm of Asking Price Difference in Logarithms of Transaction and Asking Price (1) (2) (3) (4)
Independent Variables Treatment
effect*tonne
Control
group <1km Asking Price
Transaction Price - Asking Price
FW storage ≤300m, ≤2 yrs after explosion -0.025 -0.031*** -0.015 -0.012
(0.027) (0.009) (0.010) (0.010)
FW storage ≤300m, ≤2 yrs after explosion * tonne -0.003
(0.006) Tonne 0.002 (0.003) Construction <1945 -0.185*** -0.182*** -0.169*** -0.016*** (0.016) (0.027) (0.016) (0.002) Construction 1945-1959 -0.187*** -0.199*** -0.171*** -0.016*** (0.014) (0.041) (0.014) (0.002) Construction 1960-1984 -0.159*** -0.112*** -0.152*** -0.008*** (0.011) (0.020) (0.011) (0.001) Construction 1985-2000 -0.050*** -0.040** -0.048*** -0.003** (0.011) (0.017) (0.010) (0.001) Parcel in m2 (log) 0.229*** 0.248*** 0.234*** -0.005*** (0.007) (0.018) (0.008) (0.001)
House size in m2 (log) 0.445*** 0.379*** 0.458*** -0.013***
(0.009) (0.025) (0.009) (0.001) # Floors 0.013*** 0.015** 0.010*** 0.003*** (0.003) (0.007) (0.003) (0.001) # Rooms 0.010*** 0.013*** 0.010*** 0.000 (0.001) (0.004) (0.001) (0.000) Parking Place 0.062*** 0.056*** 0.066*** -0.005*** (0.004) (0.008) (0.004) (0.001)
Quality inside - good 0.092*** 0.092*** 0.087*** 0.005***
(0.003) (0.015) (0.003) (0.001)
Quality outside - good 0.055*** 0.065*** 0.046*** 0.009***
(0.004) (0.008) (0.004) (0.001) Detached 0.110*** 0.110*** 0.126*** -0.015*** (0.007) (0.025) (0.007) (0.001) Simple house -0.218*** -0.145*** -0.217*** 0.000 (0.011) (0.018) (0.010) (0.001) Single-family -0.143*** -0.120*** -0.148*** 0.005*** (0.006) (0.014) (0.006) (0.001) Farmhouse -0.080*** -0.273*** -0.088*** 0.008 (0.023) (0.064) (0.024) (0.004) Constant 8.706*** 8.855*** 8.683*** 0.023*** (0.061) (0.149) (0.061) (0.008)
Year fixed effects Yes Yes Yes Yes
Entity fixed effects Yes Yes Yes Yes
Year*Entity fixed effects No No No No
Clustered SE on PC4 Yes Yes Yes Yes
Observations 102,078 4,270 102,078 102,078
R-squared 0.947 0.957 0.949 0.218
Note: The indicator 'FW storage ≤300m, ≤2 yrs after explosion' is one if the observation is between May 13, 2000 and May 13, 2002 and within 300 meters of a firework storage. Clustered standard errors (PC4) in parentheses, *** p<0.01, **
25 higher risk perception, but nonetheless want to sell it for the maximum price, it is expected that there will be a negative treatment effect on the asking price but smaller than the effect measured on the transaction price. The results in the third column (3) in Table 4 show that there is an estimation of the treatment effect of -1.5%. Although this effect is not significant, it is as expected lower than the baseline estimation. The control variables encounter small changes, although not in sign. The lower treatment effect after changing the dependent variable to asking price shows that the effect is due to the combination of buyer and seller forces.
Lastly, it is expected that the difference between the actual selling price and the asking price increases for the treatment group since the buyer probably has more bargaining power. He/she can demand a lower price because of the extra perceived risk and the seller’s possible rush to sell. Genesove and Mayer (2001) also argue that sellers that want to avoid a loss set a higher asking price, which would also result in a greater difference between both prices. Therefore, the dependent variable is set on the difference between the transaction and the asking price (see column 4 in table 4). The estimation of the treatment effect is -1,2%, indicating that the markup was bigger when the property was sold within two years after the explosion and within 300 meters of the storage. This effect is not statistically significant. In addition, the results in column 4 show that the bargaining power is positively correlated with the age of the property. On the other hand, good quality of both interior and exterior seem to reduce the difference between both prices.
7. Conclusion/Discussion
This study has investigated the effect of a hazardous site on property values. The explosion on May 13, 2000, is taken as a quasi-natural event for this study. By investigating the change in property values around other firework storages in that period, the effect of risk perception can be measured. It is assumed that the explosion in Enschede created more awareness of the risks of a firework storage.
The results show that properties located within a radius of 300 meters of a storage and sold within two years after the explosion experience a discount of 2.7% on transaction prices. This indicates that people perceived more risk. As stated earlier, people are willing to pay for
risk-26 reduction. Indirectly you might state that people demand a discount when risk increases. This effect is clearly visible in the results of the main regressions. The outcome strokes with the results of other related studies where they found a negative effect of a hazardous waste site.
It was expected that the treatment effect would diminish over both distance and time. Therefore, the effect is tested over several distance intervals. The results show an upgoing trend line, indicating that there is indeed a diminishing effect over distance. This can be explained by a lack of risk reception or the unawareness of the existence of the storage. In addition, the effect is tested in different time periods ranging from 1 to 4 years after the explosion. When people do no longer face the explosion daily, they might forget the impact and thus perceive less risk. The results show indeed a shift from a negative to a positive treatment over the years. Previous studies (discussed in the literature section) also find such a diminishing effect over time and distance.
To check for robustness some extra tests were executed. Since the size of the storage might influence the risk perception, an extra test is executed including a control variable for the amount of firework stored. The treatment effect decreases to -0.3% yet becomes statistically insignificant. This decrease might indicate that the baseline estimation was overestimated. Subsequently, the control group was reduced to control for local unobserved effects. This resulted in a small change in treatment effect, though still negative and significant. Thereafter, the dependent variable was set on asking prices. As expected the treatment effect is smaller comparing to the effect on transaction prices. The sellers of the properties want to receive the highest amount of money possible for their property, hence it is expected that the asking price lies above the transaction price. Lastly, the spread between the asking and transaction price is investigated. As expected, the estimation of the treatment effect is positive, meaning that the difference between the two prices is greater in the two succeeding years after the explosion, provided that the property is located within the radius of 300 meters.
This study still encounters some limitations. First of all, it is no certainty that everybody in the neighborhood is aware of the existence of the firework storage. Since the event used for this study is 15 years ago, it is not possible to conduct surveys to quantify the awareness. In future studies these surveys might emphasize the found treatment effect. Secondly, most firework storages were located in industry/business areas, resulting in fewer observations in the radius
27 of 150 meters around storages. Besides, the explosion caused a broad discussion about the storaging of firework, resulting in some adjustments of the regulations. These changes might have influenced the treatment effect in the succeeding years of the explosion. In addition, there is no information available if the storages still existed in those years. If not, the results can be distorted. Lastly, the standard errors and the entity fixed effects are clustered on the 4-digit zip code level. Since these zip codes still cover quite a large area, they might not capture all location attributes. Controlling at the 6-digit zip code level would give a better estimation of the treatment effect. Unfortunately, this was not applicable for this study due to the given dataset. For future studies it is recommended to obtain a dataset that contains these 6-digit zip codes.
The implications of this study can be viewed from different angles. From the homeowner perspective it would imply that living close to a hazardous site is coupled with the risk of an unexpected price drop. Although the negative effect diminishes over time, if the owner wants to sell it in the succeeding years of such an event, he/she still encounters a discount on the transaction price. Looking at it from the investor’s perspective, it implies that buying a house in the surroundings of a hazardous site in the period succeeding an event like an explosion can be a good arbitrage opportunity. In the case of the firework storage, the investor could have bought a property with an approximate discount of 3% and sell it after two to three years for at least the original value. Lastly, this study might have implications for the regulations regarding firework storages. Since the results show that there is no effect beyond 300 meters, the government could set a new minimum distance to the storages to reduce the volatility of the property values around these storages. Besides, the compensation for both economic and physical damage is still disputable. One could argue that homeowners wanting to sell in the succeeding years should be recompensed for their loss in value, while others could say the treatment effect is only visible in the short-term, implying that no compensation is needed.
In conclusion, the results of this study show that an event at a hazardous site has a negative impact on property values that surround similar sites. However, this effect diminishes over both time and distance.
28
Reference list
Aloui, R., Aïssa, M. S. B., & Nguyen, D. K. (2011). Global financial crisis, extreme interdependences, and contagion effects: The role of economic structure?. Journal of Banking
& Finance, 35(1), 130-141.
Bond, M., Seiler, V., & Seiler, M. (2002). Residential real estate prices: a room with a view.
Journal of Real Estate Research, 23(1-2), 129-138.
Brounen, D., Cox, R., & Neuteboom, P. (2012). Safe and Satisfied? External Effects of Homeownership in Rotterdam. Urban Studies, 49(12), 2669-2691.
Butler, R. V. (1982). The specification of hedonic indexes for urban housing. Land
Economics, 58, 94-108.
Carroll, T. M., Clauretie, T. M., Jensen, J., & Waddoups, M. (1996). The economic impact of a transient hazard on property values: The 1988 PEPCON explosion in Henderson, Nevada. The Journal of Real Estate Finance and Economics,13(2), 143-167.
Centraal Bureau voor de Statistiek/Kadaster 25-4-2016
Chau, K. W., & Chin, T. L. (2003). A critical review of literature on the hedonic price model. International Journal for Housing Science and Its Applications, 27(2), 145-165.
Cutchin, M. P., Martin, K. R., Owen, S. V., & Goodwin, J. S. (2008). Concern about petrochemical health risk before and after a refinery explosion. Risk Analysis, 28(3), 589-601.
Darrat, A. F., & Glascock, J. L. (1989). Real estate returns, money and fiscal deficits: Is the real estate market efficient? The Journal of Real Estate Finance and Economics, 2(3), 197-208.
DiPasquale, D., & Wheaton, W. C. (1994). Housing market dynamics and the future of housing prices. Journal of urban economics, 35(1), 1-27.
Dröes, M., & Minne, A. (2015). Time-varying determinants of long-run house prices. Retrieved from: http://dare.uva.nl/record/1/497597
Espey, M., & Lopez, H. (2000). The impact of airport noise and proximity on residential property values. Growth and Change, 31(3), 408-419.
Fama, E. F. (1965). The behavior of stock-market prices. The journal of Business, 38(1), 34-105.
29 Farber, S. (1998). Undesirable facilities and property values: a summary of empirical studies.
Ecological Economics, 24(1), 1-14.
Fischer, G. W., Granger Morgan, M., Fischoff, B., Nair, I., & Lave, L. B. (1991). What risks are people concerned about? Risk Analysis, 11(2), 303-314.
Freeman, A. M. (1979). Hedonic prices, property values and measuring environmental benefits: A survey of the issues. Scandinavian Journal of Economics, 81, 154-171
Gatzlaff, D., & Tirtiroğlu, D. (1995). Real estate market efficiency: Issues and evidence.
Journal of Real Estate Literature, 3(2), 157-189.
Geltner, D., MacGregor, B. D., & Schwann, M. (2003). Appraisal Smoothing and Price Discovery in Real Estate Markets. Urban Studies, 40(5-6), 1047-1064.
Genesove, D., & Mayer, C. (2001). Loss aversion and seller behavior: Evidence from the housing market (No. w8143). National bureau of economic research.
Gibbons, S., & Machin, S. (2003). Valuing English primary schools. Journal of Urban
Economics, 53(2), 197-219.
Grass, R. G. (1992). The estimation of residential property values around transit station sites in Washington, DC. Journal of Economics and Finance,16(2), 139-146.
Hansen, J. L., Benson, E. D., & Hagen, D. A. (2006). Environmental hazards and residential property values: Evidence from a major pipeline event. Land Economics, 82(4), 529-541.
Harding, J. P., Rosenblatt, E., & Yao, V. W. (2009). The contagion effect of foreclosed properties. Journal of Urban Economics, 66(3), 164-178.
Hughes, W. T., & Sirmans, C. F. (1992). Traffic Externalities and single-family house prices. Journal of Regional Science, 32(4), 487-500.
Li, M., & James Brown, H. (1980). Micro-Neighborhood Externalities and Hedonic Housing Prices. Land Economics, 56(2), 125-141. doi:10.2307/3145857
Lober, D. J., & Green, D. P. (1994). NIMBY or NIABY: a logit model of opposition to solid-waste-disposal facility siting. Journal of environmental management, 40(1), 33-50.
McCluskey, J. J., & Rausser, G. C. (2003). Hazardous waste sites and housing appreciation rates. Journal of Environmental Economics and Management, 45(2), 166-176. doi:10.1016/S0095-0696(02)00048-7
30 McClelland, G. H., Schulze, W. D., & Hurd, B. (1990). The effect of risk beliefs on property values: A case study of a hazardous waste site. Risk analysis, 10(4), 485-497.
McMillen, D. P., & McDonald, J. (2004). Reaction of house prices to a new rapid transit line: Chicago's midway line, 1983–1999. Real Estate Economics, 32(3), 463-486.
Mok, H. M. K., Chan, P. P. K. & Cho, Y-S. (1995). A hedonic price model for private properties in Hong Kong, Journal of Real Estate Finance and Economics, 10, 37-48.
Rohe, W. M., & Basolo, V. (1997). Long-term effects of homeownership on the self-perceptions and social interaction of low-income persons. Environment and Behavior, 29(6), 793-819.
Rogers, G. O. (1997). The dynamics of risk perception: How does perceived risk respond to risk events?. Risk Analysis, 17(6), 745-757.
Rossi, P. H., & Weber, E. (1996). The social benefits of homeownership: Empirical evidence from national surveys. Housing policy debate, 7(1), 1-35.
Sjöberg, L. (2000). Factors in risk perception. Risk analysis, 20(1), 1-12.
Slovic, P., Layman, M., Kraus, N., Flynn, J., Chalmers, J., & Gesell, G. (1991). Perceived Risk, Stigma, and Potential Economic Impacts of a High‐Level Nuclear Waste Repository in Nevada. Risk Analysis, 11(4), 683-696.
Theebe M.A.J. (200). Housing market risks (dissertation). Retrieved from: http://dare.uva.nl/record/1/209911
Vuurwerkbesluit. (n.d.). Retrieved from http://wetten.overheid.nl/BWBR0013360/2015-09-22
Wilkinson, R. K. (1973). House prices and the measurement of externalities. The Economic
Journal, 83(329), 72-86.
Wheaton, W. C. (1990). Vacancy, search, and prices in a housing market matching model.
