The sharing economy and externalities: the impact of Airbnb on house prices

(1)

The sharing economy and externalities:

The impact of Airbnb on house prices

December 2017

Master thesis

MSc Finance: Finance and Real Estate Finance track

Author:

Stijn Martens (10447040)

(2)

2

Abstract

Due to the growing popularity of the sharing economy, there exist disruptive effects in traditional markets. This paper addresses the rapid development of Airbnb and its impact on the housing market. It is argued that Airbnb could have a positive effect on housing prices as the hosts of an Airbnb listing can receive an extra income, which could indirectly be translated into a higher mortgage. On the other hand, the home-sharing platform is part of controversy as neighbours of Airbnb listings are complaining about nuisance from tourists. This could have a negative effect on housing prices. In this paper the development of Airbnb in Amsterdam is addressed using data from the Dutch Association of Real Estate brokers (NVM) and Airbnb data obtained from the independent organisation Inside Airbnb. Also, the effect of Airbnb on other cities like Den Haag, Rotterdam, and Utrecht will be studied. By using a hedonic regression model, the effect of Airbnb is analysed. The results indicate that Airbnb on average increases property prices in Amsterdam by 0.048% when the Airbnb density within a 1,000-meter radius increases by 1% a year prior to the transaction date. In Den Haag, Rotterdam, and Utrecht the effect on housing prices is also positive on average increase by 0.024% when the Airbnb density within a 1,000-meter radius increases by 1% a year prior to the transaction date. It must be noted that the number of Airbnb listings in Den Haag, Rotterdam, and Utrecht is considerably lower than in Amsterdam and therefore the growth potential for Airbnb is higher in these municipalities.

Statement of originality

This document was written by Stijn Martens who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is 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)

3

Acknowledgements

This thesis concludes my academic career. It started in 2009 at the Technical University of Eindhoven. After a few months in the study I was diagnosed with Hodgkin’s disease. After a successful chemo- and radiation-therapy I was cured and was able to continue my education in Eindhoven. Unfortunately, due to concentration problems left after the chemo-therapy I was unable to complete the program. After a few years I started the Economics and Business bachelor at the University of Amsterdam. This went well and eventually I received my bachelor’s degree. During the program I found that my interests moved towards the world of finance and real estate. Coincidently, I found the double specialisation program offered by the University of Amsterdam which allows to study both the finance and the real estate finance master tracks.

Now, after eight and a half years I’m ready to finalize my education. While I enjoyed most of my studentship, the last couple of months were the hardest. After a 21 year-long battle against cancer, my mother-in-law passed away which had a deep impact on my girlfriend, my family, and myself. It took some time to recover, but with the help from my girlfriend, friends, family, and my supervisor prof. dr. Marc Francke I was able to rediscover myself and finalize this thesis. I want to thank my supervisor for his support, his time, his suggestions, and his insights which were invaluable and without him I would not be able to finish this thesis. Also I would like to thank dr. Martijn Dröes for helping with the repeat sales model. Furthermore, many thanks my girlfriend Marijke Westerink for always being there to support me, keep her faith in me, and incite me to go the extra mile. Further, I would like to thank my father Peter Martens and my father in law Dick Westerink for proofreading my thesis. Finally, I would like to thank Ron Engelhart, Tomas van Gelder, Thomas de Beer, Nick van der Horst, Seppe Salari, and Dominicq Pfundt for giving me great incentives and making sure that I would go to the library. They provided a great motivation for me to keep working and writing.

Kind regards, Stijn Martens 14-12-2017

(4)

4

1. Introduction

The emergence of peer-to-peer markets, collectively referred to as the sharing economy, allows individuals to act as both the suppliers and receivers of a good or a service. This enables individuals to let others use their under-utilized goods through sharing economy platforms (Botsman and Rogers, 2010). The most notable platforms are Uber and Airbnb, a car-sharing and a home-sharing platform, respectively. In 2014, the sharing economy became a more important industry with estimated global revenue of $ 15 billion, which is expected to grow to $ 335 billion by 2025 (PwC, 2014). One of the disturbing reasons why the sharing economy is able to thrive this much is because of the lagged legislation. Policymakers are noticing the rapid development of the platforms and observe the accompanying challenges, such as unfair competition and safety, but also other side effects such as increasing housing and rental prices (European Parliament, 2015). While consumers enjoy participating in this sharing economy, the social impact might be bigger than anticipated.

Over the years, numerous reports have highlighted the negative side effects of sharing economy platforms. Uber faces legal battles worldwide as the taxi industry argues that Ubers’ taxicab operations are illegal and take away their business (Rogers, 2015). Uber’s services were banned in Germany as their drivers did not have the appropriate licenses and insurances, and therefore impose unfair competition (Cohen and Zehngebot, 2014). At the same time, users urge the governments to relax legislation in favour of Uber as it provides more efficient use of goods (Rogers, 2015). Moreover, to participate in the sharing economy users often log in via social media accounts so they have no alias to hide behind. They are then able to promote or disparage themselves and create a new kind of currency, called reputation capital (Botsman, 2012). Through reviews, consumers are able to reward good behaviour and penalize bad behaviour. In contrast to the blurred world of finance, the transparency of the sharing economy holds people responsible for their own actions.

Although this way of control aids in regulating the sharing economy, the major issues at hand are not yet dealt with. For example, due to the lagged legislation, there is a rapid increase of illegal hotels in Amsterdam. An Airbnb listing is illegal in Amsterdam when the host does not pay a tax. Therefore, in order to cope with the sharing economy platform, the municipality of Amsterdam decided to collaborate with Airbnb and is one of the first municipalities in the world to do so (Airbnb, 2014). The municipality receives data from Airbnb regarding the hosts who rent out an accommodation and who is coming to visit. In addition, some regulations are imposed which comprises a limit on the number of days a listing can be rented out per year, a limit on the number of tourists allowed per listing per time, and a tourist tax. However, even after the regulations were imposed on the first of January 2015, the number of Airbnb listings continued to increase (Appendix 1). This growth is accompanied with an increase in the number of tourists who are staying in neighbourhoods that are not designed to accommodate travellers. More than 70 percent of the Airbnb listings are located outside the areas which

(6)

6 house the majority of hotels in Amsterdam (Airbnb, 2014). This in turn leads to nuisance for the residents, and complaints at Airbnb and the municipality of Amsterdam.

Furthermore, there is a belief that Airbnb is responsible for the increase in housing and rental prices. A report from ING (2016) notes that the extra income generated through subletting a house can be used to receive an additional mortgage of € 100,000. However, not every homebuyer will sublet their house, and therefore makes the extra cash flow a potential income. Furthermore, it is recognized that homebuyers are not that rational and real estate brokers cannot take a potential income into account (Parool, 2016). Nonetheless, there are more studies that report an increase in property values due to an increase in popularity of the short-term rental market. Schäfer and Braun (2016) report an increase in housing and rental prices in Berlin. In addition, Kim, Leung, and Wagman (2016) find that the private short-term rental market has an impact on housing prices. There are also reports from Pairolero (2016) and Lee (2016) who find that Airbnb does not have an impact on housing prices, thereby adding more fuel to the debate.

The purpose of this thesis is to estimate the impact of Airbnb on property prices and the time on the market in Amsterdam, Den Haag, Rotterdam, and Utrecht. The property price is the transaction price that is paid by a buying party to acquire a property. The time on the market is the time in days the property is listed on the market. On the first day, the initial price (listing price) for the property is announced and on the sales date the transaction price becomes known. There is extensive research on the link between the transaction price and the time on the market where the majority finds a negative relationship (Sirmans, Macdonald, and Macpherson, 2010). In other words, the longer a property is on the market, the lower the transaction price. Furthermore, by estimating the impact on property prices and the time on the market, this paper tries to aid municipalities who encounter the same short-term rental sector problems.

Subsequently, the aim of thesis is to answer the following question: what is the impact of Airbnb on property prices? To quantify the impact of Airbnb, this paper will use two different analyses. The first analysis is a hedonic model which uses the transaction price as its dependent variable, and the main independent variable of interest will be the number of Airbnb listings in an 𝑥-meter radius around the property, referred to as the Airbnb density. Furthermore, the model will include listing prices, the time on the market, property characteristics, time characteristics, and location characteristics to control for fixed effects.

As part of the hedonic model, an approach comparable to a difference-in-difference approach is used. A difference-in-difference approach can be used to estimate an average treatment effect. Here, the average treatment effect is the differential effect of a treatment on a ‘treatment group’ versus a ‘control group’. The treatment here is the presence of Airbnb in Amsterdam. Normally, a treatment is measured from a given point in time. However, the number of Airbnb listings in the municipality gradually increases over time. This means that properties are already influenced by the presence of Airbnb before the home-sharing platform is widely known. To account for this, an Airbnb constant will be generated.

(7)

7 All properties and all Airbnb listings are used in order to estimate the impact of Airbnb on the housing prices and the time on the market. Therefore, all properties in Amsterdam sold over the period 2000-2016 will be subjected to the Airbnb listings, specifically the Airbnb listings in December 2000-2016. Thereby, a new Airbnb density variable is created and this will function as an Airbnb constant.

For the second analysis, a repeat sales model is used to analyse the price effect and the time on the market. The first sub analysis has the transaction price as dependent variable and the second sub analysis has the time on the market as dependent variable. The models will quantify the effect of Airbnb listings in the neighbourhood for properties that have been sold over the last couple of years since Airbnb is active.

The dataset which includes transaction prices, listing prices, time on the market, and property characteristics from every municipality mentioned above is obtained from the NVM, the Dutch Association of Real Estate Brokers. The Airbnb dataset is retrieved from Inside Airbnb, an independent non-commercial organisation.

The results of this study imply that Airbnb has a positive impact on property prices. In Den Haag, Rotterdam, and Utrecht the property prices increase by 4.4 basis points when the Airbnb density within 250 meters increases by 1 percent a year prior to the transaction of the property. In Amsterdam, property prices on average increase by 5.5 basis points when the Airbnb density within 250 meters increases by 1 percent a year prior to the transaction date.

This paper is set up as follows. Section 2 presents some background information on Airbnb, and its development will be addressed along with media reports which highlight the benefits and controversy of the sharing economy platform. Section 3 gives an overview of the existing literature surrounding the sharing economy, externalities on property prices, the effect of the time on the market, and the methodology used in order to execute the analysis. Section 4 provides the methodology used in this paper. Section 5 specifies the datasets used in this study and provides descriptive statistics. Section 6 will elaborate on the results obtained and section 7 reports some robustness checks. Section 8 provides the limitations of this research and gives possibilities for further research. Finally, section 9 presents the conclusion.

(8)

8

2. Airbnb background information

This section will provide some background information on how the Airbnb platform developed over time and identifies some issues of the home-sharing platform.

The Airbnb company was founded by Joe Gebbia, Brian Chesky, and Nathan Blecharczyk (Inc., 2010). In 2007, the roommates Gebbia and Chesky could not afford to pay rent and then decided to turn their apartment in a room that could fit three mattresses. They knew that a big conference was coming to San Francisco and that hotels would be full, so they created a simple website and provided a place to sleep for three people and also offered breakfast. Their first three guests payed 80 dollars per person to sleep on an air matrass and receive breakfast. Afterwards, the guests said they liked their stay. This led to the realization that it could be a big idea. To develop a website, they contacted an old roommate, Blecharczyk. By summer 2008, they had a working website, airbedandbreakfast.com, and allowed people to book a stay with only three clicks. Unfortunately, no investors were interested in the idea. To generate funds and keep the website running, they created special edition cereal boxes (Wall Street Journal, 2008). With the presidential elections in mind, they designed Obama O’s and Cap’n McCains and sold the cereal boxes for $ 40 on the streets. Their marketing strategy yielded $ 30,000 for the website and also got the company noticed by Paul Graham, a programmer for a start-up accelerator called Y Combinator. Graham recognized the possibilities, of Airbnb and offered cash and training to the founders in exchange for a small share of the company (Business Insider, 2016). During this time, the search for investors continued. While money was scarce, Gebbia, Chesky and Blecharczyk lived of the leftover cereal.

In March 2009, the company simplified their name to Airbnb and the site contained 2,500 listings and around 10,000 registered users (Tech Crunch, 2009). Furthermore, they expanded their listing types from air beds and private rooms to entire homes and apartments, but also more unique listings such as castles, boats, and even tree houses (Rewind & Capture, 2014). Then in April 2009, Sequoia Capital recognized the potential and invested $ 600,000. The founders note that this was the accelerator on growth (Business Insider, 2016). In November 2010, the company raised $ 7.2 million, and in February 2011 Airbnb announced its 1 millionth booking since August 2008 (SF Chronicle, 2011). Through July 2011, Airbnb had raised $ 119.8 million in funding (GigaOm, 2011). In January 2012 the apartment sharing website announced its 5 millionth booking, and the company continued its rapid growth as by June 2012 the 10 millionth booking milestone was reached (Market Wired, 2012). The next investment came from TPG Capital in April 2014 who invested $ 475 million and valued Airbnb at $ 10 billion. By March 2015, a new funding round resulted in a $ 1 billion investment and a valuation at $ 20 billion. In the summer of 2015, the apartment sharing website was compared in value with hotel brands, which are Airbnb’s main competitors. It ranked third, just behind established companies Hilton and Marriot (Skift, 2015).

(9)

9 In august 2017, Airbnb closed another investment of $ 1 billion, which brings total raisings since August 2008 to more than $ 3 billion and ups the valuation of the company to approximately $ 31 billion (CNBC, 2017a). In addition, Airbnb became profitable for the first time in the second quarter of 2016. Nowadays, the apartment sharing website offers more than 3 million accommodations worldwide, in more than 65,000 cities, in 191 countries (Airbnb, 2017a). According to Brian Chesky, Airbnb’s policy stands for something bigger than travel. It would stand for community and relationships and using technology for the purpose of bringing people together (Fortune, 2016). To support this vision, the company provided free housing when hurricane Sandy hit the city of New York (CNN Tech, 2012). The Donated Sandy Housing program was welcomed by people in need of a place to stay. This led to the development of Airbnb’s Disaster Response program, which gives homeowners the possibility to provide shelter to people who have encountered an unforeseen situation. In the beginning of 2017, the program helped people who were not able to go back to the United States because of the travel restrictions invoked by the Trump administration. The restrictions blocked travellers, refugees, visa holders, and other citizens from seven Muslim countries from entering the United States (CNBC, 2017b).

In addition to the strengthening of the community, Airbnb offers more wealth to municipalities. In a study performed by Airbnb itself, they found that on average Airbnb guests stay 3.9 nights in Amsterdam and spend € 792 during their trip, whereas hotel guests stay an average of 1.9 nights and spend € 521 (Airbnb, 2017b). Furthermore, people who stayed in 2015 through Airbnb in Amsterdam had to pay tourist taxes. At the end of 2015, Amsterdam received € 5.5 million in tourist tax from Airbnb (FD, 2015).

Despite these positive developments, there are numerous reports showing a darker side of the home-sharing website. It is suggested that Airbnb is responsible for a growth in rental and housing prices in Amsterdam. The Dutch bank ING (2016) mentioned in a report that people with an Airbnb listing can easily receive € 350 per month. This extra income could be translated into a higher mortgage amount of € 100,000 which leads to a higher property price. At the moment, the housing prices in Amsterdam are booming.

However, Barbara Baarsma, professor Economics at the University of Amsterdam, argues that it is impossible that residential property prices increase because of Airbnb (Parool, 2016). She says that homebuyers are not rational and real estate agents cannot include a fictive income when determining the maximum mortgage loan. Baarsma believes that the increase in house prices stems from an increase in demand and notes that the nuisance caused by Airbnb has a negative effect on housing prices.

More and more residents of Amsterdam are complaining about Airbnb as the tourists that stay in an Airbnb accommodation cause nuisance (Volkskrant, 2016). The main cause of the nuisance is that neighbourhoods with an Airbnb listing are not designed to accommodate tourists. Every residential property is a potential Airbnb listing, which makes a stay by using Airbnb more unique as it allows a tourist to ‘live like a local.’ Unfortunately, when tourists live like a local, they stay near locals. These

(10)

10 areas are not designed for tourism which results in nuisance to neighbours of Airbnb listings, mostly in the form of extensive noise. The lack of regulation in the sharing economy enables external effects and therefore more research is necessary in order to develop a solid regulatory framework.

In Amsterdam, a regulatory framework was necessary in order to cope with the rapid increase of Airbnb listings in Amsterdam. Therefore, the municipality of Amsterdam announced a cooperative effort with Airbnb in order to provide a collective solution for the private holiday rental market (Airbnb, 2014). The rules imposed put a restriction on the properties that could function as an Airbnb. With the new regulations, hosts are permitted to rent their accommodation no more than 60 days per year, a maximum of four guests at the same time, the hosts have to collect a tourist tax, and the hosts must have the ownership of the rented property. This last action means that renters, in the social or private sector, cannot offer their property on a short-term rental platform. The collaborative effort between Airbnb and the municipality has resulted in a closing of 284 listings in 2016 with the majority being illegal hotels (NOS, 2017). Another step in the battle against illegal hotels is the registration of hosts. By October 2017, people who want to rent out their apartment must register with the municipality, otherwise they will receive a fine which could amount up to € 300,000.

These imposed regulations are an example for other municipalities as the ‘Airbnb-problem’ is not only restricted to the capital of the Netherlands. Berlin tried to cope with the popularity of Airbnb by imposing a ban (Guardian, 2016). The municipality recognized an increase in rent and housing prices and a growing housing shortage which led to the misuse of more than 5,500 flats in Berlin (Schäfer and Braun, 2016). Therefore, the municipality introduced the Zweckentfremdungsverbot, ‘ban on wrongful use’, in May 2016. Under this law individuals are able to let no more than 50 percent of their apartment on a short-term basis. Failing to abide to this law will result in a fine of € 100,000. Unfortunately for Berlin, a few homeowners who let their second home in Berlin appealed for a withdrawal of the ban as they could not be present in two apartments for more than 50 percent of the time. In August 2016 the City Court ruled in favour of homeowners who rent out a second home via Airbnb (Phys, 2016). Therefore, secondary accommodations can be let for a longer period of time and do not have to abide to the 50 percent rule, but the Zweckentfemdungsverbot stays in effect for primary homes.

However, since December of 2017 the municipality has revoked this decision (Berliner Kurier, 2017). Homeowners, but also tenants of apartments, will be able to rent out their homes up to 60 days per year through sites like Airbnb. The municipality allows the short-term rental of homes but the hosts need to register and authorisation from the municipality. Otherwise, they could receive a fine up to € 100,000. By regulating the short-term rental market, the municipality of Berlin tries to contain the increase in rent and housing prices, thereby noting that externalities exist due to the development of the short-term rental market.

(11)

11

3. Literature review

This section will elaborate on what the sharing economy is and what kind of effect its emergence has on traditional markets. Next, studies that research the effect of Airbnb in traditional markets are discussed along with studies that test the effect of externalities on house prices. Afterwards this section concludes with an analysis of the time on the market and how it affects the transaction price.

3.1 The sharing economy

The sharing economy can be defined as “the peer-to-peer-based activity of obtaining, giving, or sharing the access to goods and services, coordinated through community-based online services” (Batsman and Rogers, 2010). It is an economic-technological event that is powered by technology developments in the information and communications sector, and growing consumer awareness (Hamari et al., 2015). By offering a platform, the sharing economy allows individuals to act as a provider of items that are costly to buy and are owned by people who do not make use of them. Botsman and Rogers (2010) note that these new platforms enable assets previously unavailable or not conveniently available are becoming widely accessible. Platforms like eBay enable individuals to act as a retailer, Uber allows individuals to be a taxi service, and Airbnb let’s individuals act as a hotel whenever it is convenient. This phenomenon, also referred to as collaborative consumption, dictates that access to goods and skills is more important than ownership of them (Stephany, 2015). With the help of peer-to-peer platforms individual parties are able to interact and carry out economic transactions themselves without interference of traditional companies, thereby evading traditional intermediaries (Nadler, 2014). This allows people access to underutilized goods and services, and increasing their utility. It appears that the average car in North America and Western Europe is in use 8 percent of the time (Sacks, 2011), which means that cars are the ultimate expensive underutilised commodity. Shelby Clark, founder of RelayRides, a platform that enables the sharing of cars owned by individuals, notes that when people’s mobility costs shift from being fixed (ownership) to variable (renting), they make more efficient decisions about when they actually need to drive. Through these developments, the sharing economy provides innovative forms of income, or maybe even welfare, to individuals.

The concept of sharing is not new, but in recent times more sharing economy companies are emerging. This can happen because sharing economy companies are united in threefold (Kaplan and Nadler, 2015). Firstly, they rely on technological advances which were not previously available in order to satisfy consumer demand. Secondly, the companies exist in and parallel to well established industries that are disrupted by the emerging of the sharing economy, and thereby offer alternative goods and services. Thirdly, as the companies operate in ‘grey’ areas of the law they present new issues that were not foreseen. This final trait makes it difficult for traditional companies to compete with these innovative sharing economy companies as they are not bound by the same rules and regulations. Innovation is

(12)

12 essential to increase the competitiveness of firms, but the regulation of innovation requires a comprehensive regulatory approach (Porter, 1990). The emergence of Uber is a good example. The company mediates through an application between travellers and individuals with a vehicle. Travelers can request a ride from and to a fore mentioned location and individuals can react to this. Due to its rapid development, regulation lags behind. This lack of regulation resulted in protests and strikes by licensed taxi companies in a number of European and American cities (Rogers, 2015). Ultimately, the resistance led to a prohibition of Uber in Germany by the court as the services offered were not deemed safe. Furthermore, the court emphasized that Uber does not have the necessary licenses and insurance, and therefore posed unfair competition.

Airbnb is the sharing-economy platform for accommodation rentals. Due to a growth in tourist demand for accommodation, the platform was able to grow rapidly (Guttentag, 2015). The listings are provided by hosts who have the ownership of the property. Some hosts stay in the accommodation, some are away, and a few use their accommodation with the sole purpose to sublet it to others. On the other hand, tourists use Airbnb because of cost reduction, but also to have the experience of another culture and the interaction with local hosts (Guttentag, 2015). However, this has become so popular that negative reports are coming in and the home sharing platform is criticized.

Because of the emergence of the sharing economy, traditional legal barriers are blurred which results in legal ‘grey’ areas, legislation uncertainty, and unfair competition (Cohen and Zehngebot, 2014). The most difficult challenge of the sharing economy is illustrated in the matter between the encouragement of innovation and the need to protect consumers from practices that might endanger public health or safety. These developments can also be seen in the short-term rental market, where Airbnb faces legal problems.

3.2 Housing prices and externalities

As mentioned in the previous chapter, the development of Airbnb can be seen both as a negative and a positive effect. The negative effect being the increase in nuisance experienced by neighbours. The positive effects are a larger supply of accommodations for tourists and the extra income for homeowners, which could have an upward effect on the housing price. It is suggested that Airbnb is responsible for a growth in rental and housing prices in Amsterdam, as people with an Airbnb listing can generate extra income. When using a valuation model, it becomes clear that an increase in income will result in a higher valuation of the entity. As modelled by Lusht (2001), when valuing a property with a fixed cash flow for a certain period of time, and an additional income can be received, the value of a property will increase. This could indicate that there is a positive effect of Airbnb listings in the neighbourhood on property prices as it allows the homeowners to provide their accommodation on Airbnb and thereby generate an extra income.

(13)

13 To support this view, Biagi, Brandano, and Lambiri (2015) estimate the impact of tourism on housing prices. Their research is based on 103 Italian cities over the period of 1996-2007 and their results confirm that an increase in tourism activity, the number of tourists that visit a city, positively affects house prices. When tourism activity increases by 1 percent, house prices will increase with 0.2 percent. The effects of negative externalities as for instance the increased crime levels do not have such a big impact. When crime levels increase by 1 percent, house prices will decrease by 0.07 percent.

A study on the effect of short-term rental regulations is performed by Kim, Leung, and Wagman (2016). They estimate the effect of short-term rental regulation on an island in Florida, United States. Anna Maria Island offers a unique dataset, as the island is home to three cities; the City of Bradenton Beach, the City of Holmes Beach, and the City of Anna Maria. The Island is a known vacation destination but there are no name-brand hotels located, which means that the majority of stays is through local offerings via short-term rental websites. In order to deal with an increase in tourism, the city of Holmes Beach imposed a length-of-stay regulation in 2007 for short-term rental accommodations. Due to a passed bill in the state of Florida in 2011, Anna Maria and Bradenton Beach could not invoke a minimum-stay requirement. What the findings of Kim, Leung, and Wagman (2016) suggest is that the ability to rent properties short-term with fewer restrictions may enhance property prices, where a rental restricting regulation may reduce property prices. A solid regulatory framework can therefore stabilize housing prices while allowing an open short-term rental market.

Legislation is also required in other cities as well. In Germany, the Airbnb platform is being criticized because it is claimed that Airbnb listings in residential flats are being misused for tourist accommodation and therefore are removed from the housing market. This leads to a decline in housing supply and increasing rents. In their study, Schäfer and Braun find that 5,555 residential flats in Berlin are being misused by Airbnb, which represents 0.30 percent of the total housing stock. In addition, they report that the rental growth is higher in the Airbnb neighbourhoods where the misuse of residential flats is higher.

In addition, Lee (2016) studies the impact of Airbnb in Los Angeles and acknowledges that Airbnb is a response to the affordable housing crisis in Los Angeles. Lee finds that 7,316 units of accommodation had been removed from Los Angeles’ rental market due to the demand for short term rental accommodations. Because of lagged legislation, these removed units were deemed illegal hotels. By removal of the units from the rental market, the study finds that the housing supply has reduced which leads to an upward shift in rents. Moreover, because of these developments, Lee suggests that the quality of life in Los Angeles decreases because of Airbnb as it would destroy the culture and identity of neighbourhoods. Lee also notes that the municipality has the responsibility to act and form legislation in order to restrict the rental of illegal hotels.

As mentioned, nuisance can be a negative effect of an increase in Airbnb listings. Through an increase in tourism, negative externalities like noise pollution, traffic, and crime will increase. These externalities can have an effect on property prices. The effect of noise on property prices is studied

(14)

14 numerously. Theebe (2004) investigates the effect of noise from planes, trains, and automobiles in Amsterdam and finds that there is a discount when properties are exposed to noise levels above 65 decibels. He notes that a maximum impact on property prices is up to 12 percent. In addition, he identifies that properties below the 65-decibel limit have a price premium of maximum 6.5 percent. Other papers support the negative relationship between airport noise levels and property prices. Nelson (1980; 2004) finds a negative impact on prices between 0.5 percent and 0.6 percent per decibel of additional noise and McMillen (2004) finds a 9 percent discount for properties.

The effect of nuisance on house prices is considered to be negative, but differs between different distances. As studied by Debrezion, Pels, and Rietveld (2007), they find that residential property within 250 meters distance to a railway station sell at a discount as opposed to residential properties that are located more than 250 meters away, which sell at a premium. This premium can be attributed to the increased accessibility and the decreasing level of nuisance when the distance to the railway station increases.

Another example results from the study of Quang Do, Wilburg, and Short (1994). Their study estimates the impact of neighbourhood churches on housing values. Findings indicate the effect of churches on sales price is negative up to approximately 260 meters and the impact decreases as distance from church increases. This research is important as it provides objective evidence in order to make more informed decisions regarding whether or not to apply regulations restricting land use. The same goes for the construction of schools in a neighbourhood. Sah, Conroy, and Narwold (2016) have studied the impact of school proximity on nearby residential properties. While other studies find a school proximity premium, their study finds strong evidence for a school proximity penalty for public elementary schools when housing is located within 1,500 meters of the school. This suggests that proximity is perceived as a negative effect.

Additionally, Dröes and Koster (2016) investigate the effect of wind turbines on house prices. By using a difference-in-differences approach they were able to find a 1.4 percent decrease in house prices located within 2 kilometres of a wind turbine. When a wind turbine is located within 750 meters of the property, the house prices decrease on average by 2.6 percent. A wind turbine provides alternative energy to reduce dependence on fossil fuels, and thereby reduces exposure to CO2, which is beneficial

for all. However, the external costs are borne by a specific group of homeowners. The methodology used in the paper of Dröes and Koster (2016) represents the basis for the methodology used in this paper. According to above studies, nuisance has a negative effect on housing prices and this effect declines as the distance to the property increases. Conversely, the increased tourism activity through Airbnb could have a positive effect on housing prices. In addition, a solid regulatory framework should aid in controlling property prices. Subsequently, the effect that Airbnb has on housing prices is ambiguous and can go either way.

(15)

15

3.3 Time on the market

Because the transaction price is a product of negotiations buyers’ and sellers’ motivations, it does not reflect the whole process. The selling party is faced with a simultaneous optimization problem: maximizing the sale price and minimizing the time on the market (Asabere and Huffman, 1993). This process is subject to negotiation and therefore subject to preferences and motivation to both real estate agents and their clients. The time on the market can be referred to as a measure of liquidity in real estate (Krainer, 2001). Liquidity is defined as the “inverse of the amount of time that elapses between the decision to sell a security and the receipt of the full market value by the seller” (Wood and Wood, 1985). Amihud and Mendelson (1986) observe that the lower the liquidity of an asset, the higher the return it is expected to yield. Which does not suggest that investors should only invest in low liquidity assets, but investors with long term holding periods benefit from holding low-liquidity assets. The more liquid a financial instrument, the higher the price for which it can be sold.

Krainer (2001) investigated the liquidity of residential real estate and used the time on the market as a measure of liquidity. He found that in periods when the market is ‘hot’ and housing is in high demand, and therefore higher liquidity, sellers do not increase their prices to take advantage. Instead the sellers make use of the greater liquidity in order to complete the transaction before the market moves against the sellers. In a market downturn, housing is in low demand, but sellers do not lower their prices in order to keep the same amount of liquidity as in a ‘hot’ market. Sellers keep their prices steady and look for a special buyer who values the property higher than other buyers in the market and therefore is willing to pay the higher price. Furthermore, in a study performed by Kluger and Miller (1990) they find that houses which are more homogenous have a higher liquidity. In their study, the property characteristics where mostly insignificant in explaining the time on the market.

Despite Krainer’s motivation and study, the majority of studies find a negative relationship between the transaction price and the time on the market as is analysed by Sirmans, Macdonald, and Macpherson (2010). They use a meta-regression on 23 studies to examine the relationship between the transaction price and the time on the market. By estimating the results, they find that the coefficients between models have varied but are mostly negative. A negative coefficient for the time on the market indicates that the longer a property is for sale, the lower the actual transaction price. Additionally, they highlight the importance of the time on the market in determining the transaction price of homes.

To account for the time on the market and the sale price, one also has to consider the listing price. The listing price is set when a seller puts his house on the market. The strategic role of setting the listing price is extensively studied by Huang and Palmquist (2001) and Anglin, Rutherford, and Springer (2003). The main argument is that the listing price can influence the number of potential buyers. By setting the price too high or too low affects the marketability of the property. It is expected that an increase in the listing price will result in an increase of the expected time on the market. Furthermore, Knight (2002) notes that the second main determinant for the transaction price is the how much a home

(16)

16 is overpriced initially. By using a price premium variable, the difference between the listing price and the actual transaction price, he estimates the time on the market. Moreover, he focusses on the changes in the price premium. His findings indicate that properties of which the listing price was revised took longer to sell and sold for less than homes that were initially priced ‘correctly’.

Additionally, Yavas and Yang (1995) argue that the listings price has a dual role. It provides a signal and a limit as well to the seller’s reservation price. Setting the listing price will have an effect on the number of buyers. Setting a high listing price with a relative low house valuation will result in a lower number of potential buyers and additionally an increase in the time on the market. Furthermore, the effect of an increase in the time on the market results in a higher selling price is ambiguous as it can be noted that there are sellers who are willing to wait longer and thereby hope to find a buyer with a high reservation price. On the other hand, a house which does not sell might be perceived as damaged and therefore will result in a lower sale price.

However, by including the transaction price and the time on the market in the same model an endogenous problem arises as both variables are simultaneously determined and are both related to the motivation of the sellers and buyers. To counter this endogenous problem, Dubé and Legros (2016) propose a two stage least squares approach. Therefore, it is necessary to find good instrumental variables for the endogenous variables, variables that are correlated with the endogenous variables. This leads to another challenge, as the instrumental variables may not be correlated with the error term. To find appropriate instruments, Dubé and Legros develop a spatiotemporal selectivity matrix which will not be explained further as it is out of the scope of this thesis. By developing the spatiotemporal selectivity matrix, Dubé and Legros have created appropriate instrumental variables and continue with the two stage least squares approach. The first stage is aimed at solving the endogenous problem. So first, they estimate the dependent variables by using the instrumental variables derived from the spatiotemporal selectivity matrix. The time on the market variable is estimated using the instrumental variables. Afterwards, this predicted value for the time on the market in the first stage is used in the second stage to estimate the transaction price. This approach is also used the other way around, by first estimating the transaction price with other instrumental variables and then using the predicted values to estimate the time on the market. The results indicate that by controlling for the endogenous problem the time on the market has a stronger negative effect on the transaction price than would be estimated by using an ordinary least squares approach. This is related to findings from Sirmans, MacDonald, and MacPherson (2010) as they show that studies using ordinary least squares produce a less negative estimate of the time on the market on the sale price as compared to a two-stage least squares approach. The methodological framework proposed here reviews the probability that the transaction price is influenced by the time on the market, and vice versa. The methodology used in the paper of Dubé and Legros (2016) is partly used in this paper to counter the endogeneity problem.

(17)

17

4. Methodology

This section will elaborate on the methodology used in this paper. First, the overall empirical approach will be discussed. Then the different methods will be explained in detail.

The purpose of this study is to estimate the effect of Airbnb on the transaction price of housing, and the time on the market of housing in Amsterdam, Den Haag, Rotterdam, and Utrecht. Therefore, two different models will be analysed: a hedonic model and a repeat sales model.

The hedonic model will include housing characteristics, time fixed effects and location fixed effects. The dependent variable will be the natural logarithm of the transaction price. The main independent variable of interest will be the Airbnb density, which represents the number of Airbnb listings within an 𝑥-meter radius a year prior to the transaction.

As part of the hedonic model, an approach comparable to a difference-in-difference approach is used. This approach will review the presence of Airbnb as a constant, i.e. the number of Airbnb listings in December 2016 will be calculated per property as if they were present all along. The result of this analysis will estimate the average treatment effect due to the presence of Airbnb near the property. The average treatment effect shows the extent to which the number of Airbnb listings located near the property is capitalized into housing prices. The analysis will also include property characteristics, time fixed effects and location fixed effects.

Finally, the effect of Airbnb is estimated by a repeat sales model. A repeat sales-model allows to estimate the first difference in house prices of a certain property between two sales dates, thereby avoiding the problem of price differences in homes with varying characteristics. In this study, the change in house prices will be estimated through the change in Airbnb listings located near the property in the period between two sales.

In order to control for serially correlated standard errors and therefore smaller standard errors in hedonic models, the standard errors are all clustered at the four-digit zip code level as studied by Bertrand, Duflo, and Mullainathan (2004). They found that hedonic models understate the standard deviation of the estimators. Not clustering the standard errors could lead to over rejection of the estimators. By using clustered standard errors, it is also possible to control for heteroscedasticity.

4.1 Hedonic model

To estimate if the number of Airbnb listings located near the property is capitalized in the property price a hedonic model is used. The study from Dröes and Koster (2016) offers the methodology needed in order to estimate the effect of Airbnb. The dependent variable will be the natural logarithm of the house transaction price, 𝑃𝑖𝑡, of property 𝑖 in year 𝑡. The 𝐴𝑖𝑡 is a density variable which represents the Airbnb density. This is the number of Airbnb listings within an 𝑥-meter radius of property 𝑖 a year prior to the

(18)

18 transaction. The 𝑥 will differ between 250, 500, 750, and 1,000 meters. In this analysis, the natural logarithm of the Airbnb density will be used. The hedonic model to be estimated is then:

ln 𝑃𝑖𝑡 = 𝛽0+ 𝛽1ln 𝐴𝑥𝑖𝑡−1+ 𝜀𝑖𝑡 (1)

Here 𝛽0 is a constant, 𝛽1 captures the Airbnb effect, and 𝜀𝑖𝑡 is an error term. Every property has its own characteristics which could have an effect on the price. Therefore, property characteristics are included to control for this.

ln 𝑃𝑖𝑡= 𝛽0+ 𝛽1ln 𝐴𝑥𝑖𝑡−1+ 𝛽′𝑋𝑖𝑡+ 𝜀𝑖𝑡 (2)

Here 𝑋𝑖𝑡 represents a vector of the property characteristics including for example house size and number of rooms. In addition, dummies for house type, parking, garden, maintenance, monumental properties, and construction year are created. 𝛽′ measures the effect of these property characteristics.

Furthermore, property prices could be influenced by time fixed effects and location specific effects (Basu and Thibodeau, 1998). The time fixed effects are controlled for per month. The zip codes in the NVM database and the Airbnb database allow to control for spatial autocorrelation. The locational effects will be controlled for at the four-digit zip code level. Furthermore, it must be noted that location is an important variable to explain house prices (Alonso, 1960). Moreover, in a research from Shoval, Mckercher, Ng, and Birenboim (2011) they highlight the importance of hotel location as their study shows that tourists spent a big part of their time in the immediate vicinity of the hotel. Therefore, location is an important variable for Airbnb listings as the density variable will be higher in more central locations.

ln 𝑃𝑖𝑡= 𝛽0+ 𝛽1ln 𝐴𝑥𝑖𝑡−1+ 𝛽′𝑋𝑖𝑡+ 𝜃𝑡+ 𝜂𝑗+ 𝜀𝑖𝑡 (3) Here 𝜃𝑡 controls for the monthly time fixed effects and 𝜂𝑗 captures the location fixed effects for location 𝑗. Furthermore, the time on the market (TOM) of a property will be included. In the model below, the time on the market is added:

ln 𝑃𝑖𝑡= 𝛽0+ 𝛽1ln 𝐴𝑥𝑖𝑡−1+ 𝛽2ln 𝑇𝑂𝑀𝑖𝑡+ 𝛽′𝑋𝑖𝑡+ 𝜃𝑡+ 𝜂𝑗+ 𝜀𝑖𝑡 (4) The relation between the time on the market and the transaction price is ambiguous. Asabere and Huffman (1993) find that a longer time on the market will result in a high transaction price, but Huang and Palmquist (2001) discover a negative impact of market duration on the transaction price. As mentioned above, numerous studies (Knight, 2002; Dubé and Legros, 2016) have shown that a challenge exists when introducing both transaction price and time on the market variables in a regression. This arises because both variables are simultaneously determined and are related to the motivation of the sellers and buyers. In order to account for this endogeneity problem, Dubé and Legros (2016) developed a spatiotemporal selectivity matrix in order to generate instrumental variables. Then a two stage least

(19)

19 squares approach was used to estimate the effect the transaction price on the time on the market, and vice versa. In the first stage, the instrumental variables were used to predict the time on the market. In the second stage, the predicted values for the time on the market were used to estimate the transaction price. This paper will use also a two stage least squares approach, but without developing a spatiotemporal selectivity matrix.

4.1.1 Two stage least squares

The first stage is used to solve the endogenous problem as both dependent variables are represented as independent variable in the other equation. As suggested by Dubé and Legros (2016), in the first stage the original variable is regressed on a list of independent variables, which is the same as equation (2).

ln 𝑃𝑖𝑡= 𝛾0+ 𝛾1ln 𝐴𝑥𝑖𝑡−1+ 𝛾′𝑋𝑖𝑡+ 𝜖𝑖𝑡 (2)

Here, 𝐴𝑥𝑖𝑡−1 represents the number of Airbnb listings within an 𝑥-meter radius a year prior to the transaction date, 𝑋𝑖𝑡 indicates a list of independent variables and property characteristics and 𝜖𝑖𝑡 represents an error term. Now, the transaction price is estimated and this value will be used to estimate the time on the market. However, the time on the market also depends on the listing price premium. This is the difference between the listing price, 𝑃𝐿, and the estimated transaction price from equation (2), ln 𝑃̂ . Therefore, the endogenous variable time on the market is estimated with the Airbnb density, 𝑖𝑡 property characteristics and the list price premium as instrumental variables.

ln 𝑇𝑂𝑀𝑖𝑡 = 𝛿0+ 𝛿1∗ ln 𝐴𝑥𝑖𝑡−1+ 𝛿′𝑋𝑖𝑡+ 𝛿2[ln 𝑃𝐿− ln 𝑃̂ ] + 𝜏𝑖𝑡 𝑖𝑡 (5) The error terms are represented by 𝜏𝑖𝑡. Here, 𝜖𝑖𝑡 and 𝜏𝑖𝑡 are independent and uncorrelated. After estimation of equation (5), the predicted value for the time on the market, ln 𝑇𝑂𝑀̂ can be used as input 𝑖𝑡 for equation (4). Then the model of interest becomes:

ln 𝑃𝑖𝑡= 𝛽0+ 𝛽1ln 𝐴𝑥𝑖𝑡−1+ 𝛽2ln 𝑇𝑂𝑀̂𝑖𝑡+ 𝛽′𝑋𝑖𝑡+ 𝜃𝑡+ 𝜂𝑗+ 𝜀𝑖𝑡 (6)

4.2 Airbnb constant

As part of the hedonic model, an additional analysis will be done. This analysis is comparable to a difference-in-difference approach. In this approach all property transactions in Amsterdam from January 2000 until December 2016 will be subjected to the presence of Airbnb in order to create a constant Airbnb density variable. For all transacted properties, the Airbnb density will be calculated for all Airbnb listings in Amsterdam in December 2016. The number of Airbnb listings in December 2016 is chosen as the most recent file retrieved from Inside Airbnb represents the most accurate situation in Amsterdam. This is because Inside Airbnb removes inactive Airbnb listings from its database. By including an

(20)

20 Airbnb density constant, 𝐴𝑖𝑟𝑏𝑛𝑏𝑥, it is able to measure time effects and control for unobserved time invariant heterogeneity. This is necessary because it is possible that the Airbnb density variable measures something which is not there, i.e. a completely different effect is estimated. By introducing a constant, it is possible to control for this time invariant unobserved heterogeneity. The following model will be estimated

ln 𝑃𝑖𝑡= 𝛼0+ 𝛼1ln 𝐴𝑖𝑟𝑏𝑛𝑏𝑥𝑖+ 𝛼2ln 𝐴𝑥𝑖𝑡−1+ 𝛼3ln 𝑇𝑂𝑀̂𝑖𝑡+ 𝛼′𝑋𝑖𝑡+ 𝜃𝑡+ 𝜂𝑗+ 𝜎𝑖𝑡 (7) Here 𝑎0 is the constant and 𝑎1 captures the time effects. As mentioned, the new Airbnb density variable does not depend on time. Therefore, the 𝐴𝑖𝑟𝑏𝑛𝑏𝑥 variable represents the number of Airbnb listings in an 𝑥-meter radius when the property would be sold in December 2016. The time on the market is again estimated with help of equations (2) and (5). The error term is 𝜎𝑖𝑡 and the other variables are already explained above.

4.3 Repeat sales model

In addition to the hedonic models, this paper will also study the effect of Airbnb on residential property prices in Amsterdam via a repeat sales model. This approach dates from Bailey, Muth and Nourse (1963); Bryan and Colwell (1982); and Case and Shiller (1987, 1989). The repeat sales methodology is a special case of hedonic price modelling as it ignores changes in the hedonic values of the property between two sale dates. This approach will be used to estimate if the emergence of Airbnb is capitalized in housing prices. In order to do this, the following equation will be estimated

ln 𝑃𝑖𝑡− ln 𝑃𝑖𝑠= 𝜇𝑡− 𝜇𝑠+ 𝛽 ∗ (𝐴𝑥𝑖𝑡− 𝐴𝑥𝑖𝑠) + 𝜔𝑖𝑡− 𝜔𝑖𝑠 (8) With 𝑡 > 𝑠, which means that the sale of property 𝑖 in period 𝑠 is prior to the sale of property 𝑖 in period 𝑡. Here, 𝛽 measures the change in the number of Airbnb listings in the 𝑥-radius in the property. As mentioned above, 𝑥 differs between 250, 500, 750 and 1,000 meters. Time dummy variable coefficients are represented by 𝜇 and the error term is represented by 𝜔.

The model assumes that property characteristics remain constant over time. If a property has been renovated and characteristics have been changed due to this renovation, the specific property is excluded from the analysis. Also, this analysis will not include properties with a holding period of less than two years, as it is assumed that these properties are bought to be resold and therefore could have a higher price difference.

In addition, the repeat sales model will be modified to estimate if the time on the market is affected by the emergence of Airbnb. Therefore, the following model will be estimated

ln 𝑇𝑂𝑀𝑖𝑡 − ln 𝑇𝑂𝑀𝑖𝑠= 𝜇𝑡− 𝜇𝑠+ 𝛽 ∗ (𝐴𝑥𝑖𝑡− 𝐴𝑥𝑖𝑠) + 𝜉𝑖𝑡− 𝜉𝑖𝑠 (9) Here again the time dummies are represented by 𝜇, and the error term is represented by 𝜉.

(21)

21

5. Data

The analyses in this thesis are based on two main datasets. The first dataset contains the location of all Airbnb listings in Amsterdam since its start-up in 2008 and is retrieved from the website Inside Airbnb. This is an independent website which is not associated with or endorsed by Airbnb. The site utilizes information compiled from the Airbnb platform. For each listing the following is known: host id, listing id, listing type (entire apartment, private room, shared room), how many people the listing can accommodate, availability per year, price per night, the total number of reviews, the origin date the host started to let his or her apartment, and the listing location (GPS location, 4-digit zip code). However, since Airbnb anonymizes location information, Inside Airbnb notes that the location for a listing will differ by approximately 150 meters from the actual address. Therefore, some location bias is inherent to the data.

Figure 1 shows the number of Airbnb listings in Amsterdam over time. In September 2008, the first host started renting out their apartment on Airbnb. In July 2014, more than 7,500 listings were in the municipality and by March 2017 this number has more than doubled, which represents the rapid growth of Airbnb in the municipality.

When comparing the number of Airbnb listings in Amsterdam with the number of potential Airbnb listings there appears to be sufficient potential for growth. As mentioned, because every residential property is a potential Airbnb listing, the number of Airbnb listings can still grow in the future. According to the Central Bureau of Statistics in the Netherlands, the number of residential properties not available for rent in Amsterdam is 125,579 (CBS, 2017). This suggests that approximately 9 percent of all residential properties which are not available for rent are currently listed on the Airbnb platform as entire home. Due to the rapid increase of Airbnb listings in Amsterdam (Figure 1), it is expected there will be more Airbnb listings in the future. When looking at the occupancy of Airbnb listings in Den Haag, Rotterdam, and Utrecht, the growth potential is even bigger. CBS (2017) shows that the number of residential properties not available for rent at the end of December 2016 in Den Haag, Rotterdam, and Utrecht are 106,782, 107,507, and 66,456 respectively. Because all these municipalities have less than 1,000 Airbnb listings (AirDNA, 2017), which means that approximately 1-1.5 percent of the potential Airbnb listings is saturated in these municipalities, which shows there is growth potential for Airbnb. It must be noted that Amsterdam is the tourism hotspot of the Netherlands which could explain the higher degree of Airbnb listings. Nevertheless, the attractiveness of other cities is increasing as well (NBTC Holland, 2015).

Table 1 presents descriptive statistics for the Airbnb dataset. This analysis will include all Airbnb listings from its origin until December 2016; therefore all new listings since the start of 2017 are excluded. The total number of Airbnb listings available for the analysis is 14,711. It appears that 1,922 listings haven not yet had a review. However, more than 25 percent of these listings are offered less than a year on the platform and therefore might not have been able to let out their accommodation

(22)

22 Table 1

Descriptive statistics Airbnb dataset (December 2016)

Number of listings in Amsterdam 14,711 Number of unique hosts in Amsterdam 12,592

Hosts with 1 listing 11,556 91.77% Hosts with multiple listings 1,036 8.23%

Types of listings

Entire home/apt 11,459 77.89% Private room 3,188 21.67%

Shared room 64 0.44%

14,711

or they did not receive a review yet. Moreover, it is assumed that all Airbnb listings in the dataset are active listings because Inside Airbnb removes inactive listings before the dataset is configured.

Furthermore, the data show there are 12,592 unique hosts and it appears that 11,556 listings are hosted by one entity. This indicates the clear majority of listings are offered by private individuals. When reviewing the hosts with multiple listings, it appears that one host offers 91 Airbnb accommodations. This is a company called BNBSTARTUP which offers multiple listings and arranges every stay. Likewise, it appears that the majority of listings are entire homes and apartments, almost 78 percent of all accommodations. This could indicate that Airbnb is a good alternative for hotels due to the higher degree of privacy as well as comfort in an apartment as opposed to a private or shared room, which can be more attributed to features of a hostel. Moreover, as mentioned above, the host must have the ownership of the apartment that is offered on the Airbnb platform. Therefore, it is more beneficial to sublet an entire home/apartment as opposed to a private or shared room, which can be seen below.

0 2000 4000 6000 8000 10000 12000 14000 16000 Figure 1

(23)

23 Table 2 shows the average price per listing type. The average price for an entire home or apartment is € 148.20 per night, which is significantly more than one would pay for a private or shared room, € 88.20 and € 88.90 respectively. However, it is still much lower than the average hotel price in Amsterdam, which lies around € 204 as of May 2017 (Trivago, 2017), thereby supporting the view that Airbnb listings are a good and cheaper alternative for hotels. The distribution of prices of Airbnb listings is given in Appendix 2.

Moreover, the number of reviews per listing indicates the use and could represent the popularity of the different types of listings. As mentioned, the entire home is the closest alternative for a hotel, but it appears that private rooms are the most popular as the average number of reviews per private room listing is 36.24.

The second dataset contains all real estate transaction in the municipalities of Amsterdam, Rotterdam, Den Haag, and Utrecht over the period 2000-2016 and is obtained from NVM, Dutch association for real estate agents. The dataset from the NVM contains more than 300,000 transactions. ` For each transaction, the transaction price per property, listing price, and numerous property characteristics like the size in square meters, number of rooms, and construction year are given. The dataset also contains information on the time on the market, the sales date, and the GPS location. When analysing the data, some variables are not complete, and these specific transactions will be removed from the dataset. The dataset is checked for irrational values. Data without an irrational value gets an indicator value of 1, and data with an irrational value gets an indicator value of 0. For 1,610 transactions, there exist irrational values. For example, there are multiple dwellings with a property size in square meters of 99,999. Therefore, these transactions are dropped. The same goes for transactions with an irrational value for the garden size, which leads to a drop of another 696 observations. Furthermore, 1,427 transactions have 0 rooms and are therefore dropped. When analysing the number of rooms, there are 15 transactions with more than 30 rooms and these are also dropped from the dataset. For 89 properties, the construction period is unknown and therefore these transactions are dropped. Finally, 927 transactions are dropped because the GPS locations are missing. The NVM dataset consists of 315,723 transactions for the analysis.

Table 3 represents the descriptive statistics for the NVM dataset. The average transaction price is € 250,930 and the average time on the market in the four municipalities is 121 days, corresponding to approximately four months. Furthermore, there are some control variables. The majority of properties types in the dataset are apartments, followed by terraced properties and flats. The proportion of (semi-)detached houses is marginal, which is anticipated as these properties are rare in big cities.

The garden, parking, and listed indicator are all dummy variables which are one if the property has a garden, a parking space near, or is listed as a cultural heritage. Moreover, it appears that more than 85 percent of the properties in the dataset have a good interior maintenance, and the proportion of badly maintained properties is minimal. When reviewing the construction year, almost 25 percent of properties are built between 1906 and 1930.

(24)

24 Table 2

Descriptive statistics Airbnb listing types’ prices and reviews (December 2016)

Prices

Mean Median Min. Max. Entire home/apt € 148.20 € 125.00 € 20.00 € 3,142.00 Private room € 88.20 € 79.00 € 19.00 € 950.00 Shared room € 89.00 € 64.00 € 23.00 € 385.00

€ 134.94

Reviews

Mean Median Min. Max. Entire home/apt 13.32 6 0 496 Private room 36.24 11 0 483 Shared room 17.45 5 0 118

Table 3

Descriptive statistics NVM dataset

Mean Median Std. Dev. Min. Max.

25th percentile

75th percentile Transaction price 250,930 195,000 203,092 14,000 6,250,000 144,756 286,000 Transaction price per m2 _2,498 _2,222 _1,089 ₆₈ _18,204 _1,724 _3,065

Time on the market 121.3 62 170.5 0 1,825 26 147 House size in m2 _100.4 ₉₀ _50.6 ₂₅ _1,540 ₇₀ ₁₂₀

Airbnb density within 250 meter 5.8 0 22.4 0 285 0 0 Airbnb density within 500 meter 19.6 0 75.0 0 741 0 0 Airbnb density within 750 meter 40.5 0 153.2 0 1,520 0 0 Airbnb density within 1,000 meter 68.0 0 254.8 0 2,434 0 0 Terraced 0.233 0.423 Semi-detached 0.016 0.125 Detached 0.010 0.098 Apartment 0.515 0.500 Flat 0.227 0.419 Number of rooms 3.821 4 1.597 1 30 3 5 Garden 0.354 0.478 Parking 0.129 0.335 Listed 0.034 0.182 Maintenance quality good 0.859 0.348 Maintenance quality moderate 0.116 0.321 Maintenance quality bad 0.025 0.155 Construction year < 1906 0.117 0.321 Construction year 1906-1930 0.248 0.432 Construction year 1931-1944 0.144 0.351 Construction year 1945-1959 0.085 0.279 Construction year 1960-1970 0.100 0.299 Construction year 1971-1980 0.049 0.215 Construction year 1981-1990 0.079 0.270 Construction year 1991-2000 0.110 0.313 Construction year > 2000 0.069 0.254 Year of transaction 2008 2008 2000 2016 Number of observations 315,723

(25)

25 The descriptive statistics for each municipality can be found in Appendix 3. There it becomes clear that the properties in Amsterdam are on average sold at a higher transaction price than the properties in the other municipalities, which can also be seen in the transaction price per square meter. Moreover, the number of properties sold in Amsterdam is the highest of all the municipalities.

Because the exact GPS location of all transacted properties and GPS location of all the Airbnb accommodations is known, it was possible to calculate the straight-line distance of each property to the nearest Airbnb. By using the program ArcGIS it was possible to generate the number of Airbnb listings in a certain radius around the property a year prior to the transaction and thereby develop the Airbnb density. The average number of Airbnb listings found in a 250-meter radius is 5. This finding is arbitrary as only the number of Airbnb listings in Amsterdam are taken in to account. When reviewing the data from Amsterdam (Appendix 2) the number of Airbnb listings within a 250-meter radius is 15 on average. Furthermore, some properties in Amsterdam have more than 2,400 Airbnb listings within a 1,000-meter radius. Given that Den Haag, Rotterdam, and Utrecht all have less than 1,000 Airbnb listings, this finding is startling.

For the second hedonic model, a different Airbnb density is calculated. This variable will use all the available Airbnb listings in December 2016 as this dataset represents the most up to date Airbnb activity. Then it will perceive the Airbnb activity in Amsterdam as a constant, as if the Airbnb listings were always there. In the period 2000-2016, more than 114,000 properties were sold in Amsterdam (Appendix 3). For all these properties, the number of Airbnb listings in an 𝑥-meter radius will be calculated. This means that a property sold in January 2000, will now have the number of Airbnb listings in an 𝑥-meter radius around the property just as it would have been sold in December 2016. Reviewing the data, it becomes clear the Airbnb density has increased. This is given in Table 4.

As mentioned before, a repeat sales analysis will be done in order to estimate the effect of Airbnb. The properties eligible for the analysis must have been transacted at least twice in the period 2000-2016. Unfortunately, in the dataset there is no specific property-id available. However, there is an address-id. This address-id shows numerous properties that are located in the same building, but properties may differ between two floors. Therefore, the address-id is not a unique identifier. However, by deduction it is possible to generate a specific property-id. Firstly, the properties are sorted by address-id. Secondly, the properties with the same address-id are grouped together when they have the same house and garden size. It is assumed that properties with the same address-id, the same house size, and the same garden size are identical. It can be argued that properties with the same address-id and the same house size can be different. For example, a homeowner could have opted to break down a wall and thereby generate a larger room. However, the total house size will remain the same and therefore the property will be assumed to be identical. Due to the same location and the same house and garden size, it is assumed that grouped properties will have the same transaction price, even though the properties might not be the same in reality. This also means that the exterior view of a property is expected to be the same, which is not always the case. For example, a house can have a

The sharing economy and externalities: the impact of Airbnb on house prices