The effect of transport innovation on property prices: A study on the new commuter line between Uppsala and Älvsjö. Student: Brikena Meha

The effect of transport innovation on property prices: A study on the new commuter line between Uppsala and Älvsjö Student: Brikena Meha Supervisor: Ina Blind Master of Science Programme in Economics Department of Economics Uppsala University, Sweden January 2017

Abstract The aim of this paper is to investigate the impact of transport innovation on housing prices. More precisely, I study the effects on housing prices of the new commuter line (J38) between Uppsala and Älvsjö that started on December 9, 2012. The properties covered in the analysis are located around the 11 station stops between Uppsala and Älvsjö. This transport innovation was initiated to increase integration between Uppsala County and Stockholm County. Using the start of the J38 line as a quasi-experiment in a hedonic price model, I compare the changes in housing prices in treated areas and untreated areas after the introduction of the new line. Separate models are estimated for properties in Housing Cooperative (Bostadsrätt), and in Home Ownership (Äganderätt). The models include contract prices at which the property is sold, house characteristics, distance to nearest rail station, bus station, water and time-fixed effects. The results suggest that apartment prices (in housing cooperative) were negatively affected by the new line, whereas the effect for house prices (home ownership) is not statistically significant different from zero. Key words: Housing Prices, Transport Innovation, Housing cooperative, Home Ownership, Distance to Commuter Station, Hedonic Price Model, Difference-in-Difference. 2

Table of Contents 1. Introduction... 4 2. Literature review... 6 3. Description on housing types and transportation in Sweden... 8 3.1. Housing types traded on the market... 9 3.2. The transport innovation... 10 4. Study perimeter... 11 4.1. Data... 11 4.2. Descriptive Statistics... 12 5. Econometric specification... 15 5.1. The model... 15 5.2. Regression results... 16 5.2.1. Baseline results for housing cooperative regression... 17 5.2.2. Baseline results for home ownership regression... 19 6. Sensitivity Analysis... 21 6.1. Anticipation effects... 21 6.2. Other thresholds... 22 6.3. Measurement error in coordinates... 22 7. Conclusions... 23 References... 25 APPENDIX... 27 3

1. Introduction Transport infrastructure plays an important role in urban development since it connects peoples residence and work, as well as influences the form, density, and expansion of urban areas. Economically active individuals make decisions that depend on the trade-off between housing and the cost of transportation to work. In big cities, some people mainly rely on rail transportation for such commuting. The number of work commutes via rail transportation within the Stockholm area is about 40% (Transport Analysis, 2011). Demand for good access to rail transportation can thus be thought to have a positive effect on housing prices close to commuter rail stations. However, there can also be negative externalities associated with commuter rail station, e.g. noise, pollution, increase in crime rates (see e.g. Bowes and Ihlanfeldt, 2001; Debrezion et al., 2007). By identifying how new transport lines actually affect property prices, policy makers will be able to support their decisions on whether to invest in transport innovation or to what extent to do so. The aim of this paper is to investigate the effect on housing prices of the new commuter line (J38) that goes between Uppsala and Älvsjö, and started on December 9, 2012. The properties covered in this analysis are located around the 11 station stops of J38 line between Uppsala and Älvsjö. The starting point of this analysis is a simple hedonic price model that relates property prices to the J38 line. This pricing model treats every property as a heterogeneous good that is determined by its features. In housing context, these features determine the price of the house and are categorized into three groups: physical, accessibility, and environmental (Fujita, 1989; Bowes and Ihlanfeldt, 2001). Rosen (1974), who popularized the theory of hedonic price model discusses that, decomposing housing prices into implicit prices is challenging since there are not two identical dwellings, or that they are not in the same location. House features cannot be sold separately, as there is no market for them, thus, it is not possible to observe them independently. 4

Given the information on changes on transport innovation over time, that is the new commuter line (J38) on December 2012, I use a difference-in-difference (DID) approach to examine how property prices respond to this new line. This approach compares housing prices before and after J38 started, for housing that are close to a rail station and those that are farther away. The key identifying assumption is that the trends of housing prices in the areas closer to the rail stations are the same to areas farther from the rail station, in the absence of the new commuter line. This study is important because it is the first of its kind to apply a difference-in-difference approach that addresses the relationship between transport innovation and property prices in Stockholm County. This quasi-experimental approach controls for unobservable location-fixed characteristics that could otherwise bias the estimation of the relation between the J38 line and housing prices. To implement this method, I will use data on housing sales in the county of Stockholm, for the time period from January 1, 2009 to December 31, 2015. The data set is very rich as it contains crucial information on the variables of interest, such as sales prices, housing characteristics, geographical coordinates for each property, accessibility to the nearest rail station (i.e. commuter, subway) and bus station, as well as to nearest water. The results for properties in housing cooperative suggest that there is a decrease in housing prices after the December event for dwellings in the treated areas, and for properties in home ownership are insignificant but positive. Further, I do four different robustness checks to test the specification of the two main regression models. The remainder of the paper is structured as follows. Section 2 provides a review analysis on the existing literature about the relationship of proximity to the station on house prices. Section 3 provides an overview of the housing types that are treated on the Swedish housing market, and a description of the J38 line. Section 4 is the study perimeter, including data, and descriptive statistics. Section 5 is the baseline empirical 5

analysis. Section 6 provides sensitivity analyses and section 7 gives the conclusions of the study. 2. Literature review Theories on land values use accessibility as a determinant factor to explain variation in property prices. A monocentric model developed by Alonso (1964) explains that jobs are located around the central business district (CBD) of the city, where most of the economic agents work. In the means of compensating for long commutes to work, housing prices fall as the distance to CBD increases (see also Muth, 1969). There is thus a trade-off between commuting to work and property prices. Similarly, people that reside in properties located around rail stations can be expected to benefit in terms of cost savings and transport time, which pushes up property prices that are closer to rail station. Furthermore, there is a considerable amount of empirical literature on the relationship between housing prices and station proximity in different metropolitan areas around the world. Most of the empirical studies are cross-sectional hedonic studies that relate spatial heterogeneity in transport accessibility to property prices at one point in time, while controlling for other observable characteristics of the property. Examples of these type of studies are Debrezion et al. (2007) and Bowes and Ihlanfeldt (2001). Debrezion et al. (2007) analyze the impact of rail station proximity on property prices in three metropolitan areas in the Netherlands, using a cross-sectional hedonic price model. They find significant effects of station proximity on properties in some metropolitan areas, where housing prices decline by 1 percent when the distance to the nearest railway station doubles. On the other hand, they find insignificant effects in some other areas. They argue that this is due to differences in urbanization levels; if less urbanized the accessibility effect is smaller. 6

Overall, in cross-sectional hedonic studies the results are mixed. Bowes and Ihlanfeldt (2001) argue that a possible reason for the mixed results is a failure to account for the simultaneous and competing effects of the rail station on nearby properties. On the one hand, proximity to rail stations is valued because of transportation cost savings and because they can be retail nodes, but on the other hand rail stations might have negative externalities such as noise, pollution, unsightliness of the station, and higher crime rates due to improved access for outsiders. Further Bowes and Ihlanfeldt (2001) note that, the net benefit from station proximity may also vary with station characteristics, median income of the neighborhood, and distance of the station from the CBD center. Different effects are found due to demographic segregation into neighborhoods based on income levels. For instance, low-income residents value more proximity to a station, as they rely more on public transport, rather than high-income residents (Nelson, 1998). Trying to sort out these different effects, Bowes and Ihlanfeldt (2001) analyze the impact of rail station proximity on property values in Atlanta (US) by estimating a crosssectional hedonic price model and auxiliary models for neighborhood crime and retail activity. Bowes and Ihlanfeldt find that properties closer to stations (within a half kilometer) sell for 19% less than properties beyond 4.8 km from a station. However, they also find that properties that are between 1.6 and 4.8 km from a station have a significantly higher value compared to those farther away. According to Bowes and Ihlanfeldt (2001), these results suggest that houses nearby stations are affected by negative externalities, but those at an intermediate distance are beyond the externality effects and benefit from the transportation access provided by the stations. However, even after controlling for many observable characteristics, the study of Bowes and Ihlanfeldt like the other cross-sectional hedonic studies, encounters problems with potential bias since there may be unobservable characteristics correlated with the distance from the station that may affect house prices. 7

In contrast, an empirical strategy that takes into account changes over time in transport innovation, and controls for potential bias from unobservable characteristics correlated with rail stations is the difference-in-difference approach. Gibbons and Machin (2005) use this method to study the effect of the transport investment made to the new London Underground and Docklands Light Railway in South East London in the late 1990s on property values. Comparing changes in housing prices in areas that were affected and unaffected by the new improvement, Gibbons and Machin find that the value of property prices increases for 2.1% for each kilometer reduction to a rail station. For the control group, that is houses located outside of the 2 km threshold from the nearest station, experience a decrease in prices. This last approach will be used for the empirical method of this paper. I will use the starting of the J38 line as a quasi-experiment in a hedonic price model and compare how housing prices changed after the starting of J38 line, in areas affected and unaffected by the new line. This method allows for a more reliable assessment of transport innovation on property values, since it controls for time fixed unobservable characteristics correlated with rail stations that could otherwise bias the estimations. This study contributes to the literature because it is the first of its kind to apply a difference-indifference approach that addresses the relationship between transport innovation and property prices for Stockholm County, using the data from Mäklarstatistik AB on housing transactions. 3. Description on housing types and transportation in Sweden Among the OECD, Stockholm is one of the most successful metropolitan areas in terms of socio-economic performance, public health, progressive environmental behavior, educational attainment and is as well the leading labour market area in Sweden. The labour market has expanded further than Stockholm County, up to Uppsala County, which is located north of Stockholm County. Furthermore, the large labour force and the 8

employment rates are driven mainly by innovation, productivity, and good macroeconomic conditions (OECD, 2006). With 2.3 million inhabitants, the Stockholm area encompasses about 22% of the total population in Sweden and nearly 40% of Sweden s GDP (OECD, 2006). With Stockholm being an attractive location to work and live in, it faces serious challenges of housing shortages. According to the OECD (2006) report, on Sweden, between 1990 and 2004, population increased by 231,000 in Stockholm County, with only 90,000 housing units being built during that time. It was estimated that, by 2015 about 9,000-12,000 dwellings/year will be needed, and by 2030 about 7,000-10,000 dwellings/year will be needed (Stockholm County Council, 2003). 3.1. Housing types traded on the market Housing provision in Sweden is about 17% in the form of tenant cooperative housing, known as bostadsrättsförening (Karlberg & Victorin 2004). The resident is a member of a housing cooperative association that owns the property (most often a block of apartments). This membership grants them the right (bostadsrätt) to inhabit the dwelling for an unlimited period of time. Tenant owners hold the responsibility for changes and repairs in their apartment, whereas the cooperative is in charge of collective services, space, and utilities (Ruonavaara, 2005). Further, the residents are subject to a monthly fee that is intended to cover operational and financial expenses of the cooperative association. Tenant owners can also transfer their right to a dwelling, thus, they can sell the right to inhabit the dwelling at the existing market price. The buyer, on the other hand, also has to be a member of the housing cooperative association. In contrast, privately owned properties are mainly in the form of detached houses for permanent living and holiday purposes, or in the form of plots. There are also a few condominiums that are privately owned apartments in apartment blocks. The latter has been available in Sweden from 2009, and is applicable in buildings that were specifically built for that purpose, or if they had not served for housing in the last 8 years. The share of population 9

living in housing cooperatives in metropolitan Stockholm increased from about 12% in 1990 to 2008. While the share of population living in privately owned houses has remained about 30% over the whole period 1990-2008 (Blind, 2015). 3.2. The transport innovation The departure point of my analysis is a simple spatial regression that relates transport access with property values. To control for time-fixed unobserved property characteristics, I will however apply a quasi-experimental approach and compare housing prices before and after a transport innovation in areas affected and unaffected by the innovation. The transport innovation is the new commuter line between Uppsala and Älvsjö, J38 that began on December 9, 2012, through a partnership between Upplands Transport and Stockholm Public Transport. The purpose of this innovation was to create easily accessible commuting and increase integration between the counties 1, but also to decrease commuting time and costs (Stockholmståg KB, 2009). Älvsjö is a district in the Southern part of Stockholm s municipality, and Uppsala is a university town north of Stockholm. The route from Älvsjö to Uppsala is about 80 km, and the J38 line covers 16 station stops: Älvsjö, Årstaberg, Stockholms södra, Stockholm Central station, Karlberg, Solna, Ulriksdal, Helenelund, Sollentuna, Häggvik, Norrviken, Rotebro, Upplands Väsby, Arlanda, Knivsta, Uppsala Central station. The advantage of this new commuter line is that the passengers commuting between any of the stations, for example between Rotebro and Uppsala, or between Knivsta and Älvsjö no longer need to change trains at Stockholm s Central station, and thus it is cost efficient and less time consuming. Since I have data only on Stockholm County, my analysis will cover only 11 stops in the J38 commuter line, from Älvsjö to Rotebro 2. 1 Stockholm County and Uppsala County. 2 Stockholm Central station and Upplands Väsby are also excluded. The reasons are explained in the descriptive statistics section. 10

Figure 1. The route of J38 line between Uppsala and Älvsjö covering 16 station stops. Source: UL 2012. 4. Study perimeter 4.1. Data For this study, I will use a dataset on housing sales in Sweden that is from Svensk Mäklarstatistik AB 3, covering the time period from January 1, 2005 to December 31, 2015. According to Mäklarstatistik, the dataset covers about 80% of all housing transactions that are made during this period. The dataset is very rich for the purpose of this study as it contains crucial information on the variables of interest such as: housing characteristics i.e., living area (square meters), plot area (square meters), number of rooms, year built, and monthly fee), final (contract) price, list (ad) prices, list (ad) date, contract date, date of possession, and address for each property sold. Additionally, the data contains geographical coordinates 4 for each property, and recently was updated with the distance from each property to geo-coded urban data, more precisely to water, to the nearest bus stop, to the nearest subway station and to the nearest commuter train station 5. This data source has been used in previous studies by Blind and Dahlberg (2015), and Blind, Dahlberg and Engström (2016). 3 Svensk Mäklarstatistik AB is a Swedish enterprise owned by two broker firms and two trade associations for brokers. 4 The coordinates in the data set are provided by two sources: Google coordinates and by a real estate agent. 5 Gustav Engström, researcher at The Beijer Institute of Ecological Economics has therefore used Google s geocoding API to convert the address for each property to coordinates. 11

4.2. Descriptive Statistics My sample will be extracted from the house price data set. I will restrict the sample only to dwellings in housing cooperative and home ownership near J38 stations, from January 1 st 2009 to December 31 st 2015. All of the dwellings in the sample are within 5 km of the J38 line stations. The structural features of houses in the hedonic model used as explanatory variables are: living area (m 2 ), number of rooms, monthly fee (only for dwellings in the cooperatives) and plot area (only for houses in home ownership). Environmental variables include distance to the commuter station, subway station, bus station, and water). I will use a distance specification of properties from the nearest respective J38 commuter station (11 stations in total) 6. For this study, I have excluded housing sales that differ by more than 100 meters between the Google coordinates corresponding to the address of a property and coordinates provided by the real estate agent. This was done in the means of improving the precision of coordinates provided by the real estate agent. Observations with missing or unusual values have also been excluded from the sample 7, for the contract price 8, living area 9, number of rooms 10, monthly fee 11, or with an unclear contract date 12. Further, newly built apartments are also excluded from the analysis as they are sold at fixed prices. I have generated year monthly variables for time fixed effects, and station dummies. After adjusting the data to these changes, my sample consists of 82,139 observations of apartments in the housing cooperative, and 234 observations for home ownership. 6 Stations in Uppsala County (Uppsala Central station, Knivsta and Arlana) are not included in this analysis since I only have data on Stockholm County. I have excluded Stockholm Central and Upplands Väsby stations, since the new commuter line did not make it easier for people travelling from one of these stations to Uppsala or Älvsjö. Their commute did not include any change either before or after the new line. 7 For the restrictions I follow Blind, I. & Dahlberg, M. (2015) 8 Housing prices below 1,000 SEK. 9 Housing surface smaller or equal to 10 square meters and larger than 400 square meters. 10 Housings with 0 rooms or more than 20 rooms. 11 Housings with monthly fee smaller than 100 or larger than 100,000 SEK. 12 Housings with contract date before the ad date. 12

In Table 1. are shown descriptive statistics for properties in Housing Cooperative and Home Ownership around respective stations in J38 line. The average price at which a property in housing cooperative was sold is 2,876,906 Swedish Krona, and the average price in home ownership is 3,664,182 Swedish Krona. Table 1. Descriptive statistics for properties in Housing Cooperative and Home Ownership around stations in J38 line. (1) (2) (3) (4) (5) VARIABLES Obs Mean S.D. Min Max Housing Cooperative (Bostadrätt) Dependent variable Contract price (in SEK) 82,139 2,876,906 1,506,684 68,597 38,900,000 House characteristics Living area (m 2 ) 82,139 61.27 24.58 11 348 Number of rooms 82,139 2.319 0.998 1 10 Monthly fee (in SEK) 82,139 3,153 1,285 105 13,952 Environmental variables Subway distance 81,136 733.8 760.3 4.141 4,963 Commuter station 82,139 1,667 1,025 2.072 5,000 Water distance 81,131 741.1 679.3 0 2,999 Buss distance 82,139 113.0 74.14 0.321 620.4 Home Ownership (Äganderätt) Dependent variable Contract price (in SEK) 234 3,664,182 1,624734 1,495,000 12,500,000 House characteristics Living area (m 2 ) 234 100.4 28.28 54 206 Plot area (m 2 ) 234 403.3 240.6 94 1,465 Number of rooms 234 4.538 1.214 2 8 Environmental variables Subway distance 234 907.0 641.1 92.91 4,355 Commuter station 234 2,502 1,061 500.0 4,983 Water distance 186 1,658 687.0 222.8 2,990 Bus distance 234 195.9 107.0 23.05 647.5 13

The graphs below are purely descriptive figures of the development of mean logarithm of price in the treatment and the control areas per year quarter from January 1 st 2009 December 31 st, 2015 (given in year quarters). In Figure 2 and 3, the red line corresponds to the treated properties that are within 2000 m from the J38 station, and the black line to the untreated properties that are beyond 2000 m from the J38 station. In both of the figures there are almost similar pre-trends in property prices per year quarter. Figure 2.. Figure 3. 14

5. Econometric specification To examine the question whether the transport innovation has an effect on housing prices around the commuter stations in J38 line, I will use a quasi-experimental approach and a hedonic price model. The difference-in-difference estimator in the hedonic model compares the logarithmic of housing prices in treated and untreated areas before and after the start of J38 line. [ ( ) ( )] [ ( ) ( )] (1) 5.1. The model Different specifications have been used to estimate the hedonic price models. The first model is on apartments that are in housing cooperative, and the second model on privately owned houses. The reason why I will do two separate regressions is because they differ in terms of property characteristics. ( ) (2) The outcome variable entering the hedonic price model is the logarithmic specification of the final contract price of property i and date t, ( ). The semi-logarithmic specification is widely applied in the property value literature, and the fact which motivates it, is that it gives robust estimates and allows convenient interpretation of coefficients (see e.g. Gibbons & Machin 2005, Bowes & Ihlanfeldt 2001). The values of the coefficients represent percentage changes on housing prices. is a constant, is a dummy variable, equal to 1 if property i is within 2 km of one of the J38 commuter stations, and 0 otherwise 13. It measures the proximity to the commuter station from property i, and controls for the effect of the commuter station on property prices and for omitted variables that are spatially correlated with the commuter station. The choice of 2 km threshold represents a walking distance of about 20-30 minutes, the maximum that 13 Properties in the control group are located between 2km to 5km from the nearest rail station in J38 line. 15

people can be expected to walk to get to a station (Gibbons and Machin, 2005). a dummy variable, equal to 1 if property i was sold after December 9, 2012, and 0 otherwise. is a vector of property characteristics for property i at date t. This vector is different in the two regressions. For apartments in housing cooperative, it includes: living area, number of rooms, and monthly fee. For houses in home ownership the vector includes: plot area, living area, and number of rooms. 16 is are dummy variables for the 11 stations of the J38 line. It controls for time-constant station effects. vector of time fixed effects (month-by-year), and controls for time variation that is not related to the new commuter line. are the error terms. The identifying assumption in a difference-in-difference method is that, in the absence of the new commuter line (J38), the trends in property prices for the treated and untreated areas would have been the same. Given the control variables, there should be nothing else systematically affecting property prices except for the new commuter line in December 9, 2012. Under this assumption, is the coefficient for the interaction term between the date and distance which measures the effect of the J38 line on property prices within 2 km of the commuter stations. 5.2. Regression results In this section I present the results from the estimations of equation (2) in the sample described in section 4. Table 2 gives the main estimation results for properties in housing cooperative around 11 station stops in J38 line. Table 4 and 5 (see Appendix) show results for properties in housing cooperative located in the north and south of the Stockholm Central station. Table 3 provides the main results for houses that are privately owned near 11 station stops in J38 line. Similarly, separate results for privately owned houses on north and south of Stockholm Central are found in Table 6 and 7 in the Appendix. The coefficients on dummies for each respective station and month-by-year fixed effects are not reported in the means of saving space. Approximately 84% of the variation in the dependent variable is explained by the independent variables in the main is a

hedonic price model for houses that belong in the housing cooperative and home ownership. 5.2.1. Baseline results for housing cooperative regression The first regression estimates the changes in property prices for apartments in the housing cooperative that are located around the 11 commuter stations in J38 line. The results are shown in Table 2. In the first column are added control variables for the physical house characteristics (i.e., living area, number of rooms, and monthly fee). In column (2) are added environmental variables (i.e., distance to nearest subway station and bus station, as well as distance to water). Commuter stations in J38 are included as dummy variables and time fixed-effects (month-by-year) are included in both specifications. The standard errors are clustered on neighborhoods 14 to allow for potential correlation within neighborhoods on property values. In other words, observations are independent across the neighborhoods, but not within neighborhoods. Clustered errors only affect the variance and the standard errors but not the coefficients. The difference-in-difference (DID) point estimates are negative and stable throughout the two specifications. The DID estimators are significant at 5% level in both columns. The coefficients show that there is a negative effect on property prices after the December event. These results suggest that the prices of apartments located nearby rail stations were negatively affected and were sold for about 2.75-3.0% less by the introduction of the new line (J38) in December 2012. The control variables are all significant at 1% in all specifications, except for the distance to the nearest bus that is insignificant. One possible explanation of the results would be that before the J38 line started, people working close to the J38 stations, (i.e., Älvsjö or Solna) needed to live in a nearby station to avoid changing trains at Stockholm central station. Now they can live somewhere else (e.g., in Knivsta or in Uppsala) and still easily get to their work in Älvsjö or Solna, or some other 14 In the data set, neighborhoods are SAMS: SAMS code identifies the SAMS district (~census tract) of an observation using a GIS map of SAMS districts provided by (maps.slu.se) and the geographical information. 17

station in J38 line. This would result in falling demand for housing close to the J38 stations and thereby falling house prices close to the J38 stations. Additionally, I have run separate regressions with same specifications for properties in both tenures located in the north and south of Stockholm Central station, to see whether the effect of the commuter line is different in both sides. Northern stations include 7 station stops 15 in the J38 line, and 4 station stops 16 in the southern part. The results (see Table 4 and 5 in Appendix) aren t very different from the main results, as they show that the properties are sold for 2.6% and 3.7% less in north and south area, and are significant at 5% level. The results for home ownership are insignificant. Other possible reasons for this negative effect could be because of the negative externalities, for example increased population mobility around the stations leads to higher crime rates. More noise and pollution could also have negative effects on house prices around the stations. Proximity to a rail station is most highly valued by lowincome neighborhoods rather than high-income neighborhoods (Nelson, 1998). The argument which supports this statement is that low-income individuals most likely rely on public transportation, therefore, choose to live closer to the station. As the proximity to the station would be 20-30 minute walk, then literature suggests that areas around the rail station are occupied by the lower-income community (Debrezion et al., 2007). 15 Karlberg, Solna,, Helenelund, Sollentuna, Häggvik, Norrviken, Rotebro. 16 Älvsjö, Årstaberg, Stockholms södra, Ulriksdal. 18

Table 2. Prices for properties in Housing Cooperative after December 9, 2012 (1) (2) VARIABLES Log price Log price Distance to station 0.136*** 0.129*** (0.0387) (0.0289) Difference-in-Difference -0.0301** -0.0275** (0.0133) (0.0130) Living area (m 2 ) 0.0144*** 0.0140*** (0.000470) (0.000529) Number of rooms 0.0227*** 0.0304*** (0.00806) (0.00751) Monthly fee -6.54e-05*** -5.57e-05*** (1.12e-05) (1.02e-05) Subway distance -8.34e-05*** (2.40e-05) Water distance -0.000173*** Buss distance (2.57e-05) -3.55e-05 (9.18e-05) Constant 12.87*** 13.58*** (0.0458) (0.107) Observations 82,139 80,128 R-squared 0.809 0.838 Commuter stations (J38) Yes Yes Month-by-Year FE Yes Yes Robust standard errors in parentheses clustered in neighborhoods *** p<0.01, ** p<0.05, * p<0.1 5.2.2. Baseline results for home ownership regression The second regression estimates the changes in property prices in home ownership that are located around the 11 commuter stations in J38 route. The results are presented in Table 3. The specification in column one includes house characteristics (i.e., plot area, living area, and number of rooms). Column (2) has environmental variables added as in Table 2. Commuter dummies for stations in J38 are included along with time-fixed effects in both columns. The DID estimates in the first specification are negative, whereas on the second column they are positive. Nevertheless, they are not statistically significant. Amongst the control variables, distance to nearest station and plot area (in both specifications) are significant at 5% level. The results suggest that there is no effect 19

on housing prices in home ownership within two kilometers from station proximity after the December event. This could be because of the low number of observations in this housing tenure model. Also, before the J38 line there was already an existing commuter line going between Stockholm Central station and Uppsala Central station that covers the same route. Therefore, the new line between Uppsala and Älvsjö might have not necessarily affected the house prices in these areas. Similar as in section 5.2.1, I have run separate analyses for privately owned houses in the north and south side of Stockholm s Central station. As the main results show, the separate analyses confirm that there was no effect of this new commuter line on properties of this tenure on both sides (see results in the Appendix in Table 6 and 7). Table 3.Prices for properties in Home Ownership after December 9, 2012 (1) (2) VARIABLES Log price Log price Distance to station 0.0109-0.000995 (0.0533) (0.0587) Difference-in-Difference -0.0253 0.0228 (0.0513) (0.0274) Living area (m 2 ) 0.00192 0.00222 (0.00160) (0.00159) Number of rooms 0.0406 0.0381 (0.0384) (0.0396) Plot area (m 2 ) 0.000310** 0.000238** (0.000126) (9.66e-05) Water distance 1.76e-06 (6.55e-05) Bus distance 0.000373** Subway distance (0.000131) 9.03e-05 (6.45e-05) Constant 13.97*** 13.66*** (0.237) (0.348) Observations 234 186 R-squared 0.856 0.850 Commuter stations (J38) Yes Yes Month-by-Year FE Yes Yes Robust standard errors in parentheses clustered in neighborhoods *** p<0.01, ** p<0.05, * p<0.1 20

6. Sensitivity Analysis To check the specifications of the two regression models explained above, I run four different robustness checks. The first analysis will control for anticipation effects. The second will check three different thresholds (proximity of property to the commuter station). The third test will control for measurement error in property coordinates in the data. Lastly, I will estimate the effect of J38 line by restricting the sample to only two years (one year before and one year after the line started). 6.1. Anticipation effects Given the fact that the new commuter line was announced two years in advance in the UL Annual Report (Årsredovisning 2010), I would assume that there could be an anticipation effect on the property market along J38 line. Further, it is very likely that annual reports are finalized and released at the end of the year, therefore, in this case that would be on December 2010. Therefore, I will look whether there were changes in prices after January 1 st, 2011, before J38 line was opened. The anticipation effects are analyzed for both tenure types when re-estimating equation (2) with the same specifications. The number of observations in this analysis lowers for both models 17. For apartments in the housing cooperative shown in Table 8 (see Appendix), the DID estimates are insignificant and negative in the two specifications. Therefore, there seem to be no effect on apartment prices when the new commuter line was announced. The results for privately owned houses are shown in Table 9 (see Appendix). These results differ comparing to the original ones. They are positive throughout the two specifications. The two columns are significant at 1% level. Prices for houses within 2 km from the station, experience a 17.3-18.6% increase in sales (in respective specifications). These results indicate that, the announcement of the new commuter line J38 has adjusted the prices of privately owned houses almost two years before the opening date. 17 For the anticipation effect, no data is used after December 9, 2012. 21

6.2. Other thresholds The second sensitivity analysis will consider three different thresholds for distance proximity between properties and commuter stations. Thus, the treated properties are defined within the perimeter of 1000 m, 1500 m, and 1800 m from the respective station, after J38 line started on December 9, 2012. Re-estimating equation (2), with the richest specification (i.e., house characteristics, environmental variables, commuter stations, time fixed effects), results in housing cooperative tenure are very close to the baseline results (when using the >2000 m cut-off) and are stable in the three specified thresholds, and are negative at 1% significant level in the first and third column, whereas, the second column gives results at 5% level. The prices fall by 3.4%-2.9%-2.7% for each specification respectively. This robustness test supports the original results. Using the same specification for housing in home ownership tenure, the property prices are positive in the three columns, but insignificant. Therefore, the new commuter line has had no effect on house prices within these three distances. The results are found on Table 10 and 11 in Appendix. 6.3. Measurement error in coordinates So far, all estimations presented in each analysis are generated from the data where the difference between the coordinates 18 is within 100 meters. In the means of reducing measurement error between the two sets of coordinates and to provide more precise location for the property, the difference will be restricted to less than 50 meters 19. For the housing cooperative model, the results are similar to the baseline results, yet, the last specification is insignificant and slightly lower. Similarly, for the model on privately 18 Coordinates are provided by Google and real estate agents. 19 Blind, I. & Dahlberg, M. (2015) 22

owned houses the results are insignificant and slightly lower than the baseline results. The result tables are found in the Appendix. 6.4. Analysis with a smaller sample for two years In the last test, the sample is restricted to only two years, which lowers the number of observations to 35,197 apartments in housing cooperative and 98 privately owned houses. This analysis covers one year before the J38 line started in December 2012, and one year after. In comparison to the baseline results in section 5 for both tenures, property prices in housing cooperative are insignificant. The estimates for houses in home ownership are different in this test, in the first specification the prices are decreasing by 10.6% after the introduction of J38 line and are significant at 5% level. In the second specification, the results are insignificant. 7. Conclusions This paper analyses the effect on property prices of the new commuter line (J38) that goes between Uppsala and Älvsjö, and started on December 9, 2012. Properties in this analysis are located within 5 km from J38 line, which covers 11 station stops in this analysis. Using the start of the J38 line as a quasi-experiment in a hedonic price model I compare the changes in housing prices in treated areas (within 2 km) and untreated areas (beyond 2 km) from the rail station after the introduction for the time period from 2009 to 2015. Separate models are estimated for properties in Housing Cooperative (Bostadsrätt), and in Home Ownership (Äganderätt), since different specifications are applied to each. The models include contract prices and dates at which the property is sold, house characteristics, and environmental variables, such as distance to nearest rail station, bus station, water and time-fixed effects. The results suggest that the apartment prices located nearby rail stations were negatively affected by the introduction of the new line (J38), and are sold for about 2.75-3.0% less 23

than comparable apartments farther away. One possible explanation would be that people needed to live close to these stations (i.e., Älvsjö or Solna), or in a nearby station, to avoid changing trains at Stockholm central. Now they can live somewhere else (e.g., in Knivsta or in Uppsala) and still easily get to their work in Älvsjö or Solna or some other station on the J38 line. This would result in falling demand for housing close to the J38 stations and thereby falling house prices close to the J38 stations. Other possible reasons for falling housing prices could be that the introduction of the J38 line increased the negative externalities that can be associated with commuter rail stations, such as noise, pollution and crime. For properties in the home ownership tenure the difference-indifference estimates are not statistically different from zero. Additional analyses for properties located north and south of Stockholm Central station are done to see whether J38 line had any effects on housing prices in both tenures. To check the specifications of the two models, I have run four different robustness checks. The first analysis controls for anticipation effects to see whether the housing prices have adjusted after the announcement of the new commuter line. In the second test I use three different location thresholds (1000m, 1500m, 1800m), to see how the housing prices vary depending on property location from the commuter station. The third test controls for measurement error in property coordinates in the data, given that there are two sets of coordinates in the data. The estimators in this test are very similar to the original ones. In the last test, I use a smaller sample for only two years, covering only one year before and one after J38 line started, and see the effect of J38 line in housing prices. This study contributes to the economic literature by adding information on whether transport investments are relevant for housing markets and commuters living nearby the rail station, in the county of Stockholm. This is the first application using a new commuter line as a quasi-experiment in a hedonic price model in Stockholm County. 24

References Alonso, W. (1964). Location and Land Use Towards a General Theory of Land Rent, Harvard University Press, Cambridge. Blind, I. & Dahlberg, M. (2015). Immigration, New Religious Symbols and the Dynamics of Neighborhoods. Ina Blind, I., Essays on Urban Economics, PhD Thesis, Uppsala University. Blind, I., Dahlberg, M. & Engström, G. (2016). Prisutvecklingen på bostäder i Sverige: En geografisk analys. Ekonomisk Debatt, 4, 16-33. Bowes, D. & K. Ihlanfeldt, (2001). Identifying the Impacts of Rail Transit Stations on Residential Property Values, Journal of Urban Economics, 50, 1-25. Debrezion, G., Pels, E. & Rietveld, P. (2007). The Impact of Rail Transport on Real Estate Prices: A Meta-analysis, The Journal of Real Estate Finance and Economics, Volume 35, Issue 2, pp 161-180.. Fujita, M. (1989). Urban Economic Theory. Cambridge University Press. Gibbons, S. & Machin, S, (2005). Valuing Rail Access Using Transport Innovations, Journal of Urban Economics, 57(1). Gravel, N., Trannoy, A. & Michelangli, A. (2002). Measuring the Social Value of Local Public Goods: A Hedonic Analysis within Paris Metropolitan Area, Working Paper 82, Universita Commerciale Luigi Bocconi, Econpubblica, Center for Research on the Public Sector. Handbook on Residential Property Prices Indices, (2013). Methodologies and Working Papers. Eurostat. European Commission. Karlberg, B, & Victorin, A. (2004). Housing tenures in the Nordic countries, in: M. Lujanen (Ed.), Housing and housing policy in the Nordic countries, pp. 57-78, Nord 2001:27 (Copenhagen: Nordic Council of Ministers). Muth, R. F. (1969). Cities and Housing, Chicago, IL: University of Chicago Press. Nelson, A.C. (1998). Transit stations and commercial property values: case study with policy and land use implications. Paper presented at the 77 th Annual Meeting of the Transportation Research Board, Washington DC. OECD (2006). Territorial Reviews. Stockholm, Sweden. OECD Publishing. 25

Rosen, S. (1974). Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition Journal of Political Economy, 82,34-55. Ruonavaara, H. (2005). How divergent housing institutions evolve: A comparison of Swedish Tenant Cooperative and Finish Stakeholders. Housing Theory and Society. Transport Analysis (2011). Commuting to Stockholm, Gothenburg, and Malmo. A current State Analysis (2011). Stockholm County Council (2003), Office of Regional Planning and Urban Transportation, Regional Development Plan for the Stockholm Region 2001. Stockholmståg KB (2009) http://www.stockholmstag.se/ Årsredovisning (2010). Upplands Lokaltrafik AB. 26

APPENDIX Table 4. Prices for properties in Housing Cooperative for J38 stations in the north side of Stockholm Central station. (1) (2) VARIABLES Log price Log price Distance to station 0.157*** 0.126*** (0.0479) (0.0368) Difference-in-Difference -0.0263* -0.0279** (0.0144) (0.0135) Living area (m 2 ) 0.0152*** 0.0143*** (0.000664) (0.000637) Number of rooms 0.0149 0.0277*** (0.0127) (0.00950) Monthly fee -8.17e-05*** -6.15e-05*** (1.96e-05) (1.25e-05) Subway distance -0.000103*** (2.18e-05) Water distance -0.000222*** (2.57e-05) Buss distance -5.03e-05 (8.77e-05) Constant 13.66*** 13.89*** (0.0396) (0.0448) Observations 82,139 80,128 R-squared 0.738 0.814 Month-by-Year FE Yes Yes Robust standard errors in parentheses clustered in neighborhoods *** p<0.01, ** p<0.05, * p<0.1 27

Table 5. Prices for properties in Housing Cooperative for J38 stations in the south side of Stockholm Central station. (1) (2) VARIABLES Log price Log price Distance to station 0.175*** 0.128*** (0.0564) (0.0374) Difference-in-Difference -0.0369** -0.0289** (0.0181) (0.0139) Living area (m 2 ) 0.0157*** 0.0141*** (0.00106) (0.000574) Number of rooms 0.0171 0.0326*** (0.0168) (0.00876) Monthly fee -0.000120*** -6.26e-05*** (2.70e-05) (1.24e-05) Subway distance -0.000154*** (1.73e-05) Water distance -0.000262*** (3.81e-05) Buss distance -0.000110 (9.44e-05) Constant 13.73*** 13.98*** (0.0945) (0.0552) Observations 82,139 80,128 R-squared 0.617 0.804 Month-by-Year FE Yes Yes Robust standard errors in parentheses clustered in neighborhoods *** p<0.01, ** p<0.05, * p<0.1 28

Table 6. Prices for properties in Home Ownership for J38 stations in the north side of Stockholm Central station. (1) (2) VARIABLES Log price Log price Distance to station -0.0592-0.0998 (0.0727) (0.0709) Difference-in-Difference -0.0683-0.0474 (0.0612) (0.0388) Living area (m 2 ) 0.00313 0.00361 (0.00177) (0.00201) Number of rooms 0.0142 0.0123 (0.0526) (0.0635) Plot area (m 2 ) 0.000489** 0.000450** (0.000187) (0.000167) Water distance 5.93e-05 (7.20e-05) Buss distance 0.000520 Subway distance (0.000324) 3.16e-05 (7.60e-05) Constant 14.33*** 14.14*** (0.163) (0.153) Observations 234 186 R-squared 0.786 0.752 Month-by-Year FE Yes Yes Robust standard errors in parentheses clustered in neighborhoods *** p<0.01, ** p<0.05, * p<0.1 29

Table 7. Prices for properties in Home Ownership for J38 stations in the south side of Stockholm Central station. (1) (2) VARIABLES Log price Log price Distance to station -0.0239 0.0118 (0.0652) (0.0809) Difference-in- Difference -0.0576 0.0115 (0.0613) (0.0471) Living area (m 2 ) 0.00219 0.00121 (0.00159) (0.00160) Number of rooms 0.0596 0.0496 (0.0327) (0.0419) Plot area (m 2 ) 0.000490** 0.000386* (0.000213) (0.000189) Water distance -4.74e-05 (6.22e-05) Buss distance 0.000165 Subway distance 30 (0.000178) -7.18e-05 (4.90e-05) Constant 14.17*** 14.74*** (0.236) (0.328) Observations 234 186 R-squared 0.787 0.784 Month-by-Year FE Yes Yes Robust standard errors in parentheses clustered in neighborhoods *** p<0.01, ** p<0.05, * p<0.1 Table 8. Anticipation effects for properties in Housing Cooperative after January 1, 2012. (1) (2) VARIABLES Log price Log price Distance to station 0.134*** 0.124*** (0.0392) (0.0349) Difference-in-Difference -0.0130-0.0102 (0.0200) (0.0241) Constant 13.09*** 13.54*** (0.0409) (0.121) Observations 40,519 39,393 R-squared 0.784 0.816 House characteristics Yes Yes Environmental variables Yes Yes Commuter Stations (J38) Yes Yes Month-by-Year FE Yes Yes Robust standard errors in parentheses clustered in neighborhoods *** p<0.01, ** p<0.05, * p<0.1

Table 9. Anticipation effects for properties in Home Ownership after January 1, 2011 (1) (2) VARIABLES Log price Log price Distance to station -0.0444-0.0756 (0.0551) (0.0652) Difference-in-Difference 0.173*** 0.186*** (0.0394) (0.0524) (6.14e-05) Constant 15.09*** 14.98*** (0.315) (0.337) Observations 158 129 R-squared 0.811 0.790 House characteristics Yes Yes Environmental variables Yes Yes Commuter stations (J38) Yes Yes Month-by-Year FE Yes Yes Robust standard errors in parentheses clustered in neighborhoods *** p<0.01, ** p<0.05, * p<0.1 Table 10. Different thresholds for properties in Housing Cooperative after December 9, 2012. 1000m 1500m 1800m VARIABLES Log price Log price Log price Distance to station (1000m) 0.106*** (0.0243) Difference-in-Difference -0.0338*** (1000m) (0.0116) Distance to station (1500m) 0.128*** Difference-in-Difference (1500m) 31 (0.0291) -0.0288** (0.0118) Distance to station (1800m) 0.119*** (0.0291) Difference-in-Difference (1800m) -0.0273** (0.0128) Constant 13.67*** 13.65*** 13.62*** (0.111) (0.112) (0.110) Observations 80,128 80,128 80,128 R-squared 0.834 0.839 0.837 House characteristics Yes Yes Yes Environmental variables Yes Yes Yes Commuter stations (J38) Yes Yes Yes Month-by-Year FE Yes Yes Yes Robust standard errors in parentheses clustered in neighborhoods *** p<0.01, ** p<0.05, * p<0.1

Table 11. Different thresholds for properties in Home ownership after December 9, 2012 1000m 1500m 1800m VARIABLES Log price Log price Log price Distance to station (1000m) -0.156 (0.153) Diff-in-Diff (1000m) 0.240 (0.133) Distance to station (1500m) -0.0697 (0.0761) Diff-in-diff (1500m) 0.112 (0.0691) Distance to station (1800m) -0.0105 (0.0612) Diff-in-Diff (1800m) 0.0265 (0.0522) Constant 14.86*** 13.37*** 13.45*** (0.200) (0.277) (0.293) Observations 186 186 186 R-squared 0.852 0.852 0.850 House Characteristics Yes Yes Yes Environmental variables Yes Yes Yes Commuter stations (J38) Yes Yes Yes Month-by-Year FE Yes Yes Yes Robust standard errors in parentheses clustered in neighborhoods *** p<0.01, ** p<0.05, * p<0.1 Table 12: Measurement error - Housing Cooperative after December 9, 2012 (1) (2) VARIABLES Log price Log price Distance to station 0.138*** 0.130*** (0.0387) (0.0289) Difference-in-Difference -0.0301** -0.0268** (0.0131) (0.0127) Constant 13.09*** 13.59*** (0.0513) (0.110) Observations 78,484 76,580 R-squared 0.809 0.838 House characteristics Yes Yes Environmental variables Yes Yes Commuter stations (J38) Yes Yes Month-by-Year FE Yes Yes Robust standard errors in parentheses clustered in neighborhoods *** p<0.01, ** p<0.05, * p<0.1 32