Same, same but different? Neighbourhood effects of accessibility on housing prices

.


Introduction
Infrastructure is generally built with the expectations of positive impacts on the surrounding economy.But despite a large and growing literature on the topic, there are still questions of to what extent, and under which circumstances, these effects materialize.Railway infrastructure has experienced a resurgence in many cities and regions, and it constitutes major investments at local, regional and national level.Understanding the impacts of such investments is therefore of great concern.One common method of analysing the effects from improved accessibility is through housing prices, since these offer information on households' revealed willingness to pay for certain attributes, including accessibility.Although there is evidence of housing prices being affected by accessibility, these effects tend to vary, both due to methodology and to the type of rail and context (Debrezion et al., 2007;Mohammad et al., 2013).There is also a smaller but growing type of literature focusing on how effects may differ depending on the socio-economic characteristics of the neighbourhoods that face changes in accessibility.Housing markets are segmented and can be understood as different submarkets that cater for different types of households.The literature distinguishes between horizontal, or geographical, segmentation, and vertical segmentation, i.e. submarkets distinguished by price or socio-economic characteristics.Households with a high dependence on or use of public transport may be more willing to pay a premium for accessibility provided by the public transport system than households that do not depend on public transport.In many countries, this translates into poorer households or poorer neighbourhoods being more affected by price changes subsequent to accessibility changes.Previous findings report price changes being higher in poorer areas, whereas there are no effects (Bohman and Nilsson, 2016) or even negative effects (Forouhar and Hasankhani, 2018) in richer neighbourhoods.In line with these findings, Nilsson and Delmelle (2018) find that poor neighbourhoods are the most likely to experience demographic and socio-economic change as a result of light-rail investments.
In order to tailor policies that contribute to inclusive and sustainable cities, it is important to understand how the effects of investments are distributed between different groups and neighbourhoods.This paper aims to enhance the understanding of how effects of new commuter rail stations on property prices are distributed.Difference-in-difference models (DID) are applied to estimate effects on different neighbourhoods both in comparison to the overall local market and to neighbourhoods that are similar to the affected areas in different respects.
Methodologically, the paper differs from previous studies in applying a DID method to different submarkets.We can thereby understand how submarkets perform by comparing similar submarkets as well as compared to the average values.Submarkets are defined taking both horizontal and vertical segmentation into account.While DID methods have been frequently applied in the estimation of effects, we know of no previous studies that use it for analyzing distributional impacts.In this study we elaborate on control groups, and use both a more general control group comparing to the rest of the local market, as well as control groups defined according to market definitions.Control groups are matched both according to price level, in order to capture market segments, but for neighbourhoods with well-defined characteristics, matching is done using both qualitative and quantitative criteria.This adds to previous research in understanding not just overall or average effects, but also to how infrastructure may change relative attractiveness between neighbourhoods.
The case of the City Tunnel Malmö is chosen partly as it provides an opportunity to study different types of neighbourhoods, with the catchment areas representing neighbourhoods ranging from wealthier inner-city areas to peripheral low-income areas.In 2000, the Öresund Bridge, connecting Sweden and Denmark, was opened for traffic, thereby considerably improving accessibility within the region through commuter train traffic (Knudsen and Rich, 2013).The bridge implied a physical integration between two large labour markets that previously were only connected by ferry (Fig. 1).On the Swedish side, the construction was complemented with the opening in 2010 of a tunnel through the city of Malmö.In addition to the existing central station, which went from a cul-de-sac solution to drive-through traffic, two new stations were built: Triangeln and Hyllie (see Fig. 2).
Malmö is a city that is struggling with significant socio-economic inequalities and, despite its strategic location in the Öresund region, high unemployment figures.One of the explicit goals of the City Tunnel is the integration of different areas, both in terms of integration with the Copenhagen region and integration with existing neighbourhoods within the city, as demonstrated by one of the targeting goals for the area surrounding the new station in Hyllie: 'Hyllie is to become an urban centre in the south of Malmö which connects and strengthens the surrounding suburban areas' (City of Malmö).A stated goal is that the catchment area should include 'mixed housing' in terms of tenure and size, but not in terms of social mix.There is no stated targets to provide affordable housing.Swedish housing policy asserts that the municipalities are obliged to create the conditions for the market to provide housing; however, there have not been any specific subsidies or initiatives to target low-income households.This makes it an interesting case to study.Traffic in the tunnel is primarily used for travelling out of town, whereas within-city travel more often is comprised of buses.
The next section contains a literature review of effects of how accessibility affects different submarkets and neighbourhoods.Submarkets are here defined as price segments, whereas neighbourhoods are defined taking also other factors into account, such as architecture and built environment.Section 3 presents the study area and the method, including previous studies using similar methods.Section 4 presents the results from DID estimate.A discussion is provided in section 5, which are then summarized and used for formulating policy implications in the final section.

Effects of improved rail accessibility: a review of the literature
There is by now a large literature finding positive correlations between accessibility and higher property prices, and the heterogeneity of results is well documented (Debrezion et al., 2007;Higgins and Kanaroglou, 2016;Mohammad et al., 2013).In addition to problems that may arise due to differences in data and methodological concerns, three contextual factors are often mentioned that contribute to this heterogeneity.First of all, different property markets are in focus.Whereas H. Bohman residential markets may be the most studied types of market, some studies (Debrezion et al., 2007;Ko and Cao, 2010) focus on commercial real estate.Secondly, different types of transport modes are in focus, e.g.high-speed rail (Andersson et al., 2010), light rail (Dziauddin et al., 2015) and bus rapid transit (Mulley and Tsai, 2016).Common findings, however, seem to be that proximity to commuter rail, which is the focus of this study, has potentially large effects (Mohammad et al., 2013).One reason for this can be that relatively small improvement in commuter-based systems can imply large individual gains on an everyday basis, since, for example, 15 min of reduced commuting one way makes a relatively large difference for an individual.Third, in addition to different transport types, the local context in terms of the built environment differ greatly, and pedestrian and transit-designed development have been found to contribute positively to house price development (Bartholomew and Ewing, 2011;Duncan, 2011).This study contributes to the literature by deepening the discussion on property markets, and especially how effects may vary according to different market segments.
Furthermore, there are evidence that effects are not evenly distributed geographically (e.g.Dziauddin, 2019) and concern of how increasing property prices affect low income households.Housing markets tend to be local and characterized by price segmentation, and preferences in terms of accessibility may differ across segments.Markets segmentation is sometimes divided into geographical/horizontal segmentation, and vertical segmentation which implies that different price ranges cater to different households that may have different preferences due to income elasticity (Baudry and Maslianskaia-Pautrel, 2015).Failure to account for this heterogeneity, they claim, may result in biased estimates.Segmentation of the housing market is generally considered from a geographic (horizontal) perspective, whereas there are few previous studies focusing on vertical segmentation, i.e. price segmentation, such as Farmer and Lipscomb (2010) and Liao and Wang (2012).Zietz et al. (2008) etsablished that several attributes of a houseincluding square footage, number of bathrooms and age -vary across segments.There is also evidence that the impact of accessibility varies across segments.Building on the market segmentation idea, Bohman and Nilsson (2016) applied spatial quantile regressions to single-family houses and determined that the effects of proximity to commuter rail stations decline over the higher price segments, and that the highest price segments of the markets revealed no effects.
The finding that poorer neighbourhoods may be more affected by new rail accessibility seems to be receiving increased support.In a study on neighbourhood change, Nilsson and Delmelle (2018) found that although it is rare that neighbourhoods change in multivariate set of characteristics, poor neighbourhoods are more likely to experience changes than their richer counterparts.In a study based on data from the Paris metropolitan region, Mayer and Trevien (2017) identified that while competition for land increased because high-skilled workers were attracted to the vicinity of the railway stations, no increase in population could be noted.In Teheran it was found that effects were positive in the poorer areas of the city, whereas the wealthier parts saw negative effects of new metro stations (Forouhar and Hasankhani, 2018).One conclusion the authors drew was that new stations must also be combined with adequate planning measures such as walkable surroundings in order to achieve the positive benefits sought from the investments.Pasha, Wyczalkowski, Sohrabian and Lendel (2020) suggest bus services as a better alternative for improving accessibility for poor households since housing prices close to transit may become prohibitively expensive for some groups.
In relation to the above discussion, recent initiatives of densification and development around transit nodes have spurred a debate on gentrification.Although gentrification may be defined in different ways, the concept often refers to displacement of low-income households in areas receiving improved transit (e.g.Dawkins and Moeckel, 2016).Price increases can be problematic if they induce residential mobility patterns that make some households worse off in terms of accessibility.However, this link is far from clear.In a survey-based study, Nilsson et al. (2020) found that the presence of light-rail does not affect residents self-stated propensity to move out of the neighbourhood.A literature review of the impact of transit-oriented development (TOD) on gentrification written by Padeiro, Louro & da Costa (2019) finds some evidence of gentrification, but state that differences in methods as well as context calls for more systematic and comparable research on the matter.Furthermore, the need to address housing and transport not as separate entities, but to consider them jointly, has been recognized as an important part in addressing sustainable housing solutions for low-income households (Coulombel, 2018).
In addition to the distributional effects, another dimension that has an impact on findings is the time dimension, which can be explained by distinguishing between direct and indirect effects (Bowes and Ihlanfeldt, 2001).Accessibility can have an impact on residential values directly, as a result of time savings for travel, but also an indirect since new stations often attract new businesses and activities than can increase prices further.
Some studies have found an impact on property prices even before the traffic is opening, as an effect of capitalization of expected future price increases.As an example Atkinson-Palombo (2010) found price effects to be significant at all stages of the process; from planning to operation.Changes of neighbourhoods should be expected to take longer, since price changes may induce mobility patterns in the long run.Nilsson and Delmelle (2018) use a time frame of 30 years in their study but still find only limited evidence of changes at neighbourhood level.

Study area
Malmö, Sweden's third largest city, was once home to several industries, including the shipbuilding and textile industries.However, as a result of increased international competition during the 1970s, many succumbed, and unemployment soared.Indeed, unemployment in Malmö is still at a very high level by national comparison; and socioeconomic differences between different parts of the city are high, despite a rise in employment in newer industries such as the service and the gaming industries.
The city has witnessed some major investments, including the establishment of a new university and the opening of the Öresund Bridge, and in 2010 the opening of the City Tunnel.Despite the low income level of Malmö, in-migration has caused a high demand for housing.The Swedish housing policy is often described as a universal policy rather than a policy targeting specific groups.However, criticism has been raised as to whether this is still the case (Grander, 2017), and there are growing concerns about the provision of affordable housing.
The tunnel is primarily used for regional traffic and can be described as providing regional commuter trains.For intra-city transportation, there are often more convenient ways to get around, including buses, cycling and driving.All three station areas -Malmö Central, Triangeln and Hyllie (see Fig. 1) -can be characterised as pedestrian-oriented, with cycling and walking infrastructure planned together with the station.In connection to the building of the tunnel, densification strategies have been present for all three stations, although the context differs in terms of built environment.However, the tunnel did change the status of the station from a cul-de-sac solution to increased possibilities in terms of regional commuting, in particular, and improved access to long-distance rails.The new stations only provide regional train traffic.With the opening of the tunnel, the area around Malmö Central station is planned to transform from light industry to residential, office and retail, thus expanding the city centre towards the north.The Triangeln station is positioned at the other end of the pedestrian area, and it has provided limited opportunities for densification, although investments were made in connecting bus services, public spaces and some high-rise buildings.
The Hyllie area differs from the other stations as it is situated at the fringe of the city, but it was subject to extensive plans already in the 1960s.Expecting a massive population growth, the city planned to build several neighbourhoods surrounding Hyllie, which was supposed to be a new city sub-centre.But given the economic development, the neighbourhoods that were built became essentially dead ends at the fringe of the city.Feeder line highways, a result of earlier car-oriented planning, provide strong physical barriers.Once the plans to build the tunnel connecting Malmö to Copenhagen started to materialize, the area received new attention due to its strategic location between the two cities.
Hyllie also differs from the other stations in that its catchment area constitutes different neighbourhoods in terms of housing characteristics, demographics and ethnic composition.One of the existing neighbourhoods, Holma-Kroksbäck, was built as a part of the Million Homes Programme, which was implemented during the 1960s and 1970s.It can be described as the most ambitious political project to improve the H. Bohman housing standards in modern Swedish history (Hall and Vidén, 2005).Many of the areas built within this programme have faced considerable socio-economic challenges with low incomes, high unemployment and low education levels.As shown in Table 1, employment figures and education level for the neighbourhood are low compared to the other areas, and also the salary of those who do work is lower than city and national average.The second neighbourhood, Lindeborg, resembles Holma-Kroksbäck in terms of architecture and landscaping, but differs in the sense that employment levels are higher than the city average, and salary almost at par with the city average.The other two station catchment areas are better off in terms of employment and salary, and education levels are well above national average.

Data and model
The data consist of a panel of property transactions, more specifically condominium data, between 2005 and 2019.Strictly speaking, the Swedish system of tenant-ownership apartments is one of indirect ownership; the 'owner' really owns the right to dispose of the apartment and the owner is obliged to be member of a tenant-owners' association, which is the real owner of the building.For this study, however, this bares little importance, and apartments are sold at market price.The database can be considered to give an almost complete representation of actual transactions since it relies on information from all brokers, and transactions without brokers are very rare in the Swedish system.Because the data are used by brokers and appraisers, for whom accuracy and coverage is fundamental, there is significant pressure to maintain high quality.The database includes street addresses that have subsequently been geo-coded and supplemented with additional data on the housing stock and housing cooperations.
Relevant catchment areas most likely differ between different contexts.For example, a metro station in a dense urban area may attract people within walking distance.In this study, we aim for different distances to the station, hypothesizing that these may differ across stations.One kilometre corresponds to a 10-min walk, whereas 3 km corresponds to a 10 min bike ride.Cycling infrastructure is by international standards well established, with bike trails and parking.Furthermore, the stations provide regional commuter possibilities often competitive to car usage, as well as close connection to a large international airport, thus implying that the gains of using the trains are potentially large.Descriptive statistics of the full dataset as well as for catchment areas defined as within a radius of 1 km are shown in Table 2.
Prices are highest for the catchment area around the Central station, followed by the Triangeln area.For the full sample, price levels are higher within the catchment area than for the city average.Also, apartments around the Central stations are larger both compared to the city average and the other catchment areas.All of Malmö has witnessed a surge in property prices since the opening of the tunnel, and the catchment areas of the older stations have been central locations with higher values than the city average before the opening.The new area of Hyllie had lower property values initially, and after 2010 when the whole market dropped, the Hyllie area dropped deeper than the rest of the market.For the full sample however, prices are slightly higher than the city average, which is a result of price changes in recent years.
The method applied is difference-in-difference (DID), as it allows for the comparison with control groups, thereby controlling for changes in the market that are unrelated to the infrastructure investment.DID has become a common quasi-experimental tool to capture effects over time.It is often attributed to Snow's (1855) famous work on a cholera outbreak in London, in which he could establish causal effects from the quality of drinking water on the spread of the disease.
In social science, many DID studies focus on regional level, by comparing e.g.municipalities after fixed-link investments (Tveter et al., 2017) or new commerce (Daunfeldt et al., 2017).Here, the focus is instead on changes within the city.The general model is applied to transaction prices for two groups (g = 1, 2), observed in two time periods (t = 1, 2), before and after changes in infrastructure as dependent variables.Time series have been found to impose a problem of serial correlation, and that addressing the problem by collapsing data into two time periods -pre and post -works well also for smaller samples correlation (Bertrand et al., 2004).In line with this, this paper uses separate time periods, containing before and after price level data, instead of panel data estimates.In order to avoid any effect caused by temporary fluctuations, different definitions of time periods have been used as robustness tests.The model can be formulated as follows: In which T is for treatment area, P is for defining post treatment period and X is a set of control variables.
A major concern for DID estimations is finding the right control groups for comparisons.Essentially, the control group should follow the same trend as the treated group before the treatment, but the literature reveals no consensus on how to choose control group.This study will use different sets of control groups in order to improve the understanding of development in different submarkets.Quantitative matching methods, such as propensity score matching, are common in the literature; but in this case, both qualitative and quantitative assessments of the local real

Table 1
Socio-economic descriptives from 2010 of catchment areas, Malmö and Sweden.Source: Statistics Sweden, MONA register data.a) high education is defined as finishing at least three years of secondary education.estate markets are used in order to define relevant control groups.The qualitative assessment is primarily a complement to the quantitative assessment.

Method: defining control groups based on market segmentation
A main focus in this study is how control groups should be formulated given previous findings that housing markets are segmented and cater to different households.The three stations represent quite different submarkets (Table 1) and will thus be assessed separately.The built environment around the two more central stations, Malmö Central station and Triangeln, consist of older buildings with less pronounced neighbourhood characteristics.Price is for these markets considered the most relevant indicator to define submarkets.Maps of the catchment areas together with their control groups are presented in Fig. 1.Before the opening, the area around the Central station had square meter prices around 60-100% above the city average, whereas the corresponding figure for Triangeln was 20-40%.Matching was based on normalized prices for each year between 2005 and 2008.
The third station, Hyllie, differs and is considered to contain two different condominium submarkets that existed before the construction of the tunnel.Both belong to the lower levels of the price spectrum, but they also have quite clear neighbourhood characteristics, and a matching based only on prices would include areas of very different types of markets.In order to construct relevant control groups for the different areas, a two-stage procedure was implemented through which price matching was complemented with additional information.The first step was a qualitative assessment in which potential control areas were identified.The assessment was based on how areas are perceived by inhabitants in Malmö, as well as on architectural styles, construction years, distance to centre, ownership versus rental housing and reputation.In the first case, Holma-Kroksbäck was constructed during the Million Home's Programme, and other neighbourhoods built during the same period were therefore obvious candidates for control groups.They share many similarities in terms of architecture, construction and planning as well as socio-economic characteristics.The second neighbourhood, Lindeborg, also belongs to the Million Home's Programme but demonstrate higher prices and income levels, as well as lower crime levels.This created a pool of potential control group areas, which was then divided and matched according to price levels.Price levels before the opening of the station were used to match groups within similar price segments.As a final step, control group areas have been mapped and revisited to exclude non-representative areas (e.g.driven by outliers).

Results
The first step of the results is to provide an overview of difference-indifference estimates, using different before and after years, as well as different definitions of catchment areas in order to test for robustness.Table 3 presents overall findings followed by separate estimates for each station area.The control group is the same for all estimates and contains all transactions in the city outside of the tunnel buffer areas.The years 2009 and 2010 are used as before-variables, as the tunnel opened in December 2010.For the tunnel as a whole there is an effect when using 2010 as the pre-treatment period.For the Central station and Triangeln,  no significant effect can be found.The main effects are valid also when different years are used for estimating the post-period.Appendix A provides estimates for the whole period after 2014, and also here the effects are only significant when compared to the situation in 2009.In this alternative specification, the Central station effects turn out negative, implying that the price development has been lower than for the control group (see Table 4).The Hyllie station is the one station for which the effects are more notable.This is perhaps not surprising, given that this station provides a large increase in local, regional, and even international accessibility compared to the pre-tunnel situation.When estimating effects for separate years, effects become statistically significant from 2017, both when using 2009 and 2010 as pre-treatment periods.The effect is also well above 0.25% in all specifications.The sharp increase in prices starting after 2014 is in line with the overall construction during which both residential and commercial properties started to develop.An overall finding is that effects tend to materialize at a late stage.There are several explanations for this.First, although price levels have increased over the years, the market is considerably slower than that of e.g.Stockholm or other capitals.Furthermore, urban development is slow both due to planning and business cycle reasons.It is therefore only after a few years that new development start to materialize.Until then, the catchment areas are often characterized by construction activities and low levels of services.In other words, it takes time for the catchment areas to become attractive to households.However, the catchment of Hyllie is heterogeneous and new constructions tend to cater for the upper market segments.A next step will therefore be to define areas, both catchment and control areas, according to market segments.Two different markets are defined for the Hyllie area, focusing on the lower market segments since these were the two well-established neighbourhoods in the area with substantial amounts of condominium buildings before the opening of the station.We will also examine whether the choice of control group makes a difference for the other stations, and formulate control groups according to market segments.
Table 3 presents findings from separate estimates using unique control groups for each sample.They reveal not only how effects may differ depending on the control group, but also how development of prices and effects of different explanatory variables differ across the city.A first observations is that the DID coefficients differ quite substantially between the different samples.DID estimates suggest the highest effect; a 15.1% increase for the Holma-Kroksbäck sample.Although this figure must be interpreted with caution, the results indicate a negative correlation between effects and initial price levels.The second highest effect is for Lindeborg, the second poorest area with a 6.3% increase, and Triangeln, with a 2.7% increase.For the Central station, where prices already were higher than average, there is no discernible effect.The findings of higher effects for poorer neighbourhoods are in line with previous studies (Forouhar and Hasankhani, 2018), also when other variables than housing prices are the focus of the study (Nilsson and Delmelle, 2018).
One possible reason for the finding can be seen in the post-treatment coefficients.This is the lowest for the Holma-Kroksbäck group, suggesting that these neighbourhoods in general have had a slower price development than the rest of the market.The other three groups seem to have developed in a more similar pattern.It means that for households investing in an apartment in the poorer neighbourhoods, the payoff is, in general, considerably lower than in the rest of the market; and in this case, the extra effect provided by being in the catchment area, can really be interpreted as a catch-up to the rest of the market.For households already owning apartments in the area, which are often low-income households, this implies that their investments have been offering similar payoffs as investments in the rest of the city, something which seems not to have been the case for the control group, for which payoffs have been lower than for the rest of the market.However, it also implies that those who are thinking of buying need to pay what could be labelled an accessibility premium compared to similar neighbourhoods, making it less affordable for some groups.
The findings also reveal something of household preferences in the different market segments.In the Holma Kroksbäck group, the premium for one extra room is strikingly higher than for the other areas, and the lowest premium is found in the richest area.Furthermore, an additional square metre makes the average square meter price drop in poorer areas, while it increases in the wealthier areas.It thus seems like an efficient use of surface is prioritized in this segment, which should be considered in the provision of housing for this group.This is not unique for transitadjacent housing, but rather a characteristic of the market segment.Nevertheless, recognizing such differences may be of importance to ensure that transit-adjacent housing also caters for lower income groups.
It is interesting to note that while the distance to CBD estimates have the expected negative signs for three of the groups, it is positive for the Hyllie group, which contain neighbourhoods belonging to the lower price segments.Given that all observations in these samples are located in more peripheral areas of the city, this is interpreted as that there are other locational factors that are more important, since all observations require some sort of bus, car or bike transportation in order to go to the CBD.Being located closer to the CBD, within the areas studied, does not provide much extra accessibility value.
When comparing Tables 2 and 3, it seems clear that the average figures also hides information that become visible when considering development in different submarkets or neighbourhoods.Although for certain purposes the average figures may be enough, they may be of less relevance when considering the development of neighbourhoods or an integrated planning of transportation and housing.

Discussion
This paper aims to provide a better understanding of how effects of improved accessibility are distributed across a city.To address this issue, DID estimates of hedonic price models have been used to analyse household valuations of accessibility over different market segments.In line with previous research (e.g.Bohman and Nilsson, 2016;Dziauddin, 2019;Forouhar and Hasankhani, 2018), the findings from this study suggest that effects on property prices are not evenly distributed across market segments.Specifically, apartments in areas where prices and incomes are low are more affected than average.Thus, increases in accessibility can have substantial impacts on residents in these neighbourhoods.In other words, the willingness to pay for accessibility is high among the households buying houses in these areas.Typically, these are lower-income households that, to a large extent, rely on public transport and may not be able to purchase and travel by car.While condominium prices may not capture all householdsparticularly the ones who cannot afford buying their own homesowners in the lower price segments are typically not high income earners.From a policy perspective, this is an important recognition since it enhances the importance of providing good accessibility for households residing in socio-economically challenged neighbourhoods.
However, this finding also highlights the concern that increases in property prices may make transit-adjacent areas less affordable to the very same households that seem to benefit the most from improved accessibility (Padeiro et al., 2019).A second important discussion therefore refers to who benefits from the value increase in the affected areas.From a housing affordability perspective, the findings above can be interpreted as a double-edged sword.On the one hand, increasing prices make the neighbourhood less affordable for households moving in.On the other hand, the beneficiaries in this case, i.e. the owners of the condominiums, are also mostly low-income households.The results show that the higher returns partly can be understood as a catch-up effect, since the low-income areas of the control group have witnessed much lower price increases than the rest of the market.This implies that returns on investments for this group are on a level with the rest of the market, which is not the case for similar areas that have not had the same accessibility improvement.

H. Bohman
Ultimately, tenure and ownership are important factors for understanding distributional effects of land value increase.In this case the focus has been on condominiums.For rental housing, the effects will depend on to what extent land value uplift is reflected in rent setting.In the Swedish case, rents are negotiated and only partly dependent on market mechanism, and it is therefore not possible at this point to know what effects there may be.
The results also indicate that the valuation of housing characteristics differs between the different submarkets.In the lower price segments, an efficient use of space, in terms of more rooms rather than larger rooms, seems to be important.In order to cater for lower-income households, it is important to recognize such differences across segments.
Whereas price effects are expected to be more rapidly observable effects compared to more general neighbourhood changes (which is the focus in e.g.Nilsson and Delmelle, 2018), the findings from this study suggest that price effects may also take several years.The Hyllie station is an area with substantial greenfield development.Consequently, the opening of the station has implied several years of construction and considerable negative externalities in terms of e.g.noise and heavy traffic in the initial stages.Positive externalities have developed over time, as agglomeration increases together with local services and amenities.Previous findings suggest that the likelihood that a neighbourhood changes drastically as a result of improved accessibility is very low (Nilsson and Delmelle, 2018).
In summary, the results stress the importance of interaction between transport planners and other urban planners and developers.To create inclusive cities, the type of housing and factors such as size and number of rooms seem to differ between market segments and need to be considered early on in the process.
Given that accessibility is important for low-income groups, a possible alternative -should certain groups be negatively affected through higher prices -may be to target these groups directly through demand-oriented policies such as housing allowances or travel subsidies.Previous research indicates that low-income households benefit considerably when transit attributes, such as fare and time, are reduced (Hasnine and Habib, 2020).Although of course this may be context dependent, it does highlight the need to address housing and transportation jointly by policy makers.

Conclusions
This study is in line with previous findings that show that the effect of increased accessibility is not evenly distributed across market segments.Using DID estimates on price developments of condominiums around stations in Malmö, Sweden, the results show stronger effects for the lower price segments.Given that the property market is segmented, different control groups were identified for different neighbourhoods.The estimates for these findings are stronger than when compared to the market as a whole.
As a consequence, these results suggest that the new station can serve as a boost for property prices, thereby making housing in the lowest price segment, in particular, more attractive.This stresses the need for integrating transport and housing policies, and, from initial stages, taking the question of affordable housing into account.While public transport seems to be highly important in the lower price segments, the risk that transit-adjacent neighbourhoods become prohibitively expensive for the very same group should be taken seriously.In the case studied, however, the beneficiaries of increased housing are owners of apartments, for whom the price increase implies a better return on investment than for similar neighbourhoods in other places in the city.From a distributional perspective, the tenure forms of housing in the affected areas therefore matter.For rental housing, the potential capitalization of increased accessibility into higher rents may result in affordability problems for existing tenants.Complementary means, such as demand-oriented policies for housing or transport targeting vulnerable groups, could be options to consider.
Furthermore, the division into sub-samples also reveals that in the lower price segments the demand for smaller apartments with more rooms, rather than larger rooms, is interpreted as a demand for efficient use of surface.New construction should take the demand from lowincome households into account also in terms layout and planning of surface use.Preferences and needs, as well as tenure structures, may, howeve,r be context-specific.Thus assessing investments from a housing affordability perspective, in which risks are identified already at the planning stage would be one way to address the overall provision of affordable housing.
As a concluding note, the results stress the need for coordination of transport and housing planning.Moreover, special attention for future provision of affordable housing should be paid attention to in the initial stages.

Declaration of competing interest
None.

Fig. 1 .
Fig. 1.Overview map of Malmö and the City Tunnel stations.

Fig. 2 .
Fig. 2. Control groups for the different stations and neighbourhoods.The treatment, or catchment areas, are marked in red and control groups are marked in dark grey.(For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Table 2
Descriptive statistics, 2005Descriptive statistics,  -2019, for full sample and catchment areas.Catchment areas defined as within 1000m radius from the station.Real prices, base year 2005.

Table 3
Difference-in-difference results for all stations and each station respectively.Real prices, base year = 2005.Post treatment dummy = 2019.Control group: rest of the city.

Table 4
DID estimates for different market segments and control groups.Std.dev. in brackets.*** significant at 1% level, ** significant at 5% level.Control variables: size, number of rooms, monthly fee, d.Real prices, base year = 2005.