1
The impact of changing the highway speed limit on the
surrounding housing market:
-The price and liquidity
effects-Friso Haitsma 10657835 May 2017 Master Thesis
MSc Finance: Real Estate Finance
University of Amsterdam, Amsterdam Business School Supervisor: dr. Erasmo Giambona
2 Statement of Originality
This document is written by student Friso Haitsma who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3 Abstract
This thesis investigates the effect of changing the highway speed limit on the surrounding housing market. This effect is measured through two channels. Firstly, this study examines whether changes in the speed limit are capitalized in house prices. Secondly, the liquidity effects of changing the speed limit are investigated. Based on the examination of 20,030 housing transactions, located within 1 km from two highways in the Netherlands, no conclusive evidence is found that changing the speed limit affects the surrounding housing market. The results of this study, found by using the hedonic pricing model in combination with a differences-in-differences estimator, suggest that lowering the highway speed limit has a negative effect on house prices of 2.2%. Furthermore, the results of this study show that lowering the speed limit does not affect the liquidity of houses significantly. However, the evidence for the negative price effect is not conclusive as its significance does not pass all robustness checks. Future research on this topic should therefore focus on improving the model by controlling for variation that was unobserved in this study, such as traffic intensity.
4
1. Introduction ... 5
2. Literature Review ... 7
2.1. The house price effects of traffic externalities ... 7
2.2. The relationship between the speed limit and traffic externalities ... 10
2.3. Noise and air pollution: the relationship with distance from a highway ... 12
3. Methodology ... 14
4. Data and descriptive statistics ... 17
4.1. Case studies ... 17
4.2. Housing transactions ... 18
4.2.1. Data comparison: Amsterdam versus Rotterdam ... 18
4.2.2. Data comparison Amsterdam: treatment group versus control group ... 21
4.2.3. Data comparison Rotterdam: treatment group versus control group ... 24
5. Results ... 27
5.1. Average treatment effect ... 27
5.1.1. Price effect ... 27
5.1.2. Liquidity effect ... 29
5.2. Radius of the effect ... 31
5.3. Anticipation and adjustment effects ... 32
5.4. Robustness of the results ... 32
6. Conclusion ... 39
5
1. Introduction
In July 2012, the Dutch government increased the speed limit on the western part of the A10 highway, the ring road of Amsterdam. Research on this decision shows that, due to increasing the speed limit from 80 km/h to 100 km/h, the live expectancy of people living adjacent to the A10-West highway dropped by 79 days, while the average car driver saves 44 seconds (Knol, 2013). Due to the negative externalities resulting from the speed limit increase, the limit was lowered almost two years later, back to 80 km/h. The A10-West is not the only highway in the Netherlands on which the speed limit has changed over the past years. Speed limits were increased and lowered again, similar to the A10-West, on highways near Rotterdam, The Hague and Utrecht. In addition, the general Dutch highway speed limit, which is applicable to almost all highways outside the Randstad, was increased from 120 km/h to 130 km/h in September 2012 to satisfy Dutch car drivers (Vermeer, 2011). These speed limit decisions illustrate the tradeoff faced by the Dutch government between car drivers and residents. Car drivers generally want to drive as fast as possible because this lowers travel time. On the other hand, vehicles produce more noise and air pollution when driving at a higher speed (Bokma, Havermans, Martens, Stoelhorst, and Tillema, 2007), which reduces the livability for residents in the surrounding area. The prevailing speed limit therefore affects the quality of a property location. However, whether changing the speed limit on a highway affects the housing market in the surrounding area has not yet been investigated. This will therefore be the main aim of this thesis. The effect of changing the speed limit on the housing market will be measured through two channels. Firstly, this study examines whether changes in the speed limit are capitalized in house prices. Secondly, the liquidity effects of changing the speed limit will be investigated.
In existing literature regarding the relationship between traffic and the housing market, researchers focused on the house price effects of several traffic externalities. Chay and Greenstone (2005) find that air pollution negatively affects house prices. The relationship between traffic noise and house prices is negative as well (Theebe, 2004). Both air quality and noise levels are affected by changing the speed limit. Bokma et al. (2007) find that lowering the speed limit results in less noise pollution and improved air quality. Based on this, the following hypotheses are derived for this thesis:
Hypothesis 1:
Lowering the highway speed limit positively affects house prices in the surrounding area.
Hypothesis 2:
6
This thesis contributes to the existing literature in several ways. As discussed before, the impact of changing the highway speed limit on house prices and liquidity has not yet been investigated. Examining the relationship between the prevailing speed limit and the housing market is therefore the main contribution of this thesis. Additionally, in previous literature regarding the relationship between traffic and the housing market, researchers focused primarily on house price effects. Including the liquidity effect is therefore another contribution of this study. The outcome of this thesis is particularly helpful for (future) homeowners who own or plan on buying a house in the proximity of a highway, but also for policy-makers, who evaluate the social benefits and costs of changing the highway speed limit.
Two highways will be examined in this thesis, the above mentioned A10-West in Amsterdam, as well as the A13 in Rotterdam. The highway speed limit on the A10-West and the A13 changed several times over the past 15 years. These two highways therefore both represent suitable case studies for this thesis. To test the hypotheses, an extensive housing dataset has been provided by the Dutch Association of Realtors (NVM). This NVM dataset contains data on 20,030 housing transactions located within 1 km from either the A10-West or the A13 highway in the period from 1999 to 2016. The variables in this dataset include transaction characteristics such as transaction price, transaction date and time on the market. The transaction price and the time on the market will be used to measure the price and liquidity effects, respectively. Furthermore, the dataset contains many housing characteristics and locational variables. Due to the availability of these characteristics, the hedonic pricing model is used. A differences-in-differences (DiD) estimator is included in the hedonic model to estimate the price and liquidity effects of lowering the highway speed limit.
The results of this study show that lowering the highway speed limit has a negative effect on house prices of -2.2%. However, this negative price effect is not very robust. After allowing time trends to differ across areas surrounding the highways, the price effect becomes very small, positive and insignificant. In addition, the price effect is insignificant in Rotterdam. The fact that the price effect is not robust could be due to unobserved variation. In addition, none of the results suggest that lowering the speed limit affects the liquidity of houses. This study therefore finds no conclusive evidence for price and liquidity effects of lowering the highway speed limit. The remainder of this thesis proceeds as follows. Chapter 2 provides an overview of all the related literature. Chapter 3 elaborates on the methodology of this thesis, followed by a description of the used data in chapter 4. The results of this study will be discussed in chapter 5. Finally, in chapter 6, a conclusion will be drawn based on these results.
7
2. Literature Review
As discussed in the introduction, the impact of changing the highway speed limit on the housing market in the surrounding area has not yet been investigated directly. However, research has been done on the effect of traffic externalities on house prices. These traffic externalities are closely related to the prevailing speed limit. This chapter first provides a literature overview of the house price effects of traffic externalities. Thereafter, the relationship between these traffic externalities and the prevailing speed limit will be discussed. Finally, an overview of the research that has been done to determine the size of the area around a highway that is affected by the highway will be provided.
2.1. The house price effects of traffic externalities
In existing literature regarding the impact of traffic on house prices, researchers focused on four externalities of traffic, namely accessibility, traffic intensity, traffic noise and air pollution. Previous literature has shown that the housing market is efficient enough to price these traffic externalities. Accessibility is found to positively affect house prices and can therefore be labeled as a positive traffic externality. On the other hand, traffic intensity, traffic noise and air pollution are found to have a negative impact on house prices and can therefore be labeled as negative traffic externalities. Table 1 provides an overview of the results from previous literature.
Table 1. Literature overview: the effect of traffic externalities on house prices.
Author Market Period Externality Effect Method
Smersh & Smith (2000) Jacksonville (U.S.) 1980-1990 Accessibility Positive Repeat sales Hughes & Sirmans (1992) Louisiana (U.S.) 1985-1989 Traffic intensity Negative Hedonic Theebe (2004) The Netherlands 1997-1999 Traffic noise Negative Hedonic Chay & Greenstone (2005) United States 1970-1980 Air pollution Negative Hedonic Levkovich et al. (2016) The Netherlands 1995-2011 Accessibility Positive Repeat sales Levkovich et al. (2016) The Netherlands 1995-2011 Traffic intensity Negative Repeat sales Levkovich et al. (2016) The Netherlands 1995-2011 Traffic noise Negative Repeat sales
The effect of accessibility on house prices has been investigated by Smersh and Smith (2000), who use the repeat sales methodology. In their study, the authors explore the construction of the Dames Point Bridge in Jacksonville, Florida. The bridge connects the southeast and northeast of the city for the first time. The southeast has more employment and shopping opportunities than the northeast. In addition, the southeast has direct access to the beach. As a result, the authors
8
expect properties in the northeast to appreciate abnormally due to increased accessibility to these amenities. On the other hand, the effect might be negative for houses in the southeast due to increased congestion and other negative traffic externalities. Their dataset covers the period from 1980 to 1990 and contains house and transaction characteristics, as well as geographic information system (GIS) codes. The availability of GIS codes allows them to calculate the exact distance from every transacted house to the constructed bridge. Over the ten-year period, the authors find that north of the bridge house prices appreciated 8.7% more than the city average. South of the bridge, the appreciation rate was 5% lower than the city average. Due to anticipation effects, the difference in price developments between the two districts is statistically significant before completion of the bridge. In addition, the authors find that the radius of the positively affected area north of the bridge is larger than of the negatively affected area south of the bridge. Hughes and Sirmans (1992) are the first to empirically investigate the price effects of traffic intensity on housing. Traffic intensity is defined as the number of vehicles using a road within a specified time period. To estimate the effect, the authors use a standard hedonic pricing model. Their dataset, covering the period 1985 – 1989, contains housing and transaction characteristics of single-family houses in two neighborhoods in the Louisiana metropolitan area. Firstly, the authors estimate the effect of traffic intensity by comparing two extremes. The authors divide their dataset into “low” and “high” traffic streets. Low traffic streets are defined as streets that only provide access to one street, such as dead-ends. High traffic streets are defined as streets that provide direct access to employment and shopping nodes, as well as its feeder streets. They find a substantial negative price effect of 8.8% for high traffic streets compared to low traffic streets. In addition, the authors have access to daily traffic counts for the high traffic streets in their dataset. Including traffic count in their model reveals that traffic intensity is not only capitalized into house prices at extremes but also on a relative level. For each additional 1,000 cars per day, properties are discounted ranging from 0.54% to 1.05%. The effect is higher for city properties than for suburban properties. In addition, the discount is greater for higher-priced houses.
In a Dutch study, Theebe (2004) uses hedonic regression with spatial autocorrelation techniques to investigate the effect of traffic noise on house prices. Similar to this thesis, Theebe (2004) uses a dataset from the Dutch Association of Realtors (NVM), which contains transaction price information and house characteristics. In the investigated period, from 1997 to 1999, this dataset comprises over 160,000 house transactions in the investigated area. The author combines the NVM dataset with noise data, such that a certain noise level is linked to each transacted house.
9
To correct for the positive effect of infrastructure, the author includes an accessibility variable in his model. Theebe (2004) finds a negative relationship between noise levels and house prices. He concludes that traffic noise only affects house prices if the noise level exceeds 65 dB. If traffic noise exceeds 65 dB, the discount on house prices ranges to 12%, with an average of approximately 5%. Finally, he states that the relationship between the house price discount and noise level is non-linear. For noise levels above 65 dB, the value of reducing traffic noise with one decibel varies between 0.3% and 0.5%.
Traffic-related air pollution results in various cardiopulmonary (heart- and lung-related) health problems, such as bronchitis and asthma (Künzli et al., 2000), a higher risk at a cardiopulmonary death (Hoek et al., 2002) and a lower heart rate variability (Schwartz et al., (2005). In addition, studies about the effect of air pollution on children’s health show that air pollution increases the risk of childhood asthma (Gauderman et al., 2005) and leads to substantial deficits in the development of children’s lung-functioning (Gauderman et al., 2007). Research by Chay and Greenstone (2005) shows that air pollution, and corresponding health problems, are capitalized in house prices. The authors combine a dataset with property values and various controls, including housing and locational characteristics, with air pollution data over the period 1980 to 1990 in the United States. Due to a new law in the United States during this period, air pollution concentrations declined in a number of counties. Using a hedonic pricing model, the authors find that the air pollution reduction is causally related to house price increases in these counties. On average, a one microgram per cubic meter (μg/m3) reduction in air pollution concentration leads
to 0.2% to 0.35% higher house prices. This result is substantially higher than the estimates found in previously conducted studies, in which the marginal willingness to pay for a 1 μg/m3 decrease
was estimated to be between 0.05% and 0.1%. Chay and Greenstone (2005) therefore conclude that individuals value clean air more than has previously been recognized.
In a more recent study, Levkovich et al. (2016) investigate the impact of two highway developments in the Netherlands on house prices over the period from 1995 to 2011. The highway completion results in sudden changes in accessibility, traffic noise, traffic intensity and air pollution in the surrounding area. Except for air pollution, the authors are able to measure these traffic externalities separately, which enables them to estimate its separate effects. Their dataset from the Dutch land registry (Kadaster) only contains transaction prices and lot sizes. As a consequence, the authors are not able to attribute house prices to house characteristics, such as size and quality. They solve this problem by using repeat sales in combination with differences-in-differences. The treatment group consists of houses located within 300 m from the highway
10
for traffic noise and within 1 km from the highway for traffic intensity. For accessibility, the authors use houses located in an area that have experienced an increase of over 2.5% in accessibility level. They find a positive effect resulting from changes in accessibility. On the contrary, the authors find that increased noise pollution and traffic intensity both result in lower house prices. Combining all effects, the development of a highway generally has a positive effect on house prices, ranging from 2.5% to 4.3%.
2.2. The relationship between the speed limit and traffic externalities
The above described literature on the house price effects of traffic externalities is useful for this thesis in several ways. Firstly, it helps to get a better understanding of the relationship between traffic and the housing market. It shows that the housing market is efficient enough to price traffic externalities. In addition, changing the highway speed limit affects the level of the above described traffic externalities. Possibly, the house price effect of changing the highway speed limit can be explained by looking at how the speed limit affects these traffic externalities. As an example, the effect of lowering the highway speed limit on accessibility, traffic intensity, traffic noise and air pollution will be discussed next.
Firstly, changing the speed limit affects accessibility as it alters travel time. Ceteris paribus, lowering the speed limit increases travel time and thus results in lower accessibility levels. Based on Smersh and Smith (2000) and Levkovich et al. (2016), this would result in lower house prices. The second traffic externality that could be affected by the prevailing highway speed limit is traffic intensity. If the speed limit is lowered on a highway and comparable alternative roads are available, individuals could start using the alternative road instead of the highway if its travel time is now shorter (Bokma et al., 2007). In this case, lowering the speed limit would thus lead to lower traffic intensity. According to Hughes and Sirmans (1992) and Levkovich et al. (2016), this would result in higher house prices. Finally, two traffic externalities that are definitely affected by the prevailing speed limit are traffic noise and traffic-related air pollution. Vehicle speed is the factor that has the most prominent influence on traffic noise (Den Boer and Schroten, 2007). Changing the speed limit on a highway thus affects the traffic noise produced by the vehicles using the highway. Ceteris paribus, a lower speed limit results in less traffic noise. Based on Theebe (2004) and Levkovich et al. (2016), this would lead to house price increases. Finally, Bokma et al. (2007) conclude that the prevailing speed limit has a significant impact on air pollution. Ceteris paribus, lowering the speed limit on a highway reduces the traffic-related air pollution produced by the vehicles using the highway. As a result, the air quality in the surrounding area improves. According to Chay and Greenstone (2005) this would result in house price increases.
11
Based on previous literature, it can be stated that the speed limit positively affects house prices through accessibility and negatively through traffic noise and air pollution. In addition, the speed limit can have a negative effect on house prices through traffic intensity if comparable alternative roads are available. According to existing literature, the effect of the speed limit on house prices is therefore ambiguous. This suggests that the relationship between the speed limit and house prices through changes in traffic externalities depends on the magnitude of these changes. In order to form a more unambiguous hypothesis for this thesis, the previous part will be related to the two highways that are investigated in this thesis.
Firstly, the two highways are relatively short. As a result, travel time is only slightly affected by the prevailing speed limit. Indeed, Knol (2013) argues that travel time for car drivers on the A10-West increases by 44 seconds if the speed limit is lowered from 100 km/h to 80 km/h. Such a decrease of the speed limit on the other investigated highway in this thesis, the A13, would alter travel time even less as this highway is shorter than the A10-West. If people would think rationally, the accessibility impact of a speed limit change on house prices would thus be negligible. However, whether people think in a rational way is questionable. It could be that their perception of the altered travel time differs from the actual changes in travel time. If this is the case, the accessibility impact of a speed limit change on house prices could be positive in this thesis. However, it is assumed that people think rationally in this thesis. Therefore, it is expected that the accessibility impact of a speed limit change on house prices is negligible in this study. Secondly, due to the absence of comparable alternative roads for the two highways, the traffic intensity impact of a speed limit change on house prices is expected to be negligible as well. Contrary to accessibility and traffic intensity, the traffic noise and air pollution impact of a speed limit change on house prices are not expected to be negligible. This is based on research by Bokma et al. (2007), who perform noise and air pollution measurements at 80 km/h zones in the Netherlands. The authors conclude that lowering the speed limit from 100 km/h to 80 km/h reduces noise emission by approximately 0.5 dB to 1.5 dB. Levkovich et al. (2016) have shown that sudden traffic noise changes can influence house prices. According to the study of Theebe (2004), such a noise reduction would lead to a house price increase ranging from 0.15% to 0.75%, depending on the initial noise level and actual noise reduction. Furthermore, Bokma et al. (2007) measured concentrations of the traffic-related air pollutants nitrogen dioxide (NO2) and particulate matter with a maximum aerodynamic diameter of 10 μm (PM10) before and after
lowering the speed limit from 100 km/h to 80 km/h. After the speed limit was lowered to 80 km/h, PM10 concentrations had declined with approximately 10%, or 0.5 μg/m3. Moreover, NO
2
concentrations were approximately 25%, or 2 μg/m3, lower than NO
12
km/h. According to Künzli et al. (2000), concentrations of traffic-related air pollutants are highly correlated. It can therefore be stated that other pollutant concentrations decline due to a lower speed limit as well. Chay and Greenstone (2005) have shown that sudden improvements of the air quality can positively influence house prices. According to the authors, a NO2 concentration decline of 2 μg/m3 would result in house price increases ranging from 0.4% to 0.7%.
Conclusively, in this thesis, both the accessibility impact and traffic intensity impact of a speed limit change on house prices are expected to be negligible. On the other hand, the traffic noise impact and air pollution impact of a speed limit change on house prices are both expected to be negative. Hence, a negative relationship between the speed limit and house prices is expected in this thesis. Finally, with respect to liquidity, economic theory would suggest that it is positively related to house prices as both increase with demand. Lower demand for houses puts a downward pressure on house prices. Furthermore, lower demand for houses results in fewer housing transactions which lead to a more illiquid housing market. Hence, similar to house prices, a negative relationship between the speed limit and liquidity of houses is expected in this thesis. 2.3. Noise and air pollution: the relationship with distance from a highway
In the previous part it was stated that changing the speed limit is expected to affect house prices through changes in noise and air pollution. The question arises to which distance from a highway the noise and air pollution extend. Answering this question helps to determine the size of the negatively affected area surrounding a highway. This is of key importance for this thesis, as this is the same area that is affected by changes of the highway speed limit. For traffic noise, Atlas Living Environment offers a map with noise nuisance from Dutch highways, based on traffic noise data from the Dutch Ministry of Infrastructure and the Environment (2011). Traffic-related noise nuisance exceeding 55 dB is shown on the map and is expressed in Level day-evening-night (Lden), a European noise indicator. Lden is defined as the average sound level in decibels over a 24-hour period. Because people are more sensitive to noise during the evening and night, penalties of 5 dB and 10 dB are added to the noise levels between 19:00 to 23:00 and 23:00 to 7:00 respectively. The map shows that, both for the A10-West highway and the A13 highway, noise levels exceeding 55 dB extend to a distance from the highway of 250 m at most. Higher noise levels, above 65 dB, extend to maximally 50 m from the highway. The research that has been done to determine the maximum distance from a highway to which traffic-related air pollution extends, shows similar results. According to the conducted research, traffic-related air pollution decays to background levels within 200 m to 250 m from a highway. Substantially
13
higher air pollution concentrations are found within 100 m from a highway. Table 2 provides an overview of the results from this research.
Table 2. Literature overview: the maximum distance from a highway that is negatively affected in terms of
the indicated air pollutants.
Author Area Period Pollutant Distance
Roorda-Knape et al. (1998) The Netherlands 1995 NO2, BS, PM2.5 & PM10 250 m
Zhu et al. (2002) Los Angeles (U.S) 2001 CO, BC & UPM 250 m
Ducret-Stich et al. (2013) Switzerland 2007-2009 NO2, PM10 & EC 200 m
Roorda-Knape et al. (1998) perform one of the first measurements of pollutant concentrations near a highway. The authors measure concentrations of NO2, black smoke (BS), PM2.5 and PM10
around the A13 highway in the Netherlands, which had an average traffic intensity of 132,500 vehicles per day during the sample period. They conclude that the pollutant concentrations decline with distance to the A13 highway. The decline is approximately 50% in the first 100 m to 150 m from the highway. After that, the concentration declines level off to background levels at 250 m.
Four years later, Zhu, Hinds, Kim, Shen and Sioutas (2002) measure the concentration of carbon monoxide (CO), black carbon (BC) and ultrafine particulate matter (UPM) near the Interstate 405, a major highway in Los Angeles. During their sample period, the average traffic intensity was 13,900 vehicles per hour. The authors find comparable results to the results found earlier by Roorda-Knape et al. (1998). The CO, BC and UPM concentrations decrease with approximately 60% after the first 100 m and are indistinguishable from background concentration after 250 m. In a more recent study, Ducret-Stich et al. (2013) measure the concentration of NO2, PM10 and
elemental carbon (EC) around part of the A2 highway in Switzerland. The traffic intensity of the Swiss A2 is substantially lower than the traffic intensity of the highways investigated by Roorda-Knape et al. (1998) and Zhu et al. (2002). During the sample period, 22,040 vehicles used the Swiss A2 per day on average. According to a study by Janssen, Van Vliet, Aarts, Harssema and Brunekreef (2001), pollutant concentrations are significantly associated with traffic intensity. The lower traffic intensity of the Swiss A2 could thus be the reason that Ducret-Stich et al. (2013) conclude that pollutant concentrations decay to background levels within 200 m from the highway, instead of the 250 m found by Roorda-Knape et al. (1998) and Zhu et al. (2002).
14
3. Methodology
Previous literature has shown that the two main methodologies to estimate the effect of traffic externalities on house prices are the repeat sales model and the hedonic pricing model. The most appropriate model depends mainly on the data availability and required number of observations. The model that will be used in this thesis is the hedonic pricing model. Making use of the repeat sales model would lead to robustness problems because the investigated areas are relatively small. In these areas, only a small amount of houses is sold multiple times in the investigated period. Using repeat sales would thus lead to too few observations. In addition, due to its assumption of fixed housing characteristics, the repeat sales model is mostly used if house characteristics are not available (Levkovich et al., 2016). The dataset used in this thesis is extensive and includes many house characteristics. Both the low number of repeat sales and the availability of housing characteristics justify the use the hedonic pricing model in this thesis. In the hedonic pricing model, introduced by Rosen (1974), house prices are attributed to a set of characteristics, including for example physical house characteristics and locational characteristics. The hedonic prices, which are the outcome of the model, represent the implicit prices for all characteristics included in the model. The model is therefore particularly useful to determine the value of goods that are not directly traded, such as the prevailing speed limit. The main disadvantage of the hedonic pricing model is that it requires an extensive dataset to prevent omitted variable bias. Ideally, all attributes that explain part of a house price are included in the model. In reality, some explanatory variables might be hard to obtain or measure. As discussed before, the number of house characteristics in the used dataset is large, which limits the risk of omitted variable bias in this study.
To estimate the impact of changing the speed limit on house prices and liquidity, a differences-in-differences (DiD) estimator will be included in the hedonic pricing model. The DiD approach was introduced by Card and Krueger (1994). The DiD estimator estimates the time effect of a treatment on an outcome by comparing the average outcome of a treatment group with the average outcome of a control group (Levkovich et al., 2016). The DiD approach is therefore perfectly suited to estimate the effect of changing the highway speed limit (i.e. treatment) on prices and liquidity of houses (i.e. outcome) by comparing the average development of prices and liquidity of houses surrounding the highway (i.e. treatment group) with the average development of prices and liquidity of houses located further away from the highway (i.e. control group). The average price and liquidity effect (i.e. average treatment effect) is calculated as the difference in price and liquidity between the two groups before treatment, subtracted by the difference in price and liquidity between the two groups after treatment (Levkovich et al., 2016). The key
15
assumption of DiD is therefore that, without treatment, both groups would have followed the same trends regarding the outcome variables. Otherwise, the DiD estimator will be biased. A violation of the parallel trend assumption could be indicated by pre-treatment trend differences (Winke, 2016). A comparison of the pre-treatment price and liquidity development between the treatment and the control group will therefore be provided in chapter 4.
In all models presented below, standard errors are adjusted for postal code (PC) 6 clusters. Here, “6” does not only indicate the four digits of each postal code, but also its two letters. The first equation, with a classical DiD estimator, will take the following form:
𝑌𝑖𝑡 = 𝛼𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑖∗ 𝑃𝑜𝑠𝑡𝑡+ 𝛽𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑖 + 𝜃𝑡+ 𝜖𝑖𝑡 (1) where 𝑌𝑖𝑡 denotes the outcome variable, which is either the log of the transaction price per m2 or
the log of the time on the market in days, of property i on day t. 𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑖 is a dummy variable
indicating whether property i is located in the treatment group, e.g. located within a specified distance from the highway, or in the control group. In paragraph 2.3, it was concluded that a highway negatively affects the surrounding area up to a distance of 250 m. In this thesis, the treatment group therefore consists of houses located within 250 m from a highway. The control group consists of properties located more than 250 m but less than 1 km away from a highway. The 𝑃𝑜𝑠𝑡𝑡 dummy takes the value one in the treatment period, e.g. when the speed limit changed.
In this thesis, lowering the speed limit from 100 km/h to 80 km/h is considered as a treatment. Hence, 𝑃𝑜𝑠𝑡𝑡 takes the value one if the speed limit equaled 80 km/h on the transaction date and
zero if the prevailing speed limit equaled 100 km/h. As a result, the DiD estimator, which is the interaction term of these two dummies, equals one if transacted property i is located within 250 m from a highway while the prevailing speed limit equaled 80 km/h. 𝛼 is therefore the coefficient of interest as it measures the average treatment effect. Finally, 𝜃𝑡 captures the monthly fixed
effects and 𝜖𝑖𝑡 is an independent and identically distributed error term. Note that the 𝑃𝑜𝑠𝑡𝑡
variable is not included separately in equation (1) as its effect is absorbed by the monthly fixed effects.
Equation (1) does not take differences in housing characteristics between houses into account. Due to these omitted variables, the results of the previous model could be biased. Therefore, the housing characteristics will be included as follows:
16
where 𝑥𝑖𝑡 represents a vector of characteristics of property i on day t, including house size,
number of rooms, garden size and number of parking spaces. In addition, the vector 𝑥𝑖𝑡 includes
dummy variables for house type, presence of a swimming pool, monumental status, inside and outside maintenance level, and construction period.
Equations (1) and (2) do not deal with different locational characteristics, which will be included as follows:
𝑌𝑖𝑡 = 𝛼𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑖 ∗ 𝑃𝑜𝑠𝑡𝑡+ 𝛾𝑥𝑖𝑡+ 𝜃𝑡+ 𝜗𝑗 + 𝜖𝑖𝑡 (3)
where 𝜗𝑗 captures the area fixed effects for each PC6 area j. These area fixed effects deal with
time invariant locational characteristics (Droës and Koster, 2016). The PC6 fixed effects are highly collinear with the treatment group dummy, which is also an area indicator. Therefore,
𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑖is not included separately in equation (3).
In the previous equations, the treatment group is compared to houses located up to 1 km from a highway. Because this area is relatively large, the results of the previous models might still be affected due to unobserved traits, such as changes in zoning regulations (Droës and Koster, 2016). To limit the impact of unobserved traits, the sample is restricted to a maximum distance from the highway of 700 m. The average treatment effect will subsequently be estimated as follows:
𝑌𝑖𝑡 = 𝛼𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑖 ∗ 𝑃𝑜𝑠𝑡𝑡+ 𝛾𝑥𝑖𝑡+ 𝜃𝑡+ 𝜗𝑗 + 𝜖𝑖𝑡 (4) Equation (4) is identical to equation (3). However, the treatment group, consisting of houses located within 250 m, is now compared to a control group consisting of houses located further away, but within 700 m from a highway. The downside is that this reduces the number of observations. However, Droës and Koster (2016) argue that this approach addresses the problem of unobserved time trends.
17
4. Data and descriptive statistics
In this chapter, the used dataset will be discussed and its descriptive statistics will be provided. Firstly, the two highways and its speed limit development over time will be discussed. Thereafter, the descriptive statistics of the two subsets in Amsterdam and Rotterdam will be provided, followed by a separate comparison of the treatment group and the control group in both cities. 4.1. Case studies
The effect of changing the highway speed limit on house prices and liquidity in the surrounding area is estimated based on two case studies, which both represent highways on which the speed limit changed over the past 15 years. Two highways are examined instead of one because this increases the number of observations in the sample which improves the robustness of the results. In addition, a comparison can be made between the two highways to see whether differences exist. The two case studies that are examined in this thesis are the highway A10-West, part of the ring road of Amsterdam, and part of the highway A13, located in Rotterdam. The investigated parts of the A10 and A13 are both relatively short, approximately 5 km and 2 km respectively. However, the highways are both located in areas with a high population density. Again, this increases the number of observations in the sample. The speed limit on the A10-West and the A13 changed several times over the past 15 years, both upwards and downwards. Speed limits are enforced on both highways by camera systems measuring the average speed. The development of the speed limit over time on the highways A10-West and A13 is shown in table 3. Originally, the speed limit was 100 km/h on the two highways.
Table 3. Development of the speed limit over time on the highways A10-West and A13.
Date A10-West A13
11th of May 2002 Decreased to 80 km/h
29th of October 2004 Decreased to 80 km/h
2nd of July 2012 Increased to 100 km/h Increased to 100 km/h 29th of March 2014 Decreased to 80 km/h Decreased to 80 km/h
In order to calculate the distance from each transacted house to the highway, Rijksdriehoeks (RD) coordinates, measured in units of length of 1 m, are collected of both highways with the software Geo Javawa. Because both highways vary in number of lanes, the coordinates of its outer borders, including ramps, are retrieved. To increase accuracy, coordinates are determined
18
every 20 m, resulting in a list of approximately 800 RD coordinates, marking the borders of the investigated parts of the A10-West and the A13 highway.
4.2. Housing transactions
To conduct this research, a dataset has been provided by the Dutch Association of Realtors (NVM). This dataset contains about 70% of all housing transactions in the Netherlands (Droës and Koster, 2016). The variables in this database include transaction characteristic such as transaction price, transaction date and time on the market. Furthermore, the database contains housing characteristics, such as house type, size, construction year and number of rooms. Locational variables in the database include postal code and its RD coordinates, municipality and province. From this database, the NVM provided a subset with housing transactions in the areas surrounding the A10-West and the A13. Over the period 1999 to 2016, this subset contains 37,599 housing transactions in the requested postal code areas. Excluding observations that are located more than 1 km from the border of one of the two investigated highways results in 20,242 observations. In addition, some observations have been removed based on their extreme values. Observations with a house size lower than 25 m2 or higher than 200 m2 have been
removed. Furthermore, observations with transaction prices exceeding €1.5 million have been excluded. Approximately 1% of the observations are deleted because of these size and price restrictions. Finally, three observations that are auctioned instead of sold after a regular negotiation are removed. As a result, the final dataset comprises 20,030 housing transactions located within 1 km from either the A10-West (Amsterdam) or A13 (Rotterdam) highway in the period from 1999 to 2016. The dataset distinguishes between 19,586 transactions for which the purchaser costs had to be paid by the buyer of a property (k.k.) and 444 transactions that excluded purchaser costs for the buyer (v.o.n.). Because purchaser costs increase the amount that has to be paid for a property, properties that are sold v.o.n. are more favourable. People are therefore willing to pay more for a v.o.n. property. Purchaser costs consist primarily of transfer tax, which equals 2% for residential properties in the Netherlands (Dutch Tax Authority, 2017). Transaction prices of properties that were sold v.o.n. are therefore divided by 1.02, such that every transaction price in the dataset represents a k.k. value.
4.2.1. Data comparison: Amsterdam versus Rotterdam
Of the 20,030 observations in the total dataset, 17,922 are located in Amsterdam, which is considerably more than the 2,108 observations located in Rotterdam. Table 4 provides the descriptive statistics of the subsets in Amsterdam and Rotterdam. Houses are on average more expensive in Amsterdam than in Rotterdam, with mean transaction prices of €222,330 and
19
€163,293 respectively. The most expensive house in the dataset is located in Amsterdam and was sold for €1,116,000. The house with the lowest transaction price, €42,500, is located in Rotterdam. In the sample, the average transaction price per m2 is €3,031.36 in Amsterdam and
€1,626.78 in Rotterdam. The standard deviation of the transaction price per m2 is also higher in
Amsterdam (€868.80) than in Rotterdam (€499.85), indicating more variation of the price per m2
in Amsterdam. The housing market in Amsterdam is more liquid than the housing market in Rotterdam as the sample average time on the market equals 111.61 days in Amsterdam and 198.24 days in Rotterdam. The sample average distance to a highway in Amsterdam and Rotterdam are 568.30 m and 291.94 m respectively. As a consequence, the share of houses located within 250 m from a highway is smaller in Amsterdam than in Rotterdam, namely 10.4% and 48.1% respectively. Table 4 shows that the subsets of the two cities differ in housing characteristics as well. Houses in Amsterdam are on average smaller, with a lower number of rooms, a smaller garden and fewer parking spaces. The house type shares vary between the two cities as well. In Amsterdam, the sample is dominated by apartments with a share of 95.9%, followed by terraced houses (3.3%) and corner houses (0.7%). The house type shares of detached and semi-detached houses are both negligible in Amsterdam. The variation in house types is higher in Rotterdam. Although apartments are most common in the Rotterdam subset as well, its share is lower than in Amsterdam, namely 60.7%. Terraced houses, corner houses, semi-detached houses and detached houses in Rotterdam account for shares of 24.1%, 7.8%, 5.5% and 1.9% respectively. The share of houses with a swimming pool is slightly higher in Rotterdam. On the other hand, the share of houses listed as monument is higher in Amsterdam, as the Rotterdam subset does not include any listed monuments. Both the average inside and outside maintenance level of houses in the dataset is higher in Amsterdam. Finally, most houses in Amsterdam were constructed before the Second World War, contrary to Rotterdam, where most houses were constructed in the period from 1945 to 1959.
20
Table 4. Descriptive statistics: transaction, highway and house characteristics in Amsterdam and Rotterdam.
Amsterdam (N = 17,922) Rotterdam (N = 2,108)
Variable Mean SD Min Max Mean SD Min Max
Transaction characteristics
Transaction price (€) 222,330 97,131 50,370 1,116,000 163,293 91,725 42,500 754,000 Price per m2 (€) 3,031.36 868.80 779.94 8,975.41 1,626.78 499.85 411.43 4,790.70
Time on market (days) 111.61 169.94 0 2,673 198.24 258.46 0 2,342
Transaction year 2008 5.06 1999 2016 2008 4.72 1999 2016 Highway distance Distance to highway (m) 568.304 245.182 16.031 999.525 291.937 198.373 12.042 883.102 Highway < 250 m 0.104 0.481 House characteristics Size (m2) 74.251 24.582 25 200 99.008 37.062 35 200 Rooms (#) 3.101 0.982 0 11 4.024 1.308 0 9 Garden (m2) 10.289 27.056 0 660 40.409 65.215 0 800 Parking spaces (#) 0.257 0.900 0 8 0.412 1.304 0 8 Apartment 0.959 0.607 Terraced house 0.033 0.241 Corner house 0.007 0.078 Semi-detached house 0.000 0.055 Detached house 0.000 0.019 Swimming pool 0.000 0.001 Listed monument 0.008 0
Maint. inside poor 0.020 0.043
Maint. inside reasonable 0.109 0.162
Maint. inside good 0.684 0.693
Maint. inside excellent 0.188 0.102
Maint. outside poor 0.040 0.104
Maint. outside reasonable 0.836 0.822
Maint. outside good 0.124 0.074
Constr. year 1500-1905 0.015 0.027 Constr. year 1906-1930 0.434 0.160 Constr. year 1931-1944 0.253 0.240 Constr. year 1945-1959 0.103 0.430 Constr. year 1960-1970 0.068 0.031 Constr. year 1971-1980 0.006 0.010 Constr. year 1981-1990 0.007 0.021 Constr. year 1991-2000 0.070 0.052 Constr. year >2001 0.043 0.029
21 4.2.2. Data comparison Amsterdam: treatment group versus control group
The subset in Amsterdam contains 17,922 observations, of which 1,864 are located within 250 m from the A10-West highway. As discussed before, these 1,864 observations are part of the treatment group, which leaves 16,058 observations for the control group. Table 5 provides the descriptive statistics of the treatment group and the control group in Amsterdam. On average, the transaction price is €2,770 per m2 in the treatment group, which is slightly less than the
€3,062 per m2 for houses in the control group. On the other hand, the time on the market is
slightly higher in the treatment group, indicating a more illiquid housing market near the A10-West highway. The differences in housing characteristics between the treatment and control group show the importance of controlling for house characteristics. If not controlled for, the treatment effect will partly reflect the different housing characteristics of the two groups and will therefore be biased. On average, the treatment group consists of larger houses, with more rooms, a smaller garden, more parking spaces, a lower amount of swimming pools and fewer houses that are listed as a monument. In addition, the treatment group has a larger share of apartments, while the control group has larger shares of terraced houses and corner houses. The shares of detached houses and semi-detached houses are negligible in both groups, similar to the total sample of houses located in Amsterdam. Maintenance levels and construction years differ slightly as well. Most noticeable is that houses in the treatment group are on average more recently constructed. For example, the share of houses constructed after 2001 is 19.4% in the treatment group and 2.6% in the control group.
As discussed in chapter 3, the key assumption of DiD is that, without treatment, both groups would have followed the same trends regarding the outcome variables. Figures 1 and 2 therefore depict the development of the monthly average transaction price per m2 and time on the market
of the treatment and the control group in Amsterdam, over the period preceding the first change of the highway speed limit on the 29th of October 2004. Figure 1 shows that the transaction price per m2 of the treatment and control group roughly followed the same trend in this period.
In the control group, the average price per m2 increased from approximately €1,750 in January
1999 to €2,500 in November 2004. The average price per m2 of the treatment group is generally
lower in this period. In figure 2, it can be seen that the average time on the market of both groups followed a similar trend as well, increasing from an on average 50 days in January 1999 to approximately 150 days in November 2004. It must be noted that, both in figures 1 and 2, the treatment group graph is more volatile than the control group graph. This can be explained by the fact that the monthly averages of the treatment group are based on a lower number of observations per month, leading to more volatility. Conclusively, based on figures 1 and 2, there is no reason to suspect that the parallel trend assumption does not hold in Amsterdam.
22
Table 5. Descriptive statistics: transaction, highway and house characteristics in Amsterdam: treatment versus control group.
<250 m from highway (N = 1,864) >250 m from highway (N = 16,058)
Variable Mean SD Min Max Mean SD Min Max
Transaction characteristics
Transaction price (€) 220,238 98,472 74,193 975,000 222,572 96,974 50,370 1,160,000 Price per m2 (€) 2,769.86 786.95 920.17 6,364.15 3,061.71 872.77 799.94 8,975.41
Time on market (days) 121.008 178.993 0 1,632 110.521 168.834 0 2,673
Transaction year 2009 4.925 1999 2016 2008 5.054 1999 2016 Highway distance Distance to highway (m) 149.521 73.846 16.031 249.874 616.916 209.135 250.098 999.525 House characteristics Living area (m2) 80.102 24.567 26 192 73.572 24.494 25 200 Rooms (#) 3.182 0.892 0 8 3.092 0.991 0 11 Garden (m2) 6.170 20.874 0 300 10.767 27.645 0 660 Parking spaces (#) 0.962 1.474 0 8 0.175 0.767 0 8 Apartment 0.986 0.956 Terraced house 0.013 0.035 Corner house 0.001 0.008 Semi-detached house 0.000 0.000 Detached house 0.000 0.000 Swimming pool 0 0.000 Listed monument 0.002 0.009
Maint. inside poor 0.012 0.020
Maint. inside reasonable 0.126 0.106
Maint. inside good 0.638 0.689
Maint. inside excellent 0.223 0.184
Maint. outside poor 0.027 0.042
Maint. outside reasonable 0.786 0.842
Maint. outside good 0.187 0.116
Constr. year 1500-1905 0.001 0.017 Constr. year 1906-1930 0.104 0.473 Constr. year 1931-1944 0.239 0.254 Constr. year 1945-1959 0.120 0.101 Constr. year 1960-1970 0.171 0.056 Constr. year 1971-1980 0.041 0.002 Constr. year 1981-1990 0.008 0.007 Constr. year 1991-2000 0.123 0.064 Constr. year >2001 0.194 0.026
23
Figure 1. Development of the monthly average transaction price per m2 of the treatment group, located
within 250 m from the A10-West highway, and the control group, located between 250 m and 1 km from the A10-West highway in Amsterdam. The graph displays the period preceding the first highway speed limit change on the 29th of October 2004. During this period, the highway speed limit equaled 100 km/h.
Figure 2. Development of the monthly average time on the market of the treatment group, located within
250 m from the A10-West highway, and the control group, located between 250 m and 1 km from the A10-West highway in Amsterdam. The graph displays the period preceding the first highway speed limit change on the 29th of October 2004. During this period, the highway speed limit equaled 100 km/h.
24 4.2.3. Data comparison Rotterdam: treatment group versus control group
The subset in Rotterdam contains 2,108 observations, of which 1,014 are located within 250 m from the A13 highway. Again, these 1,014 observations are part of the treatment group, leaving 1,094 observations for the control group. Table 6 provides the descriptive statistics of the treatment group and the control group in Rotterdam. Similar to Amsterdam, the treatment group in Rotterdam consists of houses with a lower average transaction price per m2 and a longer
average time on the market. The average transaction price per m2 equals €1,508 in the treatment
group and €1,736 in the control group. The average time on the market of houses in the treatment and control group equals 223 days and 175 days respectively. Again the housing characteristics of the two groups differ. On average, the treatment group consists of smaller houses, with fewer rooms, a smaller garden, fewer parking spaces and a higher amount of swimming pools. In addition, the treatment group has a larger share of apartments, while the control group has larger shares of all other house types. Both the average inside and outside maintenance levels of houses in the treatment group are lower than of houses in the control group. Finally, differences also exist in construction periods, although most houses in both groups were constructed in the period from 1945 to 1959.
Again, due to the parallel trend assumption of DiD before treatment, figures 3 and 4 depict the development of the monthly average transaction price per m2 and time on the market of the
treatment and the control group in Rotterdam, over the period preceding the first highway speed limit change on the 11th of May 2002. Figure 3 shows that the price trend of the treatment and control group in Rotterdam are less similar than in Amsterdam, as illustrated in figure 1. Nonetheless, over the whole period depicted in figure 3, transaction prices per m2 are generally
lower in the treatment group. In addition, average transaction price per m2 of both groups
increased with approximately €200 in the period from January 1999 to May 2002. Figure 4 shows that the development of the monthly average time on the market follows a more similar pattern than the price development in figure 3. However, the trends of the treatment and control group are again not as similar as in Amsterdam, as illustrated in figure 2. As a result, the parallel trend assumption is more questionable in Rotterdam than in Amsterdam, which should be taken into account when examining the results of this study. However, it must be noted that the monthly averages in Rotterdam, both of the treatment and the control group, are based on a substantially lower number of observations than in Amsterdam. The graphs in figures 3 and 4 are therefore more volatile and less reliable in terms of accuracy than the graphs depicted in figures 1 and 2.
25
Table 6. Descriptive statistics: transaction, highway and house characteristics in Rotterdam: treatment versus control group.
<250 m from highway (N = 1,014) >250 m from highway (N = 1,094)
Variable Mean SD Min Max Mean SD Min Max
Transaction characteristics
Transaction price (€) 143,747 70,607 42,500 605,000 181,410 104,477 58,750 754,000 Price per m2 (€) 1,508.42 442.50 411.43 4,227.27 1,736.48 524.53 728.00 4,790.70
Time on market (days) 223.010 286.821 0 2,342 175.289 226.792 0 1,908
Transaction year 2007 4.781 1999 2016 2008 4.661 1999 2016 Highway distance Distance to highway (m) 120.854 70.185 12.042 249.736 450.509 137.739 251.024 883.102 House characteristics Living area (m2) 96.173 35.180 35 200 101.635 38.555 40 200 Rooms (#) 3.954 1.318 0 9 4.090 1.295 0 9 Garden (m2) 33.455 52.509 0 600 46.856 74.544 0 800 Parking spaces (#) 0.344 1.237 0 8 0.474 1.362 0 8 Apartment 0.670 0.549 Terraced house 0.205 0.275 Corner house 0.067 0.088 Semi-detached house 0.043 0.066 Detached house 0.015 0.022 Swimming pool 0.002 0
Maint. inside poor 0.048 0.037
Maint. inside reasonable 0.172 0.154
Maint. inside good 0.684 0.700
Maint. inside excellent 0.096 0.109
Maint. outside poor 0.127 0.083
Maint. outside reasonable 0.804 0.838
Maint. outside good 0.069 0.079
Constr. year 1500-1905 0.013 0.040 Constr. year 1906-1930 0.249 0.078 Constr. year 1931-1944 0.260 0.220 Constr. year 1945-1959 0.386 0.471 Constr. year 1960-1970 0.050 0.013 Constr. year 1971-1980 0.003 0.017 Constr. year 1981-1990 0.003 0.038 Constr. year 1991-2000 0.003 0.098 Constr. year >2001 0.034 0.025
26
Figure 3. Development of the monthly average transaction price per m2 of the treatment group, located
within 250 m from the A13 highway, and the control group, located between 250 m and 1 km from the A13 highway in Rotterdam. The graph displays the period preceding the first highway speed limit change on the 11th of May 2002. During this period, the highway speed limit equaled 100 km/h.
Figure 4. Development of the monthly average time on the market of the treatment group, located within
250 m from the A13 highway, and the control group, located between 250 m and 1 km from the A13 highway in Rotterdam. The graph displays the period preceding the first highway speed limit change on the 11th of May 2002. During this period, the highway speed limit equaled 100 km/h.
27
5. Results
In this chapter, the results of this study will be discussed. Firstly, in paragraph 5.1, the results of the baseline regressions will be provided. Thereafter, in paragraph 5.2, the radius of the average treatment effect will be examined, followed by a discussion of possible anticipation and adjustment effects in paragraph 5.3. Finally, robustness checks will be performed and discussed in paragraph 5.4.
5.1. Average treatment effect
Table 7 and 8 present the price and liquidity effects of the baseline regressions, respectively. Both tables present four regression outputs, corresponding to equations (1) to (4). Column (1) therefore represents the classical DiD with monthly fixed effects, column (2) adds housing characteristics, column (3) adds PC6 fixed effects and finally, column (4) shows the results of the restricted sample. Firstly, the baseline regression results of the price effects will be discussed in paragraph 5.1.1, followed by the liquidity effects in paragraph 5.1.2.
5.1.1. Price effect
Table 7 presents the price effects of lowering the highway speed limit from 100 km/h to 80 km/h. The results of the classical DiD model in column (1) suggest an average treatment effect of -3.5%. The results in column (1) also suggest that prices of houses located within 250 m from a highway are on average 26.1% lower than of houses located between 250 m and 1 km from a highway. Both estimates are significant at the 1% level. After housing characteristics are included in column (2) the magnitude of the price effect decreases from -3.5% to -2.6%. Because of including housing characteristics, the price effect is now only significant at the 5% level. According to the regression output in column (2), houses located within 250 m for a highway are on average 18.1% less expensive than houses located further away. The share of houses located within 250 m from a highway is larger in Rotterdam than in Amsterdam. Because houses in Rotterdam are on average less expensive than in Amsterdam, the -18.1% is likely to be an overestimate, as location is not controlled for in column (2). Column (3) shows that, because of including area fixed effects, the magnitude of the average treatment effect decreases but becomes more significant. Including PC6 fixed effects results in a price effect of -2.2%, significant at the 1% level. Note that, as discussed before, the dummy for the treatment group is omitted in column (3) as it is highly collinear with the PC6 fixed effects. After restricting the sample to observations within 700 m from a highway, the average treatment effect remains unchanged. Column (4) suggests a price effect of -2.2%, again significant at the 1% level.
28
Table 7. Baseline regression results: the effect of lowering the highway speed limit from 100 km/h to 80 km/h
on house prices.
(1) (2) (3) (4)
Classical DiD characteristics Housing PC6 FE Control group 250-700 m
Average treatment effect -0.035***
(0.012) -0.026** (0.010) -0.022*** (0.007) -0.022*** (0.007) Highway < 250 m -0.261*** (0.027) -0.181*** (0.021) Size (m2) -0.003*** (0.000) -0.003*** (0.000) -0.004*** (0.000) Rooms (#) 0.005** (0.002) 0.013*** (0.002) 0.013*** (0.002) Garden (m2) 0.000*** (0.000) 0.001*** (0.000) 0.001*** (0.000) Parking spaces (#) 0.031*** (0.005) 0.013*** (0.002) 0.013*** (0.003) Corner house 0.011 (0.023) 0.045*** (0.013) 0.053*** (0.016) Semi-detached house 0.087** (0.042) 0.067** (0.032) 0.083** (0.033) Detached house 0.233*** (0.054) 0.221*** (0.056) 0.255*** (0.059) Apartment 0.098*** (0.018) -0.063*** (0.015) -0.069*** (0.018) Swimming pool 0.126 (0.084) (0.038) 0.051 (0.037) 0.056 Listed monument 0.002 (0.018) (0.014) 0.005 (0.019) 0.010
Maint. inside reasonable 0.088***
(0.011) 0.079*** (0.009) 0.081*** (0.011)
Maint. inside good 0.176***
(0.011) 0.172*** (0.008) 0.171*** (0.010)
Maint. inside excellent 0.251***
(0.011) 0.238*** (0.009) 0.238*** (0.011)
Maint. outside reasonable 0.036***
(0.008) 0.018*** (0.006) 0.016** (0.007)
Maint. outside good 0.048***
(0.009) 0.018** (0.007) (0.009) 0.015* Constr. year 1906-1930 -0.051*** (0.014) (0.010) -0.011 (0.021) 0.026 Constr. year 1931-1944 -0.099*** (0.015) (0.010) -0.011 (0.021) 0.033 Constr. year 1945-1959 -0.388*** (0.021) -0.026* (0.015) (0.024) 0.022 Constr. year 1960-1970 -0.356*** (0.023) -0.043** (0.019) (0.027) 0.012 Constr. year 1971-1980 -0.188*** (0.036) -0.073*** (0.030) (0.030) -0.015 Constr. year 1981-1990 -0.209*** (0.040) (0.047) -0.054 (0.066) -0.030 Constr. year 1991-2000 -0.155*** (0.026) (0.022) 0.012 (0.030) 0.042 Constr. year >2001 -0.127*** (0.037) (0.017) 0.016 (0.026) 0.046*
Monthly FE Yes Yes Yes Yes
PC6 FE No No Yes Yes
Only observations <700 m No No No Yes
R-squared 0.754 0.793 0.808 0.800
Number of observations 20,030 20,030 20,030 13,797
Notes: The dependent variable is the logarithm of the transaction price per m2. Standard errors are in
29
Most of the control variables in column (2) are significant and have the expected sign. A larger house size has a negative effect on the transaction price per m2. On the other hand, numbers of
rooms, garden size and number of parking spaces all have a positive effect. Semi-detached and detached houses are more expensive than terraced houses, the reference category. The difference with corner houses is not significant in column (2). Surprisingly, apartments are more expensive than terraced houses according to the model. The effect of a swimming pool and whether a property is listed as a monument are both positive but insignificant. House prices increase with higher inside and outside maintenance level. The results in column (2) also suggest that the inside maintenance level of a property is a more important house price determinant than the outside maintenance level. Finally, houses that were constructed after 1905 are on average less expensive than houses that were constructed before 1905, which is the reference category. The effects of housing characteristics on house prices in column (2) are biased if the housing characteristic is correlated with location. For example, apartments could be located in more expensive areas than terraced houses. Because location is not controlled for in equation (2), this would result in an upward bias for the effect of apartments relative to terraced houses. This explains why the effect of some of the control variables changes compared to column (2), after controlling for location in column (3). The results in column (3) show that corner houses are more expensive than terraced houses, contrary to apartments, that are less expensive. Both house type effects therefore have the expected sign in column (3), contrary to column (2). Furthermore, only houses constructed between 1945 and 1980 are significantly less expensive than houses constructed before 1905. After restricting the sample in column (4), the magnitude of the effect of housing characteristics changes only slightly.
5.1.2. Liquidity effect
Table 8 presents the liquidity effects of lowering the highway speed limit from 100 km/h to 80 km/h. Contrary to house prices, table 8 suggests that lowering the speed limit does not affect the liquidity of houses. The average treatment effects in columns (1) to (4) are all positive but insignificant. The results in columns (1) and (2) do suggest that the liquidity of houses located within 250 m from a highway is significantly lower relative to houses located further away. After controlling for housing characteristics, the average time on the market is 17.2% higher in the area surrounding a highway. Again, due to the less liquid housing market in Rotterdam, this is likely to be an overestimate as the share of houses located within 250 m from a highway is larger in Rotterdam than in Amsterdam. Table 8 also shows that the liquidity of houses is harder to predict based on housing characteristics, time and location, than house prices. This is illustrated by the low r-squared in all columns of table 8. Furthermore, the effect of only few housing characteristics is significant. Table 8 suggests that a larger house with more parking spaces is less
30
Table 8. Baseline regression results: the effect of lowering the highway speed limit from 100 km/h to 80 km/h
on the liquidity of houses.
(1) (2) (3) (4)
Classical DiD characteristics Housing PC6 FE Control group 250-700 m
Average treatment effect 0.028
(0.065) (0.057) 0.031 (0.059) 0.018 (0.061) 0.005 Highway < 250 m 0.257*** (0.057) 0.172*** (0.053) Size (m2) 0.004*** (0.001) 0.002*** (0.001) 0.002*** (0.001) Rooms (#) -0.066*** (0.012) -0.054*** (0.012) -0.042*** (0.015) Garden (m2) -0.000 (0.000) (0.000) -0.001 (0.000) -0.000 Parking spaces (#) 0.004 (0.015) 0.053*** (0.015) 0.054*** (0.019) Corner house 0.172** (0.079) (0.085) 0.051 (0.104) 0.064 Semi-detached house 0.435*** (0.114) (0.149) -0.015 (0.158) -0.026 Detached house 0.824*** (0.139) 0.503** (0.203) (0.212) 0.385* Apartment -0.339*** (0.056) (0.098) -0.052 (0.117) -0.132 Swimming pool 0.232 (0.510) (0.268) -0.278 (0.210) -0.330 Listed monument 0.006 (0.099) (0.105) 0.016 (0.139) -0.017
Maint. inside reasonable -0.006
(0.057) (0.058) 0.053 (0.070) 0.069
Maint. inside good 0.144***
(0.055) 0.145*** (0.056) (0.069) 0.129*
Maint. inside excellent -0.002
(0.058) (0.059) 0.018 (0.074) -0.028
Maint. outside reasonable -0.093**
(0.043) (0.046) 0.004 (0.052) 0.002
Maint. outside good -0.041
(0.054) (0.057) 0.061 (0.066) 0.110* Constr. year 1906-1930 0.042 (0.069) (0.080) 0.110 (0.156) 0.107 Constr. year 1931-1944 0.071 (0.071) (0.084) 0.130 (0.158) 0.144 Constr. year 1945-1959 0.062 (0.075) (0.110) -0.139 (0.177) -0.110 Constr. year 1960-1970 0.218** (0.087) (0.137) -0.065 (0.201) -0.135 Constr. year 1971-1980 -0.185 (0.147) (0.170) -0.029 (0.249) -0.045 Constr. year 1981-1990 0.166 (0.109) (0.159) 0.028 (0.221) 0.031 Constr. year 1991-2000 0.306*** (0.093) (0.163) -0.042 (0.254) -0.091 Constr. year >2001 0.209** (0.097) (0.148) 0.073 (0.224) -0.012
Monthly FE Yes Yes Yes Yes
PC6 FE No No Yes Yes
Only observations <700 m No No No Yes
R-squared 0.211 0.215 0.226 0.231
Number of observations 20,030 20,030 20,030 13,797
Notes: The dependent variable is the logarithm of the time on the market. Standard errors are in parentheses
