Valuing the risk attached with living close to a hazardous waste site

DEPARTMENT OF ECONOMICS Uppsala University

D-level thesis

Author: Kristina Svensson

Supervisor: Maria Vredin Johansson Autumn 2005

Valuing the risk attached with living close to a

hazardous waste site

Abstract

In this thesis I estimate a monetary value of the risk attached to living near a hazardous waste site in the town of Teckomatorp. This site is the result of hundreds of rusty leaking barrels of toxins being buried in the ground by the company BT Kemi in the 1970’s. Ever since then the site has been remediated in several steps and is still contaminated today. For estimating the perceived risk of living near this site I use a hedonic price model (HP) which is a form of a revealed preference approach. In a HP model the price of a market good is a function of different utility-bearing characteristics and the estimated parameters can be used to calculate the implicit prices of these characteristics. In this case I use a data set from the National Swedish Institute for Building Research (IBF) and regress property price on a number of housing characteristics. I compare an estimated town-effect for Teckomatorp with the estimates for two control towns: Billeberga and Anderslöv. I can confirm my hypothesis that, after controlling for housing characteristics, there is a negative effect on prices of property in Teckomatorp. I find that property prices are on average 46878 SEK lower in Teckomatorp than in the two control towns.

LIST OF CONTENTS

4.2 Hedonic price method (HP)... 11

5. Data... 13

5.1 Summary statistics... 13

5.2 Description of data set... 15

1. Introduction

The very nature of environmental goods makes them very difficult to valuate. Whereas regular goods are traded in the market and the equilibrium price is determined by supply and demand, environmental amenities have no market price. An interesting topic to be studied further is how hazardous waste sites are valued by the people living in their proximity. In Sweden there was an environmental scandal in the town of Teckomatorp in the 1970’s when the company BT Kemi buried tons of toxic production waste in the ground. The town and its people were greatly affected by this and so was the neighbouring town of Billeberga. My purpose with this essay is to see if, by using the hedonic price method combined with a difference-in-difference approach, I can get an estimate of the economic value of the risk attached to living near the hazardous waste site in Teckomatorp. I will study the prices of property and introduce dummy variables that give town-specific effects. The estimate I get for Teckomatorp will be compared to the estimates for two control towns: the less affected Billeberga and the completely unaffected Anderslöv. If prices are lower in Teckomatorp when controlling for other housing characteristics, I believe that this, to some extent, can be explained by the presence of the hazardous waste site. This difference in property price could also be used as a measure of the trade-off between risk and money for people living in this area.

The hedonic price method is the most common method for valuing housing quality and housing attributes.1 In Sweden most of the research using a hedonic price model seems to have been done mostly on the effects of traffic noise on near by property prices, whereas several studies have been conducted in the US where the negative relationship between property values and proximity to hazardous waste sites has been confirmed. As an example, Kohlhase finds, by using this method, that the marginal benefit of being a mile further away from a site was $2,364 (1985 US dollars) after the site had been declared as hazardous. Ihlanfeldt and Taylor (2004) finds that for the average office building, increasing distance from 0,5 miles to 2,0 miles causes a 36 percent increase in property price and for apartments a 23 percent increase. Proximity to a hazardous waste site does usually not have an impact on the property prices before the site has been officially announced as hazardous2. Another study by Kiel and Zabel (2001) finds that the benefits from cleaning up two sites in Woburn,

1 _{Eriksson, B. J. (2000). p 146}

Massachusetts would be in the range of $72 million to $122 million (1992 dollars). Studies also show that toxic waste sites have been ranked as the number-one environmental problem in the US.3

I will start by presenting some background information on the BT Kemi scandal and briefly describe the stated and revealed preference theory as this relates to the subject of interest. In section 4 I present a model that would be optimal to use as a method for estimating the value of risk in Teckomatorp. Unfortunately all of the required data for this model is not available to me and I will therefore have to use a simplified model specified in section 6. Some over-all statistics for Teckomatorp, Billeberga and Anderslöv are presented in section 5 together with a description of the data set that will be analyzed. I conclude by presenting my results and conclusions.

2. Background

In the mid 1970’s one of the biggest environmental scandals in Swedish history was revealed. The Danish company BT Kemi was manufacturing and formulating pesticides (primarily weed killer) in its factory in the small town of Teckomatorp in Skåne between 1965 and the company’s bankruptcy in 1977. Residents in the area soon began to notice a foul smell in the air, unpalatable drinking water, and children were starting to have trouble breathing and develop allergies. This created a growing debate and increased media coverage when it became known that production waste had been buried on the site. A substantial effect of this was seen in the nearby river Braån on both fish and under-water vegetation. Between 1975 and 1977 hundreds of rusty leaking barrels filled with toxics were found buried in the ground. Substances like chlorophenols, chlorocresols, phenoxyacetic acids, dintrobutyl-phenol and dioxins were found, where phenoxyacetic acids are considered to be the main pollutant.4 In 1976 a trial began which granted the gardening family Ahl in Billeberga a claim for damages of 1,25 million SEK after their greenhouse crops were damaged by the contaminated water from the river.5

3 _{Clymer, A. (1989). p 7}

4 _{Bevmo, L. and Englöv, P. (2004). p 1 f}

The industrial area in Teckomatorp is divided into two parts where parts of the southern area have been sold off by the Svalöv municipality and are now being used for different kinds of activities. It is the northern part that was used for treatment of sewage and depositing of waste that is now an inaccessible waste ground. This area is estimated to contain a total amount of hazardous waste of approximately 2-3,5 tons and cover an area of 6-7000 m2_{. It was mainly}

the soil and groundwater that became polluted and there is now a drainage system with pumps and wells that collects the ground water to stop the pollutants from spreading. However, a shut-down of this system could have severe effects on the surface water in the river. Today there are no apparent effects on the flora and fauna in the area and measurements of the water in Braån show no evident traces of leakage of pollutants from the groundwater. The substances in the ground are however of a very non-biodegradable type which means that the spreading to surrounding areas will most likely continue for a very long time.6

During the time of production there were no measurements made on the surrounding air which makes it impossible to say what the exposure to pollutants might have been. Today the air is not considered to constitute any negative effect on health or environment. It is also hard to estimate the exposure through drinking water and health examinations conducted in 1981, primarily on children in Teckomatorp and Billeberga, gave no results that can be tied to BT Kemi. Blood tests on former employees at BT Kemi however, showed unusually high levels of the toxic dioxin TCDD and several complaints were made about skin, mucous membrane and bronchi symptoms. The number of deaths is not considered to be any different from the country average and it was too soon, in 1981, to get reliable results on the carcinogenicity of chlorophenols and phenoxyacetic acids.7 The number of employees was approximately 40-45 at a time8 _{but many of the employment periods were fairly short (less than a year) and the} total number of employees for the entire period was 263 individuals.9

The site has ever since the late 1970’s been remediated in several steps and is now in its final clean up stage.10 _{The complete cleaning process is estimated to have a total cost of 200} million SEK11 _{and is, since 2002, being paid for by the government. The municipality of} Svalöv has the over-all responsibility and the goal is to remove 80 percent of all the

contaminants in certain areas, and that Teckomatorp should no longer be affected by BT Kemi.12

For the people living in this area the scandal has obviously had many negative consequences. Allergies among children, possible cases of cancer, damaged crops and a loss of 30 000 tons of soil among other things.13 Teckomatorp is still today highly associated with the BT Kemi scandal14, so besides the more health-related effects, households in this community have most likely experienced losses of a more economic character as well. Housing is probably one of the biggest investments made by a household and to have one’s house-value depreciated by an external factor, like this environmental scandal, is not just a loss of assets but also something that can be used to get an idea of how people value a clean environment - or in this case a value of the trade off between risk and money for those who decide to move into the hazardous waste site area.

Figure 1 shows the region of Skåne where Teckomatorp is located and the BT Kemi area (marked with an arrow) is situated in the outer parts of the town, but still relatively close to residential areas.

Source: BT Kemi, Huvudstudie, version 3.

12 _{http://www.arkitekt.se/s16623}

3. Theoretical framework

When it comes to valuing public goods or environmental amenities we are not really used to setting a price in the same way as we do with regular market goods. Pure public goods and environmental amenities, like air and water, are non-excludable to consumers and this therefore undermines the possibility of charging a price or a fee, and hence trading these goods in the same way as regular goods are bought and sold in the market. So how can we value a clean environment and put a monetary value on amenities, and in cases like this; a negative externality? To try to answer these questions we can choose between two different methods: stated and revealed preference.

3.1 Stated preference

The most common stated preference method is the Contingent Valuation Method (CVM). In a CVM you try to create a hypothetical market where consumers have the opportunity to buy the good in question. People are usually presented with a detailed description of the hypothetical good and then asked what their willingness to pay (WTP) for this good would be, or alternatively what their willingness to accept (WTA) would be. This method is often used to estimate the total benefit of providing a public good or improving the environment by aggregating these values to get a measure of the total price, or benefit, of the good being valued.15

Examples of questions could be how much someone would be willing to pay to, or accept if, let’s say, Statsskogen in Uppsala would be turned into a residential area. Economists often address these types of questions in cost-benefit analyses where the costs of providing a public good or a clean environment are contrasted against the perceived benefit16_.

This method is based on respondents’ willingness to reveal their true preferences, and this is also what causes the biggest problems. It is common that people behave strategically and underbid if they believe that they will actually have to pay the amount revealed by them. A reverse strategy is if people overbid when they believe that they will not have to pay the stated amount, but still believe that what they state as their willingness to pay can influence the provision of an amenity or public good. Researchers therefore often obtain larger values from

WTA surveys than from WTP surveys and this gives rise to a problem of correctly evaluating these surveys and the measurement errors can be large.17

3.2 Revealed preference

A way of trying to circumvent the problems that arise in stated preference approaches is to use a revealed preference approach that relies on consumers’ actual behaviour. In these studies, the prices of regular market are used for determining how much of the price of that good that can be associated with a certain non-priced amenity. Thus the consumers indirectly reveal how much they value different attributes that come with the good in question.18 _The price that the consumer is willing to pay represents the demand for the “bundle” of characteristics that the product has.19 In the case of Teckomatorp I wish to estimate the “value” (which is expected to be negative) for living close to a hazardous waste site. This is a characteristic feature of Teckomatorp and is most likely reflected in the prices of properties in the area. These lower prices could hence be used as an indirect measure of the economic value of the risk attached to living in proximity to such a site.

One revealed preference method is the Hedonic price method (HP) which was formalized by Sherwin Rosen in 1974. I describe this model in section 3:4.

Since it is impossible to know how the housing prices would have developed if the BT Kemi scandal had not happened, we can use a Difference in Difference (DD) approach to circumvent this problem.

4 Method

In this section I present a model that could be used for estimating the value of risk. It is constructed as a difference in difference model where the property prices are compared with results derived from a hedonic price model.

4.1 Difference in Difference (DD)

The DD approach is used to estimate the effect of a “treatment”, or an intervention, on an outcome, in this case the property price. The model is well-suited for estimating the effect of sharp changes in the economic environment and is often used in labour market research. The model requires that one finds comparison sites that can be used to estimate what would have happened in the absence of the treatment.20 Thus, this model would be used to complement the HP model. In this case these comparison sites would be two small towns in Skåne (Billeberga and Anderslöv) that are supposed to have similar characteristics to Teckomatorp with the exception that they have not been directly affected by the “treatment” (the BT Kemi scandal) in the same way as Teckomatorp.

One key identifying assumption in this model is that the interactions between these towns are zero in the absence of the treatment. This is a very strong assumption and one way to test this is to compare trends in outcomes before and after the event of interest.21

If the property price y in the treatment (d = 1) and control groups (d = 0) after (t = 1) and before the treatment (t = 0), the expected property price difference can be written as:

C = E _            = = − = = −       = = − = = 0 0 1 0 0 1 1 1 i i i i i i i i d t y d t y d t y d t y Equivalently: C = E _            = = − = = −       = = − = = 0 0 0 1 1 0 1 1 i i i i i i i i d t y d t y d t y d t y

By differencing these terms we get:

C = E

[

]

Equation (1) gives a difference between pre- and post treatment property prices in Teckomatorp and the control groups. If C < 0 the property prices in Teckomatorp have been negatively affected by the BT Kemi scandal.

The estimated property price in this model will be derived from the HP model.

Billeberga has together with Teckomatorp been used in health examinations conducted in an attempt to determine the health effects of BT Kemi. Anderslöv was used as a “control area”22 and this is why these two towns could be appropriate to use as control groups in the DD approach. The towns are relatively similar population wise, 900 and 1549 people respectively, compared to 1563 people living in Teckomatorp in the year 2000.23 Billeberga is downstream along Braån, not far from Teckomatorp, and has also been affected by the contaminated water from the river. Teckomatorp and Anderslöv’s distances to the nearest big city (Landskrona and Trelleborg respectively) are also relatively similar. It is perhaps not possible to control for all other variables that affect these towns but my hypothesis will be that, all else being equal, property prices will be at their lowest in Teckomatorp, slightly higher in Billeberga and at their highest in Anderslöv, which is completely unaffected by BT Kemi.

The advantage of the DD approach is that I do not need to control for the properties’ distances to the hazardous waste site, and macro variables like unemployment, crime rates etc. These towns are assumed to have experienced similar changes in these variables will therefore not be included.

4.2 Hedonic price method (HP)

This theory is based on the hypothesis that the price of a market good is a function of a vector

z, of different utility-bearing characteristics so that products markets implicitly reveal a function p(z) = z1,..., zn relating prices and characteristics.24 If a consumer is assumed to be a

price taker, the consumer will choose the point where its marginal willingness to pay for an additional unit of each characteristic is equal to the marginal implicit price of that characteristic. In a regression model the estimated parameters can be used to calculate the

22 _{Bevmo, L. and Englöv, P. (2004). p 101 f} 23 _www.scb.se

implicit prices of these characteristics.25 General weaknesses with the HP model are that it is very sensitive to modelling decisions and the conditions in the local housing markets.26

The model would in this case be applied to property prices and the price function yi, for

property i would take the following form:

yi = α + βti + γdi + δ(tidi) + θ'xi + εi

The parameters in the model give the marginal effects of a change in each characteristic on the property price. ti is a dummy variable that indicates whether the sale took place before or

after the BT Kemi scandal.

   = i t

di is a dummy variable that indicates whether the sale took place in Teckomatorp or not. xi is a vector of property-specific characteristics (like size of house and property, age of house and other quality attributes). α is a constant, β gives the time trend common to the treatment and the control group, γ is a treatment group specific effect that accounts for permanent differences between treatment and control groups, δ is the true treatment effect (the product of a dummy indicating observations after the treatment and a dummy indicating the treatment group) and θ is a parameter vector to be estimated consisting of all the property-specific characteristics and possibly some regional variables.

The property price would be regressed on stacked micro data for towns and years and I would test the hypothesis H0: δ = 0, HA: δ < 0 to see if there is a statistically significant negative

effect on the property prices in Teckomatorp.

There are many implicit assumptions within this model and one of these is that there are no restrictions or transaction costs associated with moving to and from an area. Together with the problem of taking all variables into account, the model is likely to suffer from omitted

25 _{Vainio, M. (1995). p 31} 26 _{Navrud, S. (2002). p 6}

variables and measurement errors.27 Since location is an important factor that influences property prices, the price of two apparently similar houses may differ greatly depending on where they are located. This in addition to the immobility of houses gives rise to regional markets.28

5. Data

In this section I will first present statistics on some macro variables for the regions we are concerned with, and thereafter I present the data set that will be used for the empirical analysis.

5.1 Summary statistics

As will be evident when presenting the model, macro economic variables are assumed to have affected Teckomatorp and the two control towns Billeberga and Anderslöv in the same way. Some variables might however, have more local variation and I therefore present some over-all statistics on the macro variables population size, unemployment and income. It has sometimes been impossible to get statistics at town-level and in these cases it will be given at either municipal or parish level. It has also been difficult to get hold of data for the same time periods so these will also vary. My main purpose with this is to see to what extent these variables vary between the regions.

Teckomatorp and Billeberga are both in Svalöv’s municipality and Anderslöv belongs to Trelleborg’s municipality. Billeberga and Anderslöv have their own parishes’ whereas Teckomatorp is part of the parish of Norra Skrävlinge. The number of residents in the parish and the town can vary substantially. In the year 2000 the number of residents in the town of Billeberga was 900 compared to 1362 in the parish. In Anderslöv the corresponding numbers were 1549 compared to 2921. In Teckomatorp the number of residents was 1563 compared to 1619 in the parish of Norra Skrävlinge.29 Since I wish to compare the towns, this difference in population size could cause a problem when interpreting the results. When it comes to Teckomatorp the measurement error should not be too great, but could be of significance in

the cases of Billeberga and Anderslöv. To clarify, I will from now on refer to the corresponding parishes when talking about Teckomatorp, Billeberga and Anderslöv.

At municipality and parish level the population development has looked as follows:

Table 1: Average population Table 2: Average population in municipalities in parishes

Year Svalöv Trelleborg Year Teckomatorp* Billeberga Anderslöv 1970 12063 36008 1970 1394 1150 1698 1975 12485 34875 1980 1510 1439 1721 1980 12963 34459 1990 1482 1483 1994 1985 12707 34103 2000 1619 1362 2921 1990 12773 35665 2001 1540 1352 2949 1995 12885 37609 2002 1586 1368 2978 2000 12590 38328 2003 1571 1375 3006 Source: SCB 2004 1619 1399 3083 Source: SCB

*Refers to the parish of Norra Skrävlinge

The population in Anderslöv almost doubles from 1970 to 2004 compared to only a slight increase in Teckomatorp and Billeberga. This should reasonably lead to a sharper increase in property prices in Anderslöv because of a higher demand for housing.

Table 3: Average income in municipalities

- in thousands of SEK

Year 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Svalöv 115' 119' 122' 126' 130' 135' 137' 144' 150' 157' 165' 171' 177' Trelleborg 121' 125' 128' 133' 136' 140' 144' 151' 157' 163' 170' 176' 183' Source: SCB

Table 4: Average unemployment rate in municipalities

Year 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Svalöv 6,0 6,4 6,4 6,5 5,1 4,7 4,3 3,0 2,9 3,6 4,1 Trelleborg 8,3 8,4 8,1 8,1 5,4 5,2 4,5 3,8 3,7 4,2 4,2 Source: AMS

It is not possible to find numbers on how the unemployment rate changed in Teckomatorp after the shut-down of BT Kemi, but a higher local unemployment rate should reasonably have a negative impact on property prices. On a municipality level I can only get data for the years after the time frame of the data which I will analyze, but what is evident is that the two municipalities sometimes differ greatly and this will make it important to take this into consideration in the analysis. Something worth noticing is that the unemployment rate in Trelleborg’s municipality is consistently higher than in Svalöv’s municipality.

5.2 Description of data set

At the National Swedish Institute for Building Research (IBF) there is data available for the different property-specific characteristics that determine the rateable value of a property, and there is also data on the number of sales and the sales price. The data is gathered for sales concerning small houses in the three parishes for the years between 1981 and 1993 and are presented in more detail below. This data is part of a larger data set previously used in a working paper by Berger (1998).

To get an overview over the data I will first present some summary statistics.

Table 5a: Number of property sales Table 5b: Average sales price

Year Billeberga Teckomatorp* Anderslöv Year Billeberga Teckomatorp* Anderslöv 1974 7 9 14 1974 107000 112000 103000 1975 10 14 17 1975 99000 126000 85000 1981 21 15 14 1981 243222 164400 202786 1982 18 13 18 1982 207111 183212 271556 1983 20 11 17 1983 159723 192182 254177 1984 21 17 23 1984 196095 140765 239261 1985 20 24 22 1985 271250 192208 245591 1986 22 18 26 1986 244046 173861 312846 1987 30 16 36 1987 287333 246188 283833 1988 27 21 26 1988 294444 255691 329923 1989 27 18 30 1989 499473 290444 406100 1990 21 15 23 1990 472619 495333 429348 1991 18 16 26 1991 564611 645000 572981 1992 7 6 10 1992 395714 282833 702000 1993 4 6 9 1993 368750 346667 590444 Total: 273 219 311 Average: 294026 256452 335256 Source: For the years 1974-1975: SCB, for the years 1981-1993: IBF

*Refers to the parish of Norra Skrävlinge

explained for the most part by the fact that Sweden was experiencing a recession in the early 1990’s30.

I run a parametric test to check if the mean sales prices between all three towns are significantly different from each other for the period 1981 to 1993. The hypothesis that the means in Teckomatorp and Billeberga would be the same cannot be rejected at the 5 percent significance level and the same goes for the test of means between Billeberga and Anderslöv. The hypothesis that means would be the same in Teckomatorp and Anderslöv can be rejected at the 1 percent significance level. This indicates that Teckomatorp for some reason differs from the Anderslöv, but so far I have not controlled for any differences in housing characteristics that could explain some of the variation in the price variable. With the model presented in section 6 I intend to control for these differences and see if there still is a negative effect associated with Teckomatorp.

5.3 Econometric issues

With this data set I should be able to control for heterogeneity when it comes to different property characteristics that affects the property’s price. There is also a possibility of heterogeneity when it comes to more geographical factors like different neighbourhoods and districts. In the data set there are no variables measuring this, but my personal belief is that neighbourhood characteristics are less important in a small town (coming from one myself) and it should be possible to get reasonable results without including these in the regression. What is also assumed is that accessibility to work places and local public services are the same in these three towns and this is also something that may vary.

Something that is difficult to control for is if there would be an incorrect perception of the danger associated with living in this area, which could result in a biased estimate. It is the perceived risk, and not the actual risk, that affects the willingness to pay for a property31. Teckomatorp is still highly associated with the BT Kemi scandal and might suffer from a “stigma” that will affect the property prices more negatively than would be expected, considering that the site is cleaner today than it was thirty years ago. Since the site is still contaminated it is hard to determine to what degree a possible stigma affects the property prices.

Another problem is that the shut-down of the BT Kemi factory might have had a grave impact on the unemployment rate in the area. If formerly employed people have lost their jobs they might be more desperate to move and be willing to sell their house at a lower price. People moving into the area might also be less willing to pay high prices for properties because of the unemployment situation. This could in such a case be an exogenous variable that affects the property price. An exogenous shift in the income level, by for example an inflow of low income people, is also something that could explain a change in property prices.

There is also the problem of not knowing if a sale took place in the town or somewhere else in the parish. As mentioned earlier this should cause the greatest measurement errors when we are dealing with Billeberga and Anderslöv, but be fairly small in the case of Teckomatorp.

Finally, the number of property sales will be equal to the sample size in the regression model and since I have a relatively small number of observations for some years this will affect the reliability of the estimates negatively.

6. Model

Ideally, if data were easily available, the hedonic price model would look as described earlier with dummy variables for both time and place of the sale. Unfortunately it is not possible to get hold of micro data that dates as far back as to the 1970’s. However, it will still be interesting to run two separate regressions with three different dummy variables indicating town specific effects and dummy variables qj for the years 1982 to 1993 (1981 will be used as

reference so j = 1982,...,1993). These two regressions should capture some of the expected difference in property price and I will be able to see if there is some support for the hypothesis that property prices are higher in Anderslöv than in Teckomatorp and Billeberga.

The two regression models will look as follows:

Regression 1: yi = α + γ1d1+ βjqj + θ'xi + εi

with dummy variable:

The expected outcome from running this regression is that the parameter γ1 is negative, that is: when controlling for different housing characteristics, property prices in Teckomatorp will be lower than in the two control towns. I will therefore test the hypothesis H0: γ1 = 0 against the

alternative HA: γ1 < 0 to see if this holds.

Regression 2: yi = α + γ2d2 + γ3d3 + βjqj + θ'xi + εi

with dummy variables:

   = 2 d    = 3 d

In the second regression I want to see if there is a joint significance of the γ2 and γ3 parameters.

This tests the null hypothesis H0: γ2 = γ3 = 0 against the alternative hypothesis HA: γ2 ≠ 0, γ3 ≠ 0

that either or both variables are not equal to zero. Since I expect property prices in Anderslöv to be higher than in both Teckomatorp and Billeberga I also expect the following outcome:

γ3 > γ2.

γ1 gives the effect of Teckomatorp and γ2 and γ3 gives the effect of Billeberga and Anderslöv

respectively. Since I cannot include data from the time before the BT Kemi scandal I will not be able to control for any initial differences between these towns. In this model it will be the treatment and control group specific effect γj that captures the differences between these

towns, and hence the difference between being situated close to the hazardous waste site or not. However, this parameter will also capture all other differences that are uncontrolled for. Macro variables like interest rate and inflation are however assumed to have affected the entire country in the same way and will therefore be reflected in the property prices to the same extent in all regions. It would be desirable to include the macro variables income level and unemployment rate, presented earlier in section 5:1, since these can obviously differ greatly between regions, but because of the lack of, and inconsistency, of this data it is unfortunately not possible. This will of course result in more uncertain interpretations

1 if Billeberga 0 if otherwise

7. Empirical analysis

Using the data set from IBF I will estimate, what I consider to be, the best model for testing my hypotheses. A description of the variables I include in the model can be found in Appendix 1. The price variable for each parish is:

Table 6: Price variable

Variable: Price Obs. Mean Std. err. 95% Conf. interval Min. Max. Teckomatorp* 196 271228,3 17562,61 236591,3 305865,4 30000 2820000 Billeberga 256 317698,8 13181,67 291740,0 343657,6 3556 1150000 Anderslöv 280 354191,1 13379,99 327852,5 380529,6 15000 1475000 *Refers to the parish of Norra Skrävlinge

Most observations are in the lower price range and there are fewer observations in the higher range. This explains why the mean is relatively far from the highest observed values. The standard errors are larger for Billeberga and Anderslöv and I believe the reason for this to be that these parishes are more heterogeneous than Teckomatorp and therefore have a larger spread in the price variable. This is probably because of what is mentioned earlier in section 5.1 about the difference in size between towns and parishes.

property significantly is if it is located near water. There is data available for this but it turns out to be unnecessary to include in this case since none of the properties sold are close to water.

Omitted variables will create biased estimates, but this should affect the estimates in the two regressions to the same extent since they all include the same variables. If the omitted variable is expected to increase the value of the property this should be reflected in an upward biased estimate for the town-specific parameter since it captures everything that is not controlled for by the other variables in the model. In other words, the negative effect we see in Teckomatorp could be even greater than the result I get from this regression. On the other hand, the positive effect from Billeberga and Anderslöv would also be lower. Omitted variables may also be expected to decrease the property’s value. But once again, the relative difference between the three should not have changed unless houses in Teckomatorp for some reason would be different than elsewhere and this is what is reflected in the town-specific variable. This however seems rather unlikely.

The final price models that I will use are linear functions of the following variables:

yi = f

Since the spread in the dependent variable is quite large I present the results from a robust regression to reduce the effect from outliers. I also present the results from an Ordinary Least Squares regression (OLS) with robust standard errors as a reference. These results can be found in Appendix 2a and b. In the next section I will refer to the results from the robust regression unless mentioned otherwise.

7.1 Results

All parameters have the expected signs except DTILE and DROOF2. It was uncertain what sign could be expected for DSTOREY and in this model it is estimated that a house with more than one storey will have a lower value. When it comes to the age of the house I could also get more ambiguous results. Old houses can be highly valued because of their architecture or other “charming” characteristics. At the same time an old house can have deteriorated and

(di, ti, AAREA, AGE, LAREA, LSIZE, DBATH, DFASAD, DFIRE, DGAR,

therefore lost some of its value32. In this model the parameter for age is estimated to be negative. The dummy variables tell me how much the existence of, or lack of, a certain attribute affects the price of the property. For example, washing and drying equipment in the house increases the value by approximately 18000 SEK. The other variables give the marginal effects of a one unit change in the explanatory variable on the dependent variable. As an example, a one square meter increase in the lot size increases the property’s value by approximately 3,5 SEK. It does not seem to matter in what year a sale took place until we get to the period after 1986. Before this period the estimates for qj are highly insignificant

whereas they are positive and significant at the 1 percent significance level for the years 1987 to 1993, and hence of significance for explaining a property’s sales price during this time period. This leads me to believe that this could be explained by the recession in the early 1990’s that was preceded by a peak in the late 1980’s33. In line with this, we can see that there was a rise in both the number of sales and the sales price at the end of the 1980’s and after that a sudden drop in the years 1992 and 1993.

In the OLS regression approximately 65 percent of the variation in the dependent variable is explained by these models and the F statistic allows me to reject the hypothesis that none of the explanatory variables helps explain this variation. The residuals in both regressions are normally distributed.

When looking at the estimated γ1, γ2 and γ3 I can clearly see that there is a difference in these

parameters. These estimates are all significant at a 1 percent significance level and I can reject the hypothesis H0: γ1 = 0 in favour of the alternative hypothesis HA: γ1 < 0 at the 1 percent

significance level. On average, the price of a property in Teckomatorp is 46878 SEK lower than in Billeberga and Anderslöv. I can also reject the hypothesis H0: γ2 = γ3 = 0 in favour of

the alternative hypothesis HA: γ2 ≠ 0, γ3 ≠ 0 and conclude that there is a joint significance

between the two parameters at the 5 percent significance level. This indicates that the two variables together are relevant for explaining some of the variation in the dependent variable. Property prices are on average 55414 SEK higher in Anderslöv compared to Teckomatorp and for Billeberga the corresponding estimate is 39269 SEK. (In the OLS regression these estimates are all higher but still significant at the 1 percent significance level. The estimate for Teckomatorp is 53445, for Billeberga 42061 and for Anderslöv 64958). Even though the

32 _{Berger, T. (1998). p 7f}

parameters γ2 and γ3 differ in the way I expected, I cannot say for certain that property prices on average are lower in Billeberga than in Anderslöv since the confidence intervals overlap, and as mentioned earlier in section 5.2 the means are not significantly different from each other.

Assuming that all heterogeneity is controlled for, the negative effect of 46878 SEK in Teckomatorp represents my estimate for how people value the risk of living close to this hazardous waste site. For some reason people pay this much less for a property in Teckomatorp and I think that this to some extent could be because of the danger associated with living there. To put this in other words, people are willing to accept a monetary compensation equal to this amount in order to live in this area. If they perceived the risk to be greater than this it should result in a higher estimate, and vice versa.

8. Conclusions

There are of course many possible explanations for the results I get from these regressions, but it is tempting, and also likely, to think that some of the difference is because of the BT Kemi scandal. What gives the strongest support for the hypothesis that there is a negative effect on property prices in Teckomatorp is that there is a clear difference between Teckomatorp and Billeberga. These towns are both in the same municipality and most macro variables should therefore be relatively similar. There is a greater difference in population between the town and parish of Billeberga than in Teckomatorp but this would to me rather indicate that property prices could be expected to be even lower in Billeberga because of a smaller town and more houses in the country side. On the other hand, Billeberga is located slightly closer to Landskrona and this could perhaps outweigh some of this effect. The small difference that is seen between Billeberga and Anderslöv could be explained by the fact that the income level in Trelleborg’s municipality is likely to have been higher than for towns in the Svalöv municipality.

the effects on property prices are usually not seen until after the site has been announced as hazardous.

Appendix 1

Variable Description

y sales price of property

d1 1=Parish of Norra Skrävlinge, 0=Parish of Billeberga and Anderslöv

d2 1=Parish of Billeberga, 0=Parish of Norra Skrävlinge and Anderslöv

d3 1=Parish of Anderslöv, 0=Parish of Norra Skrävlinge and Billeberga q82 1=year of sale 1982, 0=other

q83 1=year of sale 1983, 0=other q84 1=year of sale 1984, 0=other q85 1=year of sale 1985, 0=other q86 1=year of sale 1986, 0=other q87 1=year of sale 1987, 0=other q88 1=year of sale 1988, 0=other q89 1=year of sale 1989, 0=other q90 1=year of sale 1990, 0=other q91 1=year of sale 1991, 0=other q92 1=year of sale 1992, 0=other q93 1=year of sale 1993, 0=other AAREA 1=additional area, measured in km2 AGE age of house at time of sale LAREA living/tenement area, m2

LSIZE lot size, m2

DBATH 1=1 or more bathtubs/showers exists in dwelling, 0=does not exist DFASAD 1=brick, 0=other

DFIRE 1=fire place exists, 0=does not exist DGAR 0=garage exists, 1=does not exist

DISOGL 1=isolation glass on at least 50% of the window surface, 0=no DREC 1=recreation room (gillestuga) exists 0=does not exist

DROOF 1=roof covering/coating made of glazed brick, copper or shale, 0=other DSAUNA 1=sauna exists, 0=does not exist

DSTOREY 1=house has more than one storey, 0=no DTILE 1=tile in bathroom, 0=no

DTOWN 0=house is located in town/village, 1=no

Appendix 2a

Results from regression 1:

Robust OLS

Variable Coefficient Std. err. t-value Variable Coefficient std. err. t-valueRobust

d1** -46878.46 7694.568 -6.09 d1** -53445.09 13674.97 -3.91 q82 8840.758 17454.38 0.51 q82 -18015.65 18491.2 -0.97 q83 4723.636 17639.07 0.27 q83 -1942.046 16846.61 -0.12 q84 23178.92 16652.44 1.39 q84 13168.84 16655.6 0.79 q85 21802.52 16267.5 1.34 q85 11492.1 15105.98 0.76 q86 23674.61 16480.88 1.44 q86 8754.952 15568.66 0.56 q87** 53727.93 15691.85 3.42 q87** 45414.69 16026.07 2.83 q88** 98300.45 16026.15 6.13 q88** 96643.7 16139.03 5.99 q89** 150819.2 15968.43 9.44 q89** 194098.6 22642.6 8.57 q90** 262723 17074.98 15.39 q90** 266963.3 20443.61 13.06 q91** 326048.7 17424.66 18.71 q91** 372681.3 48999.12 7.61 q92** 259574.6 22274.36 11.65 q92** 322542.5 44213.81 7.30 q93** 230645.6 23802.86 9.69 q93** 260291.9 53295.49 4.88 AAREA** 220.8752 86.58352 2.55 AAREA* 296.5773 131.5756 2.25 AGE** -2891.21 212.9586 -13.58 AGE** -3093.652 407.7146 -7.59 LAREA** 997.6186 77.59271 12.86 LAREA** 1201.24 166.1001 7.23 LSIZE* 3.794284 1.76231 2.15 LSIZE* 6.016786 2.638459 2.28 DBATH* -22457.72 9909.298 -2.27 DBATH -16900.86 13635 -1.24 DFASAD* 15193.16 7109.49 2.14 DFASAD* 18616.32 9336.748 1.99 DFIRE 3263.415 9286.957 0.35 DFIRE 21638.04 16654.45 1.30 DGAR -12319.18 7091.41 -1.74 DGAR -1124.788 15135.42 -0.07 DISOGL** 57191.27 10710.21 5.34 DISOGL** 68619.23 25314.24 2.71 DREC** 72366.92 17131.03 4.22 DREC* 43605.58 22421.84 1.94 DROOF -17741.02 21700.87 -0.82 DROOF -23678.7 28643.31 -0.83 DSAUNA 20374.97 13308.77 1.53 DSAUNA 42254.74 25647.33 1.65 DSTOREY -14162.2 17143.21 -0.83 DSTOREY -28741.95 21959.02 -1.31 DTILE -10711.42 9050.375 -1.18 DTILE -16582.62 11393.66 -1.46 DTOWN 14507.52 7885.958 1.84 DTOWN 15579.78 11767.9 1.32 DWASH* -18555.32 7930.766 -2.34 DWASH** -39266.1 9991.467 -3.93 α** 189527.3 17989.7 10.54 α** 177031.6 27742.66 6.38 R2 _0.6466

Appendix 2b

Results from regression 2:

Robust OLS

Variable Coefficient Std. err. t-value Variable Coefficient Robust std. err. t-value d2** 39269.43 8549.911 4.59 d2** 42060.9 13402.79 3.14 d3** 55414.09 8568.916 6.47 d3** 64957.72 15881.69 4.09 q82 9307.439 17395.48 0.54 q82 -19423.4 18722.57 -1.04 q83 4563.473 17574.14 0.26 q83 -2495.868 17027.54 -0.15 q84 22381.78 16602.36 1.35 q84 11260.85 16895.37 0.67 q85 21070.54 16219.58 1.30 q85 9551.439 15160.97 0.63 q86 22937.09 16434.01 1.40 q86 6665.77 15688.94 0.42 q87** 53229.29 15650.78 3.40 q87** 43186.8 16241.01 2.66 q88** 97952.87 15969.73 6.13 q88** 95646.05 16266.5 5.88 q89** 149165.9 15919.01 9.37 q89** 192382.6 22964.41 8.38 q90** 262323.1 17024.17 15.41 q90** 264965.4 20288.87 13.06 q91** 321552.3 17397.71 18.48 q91** 369219.4 49195.71 7.51 q92** 256042.9 22207.84 11.53 q92** 319954.3 43423.71 7.37 q93** 228940.5 23772.85 9.63 q93** 255262.8 52329.31 4.88 AAREA** 213.0969 86.2652 2.47 AAREA* 299.4063 129.5824 2.31 AGE** -2892.636 212.1911 -13.63 AGE** -3083.284 405.2364 -7.61 LAREA** 994.8889 77.32732 12.87 LAREA** 1195.391 165.477 7.22 LSIZE 3.29263 1.759259 1.87 LSIZE* 5.681734 2.656894 2.14 DBATH* -22495.58 9874.041 -2.28 DBATH -17461.61 13561.45 -1.29 DFASAD* 16865.42 7177.122 2.35 DFASAD* 22100.92 9592.509 2.30 DFIRE 1508.172 9271.614 0.16 DFIRE 19837.91 16525.19 1.20 DGAR -12549.31 7068.215 -1.78 DGAR -1773.761 15207.41 -0.12 DISOGL** 55277.51 10676 5.18 DISOGL** 67560.74 25386.58 2.66 DREC** 71627.06 17067.4 4.20 DREC* 43269.6 22175.53 1.95 DROOF -15945.17 21621.56 4.20 DROOF -22850.09 29103.42 -0.79 DSAUNA 23345.63 13310.76 1.75 DSAUNA 45779.1 25837.82 1.77 DSTOREY -15563.58 17082.15 -0.91 DSTOREY -29701.48 21867.37 -1.36 DTILE -10879.57 9018.856 -1.21 DTILE -15971.51 11394.39 -1.40 DTOWN 14544.62 7870.984 1.85 DTOWN 17014.93 11852.26 1.44 DWASH* -17463.02 7905.125 -2.21 DWASH** -38511.38 9985.973 -3.86 α** 143379.2 18604.41 7.71 α** 123355.1 27498.93 4.49 R2 0.6484

References

Printed

Angrist J. D. and Krueger A. B. (1999). Empirical Strategies in Labour Economics, in Ashenfelter O. and Carl D. (eds), Handbook of Labor Economics, vol 3.

Berger, T. (1998). Priser på egenskaper hos småhus, Working Paper No. 14. Institute for Housing Research. Uppsala University.

Bevmo, L. and Englöv, P. (2004). BT Kemi, Huvudstudie, version 3, SWECO VIAK Malmö Clymer, A. (1989). Polls show contrasts in how public and E.P.A. view environment, New York Times, May 22: Section B, Column 1, p 7.

Dagens nyheter. Article: Nu ska byken tvättas, 14th of August 2005, supplement ”Kultur”. Eriksson, B. J. (2000). Bostaden som vara – egenskaper och metodproblem, in Lindh, T. (ed), Prisbildning och värdering av fastigheter, p 145-174, Research Report 2000:4, Institute for Housing and Urban research, Uppsala University.

Ihlanfeldt, K. R. and Taylor, L. O. (2001). Externality Effects of Small-scale Hazardous

Waste Sites: Evidence from Urban Commercial Property Markets, Environmental Working

Paper Series #2001-002.

Kiel, K. and Zabel, J. (2001). Estimating the Economic Benefits of Cleaning Up Superfund

Sites: The Case of Woburn, Massachusetts, Journal of Real Estate Finance and Economics,

22:2/3, p 163-184.

Kohlhase, J. E. (1991). The Impact of Toxic Waste Sites on Housing Values, Journal of Urban Economics, 30:1, p 1-26.

Mitchell, C. R. and Carson, R. T. (1993). Using Surveys to Value Public Goods – The

Contingent Valuation Method, Resources for the Future. Washington, D.C.

Navrud, S. (2002). The State-of-the-Art on Economic Valuation on Noise, Department of Economics and Social Sciences, Agricultural University of Norway.

Rosen, S. (1974). Hedonic Prices and Implicit Markets: Product Differentiation in Pure

Competition, Journal of Political Economy 82, p 34-55.

Vainio, M. (1995). Traffic Noise and Air Pollution – Valuation of Externalities with Hedonic

Price and Contingent Valuation Methods, Helsinki School of Economics and Business

Administration.

Electronic

http://www.all2know.com/sv/wikipedia/b/bt/bt_kemi.html Accessed 2005-10-26

Statistics Sweden http://www.scb.se

Sveriges arkitekter. Article: “Gammal giftmark blir attraktion”

http://www.arkitekt.se/s16623 Accessed 2005-10-26

Arbetsmarknadsstyrelsen

http://www.ams.se

Other