Effect of the North-South Line on Residential
Property Prices: Evidence from a Quasi-experiment
∗
Ties Ammerlaan
July 2017
Abstract
This research uses a quasi-experimental research design to estimate the effect of a metro line under construction on residential property prices in Amsterdam. The research design allows for a comparison of the residential properties around the metro line under construction, to a control group of similar properties. This control group consists of properties around a metro line which was in the same plan as the metro line under construction, but which will not be built. The research finds that the average effect of the metro line on property prices is negative. However, the effect is positive for properties located within certain neighbourhoods and within an 800-2000 metre range around a metro station. The results do not indicate an anticipation effect of potential future advantages and disadvantages, but they do suggest that the effect might become more positive after the line is completed.
Key words: metro network, property price, quasi-experiment JEL classification: H54, R32, R41
∗I thank Dr. Peter Berkhout for providing data on property prices and characteristics.
Addi-tionally, I thank professor Erik Plug for his supervision and am very grateful for his comments and suggestions. Ties Ammerlaan: Amsterdam School of Economics, University of Amsterdam, the Netherlands (email: t.b.ammerlaan@gmail.com).
Amsterdam’s newest metro line, the North-South Line, will open on the 22nd of
June 2018. This line stretches throughout the historical centre of the city to connect Amsterdam North and Amsterdam South. The municipality expects the metro to serve as a comfortable, safe and fast method of transportation for the increasing number of people living, working and relaxing in the city (Gemeente Amsterdam, 2017). However, the metro stations might cause a nuisance to those living nearby (Bowes & Ihlanfeldt, 2001). This research evaluates which effect dominates for home buyers; will residential properties located closer to a metro station receive a metro-premium? Or will a metro station lead to lower property prices? Thus, the aim of this research is to estimate the causal effect of the North-South Line stations on property prices.
The literature identifies both advantages and disadvantages of a metro station. The main advantage is the increased proximity to locations which provide a benefit for the home buyer, e.g. the city centre (Zhang, 2009). However, a metro station can also lead to noise, pollution, and crime. The metro itself creates noise and pollution, but so do its users. An area which was quiet before, can start buzzing with people after the opening of the metro station (Diaz, 1999). Additionally, criminals could be lured by the improved access of the area to outsiders (Bowes & Ihlanfeldt, 2001). A final disadvantage of a North-South Line metro station are the negative externalities caused by the construction process, such as noise pollution (Cobouw, 2004).
The total effect of the North-South Line metro stations on property prices de-pends on whether the advantages or disadvantages dominate and whether these are translated in a price change. If the positive effect of (expected) increased proxim-ity dominates, property prices could increase. In contrast, if the negative effect of (expectations of) noise, pollution and crime dominates, property prices should be lower.
The available literature is inconclusive on which effect dominates. Mohammad et al. (2013) performed a meta-analysis of 23 recent studies on the impact of rail invest-ments on land and property values, including 102 observations. They demonstrate that there is not a dominant effect in one direction; some research finds a positive effect, some no effect and others a negative effect. They argue that the difference in results might originate from differences in accessibility of the areas studied.
The property price effect of the North-South Line is of interest for the municipality since it reveals how the public intervention is evaluated by home buyers. First, if the benefit of the proximity of the metro dominates, home buyers should be willing to pay a premium for a residential property located closer to a metro station. Second, if home buyers are indifferent about the metro station or its benefits are balanced out by its negative externalities, no property price premium should be observed. Third,
if the negative externalities of a metro station dominate, property prices closer to a metro station should be lower. Given that the advantages and disadvantages translate in price changes.
Additionally, the municipality benefits from the impact analyses for various sub-populations. The literature suggests that the impact of a metro station can vary greatly between neighbourhoods, price categories, and the range within which a metro station is located. The analysis reveals in which subpopulations the munici-pality generates the largest positive impact, but also which subpopulations do not benefit or even experience a welfare loss. This indicates where similar interventions in the future can have the largest impact, but also where extra attention might be required.
Finally, the analysis estimates when the total impact of the public intervention has developed to its full potential. The theory suggests a different impact before, just after and a couple of years after completion. Estimates of the impact for these different periods are valuable for the evaluation of the public intervention, but also for real estate market participants who want to employ future increases or decreases in property prices.
Most research uses (a form of) hedonic price modelling to estimate the effect of public transport on property prices. Hedonic price models approach a property as a bundle of characteristics, in which each characteristic adds its own value to a property. To quantify what each characteristic adds to the total value, sufficient comparable properties and information on their characteristics should be available (Francke, 2014). A widely-used modification of the hedonic price model is a geograph-ically weighted regression. This strategy allows for a spatially varying relationship (Lu, Charlton, & Fotheringhama, 2011).
Hedonic price modelling potentially suffers from omitted variable bias. If a vari-able influences both the outcome and the explanatory varivari-able, but is omitted from the regression, it biases the estimated effect. If there is such an omitted variable, a causal effect cannot be established. Hedonic price modelling aims to correct for this problem by explaining as much variation as possible by including as many ex-planatory variables as possible. To further limit the interference of omitted variables, location-, time-, and entity-specific effects are included (Velthuijs, 2016, p. 25). For this strategy to be successful, one must assume that the omitted variables do not influence both the location-/time-/entity-specific effect and the outcome variable. This assumption is questionable.
Additionally, hedonic price modelling potentially suffers from simultaneous causal-ity. A causal interpretation of a regression estimate is wrong if simultaneous causality is present; the explanatory variable influences the outcome variable and vice versa.
To overcome this problem, researchers take a partial equilibrium perspective and assume that marginal property buyers do not have a significant effect on the ex-planatory variable and vice versa (Ohlsen, 2016, p. 9). However, this assumption is also questionable.
An instrumental variable design can correct for both risks, but is difficult to exe-cute in the context of property price research. An instrumental variable methodology requires a variable which is strongly correlated with the explanatory variable, but is not correlated with other factors and only affects the outcome variable through the explanatory variable. In the context of this research, it is a challenge to discern such a variable. Research into the effect of rail investment on property prices using this methodology could not be identified.
A second methodology is that of the repeated sales method. The repeated sales method evaluates the change in the price of the same property over a certain time frame. This methodology is used by various property price indexes, such as the house price index of the Central Bureau of Statistics (Centraal Bureau voor de Statistiek, 2014). The main critique of this method is that it only takes into account properties which are sold at least twice over the reference period. This sub-sample might be too small and unrepresentative of the whole housing market, which includes relatively more expensive, newly built properties (Nagaraja, Brown, & Wachter, 2010).
This research pioneers a more robust methodology: a quasi-experiment. To de-termine a causal effect, it compares property prices at a certain distance from the North-South Line to the prices of similar properties at the same distance from a counterfactual line. This counterfactual line was part of the same plan for a metro network as the North-South Line in 1968. Additionally, this research argues that, conditional on various control variables, a parallel trend of the development around these two lines exists and the North-South Line was randomly selected to be built. This research design corrects for omitted variables, and simultaneity. The differences in price development between the two groups can then be regarded as the effect of the North-South Line on property prices. A literature analysis suggests this methodology has never been used to estimate the effect of a rail investment on property prices.
The main finding of this research is that the overall effect of the North-South Line on property prices is negative. However, a neighbourhood analysis reveals that a benefit dominates in all relevant neighbourhoods except for Amsterdam West. Es-pecially in Amsterdam North, the positive effect is large. Additionally, relatively less wealthy home buyers evaluate the metro more positively than their wealthier coun-terparts. Finally, negative externalities seem to dominate within walking distance of a metro station, while benefits prevail outside this radius. This research finds no support for home buyer’s anticipation of either positive or negative externalities of
the metro by home buyers. However, an analysis of the constructed South-East Line suggests that the effect might become more positive after completion. This could indicate a price-discount due to the nuisance created by the construction process. These results should be interpreted in the context of potential herd behaviour and loss aversion of home owners.
The paper is structured in the following way. In Section I, the historical develop-ment of the metro network in Amsterdam is described. This section aims to provide context on the topic. Additionally, the theoretical framework explaining the effect of a metro station on property prices is discussed. Section II focusses on the datasets used in this research. The origin, interpretation, and transformation of the outcome, explanatory, and control variables are described in detail. Subsequently, the em-pirical method is discussed in Section III. In this section, the estimation method is discussed and assumptions are identified to estimate a causal effect of the North-South Line on property prices. The outcomes of the empirical method are presented in the results section, Section IV. It will begin with an overview of the baseline analysis, after which the results are reported for subgroups and their robustness is assessed. The discussion in section V addresses the shortcomings of the research and gives suggestions for future research. Finally, the research closes with section VI, the conclusion. The results are summarized and policy implications are discussed.
I
Context
A
History of the Metro Network
In 1968, the municipal executive of Amsterdam agreed to the construction of a metro network to maintain and improve the accessibility of the city. A metro was seen as fast, safe, and comfortable and had a large capacity. Additionally, construction costs were expected to be as high as those for a four-lane road and it was thought not to require large scale demolition of the city centre (Plan Stadsspoor, 1968, p. 2). The metro network was presented to the inhabitants of Amsterdam in the Plan Stadsspoor (City Transit Plan) brochure in May 1968 and agreed to by the city council later that year (Schomakers, 2016). Illustration 1 displays the proposed metro network.
It was decided to construct the Southeast Line first. First, the lines from the city centre to the outskirts were deemed most important. Second, the fewest expro-priations and demolitions were necessary to construct this line. Third, the line had the shortest (more expensive) underground portion. Finally, large scale residential projects were being developed at the end of the line in Amsterdam Southeast (Plan Stadsspoor, 1968, p. 3).
Due to large-scale public opposition, the plans for the metro network were aban-doned in 1975. The metro’s construction method required large scale demolition at the surface. It was necessary to sink large caissons, after which a city road was built on top. Additionally, setbacks caused cost overruns. This led to riots and even the failed bombing of a subway station in 1975. The Southeast Line was still under construction in 1975 and it was decided to finish the line, but to abandon the plan after that (Schomakers, 2016).
The municipality constructed the ring metro line in the 1990s. The Plan Stadsspoor (City Transit Plan) was abandoned and the metro was a taboo in the city council. However, the city still faced the problem of how to ensure accessibility. The munic-ipality decided to build a slightly adjusted version of the ring line first presented in the Plan Stadsspoor. The construction of this line was relatively easy because it did not have any underground portions (Schomaker, 2016).
In 2002, the city council decided to construct a third metro line: the North-South Line. The problems the city faced had not changed and again a metro line seemed to combine capacity, safety, comfort, and speed. However, the city council was hesitant to start an extensive metro project and therefore it organized a referendum (GVB, 2000, p.1). The referendum was declared invalid because voter turnout was too low and this cleared the way for the final decision to construct the metro in 2002 (Volkskrant, 1997).
The construction of the North-South Line was plagued by complications. The decision-making process was criticized, there were large cost overruns, and funding was uncertain. Additionally, mistakes were made during the construction, which damaged buildings in the historical centre. In the summer of 2008, multiple buildings on the Vijzelgracht sagged significantly due to a fault in the construction work. This formed the reason for the city council to pause construction. A commission advised the city council to continue so as not to trivialize all past investment, and on the 4th of July 2009, the city council agreed. The expected opening date was postponed
eight times, but is currently is the 22nd of July 2018 (Schomakers, 2016). Many historical parallels can be drawn between the construction of the Southeast Line and the North-South Line.
B
Theoretical framework
The dominant theoretical framework for explaining the effect of a metro station on property prices is expected utility theory. The basis of expected utility theory is that a home buyer maximizes its expected utility when bidding on a property. It assumes that the home buyer has well-defined preferences, considerable knowledge,
and sophisticated information processing capabilities. This means that the home buyer can enumerate its range of options, assess what utility they will deliver, and assign probabilities to all possible future states of the world (Marsh & Gibb, 2011, p. 219). Thus, the effect of a metro station on property prices originates from a change in utility a home buyer expects to receive from a property, due to either the expected increase of accessibility or the expected increase of noise, pollution, and crime.
Behavioural economics offers an alternative theoretical framework which seems to be more in line with reality. Central to this theoretical framework is the concept of “bounded reality”. The home buyer is limited by the information it has, its mind’s cognitive boundaries, and its finite amount of time (Whittle et al., 2014, p. 18). Given these limitations, a home buyer uses several “shortcuts” to come to a decision, instead of weighting all the options. Dellavigna (2009) provides a meta-analysis of evidence against expected utility theory which shows that expected utility framework is inadequate to explain decisions in the face of limited information and complexity. Additionally, Marsh and Gibbs (2011) argue that their meta-analysis shows that behavioural economics offers better insights.
Whittle et al. (2014) provide an overview of a wide range of insights from be-havioural economics which are relevant for the housing market. In the context of this research, two insights are especially relevant: herd behaviour (1) and loss aversion (2).
(1) Herd behaviour In the face of limited information and complexity, humans seem to partly base their buying or selling decision on what others do, as in behaving like a herd. Especially relevant is the role of the media in sending ‘the herd’ in a particular direction. If local news agencies report positively on the construction of the metro station and emphasizes its benefits, home buyers can form irrational expectations and are more likely to pay a metro premium (Whittle et a., 2014, p. 21). This means that an observed metro premium or metro discount, does not necessarily originate from rational expectations on advantages and disadvantages.
(2) Loss aversion Home owners are generally unable to coop with selling their property below the purchasing price. However, perfectly rational sellers would ignore the purchasing price when assessing the current market price. This explains a certain price stickiness of property prices. In the context of this research, this means that property prices might not decrease, even if the disadvantages of the metro station dominate (Whittle et a., 2014, p. 22).
II
Data
This research is based on a dataset consisting of 99,316 property sales and their characteristics in the province of North-Holland. The transactions were scraped from funda.nl, a website presenting all properties on sale in the Netherlands. The data consists of properties that were unlisted from this website between the 1st of
July 2014 and the 31st of March 2017: a total timeframe of 33 months.1
A
Outcome variable
The outcome variable of interest is the natural logarithm of the price of a residential property. This transformation is described as the best fit by most authors in the literature, e.g. (Bowes & Ihlanfeldt, 2001) and (Weinberger, 2001). Table 1 illustrates that in the final dataset prices (1) range from e62,007 to e6,776,694 with a mean of e356,112.
B
Explanatory variable
The explanatory variable of interest is the distance between the sold property and a metro station in kilometres. To calculate this distance, first the location of the sold property and the metro had to be determined, then the distance between these two points had to be calculated. Table 1 shows that the average distance (2) is 1.3 km.
The property price dataset had to be matched with a dataset of Dutch addresses to determine the location of the sold property. Longitudes and latitudes are nec-essary to uniquely determine the position of the property on a map. However, the property price dataset did not include coordinates. Therefore, the addresses of the sold properties had to be matched to their counterparts in a Dutch addresses dataset that also included coordinates. The addresses dataset is a complete overview of all addresses in the Netherlands and consists of more than 8.8 million addresses. The dataset is updated daily. To perform this matching process big data technology, the graph database NEO4J, was used. A graph database allows for a relatively easy and fast matching process because it stores data as relationships, not as strings. For example, all properties located within the same street are connected to each other in the graph database. When matching a property within that same street, the graph
1The date on which property listings were removed is not equal to the date of the sale of the
property. For example, the data includes the sale of a house on the 1stof September 2011, property was unlisted almost 4 years later, on the 25thof March 2015. However, this is an outlier. Ninety-nine
database only reviews the properties connected to each other within that street. A string database would review all properties.2 Not all properties could be matched and 7.8% of the data had to be excluded. A postal code, house number, and possibly an addition form the unique identification of property, which is necessary to match the two datasets. An addition to a house number serves to uniquely identify prop-erties with equal postal codes and house numbers, e.g. 221a and 221b. Six point five percent of all transactions did not specify a house number and/or postal code and thus could not be matched. It is unclear why the remaining 1.3% could not be matched.
Five point three percent of the observations had to be adjusted to match with their counterpart in the addresses database. Of all properties with an addition (23.6% of the total), only 77.6% could be matched within the first try. The reason for this was that the additions in the transaction database were incorrect. To be able to still use the data, the house numbers with additions which could not be matched were matched to the location of the house with the same address, but the first addition in line (ordered alphabetically or additive). For example, the property with house number 18hs could not be matched with its counterpart and therefore it was matched to the property with house number 18a. An analysis of the average difference between houses with the same address but other additions reveals that the difference between their locations is approximately one meter. Since the difference in location is small, this does not form a risk for the validity of the research. After this process, only 292 observations with an addition could not matched.
The exits of the metro stations form the basis of the analysis of the metro’s location. Exits of metro stations can be located up to 500 m apart and therefore all exits were analysed separately. This can lead to a situation in which some metro stations have up to six exits. The metro stations of the counterfactual line were described in the Plan Stadsspoor (City Transit Plan) (1968, p. 5). The plan does not include exits and thus the centre point of the station on the map was used. For the North-South Line, the exits as described on the project’s website were used (Gemeente Amsterdam, 2017). To determine the coordinates of the exits, they were identified on an interactive map using Google Earth (Google Earth, 2017).
The graph database was also used to calculate the distance in kilometres between the residential property and the metro station. Since 60 exits were identified and 91,569 transactions were uploaded, the graph database calculated 60! * 91,569 dis-tances. The literature suggests that the extent of the influence of the metro can range up to 2,000 m (RICS Policy Unit, 2002). Therefore, only properties with a
2A more detailed introduction to graph databases in general and NEO4j in particular, can be
distance equal to or less than 2,000 m were analysed.
In the resulting dataset, the same transaction can appear multiple times. A property can be located within a 2 km radius of several metro station exits. Each of those distances was saved as a separate observation, and thus a single transaction can appear multiple times: x km from metro station A and x km from metro station B. This increases the dataset to 419,295 observations.
C
Control variables
To ensure sufficient observations some control variables were manipulated. There was no observation which had information on all the control variables available in the dataset. To keep all observations in the analysis, missing numerical data was replaced by 0 and missing categorical data was replaced by a separate category. For numerical data, a dummy was created to indicate whether the data was manipulated. This manipulation made the interpretation of the coefficients of the control variables unfeasible, but this is not a problem because these were of interest for this research. This research assumes that values are missing randomly. The descriptive statistics in Table 1 describe the manipulated variables.
The dataset includes various control variables which can be grouped into several categories according to Francke (2014). Table 1 provides an overview of the mean, standard deviation and minimum and maximum values of the control variables. The dummy control variables are also included in the descriptive statistics. The reason for this is that the mean can be interpreted as the percentage of observations which meet the dummy’s requirement.
The first type of control variable is related to conditions of sale of the property: whether the sale was kosten koper (costs-to-buyer) (3) or vrij op naam (deed-in-hand) (4).
The second type of control variables are related to the legal rights regarding the property. This includes a dummy variable for the ownership of the property, whether there is full ownership (5), erfpacht (leasehold) (6), or another ownership situation (7). Related to ownership, the monthly contribution to the owners’ association of an apartment complex in euros (8) (VvEMP).
The next set of control variables relates to a property’s characteristics. First, this includes a property’s size: building-bound outside space in m2 (9), external outside space in m2 (10), plot size in m2 (11), and volume in m3 (12). Plot size
was assumed to be 15 m2 at a minimum. The median height was assumed to be
2.5m, and therefore the minimum volume was assumed to be 37.5 m3 . Lower values were marked as missing. The number of bathrooms (13) and bedrooms (14) were
also included in the regression, as well as the floor on which the property is located (15) and the number of floors within the property (16). The number of floors in the property was assumed not to be greater than eight. Information on the year of construction (17) was also available. The year of construction includes residential properties with a range of building years. For these, the middle of the range was used in the analysis.
The energy label of a property (18-25) is used as a proxy for the type of heating, insulation and water heating. The energy label is used as a proxy because it has significantly fewer categories than the other three variables combined. If the specific variables were used, this would introduce an additional 544 dummy variables in the analysis. The energy label combines information on the heating of the house and water and the type of insulation and is therefore a good proxy for the others (Milieu Centraal, 2017).
The final category of property characteristics is related to the type of property. The dataset contains an indicator of whether the property is an apartment (26) or house (27). Additionally, there are two variables which describe the property type in more detail, with 177 types of properties and 93 types of rooves. Section III discusses why these were not included in the regression.
III
Empirical method
The empirical method used to estimate the effect of the distance from the metro station on the price of a property was ordinary least squares (OLS) in a quasi-experimental setting.
A
Endogeneity
The OLS estimator of the effect of distance from the metro station on the price of property can potentially be biased by endogeneity. Endogeneity is the phenomenon in which the error term and the explanatory variable, the distance from a metro station, are correlated. If this is the case, the OLS estimator is biased because it is not solely distance from a metro station which explains the variation in the price of the property, but also the error term. Endogeneity can potentially originate from omitted variables (1), simultaneity (2), and measurement errors (3).
(1) Omitted variables The model potentially suffers from endogeneity originat-ing from omitted variables. There could be an unknown third variable affectoriginat-ing both
the presence of a metro station and property prices. This third variable is included in the error term and affects both the outcome and the explanatory variable. An omitted variable can bias the OLS estimator both upwards as downwards. For ex-ample, the municipality might decide to build metro stations in pleasant areas. The pleasantness of an area both increases property prices and makes it more likely for a metro station to be located there. The OLS estimator of the effect of the distance from a metro station would then biased upwards, since it also captures the benefit of the pleasant area. In contrast, the OLS estimator can also be biased downwards. The municipality might build metro stations in deprived areas, in an attempt to improve its quality. The low quality of the area reduces property prices, but makes it more likely for a metro station to be located there. This biases the OLS estimator downwards.
(2) Simultaneous causality Simultaneous causality can also create endogene-ity. Simultaneous causality means that the chance of a metro being present influ-ences property prices, but property prices also influence the chance of a metro being present. A change in the error term is then passed through to the distance from the metro station via the price of property. Thus, the error term and the explanatory variable are related, which creates bias. Simultaneity in both directions seems likely in the context of this research. Real estate developers could anticipate a metro sta-tion by building more expensive or more inexpensive properties and this could in its turn increase or decrease the chance of a metro station being built in that area. A change in the error term then influences property prices and in its turn also the chance of a metro station being present.
(3) Measurement errors Measurement errors form a potential risk for this re-search. Measurement errors means that an independent variable is measured with a certain amount of noise. The noise then influences both the explanatory variable and the error term. Data is often measured with a reporting error or a coding error. For example, it is challenging to precisely measure the volume of a property. Volume is therefore potentially measured with a certain amount of noise and this can create endogeneity.
B
Quasi-experiment
Both omitted variable bias and simultaneity bias can be avoided with the use of an experiment. In an experimental setting, a property is randomly assigned to one of two groups: a treatment group with a nearby metro station and a control group
without. Since the properties are randomly assigned to a group, one can safely assume that the treatment group is similar to the control group. One can then also safely assume a parallel trend in the development of property prices in the two groups in the absence of treatment. A comparison between the two groups then reveals the effect of the treatment: the effect of the metro on property prices.
The Plan Stadsspoor (City Transit Plan) allows for a experiment. A quasi-experiment is an quasi-experiment which lacks the element of random assignment. Instead, it uses an assignment method which behaves as if there were random assignment. This research uses residential properties around the West Line, a metro line which was proposed in the Plan Stadsspoor (City Transit Plan) (1968) but never built, as a control group for properties around the North-South Line.
Given certain assumptions, residential properties around the never-built West Line can serve as a credible control group for properties around the North-South Line. Firstly, the two groups of residential properties should be comparable at baseline. Secondly, the parallel trend assumption should hold: the properties in both groups should have developed similarly if the North-South Line was not built. Thirdly, the North-South Line should have been randomly selected to be built compared to the West line. There should be no selection bias.
The evidence for the assumption that the situation around the North-South Line stations and the stations around the West Line were similar in 1968 is mixed. One can reasonably assume that the areas around the North-South Line and its stations in its 1968 form are similar to those around the never-built West Line. Namely, both groups possessed the characteristics to be eligible for a metro station. Figures 1 and 2 reveal that the North-South Line in its 2002 form closely follows the trajectory of the line proposed in 1968. This is evidence for the similarity of the North-South Line and the never-built West Line. Additionally, the situation around the North-South Line and the West Line did not induce it to be built in 1968, further supporting the similarity of the two areas. However, out of the eight stations on the current trajectory of the North-South Line, only six were present in the 1968 plan. Thus, the areas around two stations are not equal to the areas around the 1968 line. This does not support the assumption of similarity at baseline.
Furthermore, it is questionable whether the parallel trend and random selection assumption are met. Forty-nine years after the Plan Stadsspoor (City Transit Plan), only the North-South Line is being constructed, and not any of the other proposed lines. While there has been discussion of a metro line from Amsterdam Centraal to Zaandam via the western harbour (AT5, 2017), there is no discussion on constructing the never-built lines. This suggests that the areas around the future North-South Line developed new characteristics which promoted the construction of a metro line.
This provides evidence against both the parallel trend and the random selection assumption. An analysis of the effect of the control variables on the likelihood that a property is located near the North-South Line and not the counterfactual West Line suggests that measurable characteristics do differ between the two groups. However, this research has assumed that the omitted variables in both areas did not develop differently. Therefore, random selection can be assumed to be conditional on certain control variables. The implications of not fully meeting these assumptions will be addressed in the discussion in Section V.
The West Line is the never-built line that is most comparable to the North-South Line. The main reason for this is that no other forms of public transport were developed on the same route to compensate for not building the metro line. If a tram line had been built to replace the never-built metro line, the control group would also have received a ‘treatment’, namely a tram line. The interpretation of the results would change from the effect of the metro alone on property prices to the extra effect on property prices of the metro compared to a tram. Figure 2 illustrates that tram 17 closely follows the track of the counterfactual West Line. However, tram 17 was built before the Plan Stadsspoor (City Transit Plan) and thus it was already extant at baseline before the random selection.
To correct for endogeneity originating from serial correlation, standard errors were clustered at the transaction level. By clustering the observations at the transactional level in the regression, serial correlation was corrected for. This way, the correct standard errors are displayed.
An additional risk for the regression, the bad control problem, can be largely mitigated by excluding all properties built after 2002. A bad control problem exists if control variables that could be outcome variables themselves are included in the analysis. In the context of this research, property characteristics could be influenced by the metro itself. A metro station could attract another type of residents to a neighbourhood. A developer can then decide to develop especially for this new type of resident, for example by increasing the size of properties. Including size as a control variable in the analysis is then wrong because it is an outcome variable in itself. Characteristics of existing properties, such as the number of rooms, are not likely to change due to the metro. In October 2002, the decision was made to build the metro. Thus, to minimize the risk of the bad control problem, all properties built after 2002 were excluded from the analysis.
C
The model
The main model used in this analysis compares the prices of comparable properties at comparable distances from the counterfactual West Line and the North-South Line. The model aims to estimate the new line’s additional effect on property prices, compared to the metro line which was not built. This is captured in the interaction term. The model looks the following:
ln (pit) = βo+ β1di+ β2di∗ N SLi+ β3Ci+ λt+ γN BHi+ i (1)
in which pit is the price of a property, β0 a constant, β1 the OLS estimator of
distance, dithe distance between the property and the (counterfactual) metro station,
and β2 the OLS estimator of the extra effect of distance from the North-South Line.
NSLiis a dummy variable which equals 1 if the distance is to the North-South Line.
Cirepresents a vector of all control variables discussed in Section II. λt are time fixed
effects which allow for market conditions prevailing at the time of sale, given that these market conditions affect all properties in the same way. NBHi is a vector of
neighbourhood dummies, which allow for differences between neighbourhoods which are constant over time, such as demographics. Finally, i is the error term. The
subscript i and t indicate a specific observation and specific time period respectively. The second model allows for non-linearities in the effect of distance on the natural logarithm of price. The meta-analysis of the literature by Mohammed et al. (2013, p. 167) suggests that there are two ranges for which a different effect can be expected. First, within a range of 200 metres, the negative externalities of the metro station might be the largest. Second, within walking distance, approximately 800 metres, the positive effect is expected to be the largest since walking is expected to be the main mode of transport to a metro station. Three dummy variables were introduced: whether the distance is between 0-200 metres (D1i), 200-800 metres (D2i) or bigger
than 800 metres (D3i). These replace the distance variable in the model.
ln (pit) = βo+ β1D1i+ β2D2i+ β3D3i+ β4D1i∗ N SLi+ β5D2i∗ N SLi+
β6D3i∗ N SLi+ β7Ci+ λt+ γN BHi+ i
(2) The third model allows for heterogeneity of the effect by neighbourhood. Since the effect of a metro station is expected to be dependent on the initial accessibility of an area before the construction of the metro (Mohammad, Graham, Melo, & Anderson, 2013), the effect is expected to differ per neighbourhood. To estimate the effect in different neighbourhoods, interaction terms were created with the use of
the neighbourhood dummies represented by the vector D1i. These interaction terms
indicate the effect of the North-South Line within a certain neighbourhood and allow for a statistical comparison.
ln (pit) = βo+ β1di∗ N BHi+ β2di∗ N SLi∗ N BHi+ β3Ci+ λt+ γN BHi+ i (3)
Model four also allows for heterogeneity in the effect, but by year. The North-South Line is not completed yet, so it could be that the effect will change as it nears completion. This would reveal any anticipation effects of home buyers. Therefore, interaction terms indicating the effect of the North-South Line by year are created. Yt indicates a vector of year dummy variables ranging from 2013 to 2017.
ln (pit) = βo+ β1di∗ Yt+ β2di∗ N SLi∗ Yt+ β3Ci+ λt+ γN BHi+ i (4)
Also a quantile regression of the main model was performed to estimate het-erogeneity in the effect by price category. Since price is the outcome variable, the effect of distance on different price categories should be analysed using a quantile regression. The quantile regression estimates the effect at the 1st, 2nd, 8th, and 9th decile. The transaction price of a property provides an indication for the wealth of the homeowner. Thus, analysing the effect for different property price categories allows for an estimation of the effect on individuals with different levels of wealth. The literature is not conclusive regarding for which group the effect is greatest. For example, Wang et al. (2015) estimate it to be the biggest for relatively wealthy home owners, while it is usually argued that public transport is used most by those who are relatively less wealthy (Paulley, et al., 2006).
Models five and six further extend the model and compare the effect of the North-South Line to the effect of the North-Southeast Line. This can shed light on the potential effect of the North-South Line after completion. The literature indicates that for most rail investments, full impact on property prices has developed itself a couple of years after completion. Full impact has not yet developed before completion because home buyers do not anticipate the metro’s positive and negative externalities perfectly. Right after completion the system needs to stabilize first, meaning that people require time to get used to the metro Mohammad et al. (2013, p. 168).
The Southeast Line, rather than the Ring Line, is used as a completed reference line for the North-South Line because it is more comparable. Both the North-South Line and the Southeast Line have underground trajectories through the historical centre of the city. Additionally, both are routed to neighbourhoods relatively far from
the centre: Amsterdam Southeast for the Southeast Line and Amsterdam North for the North-South Line. The other existing line, the Ring Line, does not share these characteristics. It must be noted that the Southeast line was built first in 1968. This does not support the similarity at baseline assumption, and thus the comparison between the two lines requires extra care.
An interaction term with a dummy variable indicating a distance to a station of the Southeast Line, SELi , was introduced to the model in the same way this was
done for the North-South Line. Including both in model five, allows for a comparison of the two effects.3 In model six, neighbourhood interaction effects are introduced
similarly to model three to compare the effects per neighbourhood.
ln (pit) = βo+ β1di+ β2di∗ N SLi+ β3di∗ SELi+ β4Ci+ λt+ γN BHi+ i (5)
ln (pit) = βo+ β1di∗ N BHi + β2di∗ N SLi∗ N BHi+ β3di∗ SELi∗ N BHi+
β4Ci+ λt+ γN BHi+ i
(6)
IV
Results
The average effect of proximity to a North-South Line station is estimated to be negative: house prices are 1% higher for every kilometre further away from a metro station. Thus, overall the negative effect of the metro stations dominates. This is indicated by the coefficient of the interaction term in the first column of Table 2. The reference group is similar properties located at a similar distance from the counterfactual line. The effect should be treated as the average treatment effect (ATE).
Additionally, the coefficient of distance indicates that property prices are 2.5% lower for every kilometre further away from an area where a metro station is or could have been located. This suggests that these areas share unknown characteristics which in general lead to higher property prices. This is consistent with municipality building metro stations in areas which are already relatively prosperous. The nega-tive coefficient does not suggest that the municipality uses the instrument of metro stations to revitalize relatively less prosperous areas.
3All houses built after 2002 were excluded to avoid the bad control problem. Following the same
reasoning, all properties built after 1968 should be excluded from this analysis. This was not done, because it excludes 60 years of residential development and 114,209 observations.
Time fixed effects especially influence the coefficient. When neighbourhood mies are excluded, the coefficient changes only slightly, but when the monthly dum-mies are excluded, the coefficient jumps up to 0.041. Column 4 of Table 2 demon-strates that the measure taken to evade the bad control problem change the co-efficient. When properties constructed after 2002 are included, the effect of the North-South Line decreases to an extra 0.8% per kilometre further away from a station.
A
Sub-samples
Table 3 suggests that the effect of the metro is negative within an 800-m range and positive thereafter. The interpretation of the coefficients in Table 3 differs from Table 2. If a property is located within a 201-800-m range of a North-South Line metro station, the price is 2.7% less than a comparable property located within a 201-800-m range of a never-built station. The difference between all consecutive coefficients is significant. Thus, the results suggest that the effect of the North-South Line is most negative within a 200-m range. However, the variable/observation ratio might be too low for the result to be robust. Within 201-800 m, there also seems to be a negative effect, but for distances between 800-2000 m, the effect is positive. This does not confirm the theory that the largest positive effect is found within walking distance of a metro station.
Table 4 illustrates that there is a trichotomy in the effect of the North-South Line within neighbourhoods. Firstly, North-South Line stations have a slight positive effect on property prices in Amsterdam Centre, Amsterdam East, and Amsterdam South, 0.8%, 2.4%, and 1.2% less respectively for properties located a kilometre away from a station. The difference between the coefficients is not significant at the 5% level. Secondly, the positive effect of a metro station on property prices in Amsterdam Noord is significantly larger: prices are 5.4% lower for properties located a kilometre away from a station. Thirdly, the North-South Line has a relatively large negative effect in Amsterdam West: prices are 5% higher for properties located a kilometre further away from a North-South Line station.
The large positive effect of the North-South Line in Amsterdam Noord is con-sistent with the theory of the largest positive effect in areas with the lowest initial accessibility. The North-South Line was built partly because accessibility of Amster-dam North was relatively low (Gemeente AmsterAmster-dam, 2017) and the largest positive effect is estimated in Amsterdam North. However, no evidence could be found for al-ready high accessibility in Amsterdam West, which would explain the negative effect. The difference in average distances between residential properties and North-South
Line stations does not seem to explain the difference in effect between neighbour-hoods. Table 5 reveals a clear dichotomy between neighbourhoods with an average distance of around 1.25 km and those with an average distance of around 1.67 km. However, it is not the case that this dichotomy relates to Table 4’s coefficients. Both Amsterdam West and Amsterdam Oost have a relatively high mean distance, but the effect is large and negative in Amsterdam West and slightly positive in Amsterdam Oost. This further suggests that the differences in initial accessibility among the various neighbourhoods might explain the differences in effect. The analysis by year does not indicate an anticipation effect. Table 6 displays the coefficients per year. The coefficients should be interpreted as the additional effect of the North-South Line compared to the control group within a certain year. The differences between the coefficients of consecutive years is never significant, thus a certain trend towards completion cannot be distinguished.
The quantile regressions indicate that the negative effect of the North-South Line is smaller for relatively lower property prices. Table 7 displays the results of the quantile analysis for the 1st, 2nd, 8th, and 9th decile. It indicates that for the 1st
and 2nd decile, the negative effect is relatively smaller than for the 8thand 9thdecile: 0.5% and 0.7% relative to 1.1% and 0.9% respectively. This confirms the theory that property buyers with relatively less wealth valuate the metro more.
A comparison of the effect of the North-South line and Southeast line indicates that the effect of the North-South line might become less negative after completion. Table 8 illustrates that the effect of proximity to the Southeast Line is a value increase of 0.5% for properties located a kilometre further away from a station. The difference with the effect of the North-South Line is significant. Since the Southeast Line is completed while the North- South is not completed yet, the results indicate that the effect could become less negative after completion. It must be noted that this comparison requires extra care because the assumption of similarity at baseline is not supported for the Southeast Line and the counterfactual line.
The difference between the North-South and Southeast lines cannot be explained by differences in geographical location. Table 9 indicates that the effects of the two lines differ substantially per neighbourhood, both in the direction of the effect as in their size. All the differences between the two coefficients are also significant, except for Amsterdam South (p=0.0638). It does not seem that differences in average distance between the neighbourhoods can explain the divergence. Table 6 reveals that the relative distance to a station does not differ that much for Amsterdam Centre, but the coefficients of the effect of the two lines in the city centre significantly differ. The mean distances in Amsterdam South differ relatively more, but the coefficients are not significantly different. Thus, no pattern can be distinguished.
B
Robustness analysis
The result is robust for an alternative specification of the research design in which only the six stations which were both in the current as well as the 1968 version of the North-South Line. Section III describes that only 6 out of 8 stations of the current North-South line match those in the original plan. In the baseline analysis, all stations of the current North-South Line are used so as not to exclude Amsterdam North from the analysis. However, the assumptions of similarity at baseline and of a parallel trend are most robust for the 1968 stations only. The second column of Table 10 illustrates that the results of the adjusted sample are similar to the baseline results.
Additionally, the results are robust for an alternative specification of the research design which secures for public investment as a potential confounding variable. The construction of the metro and its stations is accompanied by large public investments at Amsterdam Zuidas and Centrumgebied Noord. At Amsterdam Zuidas the city is developing a new central business district and in Centrumgebied Noord a new city centre for Amsterdam North is developed. The focal points of these public investments are the metro stations themselves. Therefore, an analysis of the effect of the metro stations at Amsterdam Zuidas and Centrumgebied Noord on property prices could also pick up the effect of the public investments. To correct for this risk, the two metro stations are excluded from the analysis. Table 10, column 3, demonstrates that this does not alter the results much. The negative effect changes from 1% to 1.2%.
Thirdly, the analysis is robust for another specification of the counterfactual line. Section III discusses why the West Line fits the role of counterfactual line best. However, to test the robustness of the results, the baseline analysis is replicated with the second East/West Line (the orange line in Figure 1) as a counterfactual line. The resulting coefficient is slightly higher, but still similar, as Table 10, column 4 indicates.
Lastly, the results change substantially when only the closest metro station to a property is taken into account. One could argue that solely the closest metro station to a property is relevant, since a home buyer will probably only make use of that specific metro station. The final robustness check was to limit the dataset to only include the distance to the closest metro station, not all metro stations within a 2-kilometre range. Table 10, column 5, illustrates that this change alters the results substantially. The effect changes signs and increases in size. Instead of 1% extra, 7.1% less for properties located a kilometre further away from a North-South Line station.
V
Discussion
Whether the assumptions for a valid quasi-experiment were met is debatable. Firstly, the assumption of similarity at baseline is not fully supported; two stations of the North-South Line differ from those in the original plan. Secondly, the parallel trend and random selection assumption were not completely met. Namely, the North-South Line is being built, while the others lines proposed in 1968 are not. This research must assume that the random selection was conditional on measurable control variables, but that the omitted variables did not change, in order for the methodology to hold. The quasi-experimental approach should therefore be considered with care in future research into the effect of public interventions on property prices. Before performing any quasi-experimental analysis, there should be robust evidence for its main assumptions.
Future research should focus on the various pathways through which the metro affects property prices. This research only evaluates the total effect, but the liter-ature indicates that there are various pathways through which a metro station can affect property prices, including proximity effects, noise, pollution, and crime. Ad-ditionally, herd behaviour and loss aversion could play a role. An analysis of the various pathways should be able to disentangle the total effect and pinpoint where potential positive and negative effects originate. This is valuable information for the municipality if it wants to maximize the impact of its public intervention.
Future research should also extend the time frame of the research, both into the future as well as the past. The literature indicates that the effect of metros differs across three time periods: during construction, immediately after completion, and after people are used to the new line Mohammad et al. (2013, p. 164). If the research is replicated just after completion and in 2020, the effect in the latter two periods can be estimated. Furthermore, it could be that the nuisance associated with the construction is compensated for by the anticipation of a future benefit now, but was not in 2002. An analysis from 2002 to 2012 could shed light on the detriments associated with the construction process.
An extended time frame would also allow for the research to be replicated with only taking the closest metro station into account. The robustness analysis shows that the result of the research change substantially when only the closest metro stations are taken into account. One could argue that these are the most relevant, since these are the ones which will probably be used by the home owner. However, the limited size of the dataset does not allow for an analysis of the various subpopulations in this case. If the time frame is extended, the dataset increases in size, and the effects within these subpopulations can also be estimated.
To make the research more generalizable, a measure of accessibility should be included. Both the literature analysis and the results suggest that the effect is heavily dependent on the original accessibility of a neighbourhood. If one could include a measure of accessibility to the regression, the results could be generalised to neighbourhoods with a certain accessibility level. Including a measure of accessibility might also explain the difference in coefficients for neighbourhoods. The results indicate an influence of initial accessibility, but it would good if the research could incorporate this in a formal way. Ryan (1999) provides an example of how this can be incorporated.
A final suggestion for further research is to estimate the effect for populations other than home buyers, for example commercial property buyers. The current analysis is centred around residential property prices and thus evaluates how home buyers evaluate the North-South Line. To evaluate the full impact of the public investment, the effect on other groups should be evaluated as well. For example, the literature suggests that the effect is more positive for commercial properties, since these benefit more from the increased proximity than households do (Debrezion, Pels, & Rietveld, 2007).
VI
Conclusion
This research uses a quasi-experiment to estimate that the overall causal effect of proximity to a North-South Line station on property prices is negative. Properties located farther away from the metro are worth slightly more, plus 1% for every kilo-metre farther away from a metro station. This result is robust for various alternative specification of the model, such as another hypothetical line. The result suggests that the benefit of having a safe, fast, and comfortable method of public transportation close by in the future does not outweigh the detriment of the negative externalities caused by the construction of the metro and/or the additional noise, pollution, and crime expected in the future. However, it could be that these advantages and dis-advantages do not translate fully in a price change, given herd behaviour and loss aversion. Still, the results indicate to the municipality that the public intervention might not be successful for home buyers.
However, the subpopulation analysis reveals that the effect of the North/South line is positive in all relevant neighbourhoods except Amsterdam West and for prop-erties located outside of the walking distance range around a station. Additionally, the effect of the line is more positive for relatively cheap properties. This indicates to what subpopulations the municipality should pay extra attention to for compa-rable projects in the future if it wants to generate most impact. However, it also
highlights those groups which do not benefit from the public intervention and might require extra care. For example, since properties located relatively close to a station have decreased in value, the municipality could also focus on reducing the nuisance created by the construction process or the negative externalities of the metro and its users after completion.
This research did not uncover a trend in the effect of the North-South Line over time, but the results suggest that the effect will be more positive in the future. The difference between the effect in consecutive years is never significant. First, this could suggest that there is no anticipation effect of possible negative or positive externalities of the metro; one would then expect to see a trend in the estimate. Second, it could also suggest an increasing anticipation of negative externalities in the future, which is balanced out by a decrease in negative externalities due to the construction process.
Although there is no evidence for an anticipation effect, a comparison of the effect of the North-South Line and the Southeast Line suggests that the effect might become more positive after completion of the line. The effect of the Southeast Line is less negative and the difference between the two lines cannot simply be explained by a difference in geographical orientation. This suggests that either individuals only see the full welfare benefit of the line after completion, or that there is a significant decrease in price because of the negative externalities of the construction process. This does serve as an indication to the municipality that the full potential of the public intervention might only develop in the future. Additionally, the expected future price increase does leave room for real estate participants to buy properties which might increase in value around the North-South Line.
VII
References
AT5. (2017, June 27). Stad denkt aan forse uitbreiding metronetwerk bij aanleg Haven-Stad: metroring en Zaan-IJ-lijn. Retrieved from AT5:
http://www.at5.nl/artikelen/170533/stad-denkt-aan-forse-uitbreiding -metronetwerk-bij-aanleg-haven-stad-metroring-en-zaan-ij-lijn
Bowes, D. R., & Ihlanfeldt, K. R. (2001). Identifying the impacts of rail transit stations on residential property values. Journal of Urban Economics, 50, 1-25.
Centraal Bureau voor de Statistiek. (2014). Prijsindex bestaande koopwoningen: methodebeschrijving. Heerlen: Centraal Bureau voor de Statistiek.
Debrezion, G., Pels, E., & Rietveld, P. (2007). The impact of railway stations on residential and commercial property value: a meta-analysis. 35, 161–180. Dellavigna, S. (2009) Psychology and economics: evidence from the field, Journal of
Economic Literature, 47(1), 315–372.
Diaz, R. B. (1999). Impacts of rail transit on property values. Commuter Rail/Rapid Transit Conference, Proceedings, Toronto (pp. 66–73). American Public Transit Association.
Francke, M. K. (2014). Hedonic Price Models. Amsterdam: University of Amsterdam Business School.
Gemeente Amsterdam. (2017a, April 4). Noord/Zuidlijn metrostations. Retrieved from Noord/Zuidlijn: https://www.amsterdam.nl/noordzuidlijn/stations/ Gemeente Amsterdam. (2017b). Waarom de Noord/Zuidlijn? Retrieved from
Noord/Zuidlijn: https://www.amsterdam.nl/noordzuidlijn/informatie/aanleiding Google Earth. (2017). Google Earth. Alphabet Group.
GVB. (2000). Alles over de nieuwe metrolijn voor Amsterdam. Brochure. GVB. Lu, B., Charlton, M., & Fotheringhama, A. S. (2011). Geographically Weighted
Regression Using a Non-Euclidean Distance Metric with a Study on London House Price Data. Procedia Environmental Sciences, 7, 92-97.
Marsh, A., & Gibb, K (2011). Uncertainty, Expectations and Behavioural Aspects of Housing Market Choices. Housing, Theory and Society, 28, 215-235. Milieu Centraal . (2017, June 16). Download voorbeeld energielabel. Retrieved from
Energielabel: https://www.energielabel.nl/woningen/download-voorbeeld-energielabel/ Mohammad, S. I., Graham, D. J., Melo, P. C., & Anderson, R. J. (2013, April). A
meta-analysis of the impact of rail projects on land and property values. Transportation Research Part Policy and Practice, 50, 158-170.
Nagaraja, C. H., Brown, L. D., & Wachter, S. M. (2010). House Price Index Methodology. U.S. Census Bureau.
Ohlsen, O. (2016). The Impact of the Quality of Primary Schools on Housing Prices: An Analysis of Amsterdam. University of Amsterdam, Amsterdam Business School. Amsterdam: University of Amsterdam.
Paulley, N., Balcombe, R., Mackett, R., Titheridge, H., Preston, J., Wardmand, M., White, P. (2006, July). The demand for public transport: The effects of fares, quality of service, income and car ownership. Transport Policy, 13(4),
295-306.
RICS Policy Unit. (2002). Land Value and Public Transport: Stage 1 – Summary of findings. RICS.Office of the Deputy Prime Minister.
Ryan, S. (1999). Property values and transportation facilities: finding the
transportation-land use connection. Journal of Planning Literature, 13, 412–427. Schomaker, H. (2017). De geschiedenis van de metro. Retrieved from Het
lijnkleurensysteem van Amsterdam: www.schomakers.net/index
Velthuijs, B. (2016). Effect of refugee accommodations on house prices in the Netherlands. Master Thesis. University of Amsterdam.
Volkskrant. (1997, July 26). Amsterdam kan na referendum beginnen aan Noord-Zuidlijn. Volkskrant.
Weinberger, R. R. (2001). Light rail proximity: benefit or detriment? The case of Sante Clara County, California. Journal of the Transportation Research Board, 1747, 104–113.
Whittle, R., Davies, T., Gobey, M. & J. Simister (2014). Behavioural Economics and House Prices: A Literature Review. Business and Management Horizons, 2, 2, 25-28.
Zhang, M. (2009). Bus versus rail: meta-analysis of cost characteristic, carrying capacities, and land use impacts. Transportation Research Record: Journal of
Table 1: Descriptive statistics
Number Variable Mean Std. Dev. Min Max Outcome Variable (1) Ln (price) (e) 12.783 0.608 11.035 15.729 Explanatory Variable (2) Distance (km) 1.320 0.478 0.011 2.000 Conditions of Sale (3) Kk dummy 0.987 0.112 0 1 (4) VoN dummy 0.013 0.112 0 1 Legal Rights
(5) Full ownership dummy 0.481 0.500 0 1
(6) Erfpacht dummy 0.439 0.496 0 1
(7) Other form of ownership dummy 0.080 0.272 0 1 (8) Contribution VvE (e) 74.563 84.570 0 972 Physical Characteristics
(9) Building-bound outside space (m2) 9.143 17.295 0 675 (10) External outside space (m2) 2.528 4.606 0 300
(11) Plot (m2) 8.327 42.713 0 999 (12) Volume (m3) 248.408 160.614 0 999 (13) # of bathrooms 0.842 0.634 0 8 (14) # of bedrooms 2.038 1.185 0 16 (15) Floor (´etage) 1.959 1.945 0 22 (16) Floors (woonlagen) 1.828 1.370 0 8 (17) Construction year 1.927.312 56.516 1005 2017 (18) Energy label A dummy 0.059 0.236 0 1 (19) Energy label B dummy 0.075 0.264 0 1 (20) Energy label C dummy 0.106 0.308 0 1 (21) Energy label D dummy 0.064 0.246 0 1 (22) Energy label E dummy 0.067 0.250 0 1 (23) Energy label F dummy 0.030 0.172 0 1 (24) Energy label G dummy 0.116 0.320 0 1 (25) No energy label dummy 0.482 0.500 0 1 (26) Apartment dummy 0.936 0.244 0 1
Table 2: Effect of distance from a North-South Line station on ln property price Baseline No Location FE No Time FE Including after 2002 (1) (2) (3) (4) distance -0.025*** -0.026*** -0.045*** -0.024*** (0.002) (0.002) (0.002) (0.002) distance*NSL 0.010*** 0.010*** 0.041*** 0.008*** (0.001) (0.001) (0.001) (0.001) N 130891 130891 130891 141570 adj. R2 0.847 0.831 0.806 0.838
Note: OLS regression with ln(property price) as the dependent variable. The coefficient of distance is for the counterfactual line and the coefficient of the interaction term is interpreted as the extra effect realized by a North-South line station compared to a station of the counterfactual line. Distance in km. The regression includes, unless specified differently: conditions of sale dummies, type of ownership dummies, contribution to VvE (e), building-bound outside space (m2), external outside space (m2), plot size (m2), volume (m3), no. of bathrooms, no. of
bedrooms, no. of floors, the floor the property is located on, construction year, energy label dummies, type of property dummies, monthly dummies, and neighbourhood dummies. Robust clustered standard errors in parentheses. Significant at; * 5% level; ** 1% level; *** 0.1% level.
Table 3: Effect of range of distance from a metro station on ln property price (1) <200m -0.653*** N =1,613 (0.083) 201m-800m -0.027*** N =21,956 (0.007) >800m 0.010*** N =118,001 (0.001) N 130,891 adj. R2 0.847
Note: OLS regression with ln(property price) as the dependent variable. The ranges indicate range and North-South Line interaction terms. The coefficient is interpreted as the additional effect realized by a North-South Line station compared to a station of the counterfactual line within the respective distance range. The regression includes: conditions of sale dummies, type of ownership dummies, contribution to VvE (e), building-bound outside space (m2), external outside space (m2), plot size (m2), volume (m3), no. of bathrooms, no. of bedrooms, no. of floors, the floor the property is located on, construction year, energy label dummies, type of property dummies, monthly dummies, and neighbourhood dummies. Robust clustered standard errors in parentheses. Significant at; * 5% level; ** 1% level; *** 0.1% level.
Table 4: Effect of distance from a North-South Line station on ln property price by neighbourhood (1) Amsterdam Centre -0.008*** N =34,692 (0.001) Amsterdam North -0.054*** N =4,723 (0.008) Amsterdam East -0.024* N =4,169 (0.011) Amsterdam West 0.049*** N =38,197 (0.002) Amsterdam South -0.012*** N =42,550 (0.002) N 130,891 adj. R2 0.858
Note: OLS regression with ln(property price) as the dependent variable. The neighbourhoods indicate neighbourhood and North-South Line interaction terms. The coefficient is interpreted as the extra effect realized by a North-South Line station compared to a station of the counterfactual line within the respective neighbourhood. Distance in km. The regression includes: conditions of sale dummies, type of ownership dummies, contribution to VvE (e), building-bound outside space (m2), external outside space (m2), plot size (m2), volume (m3), no. of bathrooms, no. of
bedrooms, no. of floors, the floor the property is located on, construction year, energy label dummies, type of property dummies, monthly dummies, and neighbourhood dummies. The coefficients of the interaction terms of the following neighbourhoods are not shown due to a lack of sufficient observations: Badhoevedorp, Lijnden, Amsterdam New-West, Amsterdam Southeast, Amstelveen and Landsmeer. Robust clustered standard errors in parentheses. Significant at; * 5% level; ** 1% level; *** 0.1% level.
Table 5: Descriptive statistics of distance from a North-South Line or Southeast Line station by neighbourhood
North-South Line Southeast Line Neighbourhood Mean Std. Err. Mean Std. Err. Amsterdam Centre 1.229 .003 1.258 .002 Amsterdam North 1.219 .006 1.541 .008 Amsterdam East 1.666 .003 1.286 .003 Amsterdam West 1.689 .002 1.815 .002 Amsterdam South 1.269 .003 1.413 .002
Table 6: Effect of distance from a North-South Line station on ln property price by year (1) 2014 0.006*** N =29,591 (0.001) 2015 0.007*** N =54,158 (0.001) 2016 0.004*** N =51,509 (0.001) 2017 0.007 N =6,201 (0.004) N 130,891 adj. R2 0.847
Note: OLS regression with ln(property price) as the dependent variable. The years indicate year and North-South Line interaction terms. Properties sold in the year 2013 are excluded because the number of observations is too low (N=111). The coefficient is interpreted as the extra effect realized by a North-South Line station compared to a station of the counterfactual line within the respective year. Distance is measured in km. The regression includes: conditions of sale dummies, type of ownership dummies, contribution to VvE (e), building-bound outside space (m2),
external outside space (m2), plot size (m2), volume (m3), no. of bathrooms, no. of bedrooms, no.
of floors, the floor the property is located on, construction year, energy label dummies, type of property dummies, monthly dummies, and neighbourhood dummies. Robust clustered standard errors in parentheses. Significant at; * 5% level; ** 1% level; *** 0.1% level
Table 7: Effect of distance from a North-South Line station on ln property price by price category
baseline analysis 1st decile 2nd decile 8th decile 9th decile
(1) (2) (3) (4) (5)
Distance*NSL 0.010*** 0.005** 0.007*** 0.011*** 0.009*** (0.001) (0.002) (0.001) (0.001) (0.001) N 130,891 130,891 130,891 130,891 130,891 Pseudo R2 0.847 0.528 0.570 0.684 0.699
Note: OLS regression with ln(property price) as the dependent variable. The coefficient of the interaction term is interpreted as the extra effect realized by a North-South Line station compared to a station of the counterfactual line. Distance in km. The regression includes: conditions of sale dummies, type of ownership dummies, contribution to VvE (e), building-bound outside space (m2), external outside space (m2), plot size (m2), volume (m3), no. of bathrooms, no. of bedrooms, no. of floors, the floor the property is located on, construction year, energy label dummies, type of property dummies, monthly dummies, and neighbourhood dummies. Robust clustered standard errors in parentheses. Significant at; * 5% level; ** 1% level; *** 0.1% level.
Table 8: Effect of distance from a North-South Line or Southeast Line station on ln property price (1) Distance*NSL 0.011*** N =80,013 (0.001) Distance*SEL 0.005*** N =140,120 (0.001) N 351,815 adj. R2 0.851
Note: OLS regression with ln(property price) as the dependent variable. The coefficient of the interaction term is interpreted as the extra effect realized by a metro line station compared to a station of the counterfactual line. NSL indicates the North-South Line and SEL the Southeast Line. Distance in km. The regression includes: conditions of sale dummies, type of ownership dummies, contribution to VvE (e), building-bound outside space (m2), external outside space
(m2), plot size (m2), volume (m3), no. of bathrooms, no. of bedrooms, no. of floors, the floor the
property is located on, construction year, energy label dummies, type of property dummies, monthly dummies, and neighbourhood dummies. Robust clustered standard errors in parentheses. Significant at; * 5% level; ** 1% level; *** 0.1% level
