The Fukushima Daiichi nuclear disaster : contagion effect on house price development in the Netherlands

(1)

University of Amsterdam

The Fukushima Daiichi Nuclear Disaster: Contagion Effect on House Price

Development in the Netherlands

Sven Martens - 11295783

Faculty of Economics and Business

MSc Finance – Double Specialization: Corporate Finance and Real Estate Finance

Master Thesis

Amsterdam, June 2018

(2)

Abstract

The aim of this study is to investigate whether the Fukushima Daiichi nuclear disaster in 2011 affects house price development in the Netherlands near hazardous sites. This study will distinguish between two groups of hazardous sites; nuclear sites and non-nuclear sites. To analyse the effect on house price development, a difference-in-difference model is designed. This model includes a hedonic pricing method, time fixed effects to control for time trends and zip code fixed effects to control for neighbourhood specific effects. Further, the geographical extent around the hazardous sites is investigated and the study assesses whether there were anticipation or adjustment effects. This study provides little evidence in support of the hypothesis that the nuclear disaster negatively affected house price development in the Netherlands near nuclear sites. Surprisingly, there is more evidence that the house price development near non-nuclear sites is affected by the nuclear disaster. Moreover, the estimations indicate that there is a negative effect between house price development and being close to both types of hazardous sites.

Acknowledgements

This thesis is an important milestone in my academic career. First of all, I would like to express my gratitude to the supervisor, Dr. F.P.W. Schilder, who offered me guidance, support and was the gateway to my data. Secondly, I would like to thank the Dutch Association of Realtors (NVM) for providing the data on residential housing transactions. In addition, I would like to thank the national functional manager of the ‘’Risicokaart’’ for providing the non-nuclear sites database.

Statement of Originality

This document is written by Sven Martens who declares to take full responsibility for the contents of this document: I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

1. Introduction

During the night on April 26th 1986, when most inhabitants of Chernobyl were asleep, a nuclear reactor in the nearby plant exploded. Thirty-one employees of the plant died due to the explosion and many more were exposed to radioactivity. The Soviet Union did not report the explosion for the following three days and therefore most of Chernobyl’s inhabitants were not aware of the spread of radioactivity. More than a hundred thousand individuals were evacuated after the announcement and the inhabitants near the plant were exposed to a high dose of radioactivity. The contamination spread to other regions in Europe, the increased radiation was even measured in Sweden and the Netherlands. Experts have divergent estimations regarding the exact number of humans that were exposed to the radiation in the aftermath of the disaster. According to Greenpeace1, there were almost 100,000 casualties, in contrast to the 4,000 casualties estimated by the International Atomic Energy Agency (IAEA, 2006). The different agendas of these institutions may explain the wide range in assessments. After the event, the nuclear energy programs tried to reduce risks by implementing new safety procedures, regular inspections and tests. As time progressed the risks associated with nuclear power plants decreased due to more experience in the sector and technological development. The nuclear power plants have been part of a discussion for many years and many activists raised awareness for potential risks. Twenty five years after this huge nuclear disaster, an earthquake triggered a fifteen metre tsunami in the east from Japan. The local nuclear power plant was not designed to withstand the combination of these extreme natural forces. These natural forces disabled the cooling and power supply of three Fukushima Daiichi reactors in Ōkuma, which resulted in a new nuclear catastrophe. News about the accident spread around the world and more than a hundred thousand inhabitants were evacuated. The Fukushima accident caused a great impact on nuclear power programs around the world, with governments re-evaluating their nuclear energy programs.

Past nuclear power plant disasters have caused severe long term damage to the surrounding environment. Even though nuclear disasters are not common, the potential damage is huge and initiates debates over whether to continue with nuclear programs all over the world. These debates are held on both a national and an international level as the negative externalities of a nuclear disaster can reach out to neighbouring countries. A nuclear disaster will have an impact on the entire country, but households that live close to a nuclear site will endure most consequences. The residential housing market is affected by the presence of a nuclear power plant in the vicinity, whereas the exact effect is not clear. Clark et al. (1997) argue that one cannot prior determine what the effects of a nuclear power plant will be on the residential housing market as there are different

(6)

effects. A counter argument would suggest that there are obvious risk factors, mostly associated with nuclear fission and with the storage of nuclear waste. Nuclear disasters can easily destroy properties in the vicinity and radiations threatens health of surrounding inhabitants. The risk of a nuclear disaster can have a negative impact on house prices close to the nuclear power plant. Clark and Allison (1999) indicate that the distance to a nuclear power plant negatively affects the house price development. On the contrary, nuclear power plants are often linked with positive effects on local economics (e.g. increased employment). The plant itself requires staff and different stakeholders provide the plant with supplies and services, for these reasons the economic activity increases near the plant. Clark et al. (1997) did not find a negative impact of nuclear power plants on the residential property market, which supports the theories above. Some findings even suggest that nuclear power plants have a positive effect on house prices.

Although the effect of a nuclear power plant on the housing market is not straightforward, one could argue that the severity of the Fukushima disaster changed the public opinion on safety near nuclear sites. Reports of the disaster quickly spread through news channels and remained a hot topic for months. It is obvious that the housing market around Ōkuma was affected by the nuclear accident as households were evacuated and some could not return to their home due to the enhanced radioactivity. Moreover, individuals did not want to live in the area surrounding Fukushima as a result of the disaster. One could argue that this accident changed the risk perception of households in different countries and that there is a contagion effect that stretches from the accident in Fukushima to the housing market in another country. Households outside Japan may have reappraised their risk perception and demand compensation for living near a nuclear site. Welsch & Biermann (2015) found that the disaster changed the risk perception of households in Switzerland. Similar effects were found in a study by Coulomb & Zylberberg (2016), they found that the housing market in the U.K. was negatively affected by the disaster. Properties near a nuclear power plant experienced a price reduction of 3.5% due to the disaster. The effects on the housing market are not equal in all countries though, as Fink & Stratmann (2013) did not find supportive evidence that the U.S. house prices near nuclear power plants were affected.

This study will investigate the Fukushima Daiichi nuclear disaster in 2011 to measure whether there is a contagion effect to the housing market in the Netherlands. In addition to the threat of nuclear sites, one could name different hazardous sites with threat to the surrounding inhabitants. Such examples within the Netherlands include the firework storage explosion in the city centre of Enschede in 2000 and the fire in the chemical site of Moerdijk in 2011. Households close to these sites may feel threatened as an accident can have serious consequences for their personal

(7)

health. This thesis will distinguish between two types of hazardous sites to investigate the contagion effect; nuclear and non-nuclear sites.

Firstly, the contagion effect of the disaster on house price development near nuclear sites is assessed. The goal of this part of the study is to provide insight into the relationship between house price development near nuclear sites and the nuclear disaster in Fukushima. Most studies seem to agree that individuals reassessed their risk perception regarding nuclear energy. However, the effects on the housing market differ among countries and this study can contribute to the existing literature by providing evidence of a contagion effect in the Netherlands. To my knowledge, there are no current or previous studies about the contagion effect of the Fukushima disaster on the housing market in the Netherlands. Secondly, the contagion effect of the disaster on house price development near non-nuclear sites is assessed. Previous literature merely focused on the contagion effect of the nuclear disaster on house price development near nuclear sites. This study will extend the literature by investigating the contagion effect on house price development near non-nuclear sites. The goal of this part of the study is to provide insight into the relationship between house price development near non-nuclear sites and the nuclear disaster in Fukushima. This study will endeavour to prove that contagion can cross boundaries of risk-categories with a view at time of writing to their being no current literature with respect to the contagion effect of a nuclear disaster on house price development near non-nuclear sites. This study will contribute to the existing literature when a contagion effect is revealed. The research question is:

 To what extent did the Fukushima Daiichi nuclear disaster affect house price development near hazardous sites in the Netherlands?

To examine the effect of this event on the house price development, a dataset with property transactions and a dataset with hazardous sites are merged. This study will distinguish between two groups of hazardous sites; the first group are the nuclear sites and second group are the non-nuclear sites, which are classified as ’Besluit risico's zware ongevallen’’ (BRZO) sites. The latter is meant for all companies in the Netherlands that hold large quantities of hazardous substances. The dataset of the hazardous sites consists of four nuclear sites and thirty-five BRZO sites, spread over thirty-one municipalities. To analyse the effect on house price development, a difference-in-difference model is designed. This model includes a hedonic pricing method, time fixed effects to control for time trends and zip code fixed effects to control for neighbourhood specific effects. Furthermore, the geographical extent around the hazardous sites is investigated and the study assessed whether there were anticipation or adjustment effects. This study addresses some of the endogeneity and selection effect concerns. One can argue that the nuclear and non-nuclear sites are not randomly placed

(8)

across space. This is called the selection effect and these hazardous sites are expected to be allocated outside the cities in areas with low residential house prices. To deal with these unobserved factors, our model contains zip code fixed effects. The value of a property is among others determined by the neighbourhood and this study will use zip code fixed effects to control for these unobserved variables. Unfortunately, the zip code fixed effects may not capture all unobserved differences. The four digit zip code may consist of different areas that are not homogenous and this is not observed by the model. Endogeneity concerns are addressed with various definitions of control groups. In addition, the time window to measure the effect after the disaster, the distance intervals and the distance perimeter are changed to verify that the results are robust. The following part will include the results regarding the effect near nuclear sites and afterwards the effects near non-nuclear sites will be discussed.

The first part of the study analysed the relationship between the house price development near nuclear sites and the nuclear disaster. The results of this study indicate that living close to nuclear sites has a negative effect on house price developments. One could conclude that this study was not capable in finding evidence that the event had a negative impact on house price development near nuclear sites. There are negative house price development in the year after the event and this effect is significant in the first six months. Despite these findings, one cannot conclude that the event caused the negative house price development as various model specifications find no significant effects. Most studies seem to agree that individuals reassessed their risk perception with respect to nuclear energy. The reason that this study was not capable in finding evidence that the event caused the negative house price development near nuclear sites can be explained by the low amount of nuclear power plants in the Netherlands. Similar studies in countries with more nuclear power plants provided evidence that the housing market was affected by the disaster. Therefore, the recommendation for future studies is to include the Belgian nuclear power plants and Belgian residential property data. Including these in a future study will improve the likelihood of finding significant estimations and evidence to prove that there is a causal relationship.

The second part of the study extends the scope of the research from nuclear sites to non-nuclear (BRZO) sites. This study finds some evidence of negative house price development near BRZO sites in the aftermath of the disaster. In addition there is evidence that supports the statement that living close to BRZO sites has a negative effect on house price developments. The average treatment effect for properties near BRZO sites is negative and significant, which indicates that these properties have lower house prices in comparison to properties located further away. However, due to the fact that there were negative house price development both before and after the event, one cannot prove a causal relationship between the event and the negative house price

(9)

development. The goal of this part of the study was revealing that contagion crosses boundaries of risk-categories. One can argue that the households near BRZO sites are rational and do not change their risk perception for non-nuclear sites after the disaster. However, counter to the expectations, this study finds more evidence near the non-nuclear sites compared to the nuclear sites. One of the reasons is be attributable to the small amount of observations near nuclear sites in the Netherlands. There are only three active nuclear sites in the Netherlands, of which one is producing electricity. The amount of observations near nuclear sites is smaller compared to the non-nuclear sites. In addition, the nuclear sites are often located in more rural areas with less residential transactions compared to the BRZO sites. Another reason is attributable to the heterogeneity of the nuclear sites. The facility in Borssele is the only electricity producing nuclear plant in the Netherlands. The other two nuclear sites in the Netherlands are considered to be relatively safe as they are used to create medical isotopes and to conduct research. The BRZO sites are heterogeneous in terms of the substances that are used for their processes, whereas the classification of risk is more homogenous.

The Fukushima disaster initiated debates over whether to continue with nuclear programs all over the world. The Netherlands did not change the nuclear power program following the disaster because the government wants to reduce greenhouse gas emissions by almost 50 percent in 2030. The construction of wind parks and solar energy parks are considered to be important in reducing the gas emissions. However, these forms of energy depend on the availability of wind and sun, which troubles a consistent production of energy. One can cover this inconsistency with batteries to store the energy, coal plants or nuclear energy. The latter is perceived to be more risky compared to coal plants. Despite the public opinion with respect to nuclear energy, this research demonstrated that households living further away from the nuclear sites are little affected by the presence of the site. Besides, policymakers should not forget that both the actual risks of the sites and the risk perception affect the relationship between the site and the housing market. Both risks should be taken into consideration when constructing hazardous sites or properties near these hazardous sites. In addition, if the disaster in the nuclear market affects the risk perception of individuals near non-nuclear sites, a similar contagion effects may exist in different markets. This implies that disasters in the future may change the risk perception of individuals in another market.

This study proceeds as follows. The next chapter will aim to discuss the relevant literature regarding asset pricing theories, real estate markets and negative externalities in the housing market. The third chapter describes the design of the empirical models. The fourth chapter will discuss the process of constructing the dataset and the descriptive statistics. The fifth chapter summarizes and discusses the results.

(10)

2. Literature Review

This chapter will discuss the relevant literature regarding asset pricing theories, real estate markets and negative externalities in the housing market. The first part will provide an overview of the main asset pricing theories and the contagion effect in the literature of finance. The second part will discuss the difference between asset pricing models in financial and real estate markets, followed by a further part which will provide insight into negative externalities in the housing market. The fourth part will discuss the Fukushima disaster and the impact of the disaster on international housing markets. The chapter will end with the hypotheses of this investigation concerning the contagion effect of the nuclear disaster to the house price development in the Netherlands.

2.1 Asset Pricing and Contagion in Financial Markets

This study will examine the contagion effect of a disaster in one country to the housing market in another country. To examine this effect, it is necessary to understand the contagion effect and the methods to estimate house price development. Prior to the theories in real estate being discussed, a general overview of asset pricing theories and the contagion effect in financial markets is presented. Real estate and financial markets have both similarities and differences, which makes it meaningful to mention the asset pricing theories and the contagion effect in both markets.

2.1.1 Asset Pricing in Financial Markets

A well-known theory in financial science is the efficient market hypothesis. According to the efficient market hypothesis, all public information and future expectations are incorporated in the market price of assets and securities. A market, where prices fully reflect all available information, is called efficient (Fama, 1970). All information in the market should be incorporated in the price and future development are unpredictable. The random walk theory supports the statement that future development are unpredictable (Malkiel, 1985). This theory suggests that it is not possible to forecast stock prices as the future path is random and unpredictable. The efficient market hypothesis is one of the main assumptions for many theories in financial science. Moreover, the market is assumed to be efficient in the main asset pricing theories, such as the present value model (Campbell, 1997), the dynamic Gordon growth model (Campbell, Lo & Mackinlay, 1997), the capital asset pricing model (Sharpe, 1964) and the arbitrage pricing theory (Ross, 1976). Global financial markets and economies depend on models to determine the value of assets and securities. Asset pricing theories assist investors in different markets such that they can agree on the market value. The aim of these asset pricing theories is to determine the fundamental value of an asset. The above theories are applied both in literature and businesses to value assets in financial markets. These theories are mentioned to provide insight into the differences between the models used in most

(11)

financial markets and in real estate markets. Real estate assets are valued in a different manner and the prices of real estate assets can be forecasted with specific models. The following section will present the contagion effect across economies and in the real estate markets.

2.1.2 Contagion Effect

International economies are increasingly interdependent and financial scientist are becoming aware of the economic globalization. The last decades were marked by a rapid increase in cross-border movement of services, capital and technology. This resulted in greater total welfare and one could argue that globalisation is beneficial for the world’s citizens. However, the increased interdependence of economies and financial markets can have adverse effects in economic downturns. Due to the interdependence of financial markets, economic downturns in one market can spill over to another market. This effect is known as the contagion effect.

The finance literature describes the contagion effect as the spread of market disturbances from one regional market to another. Forbes and Rigobon's (2002) define the contagion effect as a significant increase in inter-market linkages after a shock to a country. The effect can be detectable in economic crises and booms through co-movements in capital flows, stock prices and exchange rates. Global economies have become more interdependent, which intensifies the contagion effect between countries. One should make a distinction between the terms contagion and interdependence, which are sometimes used simultaneously. Samarakoon (2011) describes contagion as the incremental effect during periods of shock and/or crisis, whereas interdependence occurs in the normal market interaction during relative calm periods. A recent study by Aloui, Aïssa, & Nguyen (2011) found evidence that there was a contagion effect during the last major financial crisis. The economic downturn in the Unites States was used to show that Brazil, Russia, India, China (BRIC) were affected by the markets in the United States. Using daily returns, the results reveal that the effects exist in both bearish and bullish markets. Further, there is evidence that there was a contagion effect between some developed and emerging markets during the recent crisis (Celık, 2012; Beirne and Gieck, 2012; Ozkan and Unsal, 2012). In addition, Lee, Wu and Wanga (2007) study the contagion effect of the earthquake and the concomitant tsunami in 2004. The results of the study suggest that the stock market in surrounding countries was not affected by the potential contagion effect, whereas the foreign exchange markets of certain countries was affected.

The above literature suggests that the contagion effect exist in financial markets, whereas the effect in the real estate sector is less evident. Wilson and Zurbruegg (2004) find little effect of contagion in the real estate market during the stock market crisis of 1997. In comparison to the property markets, the equity markets seemed to have made a greater impact on other financial markets. Hoesli and Reka (2013) study the contagion in the securitized real estate market in the

(12)

period of 1990 to 2010 and they find evidence that there was contagion. Securitization will enhance the resemblance with assets in the stock market due to the ease of trading securitized real estate assets. This study will focus on the residential property market and the contagion effect in this market is extensively discussed in literature covering the effect of foreclosures on a neighbourhood (Harding, Rosenblatt and Yao, 2009; Gangel, Seiler and Collins, 2013; Gerard, Rosenblatt, Willen and Yao, 2015). These studies find evidence that the contagion effect exist around the foreclosed properties and that the effect diminishes as the distance increases. Further, the findings suggest that the price reduction for properties diminishes over time. The studies regarding the foreclosures measure the contagion effect of a foreclosure within the same market. This thesis will focus on the contagion effect of an event in one market to a different market. Although the studies are individual, this study will expect similar trends regarding the distance and adjustment effects. To examine this effect, one should first understand the real estate markets. The next section will address this.

2.2 Real Estate Markets

The real estate market is characterized by a lack of full information, a low number of buyers and sellers and a relatively slow reaction of supply and costly transactions (Lusht, 2012). These characteristics distinguish the real estate market from other financial markets. The stock and bond market meet most of the criteria set by Fama (1970). According to these criteria, all information is available with no fee for all market participants and there are no transaction costs for trading the asset. As this criteria is not met by most real estate markets, one can argue that the real estate markets are not efficient. Prior research support claims that housing prices do not reflect all available information and therefore, the market is not informationally efficient. Several studies provided evidence that future prices can be predicted through past price changes, this indicates that the market is not efficient (Guntermann and Norrbin, 1991; Pollakowski and Ray ,1997; Case and Shiller, 1989; Capozza, Mack and Mayer, 1997; Rayburn, Devaney and Evans, 1987). In addition, Pollakowski and Ray (1997) find that not all publicly available information is incorporated in the prices of houses. The standard asset pricing models that are mentioned in the previous section cannot always be applied in real estate. The appropriate models to determine the value differ and house price models or indices are constructed to estimate house price developments. There are three different types of indices: appraisal-based, property share-based and transaction-based indices. This study will use a transaction-based index because it is the most common index and transaction data is available in the Netherlands. This index can be constructed by either a hedonic price method or a repeat-sales method. The repeat-sales method requires several transactions of the same property and would be difficult with our dataset. For this reason, the hedonic method is applied and this method is examined in the following section.

(13)

2.2.1 Hedonic Method

The housing market is rather unique as it embodies the characteristics of spatial fixity, heterogeneity and durability (Chau and Chin, 2003). One way to overcome the difficulties in valuing assets in the housing market is by implementing a hedonic model. The hedonic method tries to quantify the satisfaction or utility a household derives from a specific dwelling. The price a household is willing to pay for a property depends on the inherent attributes of a house, such as neighbourhood, structural and locational attributes. One can estimate the relative importance of these variables with a hedonic regression (Rosen, 1974). Several criteria and assumptions have to be met for the hedonic price model to work effectively. One of the first assumptions of the model is homogeneity in the housing product. Chau and Chin (2003) believe that this assumption is contested as housing products seem to be heterogeneous. Houses can be differentiated in terms of neighbourhood, structural and locational attributes. The second assumption, perfect competition with numerous sellers and buyers, seems to hold in the housing market since there are many developers selling and numerous buyers looking for a house in the Netherlands. The third assumption that sellers and buyers have perfect information regarding the housing market is questionable. Chau and Chin (2003) argue that most of the information about the housing market is publicly available and accessible by households, whereas Lusht (2012) argues that there is a lack of full information. Despite these thoughts on some of the assumptions, the model is widely used in housing market literature due to it’s easy of implementation and limited information required. This study will use the hedonic method due to the fact that the repeat-sales method requires several transactions of the same property.

The hedonic price model estimates how both the external and internal characteristics of houses affect the price a consumer is willing to pay. Bourassa et al, (2006) states that by estimating the effect each characteristic has on the price of a house, the model controls for heterogeneity. The specification of the model is important to realize unbiased and accurate estimations. Further, Can and Megbolugbe (1997) state that the precision and accuracy of the estimations are affected by the assumptions regarding the error term, the selection of characteristics and the functional form of the hedonic function. One should not forget these statements when forming an empirical model.

2.3 Negative Externalities

Different types of environmental externalities often influence the housing market and some of these externalities will be discussed in this section. Positive externalities will positively affect the housing market while the opposite effect is expected for negative externalities. Environmental externalities can affect the housing market through different channels such as perception of risk, pollution and noise. Examples of negative externalities are airport noise (Cohen and Coughlin, 2008) wind turbines (Dröes and Koster, 2016), power lines (Colwell, 1990) and power plants (Davis, 2011).

(14)

Industrial sites are often mentioned to have a negative effect on the environment as pollution, noise, vision and risks may harm the surroundings. For example, a study by Li & James Brown (1980) found that an industry site has a negative impact on house prices nearby. Most studies find a negative effect whereas some studies find no clear effects attributable to the industrial site. Bléhaut (2014) finds that transaction prices do not change after an industrial accident. In a different study, the effect of a chemical plant explosion in Henderson Nevada in May 1988 is investigated by Caroll et al. (1996). The paper analyses the effect before the explosion, after the explosion and after the announcement that the plant would be relocated. The results of the paper suggest that the explosion of the plant had a negative effect on the house prices. This study will focus on nuclear and non-nuclear sites and the next section provides an overview of the literature on these subjects.

2.3.1 Hazardous Waste Sites

There is extensive literature about hazardous waste sites and these sites are comparable to the BRZO sites of this study. For this reason, the next section will provide an overview of the main literature regarding hazardous waste sites. Households incorporate the risk assessment of hazardous waste sites when purchasing a property. A hazardous waste site is expected to have a negative impact on the housing market as individuals tend to be risk averse (Pratt, 1964). The actual risk and the perception of risk can be two quite different topics as individuals may over or under-estimate the risks associated with hazardous sites. The empirical work by Cutchin, Martin, Owen and Goodwin (2008) concludes that the perceived risk differs per group in society. Research tested whether several groups in society had different risk perceptions with questionnaires regarding the large explosion in 2005 at an oil refinery in Texas City. Results indicated factors including income, gender and age were important and effective. Another study by Rogers (1997) suggests that the actual risk of a hazardous site is not as important as perceived risk. The perceived risk is largely dependent on the social processes that construct and maintain risk beliefs by individuals in the vicinity of hazardous sites.

Prior empirical studies find that the housing market is negatively influenced by hazardous waste sites. Thayer, Albers and Rahmatian (1992) concluded that households consider the proximity of waste sites when purchasing a property. Firstly, the findings suggest that the greater the distance between the waste site and the property, the lower the price reduction will be. Secondly, the findings indicate that households were willing to pay less for properties near hazardous waste sites compared to non-hazardous waste sites. Another study assessed the difference between risk assessment of experts and homeowners near a landfill site. The risk perception of homeowners and experts did not always agree. The risk beliefs of households resulted in a price reduction of the properties in the vicinity of the landfill sites, whereas risk assessment of experts was less correlated

(15)

with house prices. Furthermore, exposure to media coverage, age, young children and odour increased the risk beliefs of the homeowners (McClelland, Schulze & Hurd, 1990). The Fukushima disaster was extensively reviewed in the news and this can change the risk perception of inhabitants. Although the above literature suggests that hazardous waste sites depress property values, there are some studies where this effect does not appear. Greenstone and Gallagher (2008) study whether local changes in the housing market differ if a hazardous site is selected to be cleaned with the Superfund program. This is in comparison to sites that narrowly missed the qualification and were not selected for further cleaning. The paper finds no evidence that the clean-up of hazardous waste sites is associated with local changes in property rental rates and residential property values. In addition, the empirical work of Gayer (2000) focusses on the effect of expert’s opinions about the risk of hazardous waste sites. After the Environmental Protection Agency (EPA) published their risk assessment of a site, the property values increased, indicating that the inhabitants overestimated the risks associated with the sites. The effects of nuclear waste storage on the housing market is unclear also. Some researchers like, Metz and Clark (1987), did not find an impact of nuclear waste storage relocation on the housing market. The hedonic study compared nuclear waste sites that were relocated to those that were not relocated and found no effect. On the contrary, Smolen, Moore and Conway (1991) found that the announcement of radioactive waste sites has a negative impact of house prices nearby.

2.3.2 Nuclear Power Plants

The housing market is affected by the presence of a nuclear power plant in the vicinity, whereas the exact effect is unclear. Clark et al. (1997) argue that one cannot prior determine what the effects of a nuclear power plant will be on the property market as there are different effects. Firstly, there is the obvious risk associated with the activities and the waste of the nuclear plant. Nuclear accidents can destroy properties in the vicinity and radiation could threaten the health of surrounding inhabitants. Secondly, a nuclear power plant is often associated with positive local economic effects (e.g. increased employment). The plant itself requires staff and different stakeholders will provide the plant with supplies and services, which increases the economic activity in the vicinity of a plant. Clark et al. (1997) used a hedonic model with geographic information system (GIS) techniques to investigate what the effect of two nuclear power plants in California was on the housing market. Supportive of the above theories, Clark et al. (1997) did not find a negative impact of nuclear power plants or stored nuclear waste on the property market. Some findings even suggest (not significant though) that the nuclear power plant has a positive effect on house prices.

Nelson (1981) was one of the first to study the effect of a nuclear power plant on the housing market in the vicinity of a plant. Using a hedonic price model, he studied the effect of the

(16)

Three Mile Island accident on residential property prices in the vicinity of a plant. The findings suggest that there was no decline in residential property values and the appreciation rate for property prices remained unchanged. This is in line with the findings of Gamble and Downing (1982), who find no significant distance effect for nuclear power plants along the east coast of the United States. A different study in the United States by Clark and Nieves (1994) investigates whether the amount of noxious facilities in a region influences the wages and property values. Prior work has focused on the effect of a noxious facility within a certain region, whereas this study (Clark and Nieves, 1994) compares different regions across the United States with one another. Findings suggest that wages and housing prices in areas with more noxious facilities are lower compared to the control group. Especially, the presence of nuclear power plants and petrochemical refineries seem to contribute to this effect. One of the explanations given by the authors is that the inhabitants overestimate the risks associated with the noxious facilities. Further, a study by Clark and Allison (1999) analysed the effect of the announcement of dry storage facilities on the housing market near these facilities and their results indicate that distance to the nuclear power plant negatively affects the housing market.

The above literature describes the effect of a nuclear power station using hedonic pricing models. However, different methods can be used to measure the impact of a plant on the housing market. Schneider and Zweifel (2013) studied the relationship between the willingness to pay for increased insurance coverage and the distance from the plant. The findings suggest there is a negative relationship and that households are willing to pay less for the insurance as the distance from the plant increases. These results suggest that individuals are aware of the risks and this study will endeavour to verify if the disaster changed this risk perception.

2.4 Fukushima Disaster

The Tōhoku earthquake initiated a 15-metre tsunami on the 11th March 2011. The waves of the tsunami disabled the cooling and power supply of three Fukushima Daiichi reactors in Ōkuma, which resulted in a nuclear accident. The three operational reactors were automatically shut down after the earthquake and the damage to the normal electricity grid triggered the emergency power supply. The cooling pumps for the power stations depended on the emergency power supply and the waves damaged 12 of the 13 back-up generators. To maintain operational, the power plant was equipped with several batteries in case the emergency generators and the normal electricity would break down. However, the batteries were not designed to last more than a couple of hours as a total breakdown of both the normal electricity and emergency generators was not anticipated. The damage by the tsunami and the earthquake could not be repaired immediately as the access was blocked and the reactors started to overheat. The overheating caused a meltdown in three units and

(17)

nuclear fuel was partly melted at the bottom of the reactors. Several accidents, among which an explosion of hydrogen gas, occurred after the meltdown and it took Tepco employees nine months to control the situation. Over 100,000 inhabitants were evacuated from the area due to the radiation levels and the accident was rated 7 on the INES scale, which is the highest possible rating. Partly by the evacuation of individuals in a perimeter of 20 kilometres around the plant, no official deaths or cases of radiation sickness were reported. Among the evacuees were many cases of evacuation stress, which some argue to outweigh the radiological hazards of returning2.

The Fukushima accident made an impact on nuclear power programs around the globe and governments re-evaluated their opinions towards nuclear energy. China and India both suspended approval for new nuclear plants and the United States strengthened safety regulations. Germany announced to prematurely shut down some of its reactors and to completely quit with nuclear energy by 2022. Public debates over whether to continue with the nuclear programs emerged in France and in the Netherlands, where last month, citizens were still protesting to quit the nuclear power programs3. In a reaction to the disaster, the flood wall was raised in the nuclear power plant Tihange and additional resistance tests were performed in the Netherlands and Belgium4. Following public debate and demonstrations, the Belgian government decided last year to close all the nuclear power stations within seven years5. The Fukushima Daiichi nuclear disaster in 2011 may be used to investigate whether news about the disaster can affect the housing market in a different country. There may be a contagion effect that stretches from the accident in Fukushima to the housing market in another country. The severity of the disaster may have changed the public opinion about the safety near hazardous sites and thus, may even impact the house prices in the vicinity of nuclear power plants outside Japan.

Welsch & Biermann (2015) found that the Fukushima disaster changed the risk perception of households in Switzerland. The survey data revealed that households living close to a nuclear power plant rate their life satisfaction (proxy of utility) different compared to households that live further away. The difference in scores was influenced by the topographical and meteorological conditions (e.g. people feel safer behind a mountain). Interestingly this study is the reassessment of households following the accident as people became more aware of the risks associated with nuclear power plants. Similar effects were found in a study by Coulomb & Zylberberg (2016), who studied the effects of the Fukushima Daiichi disaster on the housing market in England and Wales with a hedonic difference-in-difference method. According to this study, properties close to a nuclear plant experienced a price reduction of 3.5% after the accident, which can be attributed to the change in 2 http://www.world-nuclear.org/information-library/safety-and-security/safety-of-plants/fukushima-accident.aspx 3 https://nos.nl/artikel/2231244-nederlands-protest-in-aken-tegen-kerncentrales-en-macron.html 4_{https://www.fanc.fgov.be/nl} 5 https://nos.nl/artikel/2225080-belgische-regeringspartij-overstag-alle-kerncentrales-in-2025-dicht.html

(18)

risk perception. Bauer, Braun and Kvasnicka (2013) concluded that the housing prices in the vicinity of nuclear power plants in Germany decreased by 5-10% after the incident in 2011. The price reduction can be attributed to either the change in risk perception or to a change in local economies. The perceived risk can be negative if the inhabitants adjust their risk perception due to the disaster in Fukushima and the local economies can be affected if the site was closed after the event. The German nuclear power policy changed after the accident and the government started closing some of the plants prematurely. The nuclear phase-out could have a negative effect on local economies and house prices. The latter seems to be the most influential as the housing markets near sites that were closed after the accident, reflected the largest drop in house prices. Although some of the above literature suggests that the disaster changed the risk perception of households along with the willingness to pay for a dwelling, not all studies find similar effects. Fink & Stratmann (2013) found no evidence that the house prices in the United States near nuclear power plants decreased in value after the Fukushima disaster. Their hedonic difference-in-difference findings even suggest that the house prices appreciated slightly and that the residents in the United States did not reappraise their risk associated with nuclear power plants after the incident in Fukushima. Different effects are mentioned in the literature and it remains unclear whether the Fukushima accident changed the risk perception of inhabitants in diverse countries.

(19)

2.5 Hypothesis

This section will present the main research question and the complementary hypothesis. The literature suggested that the Fukushima disaster had a negative impact on the housing market for some countries, whereas some countries were unaffected. Therefore, this study will test whether this effect is measurable in the Netherlands and the main research question is:

 To what extent did the Fukushima Daiichi nuclear disaster affect house price development near hazardous sites in the Netherlands?

This study will distinguish between two categories of hazardous sites located in or close to the Netherlands. The first category is the nuclear sites and the second category is the non-nuclear sites. The non-nuclear sites in this study are hazardous sites with a BRZO identification and this category is meant for all the companies that have large quantities of hazardous substances. To answer the main research question, two hypotheses are formed and empirical tests will validate the hypothesis. The two main hypotheses of this study are:

I. The Fukushima Daiichi nuclear disaster in 2011 negatively affected house price development near nuclear sites in the Netherlands.

II. The Fukushima Daiichi nuclear disaster in 2011 negatively affected house price development near non-nuclear (BRZO) sites in the Netherlands.

(20)

3. Methodology

This chapter describes the design of the empirical models applied to test whether there is a causal relationship between the event and the house price development near nuclear and non-nuclear sites. In the first part, different stages of the difference-in-difference model identification process will be presented and discussed. In the second part, the initial model will be adapted to reveal the effect of several distances to the house price development. Finally, a model to reveal the anticipation effect before the event and adjustment effects after the event will be presented.

3.1 Basic Regression Model

In this study, the average treatment effect after the disaster within d kilometres of a property will be estimated. The nuclear disaster is a quasi-natural experiment that assesses whether the house price development near hazardous sites is affected by the event. The research method of Dröes and Koster (2014) is consulted as their difference-in-difference model can be applied to this study. A difference-in-difference model is formed to disclose the relationship between the nuclear disaster and house price development. The model will estimate the differences between the treatment and the control group. The treatment group will contain houses that are located near the hazardous sites and the control group consists of houses that are located further away. Chapter 5 will elaborate on the effects of several distances (d) that are used to distinguish which properties belong in the treatment group. For now, this study will assume a distance of 1,500 meters for nuclear sites and 250 meters for BRZO sites. It is important to control for time fixed effects as the value of a property is determined by time trends. Residential property prices seem to follow time trends and property prices are correlated with interest rates, inflation and the economy (Catte et al. 2004). By adding time fixed effects in the model, one can control for these unobserved factors that change over time. The basic difference-in difference model is:

ln 𝑃𝑖𝑡 = 𝛼𝐻𝐴𝑍𝑖+ 𝛽𝐻𝐴𝑍𝑃𝑂𝑆𝑇𝑖𝑡+ 𝜆𝑡+ 𝜀𝑖𝑡 (1)

where the subscript i stands for the property in question and the subscript t stands for the date of the transaction. In this regression, ln 𝑃𝑖𝑡 is the dependent variable and reflects the logarithm of the

transaction price of property i at time t. The dummy variable 𝐻𝐴𝑍𝑖 is the treatment group dummy,

which indicates whether the house is located within the perimeter around a hazardous site. This dummy controls for the selection effect. According to the selection effect, the nuclear and non-nuclear sites are not randomly allocated across space. It is expected that these sites are placed outside the cities where house prices are low. The interaction variable 𝐻𝐴𝑍𝑃𝑂𝑆𝑇𝑖𝑡 is the variable of

(21)

coefficient captures the average treatment effect. This variable denotes ‘’1’’ if the transaction was within the 1,500 meter or 250 meter perimeter and occurred between March 2011 and March 2012. The 𝜆𝑡 variable captures the time fixed effects in year t. The identically and independently

distributed error term is denoted by 𝜀𝑖𝑡. Next, a hedonic pricing method is added to model 1 to

control for housing characteristics because both internal and external characteristics can explain differences in house prices, discussed throughout the literature review. Housing characteristics including the condition of the property, the year of construction, number of (bed)rooms, number bathrooms, lot and house size are included in the regression to control for the differences between the treatment and the control group. This results in the following regression model:

ln 𝑃_𝑖𝑡= 𝛼𝐻𝐴𝑍_𝑖+ 𝛽𝐻𝐴𝑍𝑃𝑂𝑆𝑇_𝑖𝑡+ 𝛾𝑋_𝑖𝑡+ 𝜆_𝑡+ 𝜀_𝑖𝑡 (2)

where 𝑋𝑖𝑡 captures a set of housing characteristics. Further, one may argue that the hazardous sites

are not randomly placed across space, which is known as the selection effect. These sites may be constructed in less attractive neighbourhoods and located in areas with lower residential housing prices. To deal with these unobserved factors, a model should contain location fixed effects (Dröes and Koster, 2014). The value of a property is determined by amongst other factors the neighbourhood with this study using zip code fixed effects to control for these unobserved variables. The four digit zip code is applied as a proxy for the neighbourhood as this is the most detailed information available. Unfortunately, the zip code fixed effects may not capture all unobserved differences. The four digit zip code may consist of different areas that are not homogenous and this is not observed by the model. This results in the following regression model:

ln 𝑃_𝑖𝑡= 𝛽𝐻𝐴𝑍𝑃𝑂𝑆𝑇_𝑖𝑡+ 𝛾𝑋_𝑖𝑡+ 𝜇_𝑛+ 𝜆_𝑡+ 𝜀_𝑖𝑡 (3)

where 𝜇𝑛 indicates the zip code fixed effects (zip code) in zip code n. Note that the 𝐻𝐴𝑍𝑖 variable is

omitted to prevent multicollinearity. The 𝐻𝐴𝑍𝑖 variable indicates ‘’1’’ if the house is located close to

a hazardous site, with this variable collinear with the zip code fixed effects. Finally, equation 3 is estimated with both the full sample and with a restricted sample. One could argue that properties in the treatment group should be compared to properties that are located just outside the treatment group. There may be unobserved local trends that can be accounted for when using a more restricted sample (Dröes and Koster, 2014). In addition, it makes more sense to compare transactions prices in the same municipality compared to transaction prices between municipalities. This study performed an empirical analysis to determine the correct distance to restrict the sample with. This resulted in a reduction of the control group to a perimeter of 7,500 meter for nuclear sites and a perimeter of 5,000 meter for BZRO sites.

(22)

3.2 Distance Profiles

The model to determine the correct perimeter around the sites is explained in this section. The perimeter will distinguish which properties belong in the treatment group and which properties belong in the control group. It is expected that households living closer to a hazardous site feel more threatened compared to households living further away. The effect of a hazardous site may become less pronounced when moving further away from a site. At some distance (d), households are indifferent about the presence of a hazardous site and the risk perception will be close to zero or unmeasurable. In order to identify this perimeter correctly, the model is extended to include various distance dummies. This results in the following regression model:

ln 𝑃_𝑖𝑡= ∑ 𝛽_𝑑

𝑑

𝐻𝐴𝑍𝑃𝑂𝑆𝑇_𝑖𝑡𝑑+ 𝛾𝑋_𝑖𝑡+ 𝜇_𝑛+ 𝜆_𝑡+ 𝜀_𝑖𝑡 (4)

where d indicates the different recorded distance dummies. The coefficient of interest in this model is the βd parameter, in which the treatment effect of each distance is reflected. The cut-off value is

determined by analysing at which point the treatment effect becomes insignificant. This cut-off distance is different for nuclear sites and non-nuclear sites as the risks associated with these sites differ.

3.3 Anticipation and Adjustment effects

The model to reveal the anticipation and adjustment effects is presented throughout this section. Several time intervals will be adopted to check how property prices develop before and after the event. One can assume that individuals did not forecast the Fukushima accident and that anticipation effects were zero. Although there may not be any anticipation effects, they are included as a robustness check. The absence of any anticipation effects can validate if the treatment effect is attributable for the Fukushima accident by analysing whether there is a jump in house price development around the event.

Regarding the adjustment effects, the media coverage over the event peaked in the weeks following the accident and remained in the news for months following. Effectiveness of the news over the incident may differ for the short and long term. In the short term, prices are expected to decrease for houses that are located close to hazardous sites, in comparison with the long term effect that is likely to gradually diminish as memory of the event is forgotten. In addition, Case and Shiller (1989) mentioned that it takes time for housing markets to incorporate changes. Thus, it takes time for the housing market to incorporate new information after individuals change their risk assessment. The anticipation and adjustment effects are determined by expanding the model with time dummies. This results in the following regression model:

(23)

ln 𝑃_𝑖𝑡= ∑ 𝛽_𝑝 𝑡−1 𝑝=𝑡−𝑇 𝐻𝐴𝑍𝑃𝑂𝑆𝑇_𝑖𝑡+ ∑ 𝛽_𝑝 𝑡+𝑇 𝑝=𝑡 𝐻𝐴𝑍𝑃𝑂𝑆𝑇_𝑖𝑡+ 𝛾𝑋_𝑖𝑡+ 𝜇_𝑛+ 𝜆_𝑡+ 𝜀_𝑖𝑡 (5)

where p indicates the different time period dummies. The time horizon that is used for this study is ten years and the effects of five years (T=5) before/after the event are estimated. This study will assume that the effects of the event will not be measurable beyond this time horizon. The anticipation effects are captured by the first term on the right side of the equation and the adjustments effects are captured by the second term on the right side of the equation.

Although some of the identifying assumptions of this study are addressed in this chapter, one can still question validity of these assumptions (Dröes and Koster, 2014). The parallel trend assumption is a critical assumption in a difference-in-difference model to guarantee internal validity of the model (Angrist and Pischke, 2008). This assumption states that the differences between the control and the treatment group are constant over time, in absence of an event or treatment. With respect to the study concerning the Fukushima disaster, this implies that the house price development near nuclear and non-nuclear sites should be similar to the house price development of the control group. Another aspect one should consider in this approach is the selection effect. One may argue that the hazardous sites are not randomly placed across space. These sites may be constructed in less attractive neighbourhoods and the four digit zip code may not observe differences at a different level. The four digit zip code may consist of different areas that are not homogenous and this is not observed by the model. Bertrand et al. (2004) mentioned that questions regarding the identifying assumptions is likely to remain in most difference-in-difference methods. The fifth chapter will discuss several robustness checks to address some of the highlighted issues.

(24)

4. Data

The process of constructing the dataset and the descriptive statistics will be discussed in this chapter. This study will distinguish between two groups of hazardous sites; nuclear sites and non-nuclear sites. Firstly, these two groups of hazardous sites shall be presented and the data from the Dutch Broker Association discussed. Secondly, two datasets will be merged and the final dataset presented. Finally the descriptive statistics of the treatment and control group are compared with one another.

4.1 Nuclear Sites

The locations of the active nuclear sites should be determined to investigate whether living near a nuclear site affects house price development. There are three sites in the Netherlands with nuclear activity, whereas only one site is producing electricity. Nuclear power plant Borssele (KCB) generates electricity with a mixture of uranium and plutonium and supplies a maximum of 495 megawatts6. The other two nuclear sites are not used for producing electricity and this reduces the risks associated with the sites. The Hoge Flux Reactor (HFR) in Petten produces medical isotopes that are used for research and the Hoger Onderwijs Reactor (HOR) in Delft is used by students of the TU Delft for educational purposes and research. The reactor in Delft is situated near the University and approximately three kilometres from the city centre, which endorses the previous statement that risks association with this site is considered to be small. There are several nuclear sites in Germany and Belgium that can threaten the inhabitants close to the border. This study will include one nuclear power plant that is situated three kilometres from the Belgian border. Nuclear power plant Doel compromises four nuclear reactors and may affect the house price development in the Netherlands. News about leakage at the plant intensified the public debate whether to close the facility7 and on the 10th of June, the reactors were stopped due to a malfunction8. It is important to note that the nuclear sites are not randomly placed across space. The nuclear site in Delft is situated near the Technical University of Delft to enable the students and researchers of the university to conduct experiments. The nuclear sites in Borselle, Petten and Doel are situated in rural areas with lower residential property prices. The fact that these sites are not randomly assigned across space results in a selection effect, which may influence the estimations. This study will include these four nuclear sites with more information of these sites reflected in Appendix 1 and in Figure 1.

6

https://www.autoriteitnvs.nl/onderwerpen/kerncentrale-borssele-epz

7_{http://www.omroepbrabant.nl/?news/2777081133}

(25)

Figure 1 – Spatial Distribution of Nuclear Sites

Note: The signs on the figure indicate where the nuclear sites are situated within the sample.

Figure 2 – Spatial Distribution of Non-Nuclear (BRZO) Sites

(26)

4.2 Non-Nuclear Sites (BRZO)

The locations of the non-nuclear sites should be determined to investigate whether living near a nuclear site affects house price development. The Dutch government continues to track varying hazardous sites and activities throughout the country and created the platform ‘Risicokaart’ to inform the public of risks in their environment. The data on the platform contains the locations of all hazardous sites and routes in the Netherlands. This study will focus on the most dangerous companies/sites due to nuclear sites being most comparable. These sites are categorized as ‘Besluit risico's zware ongevallen’ (BRZO). This category is meant for all companies that have large quantities of hazardous substances and are not working with radioactive materials. These companies are a source of risk as they store toxic, flammable or explosive substances. The locations of most hazardous sites are publicly available and can be obtained from the ’Risicokaart’ platform of the Ministry of Security and Justice. However, the dataset of the hazardous sites does not differentiate between the various types of hazardous sites, routes and potential natural disasters. The national functional manager of the ‘’Risicokaart’’ was approached and the manager was willing to provide the necessary codes to extract the BRZO sites from the data. Important to note is that the BRZO sites are not randomly placed across space. One would expect that the BRZO sites are situated in rural areas with low house prices. However, the properties near the BRZO sites are more expensive when compared to the control group, which will be discussed further in the next section. This study will include thirty-five BRZO sites and more information about these sites is reflected in Appendix 1 and in Figure 2.

4.3 Descriptive Statistics

The Dutch Association of Realtors (NVM) provides a monitoring database with information of transaction prices and housing characteristics in the Netherlands. The NVM was requested to share their data and the association provided data on 31 municipalities from 2006 until 2016. These municipalities are selected due to the location of nuclear sites within these areas. This time horizon is chosen to make sure the anticipation and the adjustment effects can be revealed. The full list of municipalities in this study is shown in Appendix 2. The NVM dataset and the dataset within which locations of the hazardous sites are merged with support of the Pythagoras’ theory. In order to split the sample in a treatment and control group, the distance between the hazardous sites and the observations (house transactions) is calculated. The coordinates of the nuclear and the BRZO sites are manually obtained with support of a GPS tool9. The coordinates of the properties in the NVM

(27)

dataset are known and equation (6) and (7) are applied to calculate the distance between the hazardous sites and each property.

𝑎2_{+ 𝑏}2_{= 𝑐}2 ₍₆₎

𝑑 = √((𝑥2− 𝑥1)2+ (𝑦2− 𝑦1)2) (7)

Following the calculation of distance between the property and the nearest hazardous sites, observations can be divided into two groups; the treatment and the control group. The distance to the nearest hazardous site will determine whether an observation is part of the treatment group or not. The NVM provided a dataset of 75,821 property transactions in four provinces. To control for outliers several variables (e.g. parcel size, house size and the transaction price) are windsorized at the 1% level. Furthermore, the dataset that was not complete due to missing data and these were dropped to endeavour the dataset was complete. If missing data is random, there should be no issues with deleting this data. However, if the missing data is not random, deleting the data may bias the results. One could argue that the missing NVM data is not random as the realtors may not complete some parts of the questionnaires for particular reasons. For simplicity, this study will assume that the missing data was random and the missing data is dropped. Table 1 provides an overview of the most important characteristics of the dataset. It is important to note that the dataset is split in two treatment and two control groups. The control groups are almost identical to one another and therefore, only one control group is presented in Table 1. Further, the descriptive statistics of the full dataset can be seen in Appendix 3.

The average transaction price is € 238,869 and ranges from € 73,000 to € 740,000. The average transaction price for properties located near a nuclear site is lower compared to the control group. This can be attributed to two reasons, either the nuclear sites depress property values or the nuclear sites are placed in areas with low house prices. The latter can be described as the selection effect and this implies that nuclear sites are not randomly placed across space. Governments are aware of the risks associated with nuclear sites and will place the sites in rural areas with low residential house prices. In contrast, the average transaction price for properties located near a BRZO site is higher compared to the control group. The BRZO sites in this dataset are located in industrial areas at the outskirts of cities, which would imply lower housing prices compared to control group. However, the statistics do not confirm this statement and this could be due to the fact that the control group contains more houses in rural areas compared to urban areas. One could argue that properties in the industrial areas have higher transaction prices compared to properties

(28)

in rural areas. Another argument could be that the sample of properties near the BRZO sites is not representable for the population due to the low number of observations.

The descriptive statistics reveal that more than 85% of the properties are well maintained, both within and outside the property. Furthermore, the average house and parcel size are respectively 127 and 349 square meters. Interesting to note is that most of the houses are classified as single-family properties, this could be due to the fact that most of the municipalities are located in rural areas. The fact that over 20% of the properties are classified as detached houses supports the above statement. The descriptive statistics provide more insight on the minimum distances between a property and a nuclear sites. The properties in the dataset are on average 18 kilometres away from the nearest nuclear site, ranging between 548 meter and 40 kilometres. Approximately 0.28 percent of the transactions are located within 1500 meter of a nuclear site, which represents a total number of 153 transactions. Only 4 (2.6 percent) of these transactions occurred in the year following the event. The properties in the dataset are on average 7 kilometres away from the nearest BRZO site, ranging between 189 meter and 24 kilometres. The treatment group of the BRZO sites consists of 29 observations and only 1 (3.4 percent) occurred in the year after the event. Both the treatment group of the nuclear and the BRZO sites consist of few observations and the number of observations in the year after the event is even smaller. This could be attributable to the fact that households prefer to live further away from these sites and therefore, only few properties were built in the vicinity of these hazardous sites. Another explanation could be the size of the dataset, as only 31 of the 380 municipalities are included in this study.

The Fukushima Daiichi nuclear disaster : contagion effect on house price development in the Netherlands

University of Amsterdam