Page | 1
University of Amsterdam
The effect of Airbnb on the housing prices for different areas
in Amsterdam
Immanuel de Potter, 10797130
BSc Economics & Finance, Thesis
Faculty of Economics & Business
Supervisor: A. Akdeniz
Page | 2
Abstract
Recently, there has been a lot to do about Airbnb in Amsterdam. The company is growing in the city and this causes some problems for citizens. The housing market in Amsterdam is already tensioned and due to the Airbnb activity in Amsterdam this is only increasing. This paper investigates the effect Airbnb has on the housing prices for different Areas in Amsterdam. It uses data on housing
transactions for the period 2006-2016 provided by the NvM. The Airbnb activity is measured based on data from insideairbnb.com. A hedonic price regression is used to calculate the relation between the variables.
Statement of originality
This document is written by Immanuel de Potter 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.
Page | 3
Content
Introduction
4
Literature review
5
Data
7
Methodology
11
Results
12
Conclusion
15
References
16
Page | 4
Introduction
Airbnb was founded in 2008 by Brian Chesky, Joe Gebbia and Nathan Blecharczyk. The idea all started in 2007 when Chesky and Gebbia were roommates in San Francisco but could not afford their rent anymore. That is when they came up with the idea to turn their room into a space where 3 air mattresses could fit so that they could rent it out. As a service to their guests they provided a breakfast every morning and that’s where the name airbedandbreakfast comes from. This would later be turned into Airbnb in March 2009. When a big design conference was coming up in San Francisco hotels were struggling to provide enough rooms. Checky and Gebbia took their chance and that’s when the first Airbnb ever was rented out. After the conference they realised that this idea could become something big. Together with their old roommate Blecharczyk they worked out the plans to turn their idea into a business. After seeing their business going bankrupt after the first two launches they refined their business plan. The third time was a charm. Airbnb was created (Business Insider, 2016, February 3).
In 2009 Airbnb launched in Amsterdam and has been growing very rapidly. The popularity is ever since so high that a company called Open Huis wants to start buying houses that would only be used for Airbnb rental, creating the first ever Airbnb neighbourhood (Parool, 2017, June 12). This puts more pressure on the tension that is already existing in the Amsterdam housing market. The price per square meter is almost two times higher compared to the rest of the country. In 2016 there was an increase in prices of almost 23% and houses were sold in an average of 26 days causing a high level of tension in the housing market in Amsterdam (Infonu, 2017, January 12). On the Airbnb market in Amsterdam there is a high tension as well. Citizens complain about nuisance caused by guests and they blame that Airbnb houses are taking over their place to live. Therefore, the city council of Amsterdam has introduced a rule that requires hosts to mention their listing at the council if they want to rent out their house. Also, they have put a maximum of 60 days on the amount of nights (NOS, 2017, September 29).
The main goal of this thesis is to see if there is a relation between the tension on these two markets. Are they related and if so, how? The research question of this thesis is: what is the effect of Airbnb listings on the housing prices for different areas in Amsterdam? On both the housing market and the Airbnb market, extensive research has been done. But about the relation of the two in Amsterdam not so many relevant papers have been published. Glaeser, Gyourko, & Saks (2004) have investigated the supply and demand mechanisms of the housing market in New York. They found that a lack in supply was the cause of the high prices on the market. However, Eichholtz, Huisman, & Zwinkels (2015) argue that only a small part of the housing prices can be explained by these supply and demand fundamentals. Also, on Airbnb, quite an amount of research is done but the majority is focused on the operations of the company itself. Nonetheless, there is some literature focussing on how Airbnb influences other markets. For example, on how Airbnb has affected the housing rental market in L.A. by Lee (2016). As well as by Zervas, Prosperio, & Byers (2015) who evaluate how Airbnb is affecting the prices of hotel rooms in Texas. The same goes for research on the effect externalities have on housing prices. Research has been done on how natural amenities are affecting housing prices by Gibbons, Mourato, & Resende (2014). Furthermore, Linn (2013) has investigated how the density of certified brownfields affect housing prices. But the relation with Airbnb has not been made very often.
This thesis will investigate this relation for the Amsterdam market. Will limiting Airbnb in the city also help solving the scarcity on the housing market? Or will it just help to resolve the nuisance caused by Airbnb guests? The results of this research could be used to base policies for these two markets on or test their effectiveness. We used the same hedonic approach as is used by most of the
Page | 5 literature about the effect of externalities on the housing market. We found that there is a significant difference between the effects of Airbnb listings on the housing prices for different areas in
Amsterdam. For 8 of the 11 areas this effect was positive with the highest value for Amsterdam Oost. The next section in this thesis describes the existing literature on the two topics. Next, the datasets used are described. Thereafter, the methodology is explained in detail and the final section reports the results and give a discussion.
Literature review
To get a better view on how Airbnb is influencing the Amsterdam housing market we first have to analyse how it operates. From now on, a person who offers accommodation on Airbnb is defined as a host and the accommodation offered on Airbnb is defined as a listing. After a slow start, Airbnb now has a bigger market share which is still increasing on the traditional accommodation market. Also, started it to be seen more and more as a disruptive product in this industry (Guttentag, 2015). Therefore, its current growth is also expected to flatten out over the upcoming years, but it will probably already have obtained a big market share by then (Guttentag, 2015). To be able to analyse where Airbnb listings will have the biggest impact it is important to know where they are located the most and what factors influence this. This research is done on Airbnb data in London by Quattrone, Proserpio, Quercia, Capra, & Musolesi (2016). From their results, we can conclude that listings are likely to be in the centre of the city. This is in line with the findings that listings appear to be in areas that are easily accessible by public transport and have more flats than houses. It also seems to be the areas where the citizens are younger and employed (Quattrone et al., 2016). Moreover, the question that raises is what determines the price of Airbnb listings. Wang & Nicolau (2017) use a hedonic price regression to estimate what factors are influencing the price of sharing economy-based
accommodation. These are basically the same property factors determining the prices of houses plus some additional factors such as host and online review aspects, amenities, services and renting rules. Data on Airbnb in this research is retrieved from insideairbnb.com (Wang & Nicolau, 2017).
Because of its exponential growth, Airbnb has been disrupting other markets which it is related to. One of them is the hotel industry. Since Airbnb listings are mainly competing with hotel rooms, it is likely that there exists a relation between the concentration of listings and hotel prices in these areas. This effect is investigated by Zervas, Proserpio, & Byers (2017). In their research Zervas et al. (2017) divide the hotels into 5 different segments to see if there is a different effect on prices for different price tiers. They find reduced hotel revenues since the entry of Airbnb. This effect is stronger for the lower priced hotel rooms. Since Airbnb is using residential houses to offer accommodation it is reasonable to think that this also has its effect on the residential housing market. How this works and what effects it has on the housing market in Los Angeles is investigated by Lee (2016). In his paper, multiple channels are described in which Airbnb is influencing rental prices in the housing market in L.A. One mechanism is that every Airbnb listing in L.A. is a residential house that is removed from the market. If a property owner chooses to use a house for Airbnb rental it can no longer be used as a residential house. Therefore, there are less houses available for Los Angeles’s citizens to live in. Second, if Airbnb prices are lower than hotel prices but still higher than residential rents, property owners have no incentive to rent a house to Los Angeles residents. This is decreasing the supply of the residential houses and so prices are moving more up (Lee, 2016). About this same effect, Levendis, & Dicle (2016) published a report on the rental housing market in New Orleans, accounting for different zip codes. They try to distinguish what part of the rental increase is caused by a raise in demand and what part is due to Airbnb activity. However, after controlling for general housing demand and supply they were unable to find a correlation between rental rates and Airbnb activity in any neighbourhood of New Orleans (Levendis et al., 2016).
Page | 6 Housing prices worldwide have gone up very rapidly in the recent years and there are many theories and opinions about the cause of this increase. Liu, Chang, Su, & Jiang (2016) argue that one of the reasons for this effect is that real estate has become more and more of an investment field, even for individual home owners. People are starting to buy houses at prices they cannot actually afford. This trend is driven by the expectation that prices will continue to rise. Therefore, an
opportunity to sell against a profit in the future is created (Liu et al., 2016). This could lead to major macroeconomic fluctuations and even to the creation of a bubble in the housing market. This has already happened before in the American housing market in the years prior to the 2007 financial crises. Case & Shiller (2003) have investigated the existence of such a housing bubble. They claim that fundamentals such as increasing incomes and low interest rates have caused the biggest part of the increase in housing prices. However, Eichholtz, Huisman, & Zwinkels (2015) claim that only a small part of the house prices, 10 to 40 percent, is explained by these fundamentals and that the biggest part depends on trends on the housing market. Therefore the explanation for the increase in house prices cannot be found in supply and demand models (Eichholtz et al., 2015).
However, Glaeser, Gyourko, & Saks (2004) try to explain the rise in housing prices by looking at the supply and demand mechanism. They claim that too often the focus is only on factors on the demand side when analysing housing prices. Therefore, they try to use supply side factors to explain for the sharp increase in housing prices. As supply side factors they take construction costs, land prices and regulations on new constructions (Glaeser et al., 2004). In another research, Glaeser et al. (2005) use the same model to try to explain the extraordinary high prices in New York’s Manhattan. They claim that not an increase in the demand is responsible for this effect but that it’s caused by factors on the supply side. Since there is a big gap between construction costs and housing prices there, Glaeser et al. (2005) claim that in this case the power of land limits new constructions which had led to the high prices in Manhattan. In their research on the prices of single family homes, Case & Shiller (1987) present a different method for estimating housing prices. They use the weighted repeated sales method to control for property characteristics since they only use data on properties that have sold twice or more. This is an alternative to the hedonic price method.
The hedonic price method is often used to measure the effect of externalities on housing prices and this is exactly what is done by Gibbons, Mourato, & Resende (2014). They investigate which effect the amenity value from living close to natural areas and environmental resources has on property prices. Their data sample covers information about housing transactions from over more than 13 years. Furthermore, they control for internal house characteristics and fixed effects such a travel to work area to control for several labour market differences across space. They find that the majority of these environmental variables have a significant effect on the English property prices (Gibbons et al., 2014). The hedonic price method is also used by Wilhelmsson (2000) to estimate the negative externality traffic noise has on single family homes in Sweden. Moreover, he also gives basic assumptions and propositions of the model. For example, the model assumes that there is an
equilibrium on the housing market and that buyers and sellers have symmetric information.
Furthermore, it relies on the proposition that a house price is composed out of single attributes that give utility to a person and that therefore, individual prices can be assigned to those attributes (Wilhelmsson, 2000). Finally, we will discuss Linn's (2013) paper. Using the hedonic price method, he investigates how the certification of brownfields affects housing prices. To be able to estimate the effect certified brownfields have on the housing prices in Illinois he must implement the
concentration of such field in the regression equation. However, this regression could yield biased estimates. This potential bias is caused by the fact that the location of brownfields is likely to be correlated with other local variables. Moreover, it could also lead to reverse causality since
brownfields located in areas with increasing property values are more likely to be certified. To avoid biased estimates, Linn (2013) chooses not only to use a density variable. Therefore, he also
Page | 7 constructs a gravity variable which puts more weight on brownfields that are located closer to the property than the ones that are further away (Linn, 2013). This same method is used by van der Bijl (2016) in his research in the effect of Airbnb on housing prices. He uses data from insideairbnb.com to define the Airbnb activity. The methodology used in this thesis is based on the research method by van der Bijl (2016). He uses the amount of reviews to represent the level of Airbnb activity. He calculates this density with the same count variable as is used by Linn (2013). He finds a positive effect of 0.27% on the average housing prices in Amsterdam if the Airbnb density increases with 10,000 reviews posted the year before the transaction date in a 1-kilometre radius.
Data
The data used in this research comes from two different datasets. The first dataset contains information about Airbnb in Amsterdam and is retrieved form insideairbnb.com. This is an organisation which claims to be independent and non-commercial. It uses scrapers to get information and figures about Airbnb from its website for cities all over the world. The second dataset contains information about all the housing transactions conducted by NvM related real estate agents in Amsterdam for the period 2006-2016. The data is provided by the NvM itself and covers 95% of the total market.
The first dataset about Airbnb consist of two sub datasets. The first sub dataset contains information about all the separate listings in Amsterdam. It starts in the year 2009 when Airbnb was first introduced in Amsterdam and ends in 2016. That year is chosen because this is the final year covered in the NvM dataset. The total amount of listings is 15,165 observations. We choose to drop the inactive listings without any reviews. Also, observations including missing values are dropped. Together this accounts for 2,200 observations so that the final dataset contains 12,965 listings. These listings are placed by 11,278 unique hosts.
Furthermore, this dataset contains information about the host, neighbourhood, and exact location of the listing. For the exact location of the listing, Airbnb uses a 150 meters standard error to protect the privacy of the host. Since this thesis only uses the neighbourhood of each listing it is not necessary to control for this deviation. In addition, it contains information about the price, the minimum amount of nights the listing must be booked, number of reviews and the type of listing. In table 1 below, you can find some descriptive statistics about the listings in Amsterdam. You can see that 77.42% of the listings in Amsterdam are entire homes or apartments. Guests spend on average 2.83 nights at a listing for an average price of €132.48 per night.
Table 1
Descriptive statistics of the active listings from Airbnb
Variable Frequency Percent
Entire home/apartment 10,038 77.42% Private room 2,877 2.19% Shared room 50 0.39% Total 12,965 100%
Descriptive Statistics about the active Airbnb listings in Amsterdam for the period 2009-2016.
Variable Mean Std. deviation Min. Max.
Price €132.48 78.48958 €9 €2100
Number of nights 2.82 7.621482 1 523
Number of reviews
Page | 8 The second Airbnb sub dataset is the dataset containing all the reviews posted about the listings in Amsterdam for the period from 2009 to 2016. This dataset accounts for the activity of each listing. The necessary assumption that needs to be made here, is that every Airbnb user posts on average the same amount of reviews per visit. In that way every review posted represents the same amount of Airbnb stays. Figure 1 below gives the amount of reviews posted per year. The total amount of reviews posted reaches 218,863. As you can see, the amount reviews posted per half year is
increasing. It is growing so fast that the second half year of 2016 almost accounts for one third of the total amount of reviews ever posted since the introduction of Airbnb in Amsterdam. The original dataset contains information about the reviewer, the host and the listing, the date of the review and the actual review self. For this research, only the date of the review is being used. Using the listing id’s in each Airbnb dataset the two datasets are merged into one Airbnb dataset containing all the information about the location and date of a review. This information is needed to account for Airbnb activity per area that will be discussed later in this paper.
Figure 1
Amount of Airbnb reviews posted in Amsterdam per half year
Next, we discuss the NvM dataset containing information about the housing transactions in Amsterdam. As said before, this dataset covers the period from 2006 until 2016. It contains
information about the transaction price and date of housing transactions conducted in Amsterdam for this period. Furthermore, it contains variables giving information about all kinds of internal housing characteristics. This can be factors such as construction year, the presence of a garden or parking, the state of maintenance all other kinds of specific housing characteristics. After removing observations that miss essential values the dataset has 84,481 observations. An analysis looking for potential outliers showed that observations with a price higher than €3 million need to be deleted. Therefore, 18 observations were dropped. This same analysis on house size in square meters showed us that 861 observations with a size higher than 287 square meters needed to be dropped. As well as all the 57 observations with a size lower than 22 square meters. Furthermore, 7 observations with more than 20 rooms were deleted from the dataset. In addition, all observations without any rooms at all were dropped accounting for 208 observations. As a result, a total of 1,145 observations were dropped leaving us with a dataset containing 83,330 observations. In table 2 you can see descriptive statistics about the NvM dataset.
In this paper, the densities are the main variables of interest. They are constructed as follows. All the observations from the Airbnb dataset as well as the observations from the NvM
Page | 9 dataset were categorised on half years. This was done by giving a value h1 to every observation with a date between January 1 and June 30 and h2 otherwise. Dividing the observations in half years reduces the interval between the date of a review included for a certain density and the date of a housing transaction. The problem of taking reviews into account that were posted after a particular housing transaction stays but since this thesis uses the proposition that the Airbnb density gives a measure of popularity for a certain area this will not cause any problems. In the end it’s not the actual Airbnb listing that adds value to the selling price of a house but the popularity of the
neighbourhood it is in. The period of a half year fitted best for capturing changes in these densities. After categorising both datasets in half years the amount of reviews posted per half year per neighbourhood was calculated in the Airbnb dataset. This number will represent the density that is used in this thesis. Next, the densities coming out of the analysis in the Airbnb dataset are put into the NvM database. This is done by matching the densities on the same neighbourhood and half years as the housing transactions. At the end, all the densities per neighbourhood are merged creating 11 different density variables.
To be able to match neighbourhoods for the two different datasets some areas had to be merged. The Airbnb dataset is using the newest areal classification for neighbourhoods in
Amsterdam dividing it into 22 different areas. However, the NvM uses the CPB 2006 areal
classification. This classification contains 14 different areas. To avoid any overlap of neighbourhoods the Centrum area and Westerpoort from the NvM dataset were merged containing Centrum-West and Centrum-Oost from the Airbnb dataset. Second, Oud-West and Baarsjes had to be merged in the NvM dataset because this is seen as one area in the Airbnb dataset. Zeeburg in the NvM dataset contains both Ij-Zeebrug and Oost-Indische from the Airbnb dataset. In the NvM dataset Osdorp contains Osdorp and Aker-Nieuw-Sloten from the Airbnb dataset. Zuid-Oost in the NvM dataset represents Bijlmer-Centrum, Bijlmer-Oost and Gaasperplas-Driemond from the Airbnb dataset. Then Oud-Zuid and Zuider-Amstel from the NvM dataset were merged into Zuid containing Buitenveldert-Zuidas, Pijp-Rivierenbuurt and Zuid from the Airbnb dataset. Finally, Slotermeer-Geuzeveld and Slotervaart from the NvM dataset were merged into Nieuw-West.
For every area a dummy variable was created to link the right amount of Airbnb reviews posted per half year from the Airbnb dataset to the matching observation in the NvM dataset. Next, dummies were created for the construction year of a house starting from the year 1906. Moreover, the dataset contained information about if a house had a garden, parking if it was detached and if it was an apartment or not. Dummies for these factors were created as well. This had to be done because the original dataset used categorical dummies giving also information about the
specifications of these factors. Finally, a dummy was created to check if the maintenance of a house was good in the period of the transaction. The original dataset contained 2 separate categorical variables giving the state of the in and outside maintenance on a scale of 1 to 9. Where 9 represents the highest level of maintenance. 5 was the average and so all the values between 6 and 9 stand for a higher maintenance level than the average houses. Next, the average of inside and outside
maintenance was calculated for every observation. After that, the maintenance dummy variable was generated giving 1 to every observation with an average maintenance level of 12 or higher and 0 otherwise.
Page | 10
Table 2
Descriptive statistics of the housing transactions
Variable Mean Std. deviation Min. Max.
Transaction price 302,919.5 209,305.6 48,500 2,980,000 House size in m2 84.60019 39.20342 22 287 Number of rooms 3.248734 1.295096 1 19 Number of bathrooms 0.9272051 0.4325446 0 7 Apartment 0.8915757 0.3109174 0 1 Parking 0.1286331 0.4008323 0 1 Detached 0.0061442 0.0781445 0 1 Garden 0.2224289 0.4158802 0 1 Maintenance Good 0.9681627 0.1755677 0 1 Construction year 1906-1930 0.2984639 0.4575869 0 1 Construction year 1931-1944 0.0885515 0.2840971 0 1 Construction year 1945-1959 0.0452178 0.2077828 0 1 Construction year 1960-1970 0.0887676 0.2844096 0 1 Construction year 1971-1980 0.0341054 0.1815009 0 1 Construction year 1981-1990 0.1016081 0.302134 0 1 Construction year 1991-2000 0.948398 0.2929953 0 1 Construction year 2001 or later 0.0771751 0.2668707 0 1 Density Centrum 441.34 2247.918 0 18,872 Density Westerpark 48.8006 368.0726 0 4,648 Density Oud-West 243.4178 1263.29 0 10,811 Density Zeeburg 54.15013 356.0928 0 3,775 Density Bos en Lommer 27.53114 232.1729 0 3,318 Density Noord 54.74824 371.2037 0 4,128 Density Osdorp 11.63306 108.8594 0 1,538 Density Nieuw-West 12.55826 92.63651 0 1,255 Density Zuid-Oost 9.324697 93.50291 0 1,385 Density Oost 67.83532 459.6255 0 4,903 Density Zuid 528.8125 1943.965 0 12,203 Year 2011.252 3.332564 2006 2016
Page | 11
Methodology
In this analysis we use a hedonic price regression as mentioned in the literature review to estimate the effect of Airbnb listings on the housing prices in Amsterdam. As said before, the main variables of interest here are the Airbnb densities per area. To make it easier to analyse the effect of a change in Airbnb density on housing prices, we take the density per 1000 reviews per half year for every single neighbourhood. Therefore, the effect of a unit change in density on housing prices has an easy interpretation. We get the following regression where the dependent variable is LnPijt which is the natural logarithm of the transaction price ‘i’ for area ‘j’ in period ‘t’. Where ‘i’ stands for a particular housing transaction. The constant is given by α and Airbnbdensityijt gives the Airbnb density per 1000
reviews per half year for the area of the observation.
𝐿𝑛(𝑃
𝑖𝑗𝑡) = 𝛼 + 𝛽
1𝑖𝐴𝑖𝑟𝑏𝑛𝑏𝑑𝑒𝑛𝑠𝑖𝑡𝑦
𝑖𝑗𝑡+ 𝜀
𝑖𝑡(1)
First, to control for internal housing characteristics we include vector Xit. This vector controls forfactors such as the size of a house which is given in natural logarithm in the regression, construction year and multiple dummies controlling for things like a garden or parking. All the housing
characteristics can be found in the descriptive statistics of table 2.
𝐿𝑛(𝑃
𝑖𝑗𝑡) = 𝛼 + 𝛽
1𝑖𝐴𝑖𝑟𝑏𝑛𝑏𝑑𝑒𝑛𝑠𝑖𝑡𝑦
𝑖𝑗𝑡+ 𝛽
2𝑖𝑋
𝑖𝑡+ 𝜀
𝑖𝑡(2)
Second, we must control for year and location fixed effects. Prices change over time and can be quite different per area. The year fixed effects are given by 𝛾t and the location fixed effects are given by λj.The location fixed effects are based on the PC6-codes that are included in the NvM database. This gives the following equation.
𝐿𝑛(𝑃𝑖𝑗𝑡) = 𝛼 + 𝛽1𝑖𝐴𝑖𝑟𝑏𝑛𝑏𝑑𝑒𝑛𝑠𝑖𝑡𝑦𝑖𝑗𝑡+ 𝛽2𝑖𝑋𝑖𝑡+ 𝛾𝑡+ 𝜆𝑗+ 𝜀𝑖𝑡
(3)
Finally, an interaction variable between the year and location fixed is included. This is done
because both effects can be dependent on each other. This is given in the final regression 4.
Page | 12
Results
In column 1 the results of the basis regression are described. Only the effect of Airbnb
density on the housing prices is analysed here. The densities for Centrum, Westerpark,
Zeeburg, Oud-West, Oost and Zuid are all positive at significant at a 1% level. A negative
effect is found for Bos&Lommer, Noord, Osdorp, Nieuw-West and Zuidoost. The coefficient
of the density for Nieuw-West is not significant whereas the coeffiecient for Bos&Lommer is
only significant on a 5% level. The other 3 coefficients are significant at a 1% level.
In regression 2, there is controlled for internal housing characteristics. Except the
variable for the number of rooms that is not significant, all the other variables are significant
at a 1% level. The sign of the Airbnb density for Bos&Lommer and Nieuw-West changes from
negative to positive. The size of the coefficient for the areas Centrum, Zeeburg, Osdorp,
Zuids-Oost and Zuid decreases compared to regression 1. For the other areas the effect is
stronger compared to regression 1. The house size has linear effect on the housing prices. A
1% change in house size would lead to a 0.95% increase in the house price. The effect of the
number of rooms is negative but as is said before it is not significant. The variables for the
dummies apartment, parking, detached, garden and good maintenance are all positive. Their
effects on the housing prices are 11.3%, 4.9%, 3.0%, 6.3% and 13.3% respectively. The signs
of all the coefficients for the construction years are negative. Houses build within one of
these time intervals have on average a 11.1%, 19%, 40%, 55.2%, 58.4%, 28.6% or 27.6%
lower price than house build before 1906. Controlling for all these internal housing
characteristics has increased the adjusted R-squared from 0.1093 to 0.8100 implying a much
better fit of the model.
In regression 3 there is also controlled for year and location fixed effects. The year
fixed effect controls for natural prices changes in the market and the location fixed effects
control for local developments per area that could affect housing prices. Compared to the
results in regression 2 there are some slight changes in coefficients for some variables. All
the coefficients are significant at a 1% level and the adjust R-squared increases to 0.8172.
The results of the final regression are given in column 4 of table 3 below. Now also
the interaction variable between the year and location fixed effects is included. As you can
see, all the coefficients from regression 4 are significant at a 1% significance level. As one
would expect, the internal housing characteristics such as the house size, number of rooms
and number of bathrooms are all positive. The dummies controlling for if a house has a
parking, garden, good maintenance and if a house is detached are all positive. This all makes
sense since it’s logical to think that these factors would add value to a house which would be
reflected in the transaction price. All the dummy variables for the construction year are
negative. This could be explained by the fact that Amsterdam has a lot of old and historical
buildings and houses that are mostly built before 1906. These buildings carry historical value
and history which adds value to the price of the property. This could explain the fact that in
general the houses that are built later have a lower value. Also, some of these historical
houses are part of the cultural heritage but unfortunately the NvM dataset didn’t contain
any information about this so for these effects couldn’t be controlled for.
If you look at the density variables for the 11 areas, you can see that the influence of
Airbnb listings is quite different. There are 8 areas that experience a positive influence of
Page | 13
Airbnb listings on the housing prices as can be seen in table 3. The coefficient of the density
variable varies from 0.0322 for Centrum which is the Centre of the city to 0.1256 for Oost
which is the East of Amsterdam including the Watergraafsmeer. The result is remarkable
since the area Centrum accounts for the most Airbnb reviews posted in Amsterdam, so you
would expect the highest influence here. An explanation for this could be that Airbnb density
has a less stronger effect on houses form higher segments. Such an effect was also find by
Zervas et al. (2017) in their research on the effect of Airbnb on hotel room prices. The
increase of housing prices, if the amount of reviews per halfyear raise with 1000, with
12.56% in Oost is also very remarkable since this effect is very high since the number of
reviews posted here is not so big. One explanation for this could be that the centre of
Amsterdam is expanding and that this part of the city is becoming like a second Centrum. It
is especially popular with younger people who could value the presence and opportunities
Airbnb brings more.
For 3 of the areas the effect is even negative. These areas are Noord, Osdorp and
Zuid-Oost. The negative effect could be explained by the fact that these areas are all known
to have a big difference in housing classes. Since most of the houses that are normally active
and suitable for Airbnb are houses in the higher price classes. Therefore, Airbnb listings
could add value to these houses but make the rest of the houses in the lower segments less
attractive. As a result, prices for these houses drop or increase but less compared to the
other houses and the general effect of Airbnb for these areas could negative. The
coefficients for the 3 areas are, -0.0224, -0.0560 and -0.1087 respectively. This means that
an increase with 1000 listings per half year in these neighbourhoods on average would lead
to a decrease in the price of a house with -2.24%, -5.60% and -10.87% respectively. This is
quite a strong effect. Another explanation for the direction of the sign could be that Airbnb
in these neighbourhoods also has brought a lot of negative externalities such as nuisance
caused by Airbnb guests or a neighbourhood that becomes too crowded. If these negative
externalities are bigger than the positive effects Airbnb brings this could also be an
explanation for the negative sign of the coefficients. The adjusted R-squared of the final
regression remains at 0.8173. This means that 81.73% of the variation in the housing prices
is explained by all the variables in the final regression.
Page | 14
Table 3
Variable Regression (1) Regression (2) Regression (3) Regression (4)
Density Centrum 0.0457754*** (0.000762) 0.0370874*** (0.0003589) 0.0333962*** (0.0004067) 0.032269*** (0.0004694) Density Westerpark 0.0447048*** (0.0046394) 0.0895796*** (0.0021568) 0.0887972*** (0.0022324) 0.086687*** (0.0022725) Density Oud-West 0.0352923*** (0.0013556) 0.0390396*** (0.0006342) 0.0423409*** (0.0006847) 0.0420574*** (0.000687) Density Zeeburg 0.0992922*** (0.0047997) 0.0677203*** (0.0022742) 0.0886251*** (0.0023757) 0.0897799*** (0.0023868) Density Bos en Lommer -0.0184831** (0.0073508) 0.077363*** (0.0034608) 0.0861299*** (0.0035287) 0.0852187*** (0.003533) Density Noord -0.0534812*** (0.0046033) -0.0056583*** (0.0021682) -0.0185125*** (0.0023117) -0.0223983*** (0.0024413) Density Osdorp -0.1142184*** (0.0156706) -0.0986935*** (0.0073488) -0.057419*** (0.0074823) -0.0559692*** (0.007487) Density Nieuw-West -0.0136123 (0.0184362) 0.0252412*** (0.0087355) 0.0778074*** (0.0090312) 0.0785451*** (0.0090311) Density Zuid-Oost -0.4510274*** (0.0182397) -0.2296497*** (0.0086062) -0.1233827*** (0.0088167) -0.1087011*** (0.0093021) Density Oost 0.0919274*** (0.0037178) 0.0935037*** (0.0017241) 0.1225912*** (0.0018342) 0.1256239*** (0.0093021) Density Zuid 0.0611717*** (0.0008844) 0.0491647*** (0.0004148) 0.0563806*** (0.0004849) 0.0568786*** (0.0004952) House size m2(ln) 0.9535372*** (0.0032193) 0.9414424*** (0.0031816) 0.9415062*** (0.0031812) Number of rooms -0.0015491 (0.0010249) 0.0045859*** (0.0010138) 0.0045219*** (0.0010137) Number of bathrooms 0.0354049*** (0.0018875) 0.0339681*** (0.0018519) 0.0339001*** (0.0018517) Apartment 0.1133964*** (0.0032483) 0.1209552*** (0.0031893) 0.1206651*** (0.0031894) Parking 0.0491682*** (0.0023148) 0.0459893*** (0.0022715) 0.0457246*** (0.0022718) Detached 0.2952554*** (0.0104543) 0.2925095*** (0.0102551) 0.292721*** (0.0102538) Garden 0.0625686*** (0.0021684) 0.0662876*** (0.0021282) 0.0663527*** (0.0021279) Maintenance Good 0.1330823*** (0.0045361) 0.1332914*** (0.0044502) 0.1331939*** (0.0044496) Construction year 1906-1930 -0.111188*** (0.0024515) -0.0867073*** (0.0024475) -0.0871951*** (0.0024492) Construction year 1931-1944 -0.1908188*** (0.0033677) -0.1734599*** (0.0033188) -0.1738906*** (0.0033194 Construction year 1945-1959 -0.3966039*** (0.004293) -0.3681313*** (0.0042412) -0.3685025*** (0.0042413) Construction year 1960-1970 -0.5520333*** (0.0033772) -0.5275074*** (0.0033429) -0.5276491*** (0.0033425) Construction year 1971-1980 -0.583902*** (0.0047994) -0.5442038*** (0.0047633) -0.5447694*** 0.004764
Page | 15
Table 3 continued
Variable Regression (1) Regression (2) Regression (3) Regression (4)
Construction year 1981-1990 -0.4506107*** (0.0031977) -0.4133186*** (0.0032062) -0.4135577*** (0.003207) Construction year 1991-2000 -0.2864451*** (0.0033976) -0.2800721*** (0.0033351) -0.2805535*** (0.0033361) Construction year 2001 or later -0.2755662*** (0.0037583) -0.245254*** (0.003725) -0.2449545*** (0.003725) Housing characteristics
No Yes Yes Yes
Year fixed effects No No Yes Yes
Location fixed effects No No Yes Yes Interaction No No No Yes Observations 83,330 83,330 83,330 83,330 R-squared 0.1094 0.8100 0.8175 0.8173 Adj. R-squared 0.1093 0.8100 0.8172 0.8172 Table reports coefficients and standard errors (between parentheses) of OLS regressions. Significance level of 10%, 5% and 1% is respectively given by * , ** and ***.
Conclusion
In this thesis we studied the effect Airbnb has on housing prices for different areas in Amsterdam. Therefore, we used 2 databases from insideairbnb.com about Airbnb. One includes information about every listing in Amsterdam and the second contains information about the all the reviews ever posted about Airbnb listings in Amsterdam. For data on housing transactions we used a dataset that was provided by the NvM containing information about housing transactions for the period 2006-2016. After adjusting these datasets, we had to control for multiple factors. Finally, a hedonic price regression is used to calculate the effect of the 11 different densities.
To conclude, it can be said that Airbnb has an influence on the housing prices in Amsterdam. How big this effect is and in which direction it goes depends on the area. For the neighbourhood Noord, Osdorp and Zuid-Oost, the effect is negative. For the other 8 areas, it is positive. The
strongest effect on the housing prices was in Oost where an increase of 1000 reviews posted per half years adds 12.56% to the housing prices. The weakest relation between Airbnb listings and housing prices is in Noord where prices decrease with 2.24% if the reviews posted per half year increase with 1000.
One limitation of this study is that it hasn’t made use of a count variable. Instead it has categorised the observations of the datasets in half years to calculate the density of the housing observations. Hence, reviews posted after a housing transaction took place could be included in the density for that observation. This is still within the time interval of 6 months. To resolve this problem, a count variable such a constructed by Linn (2013) could be used to calculate the Airbnb density per housing observation. This count variable would count all the reviews, posted within 1 kilometre of the housing transaction, that were posted in the year before the transaction date. In this way, more precise results will be generated. Unfortunately, during this research constructing such a variable didn’t succeed due to time constraints.
Page | 16
References
Bijl, V. (2016). The effect of Airbnb on House prices in Amsterdam. A study of the side effects of a disruptive start-up in the new sharing economy. University of Amsterdam
Carson, B. (2016, February 3). How 3 guys turned renting an air mattress in their apartment into a $25 billion company. Business Insider. Retrieved from:
http://www.businessinsider.com
Case, K. E., & Shiller, R. J. (1987). Prices of Single Family Homes Since 1970: New Indexes for Four
Cities (Working Paper No. 2393). National Bureau of Economic Research.
Case, K. E., & Shiller, R. J. (2003). Is There a Bubble in the Housing Market? Brookings Papers on
Economic Activity, 2003(2), 299–342.
Eichholtz, P., Huisman, R., & Zwinkels, R. C. J. (2015). Fundamentals or trends? A long-term perspective on house prices. Applied Economics, 47(10), 1050–1059.
Gibbons, S., Mourato, S., & Resende, G. M. (2014). The Amenity Value of English Nature: A Hedonic Price Approach. Environmental and Resource Economics, 57(2), 175–196.
Glaeser, E. L., Gyourko, J., & Saks, R. (2004). Why Have Housing Prices Gone Up? American
Economic review, 95, 329-333.
Glaeser, E. L., Gyourko, J., & Saks, R. (2005). Why is Manhattan So Expensive: Regulation and the Rise in Housing Prices. Journal of Law and Economics, 48, 331–370.
Guttentag (2015). Airbnb: disruptive innovation and the rise of an informal tourism accommodation sector. Current Issues in Tourism, 18:12, 1192-1217.
Kraniotis, K. (2017, September 29). De explosieve groei van Airbnb in Amsterdam is voorbij. NOS. Retrieved from: https://nos.nl
Levendis, J., & Dicle, M. F. (2016). The neighbourhood Impact of Airbnb on New Orleans. SRRN Retrieved from: http://ssrn.com/abstract=2856671
Lee, D. (2016). How Airbnb Short-Term Rentals Exacerbate Los Angeles’s Affordable Housing Crisis: Analysis and Policy Recommendations Student Notes. Harvard Law & Policy Review, 10, 229– 254.
Page | 17 Linn, J. (2013). The effect of voluntary brownfields programs on nearby property values: Evidence
from Illinois. Journal of Urban Economics, 78(Supplement C), 1–18.
Lonica. (2017, January 12). Stijging in huizenprijzen Amsterdam (2013 – 2017). Infonu. Retrieved from: https://financieel.infonu.nl
Liu, T.-Y., Chang, H.-L., Su, C.-W., & Jiang, X.-Z. (2016). China’s housing bubble burst? Economics of
Transition, 24(2), 361–389.
Meijden, N. (2017, June 12). Amsterdams bedrijf wil eerste Airbnb-wijk ter wereld oprichten. Parool. Retrieved from: https://www.parool.nl
Quattrone, G., Proserpio, D., Quercia, D., Capra, L., & Musolesi, M. (2016). Who Benefits from the ‘Sharing’ Economy of Airbnb? In Proceedings of the 25th International Conference on World
Wide Web (pp. 1385–1394). Republic and Canton of Geneva, Switzerland: International
World Wide Web Conferences Steering Committee.
Wang, D., & Nicolau, J. L. (2017). Price determinants of sharing economy based accommodation rental: A study of listings from 33 cities on Airbnb.com. International Journal of Hospitality
Management, 62(Supplement C), 120–131.
Wilhelmsson, M. (2000). The Impact of Traffic Noise on the Values of Single-family Houses. Journal of
Environmental Planning and Management, 43(6), 799–815.
Zervas, G., Proserpio, D., & Byers, J. W. (2017). The Rise of the Sharing Economy: Estimating the Impact of Airbnb on the Hotel Industry. Journal of Marketing Research (JMR), 54(5), 687– 705.
