About the importance of planning the location of recycling stations in the urban context

: Recycling is essential to the circular economy and reduces our consumption's impact on the environment. Creating conditions for recycling in new residential areas is relatively easy, but finding good recycling opportunities in existing residential areas is more complicated. The recycling of newspapers, plastic and glass must be relatively close to where people live; at the same time, the locations must be relatively discreet and not disturb the residents in the area. The purpose of the article is to analyse the effect of small and local recycling stations (RCS) on the attractiveness of residential areas. This has been made possible by analysing housing values for almost 200,000 housing units near 250 RCS in Stockholm, Sweden. Using an identification strategy that relies on postal code fixed effects, we find evidence that the proximity to RCS affect housing prices on average in both owner-occupied single-family houses and cooperative owner-occupied apartments (condominiums). The results indicate that proximity to the RCS is negatively capitalised in housing values, which indicates that the city should consider this in its planning.


Introduction
If net-zero emission by 2045 is to be achieved, more raw materials will be need to be recycled more efficiently.There has been producer responsibility for recycling packaging in Sweden since the early 1990s, and the climate benefits of recycling are significant.Calculations indicate, for example, that for every kilogram of recycled plastic, carbon dioxide emissions are reduced by up to two kilograms (Eriksson, 2012).Even when considering the resources used for collection, transport, and sorting, the reduction in CO 2 is still nearly 1.5 kilograms.According to the Swedish Environmental Protection Agency, Sweden must have no net emissions of greenhouse gases into the atmosphere by 2045 to have negative emissions.As emissions per person amount to nine tonnes per year, recycling is only one measure among many needed to reduce the total climate impact.However, material recycling is an integral part of the overall transformation.
A circular economy with efficient recycling permeates all consumption and production; such a society will achieve high sustainability goals.Recycling is of utmost importance for the circular economy through increased recycling in production and consumption (see Gallego-Schmid et al., 2020, Joensuu et al., 2020, Mansuy et al., 2020).Recycling materials such as glass, paper, and plastic are essential for a more circular economy and crucial for preventing material from being deposited or thrown away in nature.For example, recycling plastic is vital to reducing microplastics in nature and water (Prata et al., 2019).
To increase recycling in households, recycling stations (RCS) must be accessible.Recycling opportunities must be close to households, and as long as it is not possible in existing residential areas to cost-effectively create the opportunity to recycle in the property, RCS must be placed locally in the residential areas (Chang andWei, 1999, Purkayastha et al., 2015).According to Sidique et al. (2010), recycling behaviour is affected by the distance to RCS.The closer the distance is between the recycling station and the dwelling, the more frequently the recycling station will be used.
Increasing the number of RCS and placing them close to residential areas has already occurred in Sweden.There are approximately 5,000 RCS in Sweden and as many as 250 RCS in a city of a million inhabitants, such as Stockholm.The location of the stations is vital for increasing the recycling rate of households, but they are also potentially an externality in the urban environment.RCS entail traffic, noise, dirt and a potential health risk.The location is thus essential not only based on the recycling volume but also the attractiveness of the residential areas.However, few studies have been conducted on the negative externality of RCS.
Several studies analyse the effect of waste disposal by incineration, such as a recent article by Zhao et al. (2016) and an older study by Kiel and McClain (1995).However, those facilities are much larger and expected to have a significant negative capitalisation on housing values.For example, Zhao et al. (2016) found that housing prices fell by as much as 25 percent near an incineration plant, and Kiel and McClain (1995) found that this capitalisation starts even before the plant is in place and that the effect is persistent for several years after the plant is in operation.Eshet et al. (2007) analyse the capitalisation effect of living near waste transfer stations using the hedonic methodology.Their explanation for why we can see this negative externality is that the RCS bring disamenities such as noise, odour, litter, vermin, visual intrusion and perceived discomfort.The effect is apparent, and the closer the dwelling is to the RCS, the more negatively it affects the housing value.The size of these waste plants is significantly more extensive and has a much greater impact on the surrounding environment than the RCS analysed here.
This study aims to contribute knowledge about RCS' impact on the attractiveness of residential areas.
Recycling is important and will become increasingly important in the future to create a sustainable society; nevertheless, the RCS, as they are designed and located today, may have local costs that could and should be avoided to gain greater acceptance and higher welfare gains.
The location of RCS is important for creating opportunities for the efficient collection and recycling of packaging materials.Local installation of RCS must be cost-effective for emptying and cleaning, both in number and location.At the same time, the RCS must create as little damage as possible to residential attractiveness.It is essential for housing satisfaction, yet to our knowledge no studies have analysed the possible effect of RCS location on housing attractiveness.A significant contribution is thus that we produce a basis for planning where these RCS should be located to minimise the socioeconomic cost.
An important issue in these analyses of capitalisation effects is the question of endogeneity.Have RCS been located in areas with low residential attractiveness, or has the location of the RCS created areas with lower residential attractiveness?One contribution to the analysis is how we have addressed this problem by using a few excluded yet relevant explanatory variables, namely fixed effects, a treatment effect model, a propensity score method and a micro-analysis of the immediate area around each recycling station.
Section 2 will present the theoretical model and methodological approach.Decisive for interpreting the relationship between housing prices and RCS is that we do not have problems with endogeneity, which will permeate the chosen method.Section 3 presents the selected case study focusing on locating RCS in Stockholm.This is followed by section 4, where data is presented and described.The empirical economic analysis is presented next in section 5, and the article concludes with a conclusion and policyrelevant questions in section 6.

The Theoretical and Methodological Framework
The theoretical starting point is Rosen (1974), where the housing price is a function of its attributes.In the first step, a hedonic price equation is estimated where estimated parameters can be interpreted as marginal willingness to pay for each attribute.The estimated parameters can thus be interpreted as implicit prices (hedonic prices).The attributes consist of the characteristics of the apartment or property and the residential area.The methodology is often used to estimate implicit prices for different types of negative externalities, such as traffic noise (Wilhelmsson, 2000), or positive externalities, such as shopping malls (Long and Wilhelmsson, 2020).Here we will test the hypothesis that proximity to RCS negatively capitalises on housing prices by including the distance to the recycling station in the hedonic price equation and, as an alternative, a binary variable that indicates whether the dwelling is within a specific range of the recycling station.The hedonic price equation that we will estimate looks like equation 1, where HP is equal to house prices (all models are estimated with a price as a natural logarithm based on a Box-Cox transformation), and the matrix X represents all value-affecting attributes such as size, age, and location.The variable RCS represents proximity to a recycling station.We used proximity to an RCS as a binary variable, or the shortest distance to an RCS, in the empirical analysis.We hypothesise that 2 is negative, and the vector T is a binary variable measuring the month the dwelling was sold What can we then expect when it comes to the capitalising of proximity to RCS on housing prices?A large part of the negative impact depends on where and in what context the RCS are located.If they are located on a minor street close to parks and green areas, the recycling station may be considered polluting the environment, but if it is located on a major road with much traffic and near a gas station or adjacent to a shopping centre, it can be expected that the effect is significantly less or negligible.The management of the RCS is also essential.How often they are emptied, and if it is often messy, has an expected adverse effect.For example, Mattsson Petersen and Berg (2004) asked some individuals about the importance of managing their RCS.The majority stated that they thought the care was good or better, but the variation was large between them.At one station, as many as 48 percent stated that the cleanness of the area around the recycle stations was bad or very bad.
Another reason to consider may be how much traffic is generated to and from RCS.It may also be reasonable to expect that the capitalisation effect regarding RCS will vary depending on whether it is near single-family houses compared to condominiums.Mattsson Petersen and Berg's (2004) survey results indicate that visitors to the RCS are on their way to other activities, i.e., the visit is not the trip's primary purpose.Furthermore, they observed that 90 percent of the visits took place by car, which indicates that the location of the RCS is not only important in terms of logistical and cost-effectiveness but that the location itself generates traffic that can be disruptive.We know that car traffic has a negative impact on housing prices (e.g., Wilhelmsson, 2000), but the increased traffic generated by the RCS can be marginal, depending on the existing traffic volume.Minor streets in single-family housing areas can significantly increase traffic volume due to the recycling station, while if the street is already a major road, the traffic from and to the recycling station is relatively insignificant.
Underlying factors that positively affect price are the size of the dwelling, measured as the total square meters of living space, and the number of rooms.We analyse both the owner-occupied condominium housing market and the single-family housing market.In the condominium case, the fee to the tenantowner association has a negative price effect.There is also an expected price premium for houses closer to the CBD (Voith, 1993;Herath and Jayasekare, 2021), and the same applies to the proximity to public transportation (Herath and Jayasekare, 2021;Cordera et al., 2019) and shopping centres (Long and Wilhelmsson, 2020).
The independent attributes must be exogenously given to interpret estimated parameters as implicit prices and thus marginal willingness to pay.In the presence of endogeneity, estimated relationships are just relationships, not causal ones.It is usually no problem to assume that they are simply exogenously given for all apartment and property attributes, but for many residential area attributes it is more difficult.This is perhaps the case with the attribute of primary interest in this study, namely proximity to RCS.
There are several reasons why the attribute might be endogenous, for instance, reverse causality, omitted variables, and measurement errors (Hill et al., 2021;Sande and Gosh, 2018;Bascle, 2008).

(a) Reverse causality
One reason for endogeneity is that RCS are located in low-priced locations rather than in the surrounding areas.For example, RCS may be located near major roads, petrol stations or similar places.The estimated relationship between housing prices and proximity to RCS will not, in those cases, be causal, and it can even denote reverse causality.
We have used a methodology that assigns certain transactions as treatment and compares these with other transactions that we assign to the control group, a quasi-experimental design.It is a treatment effect model similar to Heckman's treatment effect model without instrument variables (see, e.g., Bascle, 2008;Hill et al., 2021), but the group is not randomised.Hence, there is undoubtedly a treatment selection bias.We have tried to mitigate this effect by using a propensity score methodology (see Rosenbaum and Rubin, 1983), where the observations in the treatment area are as similar as possible to those in the control group in all respects other than proximity to the RCS.The methodology has been used in previous analyses such as D'Elia et al. (2020) and Bilbao and Valdés (2016).The optimal way to handle the problem would have been to use the difference-in-difference, instrument variable or regression discontinuity design methodology.Unfortunately, we do not have data on transactions before the current location of the RCS, which makes these methods impossible to utilise.
The treatment effect model is a two-step model where in the first step we define treatment and control groups and calculate the probability that the observation is included in the treatment group.In the second step, a weighted least square model will be estimated where the probability is the inverse of the weights.
In this way, we analyse whether the observations in the treatment group are as similar as possible to the observations in the control group.The method is suggested by Hirano et al. (2003) and used in, e.g., D'Elia et al. (2020).The propensity score equation looks like equation 2, where the propensity score (PS) is the probability of treatment (T) given the covariates Z, with 0≤P(Z)≤1, and the weighted hedonic least square model looks like equation 3, where PS is the estimated propensity score.The higher the probability that the dwelling is similar to the properties close to RCS, the greater the observation's weight in the estimate.
Moreover, we have visually inspected all RCS to ensure no justification for reverse causality between housing prices and proximity to RCS.The inspection has also allowed us to classify the RCS based on characteristics in the geographical location.This, too, has been done to minimise the potential endogeneity problem.All RCS locations have been classified as good or bad locations.The hedonic price equation that has been estimated looks like equation 4: We hypothesise that 2 and 3 are negative and that |2|>|3|.

(b) Omitted variables
The empirical analysis will not use panel or pure cross-sectional data but instead pooled cross-sectional data.We have implemented fixed effects for time and fixed effects for residential areas, and we have done this by using postal code information.Our goal is to reduce the problem of omitted variables and thus the endogeneity problem by including fixed effects.They will be effective if we assume that the spatial effect is constant within the group or invariant over time (Hill et al., 2021).There are a large number of hedonic studies that address spatial heterogeneity by including fixed effects, such as Heintzelman and Tuttle (2012), Gibbons et al. (2014) and Czembrowski and Kronenberg (2016); nevertheless, as Helbich et al. (2014) point out, a large number of included fixed effects can result in few degrees of freedom, which impairs the model's accuracy.
Hence, we use spatial fixed effects to control spatial dependency and omitted variables.However, we have also included fixed property effects in the analysis of condominiums, as condominiums in the same properties have the exact same coordinates and information about repeated sales in the single-family housing data.This is another possibility to minimise the problem of omitted variables and thus the potential endogeneity problem.The inclusion will also mitigate the effect of spatial dependency to some degree, as the fixed property effect aims to check for heterogeneity at the property level.The methodology will be effective if properties at the property level are constant over time, although this may be too heroic an assumption.Another weakness is that the inclusion of the fixed property effect dramatically reduces the degree of freedom and makes the estimation significantly more computationally challenging.If the number of analysed observations is large, the problem of many fixed effects is not severe.

(c) Measurement error
Of course, it can also be the case that we have an endogeneity problem due to measurement errors in the variables examined and then mainly in the RCS variable.In order to eliminate problems with measurement errors regarding RCS, each location, according to the FTI's register, has been checked in Google Map and Google Street View to coordinate correctly the location of the RCS.Street addresses tend to place the property's coordinates some distance from the roadside where the RCS are often located.Thus, we have minimised the endogeneity problem caused by measurement errors in the location of the RCS.

(d) Robustness test
As a robustness test, we have also (1) changed the assumptions about the treatment and control area, respectively, and (2) randomised where the RCS are located.We have done the latter by "moving" RCS 0-500 meters from their original location.After that, the hedonic price equation was been estimated again, with the assumption that RCS should not have a negative capitalisation on housing values.The test is similar to the placebo test commonly used in the regression discontinuity design methodology (see, e.g. de la Cuesta and Kosuke, 2016; Chen et al., 2019) and Bertrand et al.'s (2004) difference-indifference context.
In summary, we have tried to control for endogeneity by including fixed area variables.In this case, these included postal codes, visual inspection of the vicinity around the RCS, and any additional area attributes.In a sub-analysis, we used property fixed effects after restricting the area to a smaller innercity location where RCS are more commonly found on smaller streets in residential areas.

Recycling in Sweden
In Sweden, there are more than 5,000 RCS.In 2020, the collection result per inhabitant in Sweden was just over 22 kilograms of glass, 17 kilograms of packaging paper (cardboard), almost 9 kilograms of plastic, just under 2 kilograms of metal and 14 kilograms of newspapers.An essential prerequisite for this accomplishment is that there are RCS in the built environment that are accessible and make recycling easy.We have had a far-reaching producer responsibility since 1994 regarding the collection and reuse of packaging materials, and the RCS that we analyse are an essential part of this responsibility.
Recycling behaviour across municipalities in Sweden varies.Hage and Söderholm's (2007) results indicate that the variation between municipalities regarding collection can be explained by differences in demographic and socioeconomic factors together with environmental preferences, geographical differences, and local policies.Mattsson Petersen and Berg ( 2004) examined RCS in a small town in Sweden at the end of the last century.The purpose was to create a basis for planning future locations of RCS by examining attitudes toward collection and how much was collected at each station, i.e., the volume of recycled material.Both are important questions to determine where and how many RCS the city should plan for.
There is an ongoing discussion in 2021/22 about what the system of RCS should look like in the future.
A proposal discusses how recycling can be made closer to the property by collecting it in the property or its vicinity.The proposal is out for consultation, and the government is expected to decide about the future system in June 2022.

Recycling in Stockholm
Our case study is the city of Stockholm, the capital of Sweden, which as of 2020 has nearly 1 million inhabitants.Stockholm is divided into 13 different districts.By population, Södermalm is the largest district, followed by Hägersten-Älvsjö and Enskede-Årsta-Vantör.In the year 2020, Spånga-Tensta and Skärholmen were the smallest districts, with around one-third of the population of Södermalm.should be emptied.The ownership of FTI consists of four material companies and was formed in connection with the government's decision on producer responsibility for packaging in 1994.Table 1 shows the number of RCS in different parts of Stockholm per 100,000 inhabitants and 1,000-hectares of land.The number of RCS varies between the different districts in Stockholm, from only 6 to as many as 35.
Of course, this is mainly due to the number of residents in the district.More densely populated neighbourhoods also have more RCS, but that is not the whole explanation.The number of RCS per 100,000 residents in the district varies from just under 11 to as many as 33 RCS per 100,000 inhabitants, and on average there are nearly 26 RCS per 100,000 inhabitants in Stockholm.
If we instead analyse the number of RCS per 1000 hectares of land area, we can observe that the spread is significantly greater between the districts.The inner-city districts of Södermalm, Norrmalm and Kungsholmen all have a significantly higher RCS density than those in the suburbs.The only exception is the inner-city district of Östermalm, which has relatively few RCS, both measured as a proportion of the population and measured as land area density.The difference between Södermalm and Östermalm is surprisingly significant.
Through FDI, we have received a list of addresses where the 250 RCS in the municipality of Stockholm are located.All surveyed RCS are small and aim for recycling in the local residential area, and the space on the street or sidewalk is about 10 x 3 meters.These addresses have been coordinated via Google Maps, and each location has been visually inspected through Google Street View.The pictures show the design of the recycling bins.All RCS are located by a road to allow for the emptying of the containers.The majority of RCS in the inner city are located on streets in residential areas, while outside the inner city, many stations are located on the exit road from residential areas.This means that many are located with other urban area disamenities such as roads, petrol stations and subway stations.However, several are also located near amenities such as parks and playgrounds.In Table 2, all RCS are categorised depending on where they are located in terms of micro-location.
The majority of RCS is located by a road to make it easier for residents to recycle and facilitate the emptying and maintenance of the facilities.We have classified the streets into three groups: A minor road means a smaller street with traffic that mainly consists of the area's residents; a road is a street with some pass-through traffic; and major roads are those with significant traffic volumes.RCS are primarily located on minor roads, but several are located in car parks in residential areas.Relatively few are colocated with subway stations or local shops in the area.Only seven of 250 RECS are found at or near petrol stations.It can also be noted that significantly more are located at parks and other green areas with playgrounds or directly adjacent to water.More RCS are found in residential areas with multi-family houses than those with single-family houses.The conclusion that can be drawn is that RCS are, above all, co-located with amenities rather than with disamenities.
Recycling stations and micro-location.

Micro-location Number
Minor road 167

Major road 22
Patrol station 7

In a multi-family housing area 143
In a single-family housing area 50 Office, light industry 23 Note.Each recycling station can be located in multiple micro-locations.

Importance of RCS location
Hage and Söderholm's ( 2007) results indicate that proximity to RCS, population density, and the proportion of residents in urban areas have a small economic and statistical impact.Moreover, one recently published article by Li et al. (2020) shows that distance is not as crucial as might be expected.
In an experiment in Shanghai, households had to register their interest in a recycling programme, and they did not find that the distance to the nearest recycling station had any effect on the households that chose to sign up for the programme.These results contradict Chang and Wei (1999), Purkayastha et al. (2015), and Rousta et al. (2015).Moreover, Hage et al. (2018) show that increased density (RCS per capita) increases the degree of recycling in the municipality.
Hence, Chang and Wei (1999), Purkayastha et al. (2015), Rousta et al. (2015), and Hage et al. (2018) all show that the distance to RCS is essential and that reduced distance significantly increases the sorting of recycled materials.Proximity to RCS is therefore crucial in city planning in both new and existing residential areas.
Moreover, there is a trade-off between how much space the city wants to cover, or population they want to reach, and the collection cost.Cubillos and Wøhlk (2021) present a mathematical model where the problem is to increase simultaneously the degree of recycling, while we, through a strategic location of RCS, minimise installation and collection costs.To this problem, the effect on housing attractiveness could be added.Li et al. (2020) analyse the optimal number and distribution of RCS, and they also do not consider the negative externality of the stations in the form of deteriorating residential area attractiveness.They only optimise recycled material as a function of population density and consumption to transport costs.

Data and Descriptive statistics
Data regarding housing transactions come from Svensk Mäklarstatistik.A high proportion of brokers report contract data to them, who then compile and publish statistics on price development in Sweden and different geographical areas.The transactions we have access to for this research project are raw data included in all sales, and we also have data regarding owner-occupied cooperative apartments and single-family houses for 2005-2019.In total, we have access to over 200,000 transactions, and the majority of these are condominiums sales.
Available data regarding condominiums are transaction price in SEK, contract date, the living area measured in square meters, number of rooms, monthly fee to the housing association in SEK, year of construction, and floor plan.Apart from the variables monthly fee and floor plan, available data for single-family houses are the same, with the addition of information about plot area in square meters.For all transactions, we also have access to longitude and latitude.We have then calculated distances in kilometres to the Central Business District (CBD), nearest metro station and seventeen largest shopping centres in the Stockholm area (the same as in Long and Wilhelmsson 2020).The descriptive statistics are exhibited in Table 3.
In the analysis regarding condominiums, we will use approximately 165,000 transactions.
Approximately 25,000 transactions are missing information about the apartment floor level, so fewer observations are used.Potential outliers have been excluded (see information in the note in Table 3).
The number of transactions regarding single-family houses amounts to just over 15,000.The average price for single-family houses is almost SEK 2 million higher, with an average price of approximately SEK 5 million.However, the standard deviation is high both in the case of condominiums and singlefamily houses.Nevertheless, the variation concerning the price is higher for condominiums, as there is a more significant variation in price ranges for condominiums than for single-family houses.The higher price is reflected in the fact that the size of single-family houses is almost twice as large as condominiums.Condominiums exist for natural reasons in the inner city, and thus the distance to the CBD is significantly higher for single-family houses than for condominiums.The same applies to proximity to metro stations and shopping centres.It is also clear that the distance to the nearest recycling station is significantly shorter for tenant-owner apartment transactions than for single-family housing transactions.For condominiums, the average distance to a recycling station is about 300 meters compared to 500 meters for single-family houses.However, the variation is significantly greater for single-family houses.Proving the capitalisation effect on housing values from RCS exclusively will be challenging.Each recycling station has been related to housing transactions, and the shortest distance (measured as the bird flies in meters) to a recycling station has been registered for each transaction.This distance to RCS (Dist RCS) will be one of the area's attributes included in the hedonic price equation.The average distance equals 290 meters for condominiums and 490 meters for single-family houses.The variation around the average distance is considerable.We have also included a binary variable (Binary RCS) indicating whether the house is within 2 kilometres of the RCS.Nearly 40 percent of the condominiums are within 2 kilometres, but only 15 percent of the single-family houses.
We have included fixed effects to minimise the risk of omitted variable bias.All models include fixed effects for time and postcode areas.The postcode areas are relatively small, and thus they are manyjust over 700.We will also use the information concerning transactions with the same location, i.e., on the same property.In some cases, it is the same apartment, but in others, it involves apartments on the same property.Just under 50 percent of condominium transactions are transactions on the same property.
Most of these relate to 2-4 transactions per property, but there are some with significantly more transactions.In fact, one property reports just over 300 transactions.It is a sizeable tenant-owner association, and the sales cover an extended period, so the number is not unreasonable.However, 95 percent of sales regard properties with 0-10 transactions.When it comes to single-family houses, we know that sales of the same property are repeated sales.Significantly fewer have been sold more than once -as many as 78 percent have only been sold once, and about 22 percent have been sold 2-4 times.
With the help of this information, fixed property effects have been created and included in the hedonic price equation.When it comes to the condominium apartment market, it has not been possible to include 100,000 fixed property effects; we have, therefore, only estimated one model regarding the inner city.

Empirical Analysis
The estimation of the hedonic price equation has been made in four different steps.In the first step, we have estimated default models regarding condominiums and single-family houses.The models include residential and area attributes previously presented.In addition to these, fixed time and area effects are also included, and the latter refers to postcode areas.Proximity to RCS is included partly as a continuous variable that measures the distance between the recycling station and the apartment, and partly as a binary variable in a simple treatment model where the variable is equal to 1 within 200 meters and the control area consists of 200 to 500 meters from the recycling We have used the propensity score method to control for nonrandomness.Results are shown in Table 4.
In step two, we tested the hypothesis that the implicit price for a recycling station is affected by microlocation.The RCS are classified into two groups based on the micro-location characteristics.The first refers to locations with disamenities (such as major roads or petrol stations) in their vicinity, and the second refers to whether they are located where there are amenities (such as parks).The results are shown in Table 5.In step three, we test if the estimates are affected by including fixed property effects, and those results are exhibited in Table 6.Finally, in step four, we are robust in testing our assumption about treatment and control areas, as well as performing a placebo test concerning the location of RCS.
The results for this are presented in Tables A1 and A2 in the Appendix.

(a) Default models
The estimated default models are presented in Table 4.The owner-occupied condominium market is analysed in the first two columns, and in the last two columns, the results from the single-family housing market are presented.Proximity to RCS has been estimated in two different ways, namely (1) as a continuous variable (Dist RCS) and (2) as a binary treatment variable (Binary RCS).
(  4 shows the ordinary least square estimates (OLS) concerning models 1 and 3 and weighted least square estimates (WLS) concerning models 2 and 4. The weights are based on the propensity score estimates belonging to the treatment group.Models 1 and 2 are the condominium apartment market, while models 3 and 4 address the single-family housing market.All models include fixed postal code effects and fixed monthly effects.Only observations within 500 meters from the RCS are included in the estimations.The treatment (RCS) group are observations within 200 meters of RCS, and the control group observations within 200 to 500 meters.t statistics in parentheses, * p < 0.05, ** p < 0.01, *** p < 0.001.
All models include fixed postal codes and monthly effects, and the degree of explanation is generally high.The explanation rate in the condominium models is approximately 95 percent, while the explanation rate is approximately 87 percent for the single-family housing market.The degree of explanation is slightly higher in the treatment effect models, where the proximity to RCS is a binary variable.We have weighted the model based on the probability that the transaction has a treatment.All estimates have an expected effect on prices and reasonable magnitude.Increased living area in square meters or number of rooms raises prices.Older houses have a lower expected price, and for condominiums, a higher monthly fee to the housing association has a negative impact on the price.
Proximity to the metro station positively impacts apartment prices but has a negative impact on singlefamily housing prices.Proximity to the CBD positively affects prices, while proximity to the shopping mall does not.
The variable of interest here is proximity to RCS.All estimates are statistically significant at a 5% significance level with t-values of around 10 (absolute values) in the condominium but slightly lower in the single-family housing model (around 2-3 in final values).Proximity also has an expected effect to the extent that it can be regarded as a disamenity.The closer to the RCS the dwelling is located, the lower the price, everything else being equal.In the binary models, the effect amounts to approximately 1.3 percent of the housing values, which can also be regarded as economically significant.
Measured in SEK, the capitalisation amounts to approximately 40,000 for condominiums and 70,000 for single-family houses (around 3,800 and 6,800 EUR).Our results are in line with Eshet et al. (2007) but significantly lower than, e.g., Zhao et al. (2016) and Kiel and McClain (1995).

(b) Micro-location models
To test the significance of the micro-location, we have included two interaction variables in the model where we integrate RCS with amenities and disamenities.Amenities mean that RCS are located close to a park or other green area, while disamenities mean that RCS are co-located near, for example, major roads, petrol stations and retail trade.The results from the estimates are shown in Table 5.
Results indicate that the co-location of RCS with amenities and disamenities has some impact on capitalisation.Both the condominium (column 1) and single-family (column 2) housing analyses have been estimated with weighted least square, where the weights consist of the probability of being included in the treated group.The degree of explanation is, as the default model, high.Around 95 percent of the variation in condominium prices can be explained by the included variables, as well as around 88 percent of the variation in single-family house prices.5 shows the weighted least square estimates (WLS) concerning models that address both the (1) condominium and (2) single-family housing market.The weights are based on the propensity score estimates belonging to the treatment group.All models included fixed postal code effects and fixed monthly effects.Only observations within 500 meters of the RCS are included in the estimations.The treatment (RCS) group are observations within 200 meters of RCS, and the control group observations within 200 to 500 meters.RCS is a binary variable measuring whether the observation is close to any RCS, an RCS with amenities (good), or and RCS with disamenities(bad).t statistics are in parentheses, * p < 0.05, ** p < 0.01, *** p < 0.001.
Proximity to RCS has a negative impact on prices.The effect is more evident in the single-family house segment than in the condominium one.In the single-family housing market, we can also note that the negative capitalisation effect is independent of whether RCS is co-located with other disamenities or, for that matter, with amenities.This is not the case in the condominium segment, where there is a negative capitalisation regarding proximity to RCS.However, it is higher if RCS are co-located with positive characteristics in the residential environment, such as proximity to parks and other green areas.
Moreover, it is also significantly higher negatively capitalised if RCS are co-located with negative properties in the residential environment, such as proximity to major roads or gas stations.Our results indicate that RCS have a negative impact on condominium prices, but this is especially clear if there are other disamenities in its vicinity.Thus, one could that in an urban environment, the effect of RCS is relatively limited if one has not placed RCS in environments that are otherwise considered attractive.In the single-family housing market, the proximity to RCS is negatively capitalised into housing values regardless of its micro-location.

(c) Property fixed effects
We have added additional fixed effects at the property level to reduce the risk of omitted variables and endogeneity problems.For the owner-occupied apartment market, we have added fixed effects for the building, which in most cases refers to neighbouring condominiums in the same properties but also repeated sales.For the single-family housing market, we have added fixed property effects that, in all cases, refer to repeated sales.The results are presented in Table 6.
The number of independent variables increases dramatically from just over 200 in the model with fixed postal codes and monthly effects to over 700 in the model with fixed property effects.The degree of explanation in the model increases slightly, but it is not a statistically significant difference.The capitalisation of proximity to RCS increases in the inner city, where we have more condominiums, while the effect in the single-family areas is equivalent to the model without the property fixed effects.An RCS within 200 meters doubles from 1.3 percent to 3.2 percent in the condominium market.The inclusion of the fixed property effects thus has a substantial effect on capitalisation.It can also be stated that the implicit prices regarding proximity to the subway station and CBD change dramatically in the model that explains the condominium prices but not in the single-family house model.The fixed property effects effectively pick up the effect of proximity to the CBD (no longer statistically significant) and subway station (statistically significant but with a changed sign).6 shows the condominium (1) and single-family (2) housing market's weighted least square estimates (WLS).The weights are based on the propensity score estimates belonging to the treatment group.All models include fixed postal code effects, fixed monthly effects, and fixed property effects.The condominium sample can be repeated sales or other sales in the same building, and in the single-family sample, it is repeated sales.Only observations within 500 meters of RCS are included in the estimations.The treatment (RCS) group are observations within 200 meters of RCS, and the control group observations within 200 to 500 meters.t statistics are in parentheses, * p < 0.05, ** p < 0.01, *** p < 0.001.

(d) Robustness test
It is assumed that the effect of RCS is local to the extent that we have assumed that the treatment range is from 0 to 200 meters, and the control group consists of 200 to 500 meters.To test how robust the estimates are, we have tested alternative intervals, namely 0-100 meters and 0-300 meters as a treatment group and 100-400 meters and 300-600 meters as a control group, respectively.The results are shown in Table A1 in the Appendix.
The estimates are robust concerning assumptions the size of the treatment and control group.A narrower treatment group results in equivalent estimates in the condominium analysis, even if the statistical significance is somewhat greater in the narrower range than in the broader one.The estimates in the single-family housing sample are slightly larger in the broader range than in the narrower one, but the difference is not statistically significant.However, the robustness test indicates that the capitalisation effect is not dependent on the assumption of treatment and the size of the control group.The degree of explanation is somewhat higher in the narrower range, but the differences are minimal in the condominium sample and slightly larger in the single-family house sample.
As a further robustness test, we have randomised where the 250 RCS are located.We have randomly "moved" the RCS 0-500 meters from its actual location.We then calculated the distance between the housing transactions and the "new" location.The methodology is inspired by the so-called placebo test used in the regression discontinuity design methodology.The hypothesis is that this random movement of RCS does not have a negative capitalisation on prices.
The results from these analyses can be found in Table A2 in the Appendix and show that the effect of RCS is now statistically insignificant.This strengthens our interpretation that the proximity of RCS has a statistically significant negative causal effect on housing value, i.e. no placebo effect, which means that RCS harm housing attractiveness.

Conclusion and Policy Implications
This paper discusses the importance of planning where RCS should be located in the urban context.In our analysis of whether the location of RCS has a negative effect on the residential environment, we have used the traditional hedonic methodology where the value of the dwelling is built up by the housing attributes.The most important of these are attributes and characteristics associated with the house and its location in the city.However, attributes in the residential environment also impact the value, including attributes such as proximity to green areas and various private and public services.There are also many negative externalities in the city that can affect the attractiveness of the residential area.
Proximity to RCS is one of these disamenities as they can cause disturbances in litter, noise, odour, and pollution.
We have used almost 200,000 housing transactions in Stockholm, Sweden, to estimate the hedonic price equation.We have taken great care to ensure that the relationships we have estimated are also causal relationships by including different types of fixed effects, analysing the environment where the RCS are located and including as many value-adding attributes as possible in the model.This has been done to minimise omitted variable bias, control for reverse causality and selection bias in treatment and minimise the extent of measurement error.A1 shows the condominium (1 and 2) and single-family (3 and 4) housing market's weighted least square estimates (WLS).The weights are based on the propensity score estimates belonging to the treatment group.All models include fixed postal code effects and fixed monthly effects.Only observations within 4-600 meters of the RCS are included in the estimations.The treatment (RCS) groups are observations within 100 and 200 meters of RCS, and the control group observations within 100 to 600 meters.t statistics are in parentheses, * p < 0.05, ** p < 0.01, *** p < 0.001.

(
fixed time effects).The subscripts i and t indicate transaction and time.All Greek letters indicate parameters that are estimated.The parameter has a subscript of j for a postal code, indicating that fixed urban effects are included in the model.
Map 1. Map of Stockholm and its districts Source: The City of Stockholm In the city of Stockholm, there are 250 RCS.This system is an integral part of recycling paper, glass, plastic, metal packaging, and batteries.Mattsson Petersen and Berg's (2004) results show that the most common visitor to the recycling station brought paper, newsprints, and glass packaging; the least common were batteries and textiles.Of course, this may have changed since the survey was conducted.FTI (Förpackning och Tidningsinsamlingen) owns and operates these RCS.Together with the municipality, they decide where they should be placed, what should be collected, and how often they

Figure 1 .
Figure 1.Pictures of recycling stations in the city of Stockholm.

Table 1 .
Number of recycling stations (RCS) in relation to population Source: FTI and the City of Stockholm.Own calculations.

Table 3
Table3shows the descriptive statistics concerning the two forms of housing transaction data used in the study: condominiums (Panel A) and single-family houses (Panel B).Shown in the table are the number of observations, mean value, standard deviation (Std.Dev.), minimum (Min) and maximum (Max.)values.Price is measured in Swedish krona (SEK), living area and plot area in square meters and monthly fee in SEK.Built equals the building year.RCS (recycling station) is a binary variable measuring the treatment that the transaction is within 200 meters from the recycle station.Distance to RCS, subway station, shopping mall and Central Business District (CBD) is measured in kilometres.CBD in Stockholm is Sergel Torg.Potential outliers that have been excluded are observations with prices below the one percentile of the price distribution: the same concerns living area, monthly fee and building year.Condominiums with an apartment floor below 0 have been excluded.

Table 5 .
Empirical results -Micro location models

Table 6 .
Empirical results -Property effects models

Table A1 .
Robustness test (treatment control groups)

Table A2 .
Robustness test (placebo effect)Note.TableA1shows the condominium and 2) and single-family (3 and 4) housing market's weighted least square estimates (WLS).The weights are based on the propensity score estimates belonging to the treatment group.The variable binary Placebo RCS is based on the distance from the dwelling to the recycle stations randomly moved from 0-500 meters.All models include fixed postal code effects and fixed monthly effects.t statistics are in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001