Effect of Airbnb on hotel industry in Amsterdam
Dave Jacobs, 10588426 University of Amsterdam
MSc in Finance - Double specialization: Finance and Real Estate Finance Master Thesis
July 2017
Abstract
Since its founding in 2008, Airbnb has evolved into the second most valuable private company in the United States. However, the company’s fast growth has led to controversy about the side effects of the start-up. This thesis focuses on the effect of Airbnb on the hotel industry in Amsterdam. This effect is analyzed by using a time series regression based on the hotel room database provided by the Research, Information and Statistics (RIS) Amsterdam, the Airbnb database retrieved from Inside Airbnb, and the house price database made available by the Dutch Association of Realtors (NVM) over a period from 2009-2016. The regression results show that, on average, every one Euro increase in the price per night of an Airbnb listing in the previous year leads to a decrease of 0.45 Euros in the price per night of a hotel room in Amsterdam.
Acknowledgements
This thesis is an important milestone in my academic career. First of all, I would like to thank my thesis supervisor dhr. dr. M.I. Dröes whom put in great effort and offered innovative suggestion during the semester. Secondly, I would like to thank the Dutch Association of Realtors (NVM) for making the data available on housing transactions. In addition, I would like to thank the Research Information and Statistics (RIS) Amsterdam for providing the hotel room database.
Statement of Originality
This document is written by Student Dave Jacobs 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.
Table of Contents
1. Introduction 3 2. Literature Review 6 2.1. Airbnb in Amsterdam 8 2.2. External effects 11 2.3. Airbnb on hotel industry 13 3. Data 15 4. Methodology 21 5. Results 23 5.1. Effect of Airbnb on hotel room prices 23 5.2. Effect of Airbnb on revenue per available room 25 5.3. Effect of Airbnb at the district level 27 6. Limitations and Future Research 32 7. Conclusion and Discussion 34 References 36 Appendices 381. Introduction
Since the financial crisis of 2008 the economy has undergone some changes. The emergence of platforms like Uber, Lyft, and Airbnb led to a peer-to-peer economy. Technology innovations and supply-side flexibility are two key factors that made it possible for such peer-to-peer platforms to grow rapidly (Zervas, Proserpio & Byers, 2016). For example, Airbnb has enabled individuals to collaboratively make use of under-utilized houses through fee-based sharing. As a result, houses are used more efficiently and Airbnb hosts generate some extra income. According to a study by Airbnb, 36 percent of Airbnb hosts in Amsterdam used the income they earned via Airbnb to help make ends meet1. Airbnb derives revenue from both guests and hosts for the home-sharing service. Depending on the length of the stay, guests pay a 9-12 percent fee for each booking, and hosts pay a 3 percent service fee. Since its introduction in 2008, Airbnb has evolved into a $31 billion valued start-up company offering over 3 million short-term rentals across 65,000 cities and 191 countries2. Almost fourteen thousand of those listings are located in Amsterdam, the capital of the Netherlands, where the start-up company has been active since 2009. In this thesis, the focus is on the effect of Airbnb on the hotel industry in Amsterdam. This leads to the following research question:
What is the effect of Airbnb on the hotel industry in Amsterdam?
Airbnb has served over 150 million guests worldwide and is valued at $31 billion. They claim to have generated 380 million Euros in economic activity in Amsterdam in 20153. In addition, Airbnb hosts welcomed 575,000 guests in Amsterdam in 2015. I therefore hypothesize that Airbnb has a significant effect on the hotel industry in Amsterdam. In line with Zervas et al. (2016), I hypothesize that some stays with Airbnb serve as a substitute for certain hotel stays in Amsterdam, and therefore negatively affecting the hotel industry in Amsterdam.
Since peer-to-peer platforms like Airbnb are a new phenomenon, there is relatively little literature on the subject. There is one study that investigates the effect of Airbnb on the hotel industry in Texas. Zervas et al. (2016) explore the economic impact of the sharing economy by analyzing the impact of Airbnb’s entry into the state of Texas on the Texas hotel 1 http://blog.atairbnb.com/economic-impact-airbnb/ 2 https://www.airbnb.nl/about/about-us 3https://www.airbnbcitizen.com/wp-content/uploads/2016/05/AmsterdamDataRelease.pdf 2 https://www.airbnb.nl/about/about-us 3https://www.airbnbcitizen.com/wp-content/uploads/2016/05/AmsterdamDataRelease.pdf
industry. They find that Airbnb supply has decreased hotel room revenue by 10 percent during the research period 2010-2014. The literature covering external effects affecting hotel revenues can be divided into two groups; positive and negative externalities. Hotel star rating and customer reviews are concluded as important determinants of hotel room prices (Schamel, 2012). Another hotel star rating or point in the online rating positively affects room prices. On the other hand, higher distance to the city center leads to less accessibility to attractions and public goods, leading to lower hotel room prices (Hung, Shang & Wang, 2010).
Although little academic literature is written about the subject, a large amount of press reports about the side effects of Airbnb is published. It started with a research into the short-term rental sector by the municipality of Amsterdam (2013). The report states conditions on nuisance, safety and a maximum number of guests per property for renting out houses. In addition, Airbnb agreed to collect tourist taxes and is expected to pay 8.2 million Euros in taxes to the municipality of Amsterdam in 20164. As a sequel to the agreements made in 2014, Amsterdam has reached a recent agreement with Airbnb to limit the sharing of private homes to 60 days a year. This agreement went into effect on January 1, 2017 and will last until the end of December 2018. Starting from October 1, 2017, Airbnb hosts also have a reporting duty to the municipality of Amsterdam. Despite the growing regulations, there is still a lot of controversy about the side effects of Airbnb.
In a report by the ING bank (ING, 2016) it is stated that Airbnb positively affects house prices in Amsterdam. The extra income generated by Airbnb hosts can cover interest and mortgage payments of an additional mortgage of almost 100,000 Euros, leading to higher house prices. As a response to the report of ING, Barbara Baarsma, professor at the University of Amsterdam, and Pieter van Dalen, Economist at Rabobank, wrote an opinion article. Baarsma and Van Dalen (2016) disagree with the report of the ING (2016) and state that Airbnb has no effect on house prices.
To solve the controversy about the side effects of Airbnb, this thesis aims to answer the research question: What is the effect of Airbnb on the hotel industry in Amsterdam? The research in this thesis uses three datasets. Data on Airbnb listings is collected from Inside Airbnb, an independent, non-commercial organization which offers public information compiled from the Airbnb website. Secondly, the hotel data is made available by the Research, Information and Statistics (RIS) Amsterdam. This dataset contains data on hotels in
4
https://www.nrc.nl/nieuws/2016/12/01/airbnb-en-amsterdam-sluiten-deal-over-illegale-woningverhuur-a1534433
Amsterdam for the period 2009-2016, including the number of hotel rooms, hotel room prices and occupancy rates. At last, data on house prices is made available by the Dutch Association of Realtors (NVM). This dataset includes transaction prices, time on market and house characteristics of houses sold in Amsterdam for the period 2000-2016. The data on house prices is used to account for neighborhood fixed effects. The main variable of interest is the price per night of an Airbnb listing.
This thesis contributes to the existing literature by investigating the effect of Airbnb on the hotel industry in Amsterdam. There is little literature on the subject and only one academic study examines the effect of Airbnb on the hotel industry. Zervas et al. (2016) study the impact of the sharing economy on incumbent firms. However, they analyze the effect of Airbnb’s entry on the Texas hotel industry. Secondly, this thesis contributes to the controversy about the side effects of Airbnb. The results may benefit the municipality of Amsterdam and the RIS Amsterdam in their research on the effect of Airbnb.
The results show that Airbnb negatively affects the hotel industry in Amsterdam. More specifically, every one Euro increase in the price per night of an Airbnb listing in the previous year leads to a decrease of 0.45 Euros in the price per night of a hotel room in Amsterdam. Furthermore, Airbnb presence leads to lower revenue per available hotel room, implying that Airbnb affects the hotel industry through both lower occupancy rates and decreased room prices.
The remainder of this thesis proceeds as follows. Section 2 presents the history of Airbnb and a literature review of prior empirical evidence followed by the hypotheses. Section 3 describes the data used in this thesis. In section 4 the applied methodology is explained, where section 5 analyzes the results. Section 6 discusses the limitations of this thesis, and section 7 concludes.
2. Literature Review
This section starts with an overview of the beginning of Airbnb followed by a literature review of prior empirical evidence regarding the side effects of Airbnb. In addition, studies investigating other determinants of hotel revenues are discussed. At last, an empirical study investigating the effect of Airbnb on the hotel industry is reviewed.
What started in 2007 with an idea of renting out three airbeds grew to a $31 billion valued start-up company. During the Industrial Designers Society of America (IDSA) conference in San Francisco in October 2007, all the city’s hotels were fully booked. In order to pay their rent, Joe Gebbia and Brian Chesky came up with the idea of renting out three airbeds on their living-room floor. They recruited Nathan Blecharczyk, Gebbia’s former roommate, to develop a website to get local people to list their rooms and travellers to book them. In the summer of 2008, the three co-founders found the perfect kick-start for their business. Taking advantage of the presidential election, they designed cereal packaging depicting Barack Obama and John McCain. They sold 800 boxes generating more than $30,000. This was just the beginning of the cash inflow.
In early 2009 Gebbia, Chesky and Blecharczyk raised their first funding of $20,000 from Y Combinator in exchange for a small interest in the company. This first funding led to a further $600,000 seed investment from Sequoia Capital in April 20095. By this time Airbnb
became profitable and the founders were making enough money to pay for living expenses. The company started growing rapidly and reached 700,000 nights booked in November 2010. In addition, the company raised another $7.2 million in Series A from Jeremy Stoppelman, Elad Gil, SV Angel, Y Ventures, Keith Rabois, Greylock Partners, and Sequoia Capital6. In one year, the company had seen 800 percent growth in nights booked, reaching its 1 millionth booking in February 2011. Half a year later, in July 2011 Airbnb received funding of $112 million from venture capitalists leading the company to be valued at $1.3 billion. As the company kept growing, Airbnb reached 10 million nights booked in June 2012. After raising another $475 million in funding in April 2014, Airbnb was at a $10 billion valuation7. Less than a year later, in June 2015, the company has closed a new $1.5 billion funding round8. 5 http://www.telegraph.co.uk/technology/news/9525267/Airbnb-The-story-behind-the-1.3bn-room-letting-website.html 6 https://techcrunch.com/gallery/a-brief-history-of-airbnb/slide/24/ 7 http://fortune.com/2014/08/01/airbnb-closes-475-million-funding-round/ 8http://fortune.com/2015/06/27/airbnb-raises-1-5-billion/
This round values the company at $25.5 billion. Later on, in September 2016, Airbnb has raised $555 million in new funds as the company expands around the world9. This funding round values the San Francisco-based online room rental service at $30 billion.
The most recent fund raising appeared in March 2017. According to a Securities and Exchange Commission, Airbnb has closed on a $447.8 million round of funding10. The company is now worth approximately $31 billion. Since its start in 2008, Airbnb has raised over $3 billion of funding. With Airbnb’s latest valuation exceeding all other hospitality companies, the start up is now listed as the second most valuable private company in the United Stated, following Uber, as CB Insights reported11. In addition, the company also turned profitable in the second half of 2016 and is expected to be profitable in 2017. This profitability is on the basis of the EBITDA of the company, meaning that expenses like interest, taxes, and depreciation are not taken into account.
Since its founding in 2008, Airbnb has grown to offer more than 3 million short-term rentals across 65,000 cities in 191 different countries12. However, not only positive effects of Airbnb are discussed in published articles. Also negative consequences of the online room rental service are dealt with. Bruinius (2015) states that the booming home-sharing website Airbnb has received a fair amount of backlash. In New York, affordable housing is one of the central pillars of the government. Rent regulations have been in place to keep many housing units affordable for a variety of income levels, but landlords can sometimes generate a higher income with short-term rentals through sites like Airbnb. For this reason, Airbnb is contributing to the affordable housing crisis. In addition, state attorney general Eric Schneiderman found that 75 percent of the Airbnb listings in New York City violated state and city laws (Bruinius, 2015). As a result, short-term rentals through the home-sharing website are mostly prohibited in New York. Also Berlin, the capital of Germany, started restricting private property rentals through short-term rental services like Airbnb in 2016. Since long-term rents in Berlin rose 56% during the period 2009-2014, affordable housing has become scarce13. Therefore, home-sharing websites like Airbnb have received a fair amount of criticism. Furthermore, as Berlin has become one of Europe’s most popular travel destinations, Airbnb has affected the local traditional hotel industry. From 1 May 2016, Berlin 9 https://www.bloomberg.com/news/articles/2016-09-22/airbnb-raises-at-least-555-million-in-funding 10 http://www.cnbc.com/2017/03/09/airbnb-closes-1-billion-round-31-billion-valuation-profitable.html 11 https://www.nytimes.com/2017/03/09/technology/airbnb-1-billion-funding.html?_r=1 12https://www.airbnb.nl/about/about-us 13 https://www.theguardian.com/technology/2016/may/01/berlin-authorities-taking-stand-against-airbnb-rental-boom
effectively banned short-term rentals of entire apartments to tourists without a license by introducing a new law entitled “Zweckentfremdungsverbot”. This law states that landlords can rent out up to 50 percent of the home they live in, and entire apartments can be rented out only if the apartment is officially registered as a holiday apartment14. The law was already passed in 2014 with a two-year transition period before coming into effect, which ended on April 30, 201615. From March 2016 on, the amount of Airbnb listings fell by 40% within two months as a result of the April 30 deadline. Many hosts removed their listings from the website preventively, in order to abide by the law before it came into effect16.
2.1. Airbnb in Amsterdam
Out of the more than 3 million listings worldwide, 14,000 are located in Amsterdam, the capital of the Netherlands. The city is in third place on the occupancy rates of hotels in Europe in 2016. Bernard D’heygere, spokesman for Airbnb, describes Amsterdam as a special city for Airbnb. The Dutch capital was one of the first cities in the world to embrace home sharing and is currently a world leader for the sharing economy. The city is home to the longest and strongest partnership between Airbnb and any city in the world, and one of the first cities in the world introducing clear and simple rules for hosts, and partnering with Airbnb to simplify tax payments for hosts17. In the Economic Impact Report for Amsterdam published by Airbnb in 2016 it is stated that the company generated 380 million Euros in economic activity in Amsterdam in 201518. Also, Airbnb paid the municipality of Amsterdam 5.5 million Euros in tourist tax. In a more recent report covering the whole country, Airbnb announced that the Airbnb community generated approximately 795 million Euros in total economic activity in the Netherlands in 2016, which takes into account guest spending and host income19.
Nevertheless, also in Amsterdam there have been concerns about the side effects of the hospitality company since its introduction in 2009. The concerns started with the research into the short-term rental sector by the municipality of Amsterdam (2013). The report covered 14 https://www.citylab.com/equity/2016/12/berlin-has-the-worlds-toughest-anti-airbnb-laws-are-they-working/509024/ 15 http://www.independent.co.uk/news/world/europe/airbnb-rentals-berlin-germany-tourist-ban-fines-restricting-to-protect-affordable-housing-a7008891.html#gallery 16 http://www.settle-in-berlin.com/airbnb-ban-berlin/ 17 https://amsterdam.airbnbcitizen.com/setting-priorities-for-airbnb-in-amsterdam/ 18https://www.airbnbcitizen.com/wp-content/uploads/2016/05/AmsterdamDataRelease.pdf 19 https://netherlands.airbnbcitizen.com/wp-content/uploads/sites/50/2017/02/The-Airbnb-Community_The-Netherlands.pdf
illegal hotels and the nuisance the guests cause. It states that the use of short-term rental services like Airbnb is permitted as long as there is no excessive nuisance and a maximum of four persons during a stay (Municipality of Amsterdam, 2013). In addition, the landlord has the obligation to pay tourist tax and has to ensure fire safety. In December 2014, a first Memorandum of Understanding was concluded with the City Council of Amsterdam and Airbnb20. This agreement made Amsterdam one of the first cities to embrace home sharing and the sharing economy. The agreement provides more accessible information on the rules for home sharing and states that Airbnb will collect tourist tax on behalf of hosts. However, since a home is primarily meant to live in, not to use as a hotel, Airbnb and the municipality of Amsterdam agreed upon a new agreement in November 2016 to stop illegal rents. The new pioneering agreement promotes responsible and sustainable home sharing and combats illegal hotels21. The agreement took effect on 1 January 2017 and expires on 31 December 2018 and focuses on four key priorities for home sharing in Amsterdam. First, to ensure entire home listings are not shared for more than 60 days, Airbnb introduced a new automated limit. Secondly, new tools to promote responsible home sharing are introduced. The home-sharing website introduced a day counter to help hosts track and limit home sharing activity. In addition, the new neighbor tool will be promoted22. This tool enables other residents to share
concerns about a specific listing, including noise complaints. Thirdly, Airbnb shares aggregated data of listings, including the total number of active hosts, total number of nights booked during the past calendar year and the number of listings in which the home is offered to more than four persons (Municipality of Amsterdam & Airbnb, 2016). At last, the pioneering agreement between Amsterdam and Airbnb strives towards making comparable agreements with comparable platforms.
Higher house prices as a side effect of Airbnb presence in Amsterdam is discussed in a study by the Dutch bank ING (2016). The report states that homeowners in Amsterdam could increase mortgage size by 100,000 Euros based on future Airbnb income. ING (2016) assumes an average rate per night of 130 Euros and the full use of the 60 days per year home sharing limit. Besides a place to live, owning a house is also a way to generate income. This extra income can cover interest and mortgage payments of an additional mortgage up to 100,000 Euros. Therefore, buyers are willing to pay more leading to higher house prices in 20 https://www.airbnb.com/press/news/samenwerking-tussen-airbnb-en-gemeente-amsterdam?locale=en 21 http://www.iamsterdam.com/en/media-centre/city-hall/press-releases/2016-press-room/amsterdam-and-airbnb-announce-new-unique-agreement 22http://www.dutchnews.nl/news/archives/2016/12/amsterdam-airbnb-agree-new-deal-to-stop-illegal-rentals/
Amsterdam. Homeowners profit from this increase, whereas people looking to buy a house face more difficulties finding a suitable and affordable house. The study by ING (2016) estimates that house prices in Amsterdam have increase by 2-4% due to Airbnb. As a response to the report of ING, Barbara Baarsma, professor at the University of Amsterdam, and Pieter van Dalen, Economist at Rabobank, wrote an opinion article for Het Parool, a Dutch newspaper. They state that Airbnb has no effect on the house prices in Amsterdam. Baarsma and Van Dalen (2016) discuss the validity of the two main assumptions the ING made. First, they argue that homebuyers do not take future Airbnb income into consideration when buying a house. Potential Airbnb hosts are thus not willing to pay more for a house compared to potential homebuyers who do not use Airbnb. Secondly, banks are not legally allowed to take into account future Airbnb income when calculating the maximum mortgage amount. Therefore, a potential Airbnb host cannot get a higher mortgage than a potential homebuyer who does not use Airbnb. In addition, applied regulation and conditions on short-term rentals limit the ability to generate extra income through Airbnb. Baarsma and Van Dalen (2016) conclude that Airbnb does not positively affect house prices in Amsterdam. They even state that nuisance caused by Airbnb guests has a negative impact. Both statements are supported by two studies of Pairolero (2016) and Lee (2016). Pairolero (2016) examines the effect of Airbnb on the housing market in Washington D.C. and finds no significant effect of Airbnb on house prices. The study by Lee (2016) on the other hand, concludes that Airbnb has a negative effect on house prices through nuisance. The latter study states that Airbnb increases nuisance and therefore decreases the quality of life in certain neighborhoods, leading to lower house prices.
The negative impact of nuisance on house prices is shown in other scientific research as well. Since the Netherlands is one of the most densely populated nations in the world, the available infrastructure is used very intensively. One of the unwelcome side effects of intense traffic in high dense areas is noise nuisance. Theebe (2004) estimates the impact of traffic noise on property prices using a data set that covers over 100,000 property sales transactions combined with noise levels for 2 million small 100 by 100m areas. The analysis uses spatial auto regression techniques to yield more accurate estimates. In addition, accessibility variables are included to account for the positive effects of the infrastructure (Theebe, 2004). The study finds that traffic noise is capitalized into property prices, leading to a maximum discount of 12 percent. If the property is located in a quiet area, it might sell at a premium up to 6.5 percent. Moreover, Theebe (2004) concludes that properties in high-income neighborhoods are affected more by traffic noise nuisance than properties in low-income
areas. These findings are in line with a study done by Dekkers and Van der Straaten in 2009. Dekkers and Van der Straaten (2009) analyze the effect of transport noise on house prices. More in particular, using a hedonic pricing method, the study examines the effect of aircraft noise on house prices around Amsterdam airport. A key point in the analysis is that it accounts for background noise by setting threshold values for multiple sources of traffic noise, including road, railway and aircraft noise. The study finds negative signs for all three noise variables, meaning higher noise levels leading to lower house prices. An additional decibel of nuisance caused by road, aircraft or railway decreases house prices by 0.14, 0.80 and 0.72 percent, respectively (Dekkers & Van der Straaten, 2009).
2.2. External effects
The aim of this thesis is to find the impact of Airbnb on the hotel industry in Amsterdam. However, not only Airbnb affects the hotel industry, also external effects have an impact. Besides hotel specific characteristics, Yang, Mueller and Croes (2016) investigate the influence of market accessibility on hotel prices. In addition, Zhang, Zhang, Lu, Cheng and Zhang (2011) examine how site and situation factors affect hotel room prices in Beijing. The latter research measured accessibility as S-distance and T-distance to investigate the effect of location on hotel room price. S-distance is the straight linear distance between hotel and its nearest scenic spot. T-distance refers to that distance between hotel and its nearest transport hub. T-distance has significant but negative impact on hotel room prices in Beijing, meaning lower hotel room prices when nearest transport hub is further away.
Yang et al. (2016) investigate the relationship between market accessibility and hotel prices using a randomized sample of hotels in the Caribbean islands. Market accessibility is measured by the average cost of a round trip flight from New York, Chicago, and Los Angeles to the island where the hotel is located. The study contributed to the existing literature since past literature only considered accessibility to attractions (Hung et al., 2010) and public goods (Zhang et al., 2011a) without looking into the market accessibility. Accessibility to attractions is mostly measured by the distance of hotels from the city center to investigate the effect of location on hotel prices. Yang et al. (2016) find that the level of market accessibility, hotel class, online quality-signaling factors, availability of free breakfast, and the presence of a fitness center and a swimming pool affect hotel prices. A lower price for a flight from major source markets to the island leads to higher hotel prices. Moreover, the study concludes that hotel room prices are less sensitive to the influence of accessibility than
the influence of a positive reputation. The relationship between reputation and hotel prices is also confirmed by studies of Ye, Law and Gu (2009), and Schamel (2012).
Customer reviews and hotel star rating are important determinants of hotel room prices. Ye et al. (2009) conduct a study to empirically investigate the effect of online consumer-based reviews on hotel room sales. Using data retrieved from Ctrip, the largest travel website in China, the study finds a significant relationship between the average user rating and the number of bookings of a hotel. The results show that positive online reviews can significantly increase the number of bookings of a hotel. More in particular, a 10 percent improvement in reviewers’ rating increases sales by 4.4 percent (Ye et al., 2009).
Schamel (2012) investigates consumer willingness to pay for different hotel characteristics using data generated through the online meta-booking engine for hotel rooms trivago.com. Among other hotel characteristics, the study concludes that popularity ratings derived from customer reviews and the hotel star rating significantly affect hotel room prices. The results are distinguished into two hotel room types: single and double room. Hotel star rating is the most significant variable affecting room prices. Each additional hotel star rating would add 21 Euros to the rate for a single room, and 27 Euros for a double room. In addition, the study states that an additional point in the online rating adds 2 Euros to the rate for a single room, but only 54 cents for a double room. Furthermore, Schamel (2012) concludes that distance to the city center is another important determinant contributing to the willingness to pay for a hotel room. The latter finding is in line with the findings by Hung et al. (2010), who also study the major determinants of hotel room pricing strategies.
In addition, Andersson (2010) shows the significant impact of online customer ratings on hotel room prices. Hoteliers tend to charge higher prices when their rooms have good reputations. Online reviews are perceived as more up-to-date and reliable information compared to information provided by travel service providers (Zhang, Ye, Law & Li, 2010). The effective and reliable customer reviews degrade the uncertainty of purchase, leading to buyers willing to pay a higher price for rooms with positive reviews (Yang et al., 2016). Therefore, a positive relation between reputation and hotel room price is expected. Andersson (2010) estimates a hedonic model for Singapore’s hotel market using online search results and customer ratings for facility and food-and-beverage quality. Also Zhang, Ye and Law (2011) evaluated the effect of online reviews on hotel prices, and find that the online rating scores on room quality and location positively affect hotel prices. On the other hand, Wang and Nicolau (2017) state that the number of online reviews per year has a negative effect on Airbnb listing prices. Previous studies reported that most tourists rent sharing economy based
accommodation to reduce costs (Guttentag, 2015). As a result, cheaper listings tend to receive more bookings and consequently more reviews. Wang and Nicolau (2017) find a price decrease of 0.01 percent per each additional online review.
Traditional e-commerce experiences an increase in products sales due to a positive reputation of the product. However, Airbnb hosts are limited in terms of sales volume of their products and services (Teubner, Hawlitschek, & Dann, 2017). More specifically, an apartment can at most be rented out 60 nights per year in Amsterdam. Previous empirical research by Ert, Fleischer and Magen (2016) does not find a significant price effect of the number of reviews on Airbnb listings. The study uses a dataset including 395 posted listings in the city of Stockholm, of which 185 received guest reviews. The online review scores were not found to lead to price increase. Gutt and Herrmann (2015), on the other hand, show that Airbnb hosts capitalize high reputation by demanding higher prices. Using a treatment and control group, the study analyzes price differences before and after a host’s average rating is publicly displayed for the first time. The dataset consists of 14,871 Airbnb listings in New York City and contains data for the price of the listing per night, the total number of reviews for the listing, and listing specific characteristics such as the number of rooms and the neighborhood in which the listing is located. Gutt and Herrmann (2015) conclude that the owner of a listing that receives reviews tends to raise the price per night by 2.69 Euros more compared to the owner of a listing that does not receive reviews.
2.3. Airbnb on hotel industry
Since peer-to-peer platforms like Airbnb are relatively new, there is little literature on the subject. Especially the impact of Airbnb on the hotel industry is not investigated often. As already discussed in section 2, the side effects of Airbnb include increasing house prices, nuisance and economic activity in Amsterdam. Due to the company’s innovative internet-based business model and its unique appeal to tourists, Airbnb’s growing popularity has a great impact on the traditional tourism accommodation sector (Guttentag, 2015). This traditional sector includes hotels and other formal businesses renting out rooms to tourists. With its disruptive business model, Airbnb has shaken up this sector providing an online market place for peer-to-peer accommodation. Zervas et al. (2016) investigate the economic impact of the sharing economy by investigating the impact of Airbnb’s entry into the state of Texas on the Texas hotel industry. Airbnb is active since 2008 in the state of Texas. The study hypothesizes that some stays with Airbnb serve as a substitute for certain hotel stays.
Consequently, an impact of Airbnb on hotel revenue is expected. Zervas et al. (2016) conducted a dataset consisting of 13,935 Airbnb listings spanning a period from 2008 to August 2014. Data on hotels is obtained from public records provided by the Texas Comptroller of Public Accounts, and includes monthly hotel room revenue, hotel names, and capacity. Using a difference-in-difference strategy, the study estimates Airbnb’s impact on hotel room revenue. This identification strategy is not used in this thesis due to limitation on data availability. Zervas et al. (2016) find that Airbnb supply has decreased hotel room revenue by 10 percent during their research period 2010-2014. In addition, they conclude that a large population of individuals has benefited from Airbnb: not only Airbnb hosts generating extra income, but also consumers who benefit from lower prices and increased competition in the traditional hotel industry.
3. Data
The analysis in this thesis is based on three databases. The first database contains information about all Airbnb listings in Amsterdam from 2009-2016 and their reviews. Secondly, the hotel database consists of facts about hotels in Amsterdam from 2009-2016. At last, the house price database covers all housing transactions in Amsterdam from 2000-2016.
The Airbnb database is retrieved from Inside Airbnb, an independent and non-commercial organization. This database includes two sub-datasets: a dataset with information about all Airbnb listings in Amsterdam and a dataset with reviews on these listings. For each listing the room type, neighborhood, price per night and the number of reviews is known. Table 1 shows the descriptive statistics for the Airbnb dataset. The first sub-dataset contains 15,140 listings in Amsterdam that are posted on Airbnb since the launch in 2009 and provides information about the location, room type and price of these listings. The average price per night is 135 Euros, with a minimum of 19 Euros and a maximum of 3,142 Euros. About 80% of Airbnb listings are entire apartments and 20% are private/shared rooms. The average price per night for an entire apartment is 148 Euros, whereas the average price for a private and shared room is 88 and 89 Euros, respectively. In addition, the descriptive statistics show that the average number of listings per host is 1.2, with a minimum of a single listing and a maximum of 91 listings. There are 12,965 unique Airbnb hosts in Amsterdam of which 1,063 offer more than one listing.
Table 1
Descriptive statistics: Airbnb listings in Amsterdam.
Full Sample
Listings
# of Obs In % Avg. Price per Night Mean Min Max
Total listings 15,140 100% € 135
Entire Home/Apartments 11,780 77.8% € 148
Private Room 3,298 21.8% € 88
Shared Room 62 0.4% € 89
Total unique hosts 12,965 100% 1.2 1 91
Host with single listing 11,902 91.8% 1 1 1
Fig. 1 displays the spatial distribution of Airbnb listings across Amsterdam since the launch of Airbnb in 2009. Airbnb listings are mostly located in the city center. The red, green and blue dots indicate entire apartments, private rooms, and shared rooms, respectively. The figure shows that location information for listings is anonymized by Airbnb. According to the map, some of the listings are located in the water called ‘Het IJ’, which separates the city center from Amsterdam-Noord. The random error Airbnb applies to the location of its listings is 0-150 meters. This means that even if there are listings in the same building, both listings may appear on different locations in Fig. 1.
The second sub-dataset of the Airbnb database covers all the reviews on the Airbnb listings from 2009-2016. It contains 270,477 reviews for 12,967 of the listings available in the first Airbnb sub-dataset. The dataset provides a variety of variables ranging from the review itself to the star ratings per section. However, this thesis only uses the number of reviews per listing. The total amount of reviews for each listing is merged to its price, room type and neighborhood using the listing’s id number. Fig. 2 displays the number of Airbnb reviews over time. The graph shows that the total amount of reviews more than doubles each year. In addition, starting from 2013, per year the third bar is the highest, indicating that in the third quarter of each year the most reviews are written. This can be explained by the seasonal variation in demand. During the peak season there are more tourists in Amsterdam using Airbnb leading to a higher amount of reviews. The figure also indicates that in 2016 a total amount of just over 125,000 reviews is written.
The hotel room database is the second database used in this thesis. This database includes hotel room prices for each star category per neighborhood in Amsterdam from 2009-2016 and is provided by the Research, Information and Statistics (RIS) Amsterdam. In addition to hotel room prices, the dataset includes the occupancy rate and basic information such as the total number of hotels per star category in each neighborhood. The sample contains 280 observations of hotel room prices and 276 observations of occupancy rates. It is not a complete sample since there are missing observations. This is one of the limitations that are discussed in section 6.
Table 2 provides the descriptive statistics for the hotel room database. The table shows the averages of both the prices and occupancy rates of hotels per neighborhood of Amsterdam. In addition, the revenue per available room and the average star category per neighborhood are displayed. The revenue per available room is the product of average room price and occupancy rate. Airbnb can affect the hotel industry through lower occupancy rates,
Fig. 1. Spatial distribution of Airbnb listings in Amsterdam. Notes: Airbnb applies a 0-150 meters random error to the exact location of its listings. Source: Inside Airbnb.
Fig. 2. Number of Airbnb reviews in Amsterdam.
0% 20% 40% 60% 80% 100% 0 10000 20000 30000 40000 50000 2009 2010 2011 2012 2013 2014 2015 2016 % of total r evi ew s # of r evi ew s Date # of reviews % of total reviews
decreased hotel room prices, or a combination of these two components. Within the hotel industry, this combination is regularly reported as the revenue per available room. The neighborhood Westerpark has the highest average price per night of 150 Euros. However, in the dataset there is only one observation for the neighborhood Westerpark, which represents a four-star hotel. Therefore, this result can be considered biased. The neighborhood Centrum-Oost represents the second highest average price per night with 141.69 Euros per night. There are 46 hotels in the hotel room database that are located in the neighborhood Centrum-Oost, and therefore this average price is considered more reliable than the average price of the neighborhood Westerpark. On the other hand, the neighborhood Noord-Oost has an average price per night of 67.50 Euros, which represents the lowest average rate in Amsterdam. Also
Table 2
Descriptive statistics: hotel rooms in Amsterdam.
Average
Neighborhood Rate in € Occupancy in % RevPAR Star category # of Obs
Centrum-Oost 141.69 73.03 103.00 1.8 46 Centrum-West 136.40 75.69 102.52 1.8 48 Noord-Oost 67.50 46.50 31.88 2.1 2 Noord-West 0 Oud-Noord 92.88 76.13 71.31 2.5 4 De Aker/Nieuw Sloten 97.15 68.79 66.77 4 8 Geuzenveld/Slotermeer 83.33 76.71 61.64 2.4 5 Osdorp 76.00 84.00 63.84 1.8 1 Slotervaart 85.48 79.07 65.27 3.5 8
Indische Buurt/Oostelijk Havengebied 125.43 70.95 89.10 1.2 7
Oud-Oost 108.93 79.29 89.84 1.8 7 Watergraafsmeer 88.31 76.85 69.04 3.4 4 IJburg/Zeeburgereiland 110.00 50.00 55.00 1.0 1 Bos en Lommer 91.63 83.14 78.01 3.1 7 Oud-West/De Baarsjes 93.36 79.43 74.88 2.6 36 Westerpark 150.00 75.00 112.50 3.6 1 Buitenveldert/Zuidas 117.26 77.46 90.05 2.7 18 De Pijp/Rivierenbuurt 80.89 75.91 64.40 2.1 26 Oud-Zuid 132.82 69.97 93.53 2.5 44 Gaasperdam/Driemond 80.78 69.93 56.25 2.5 6 Bijlmer-Centrum 94.43 67.13 70.44 2.6 5 Bijlmer-Oost 0
the lowest average occupancy rate can be found in the neighborhood Noord-Oost. However, only two observations in the dataset correspond to the neighborhood Noord-Oost, leading to biased results in both the average of the price per night and occupancy rate. Moreover, table 2 shows that the hotel room database is not equally distributed over all the 22 neighborhoods of Amsterdam. Most of the observations occur in Centrum-Oost, Centrum-West, Oud-West/De Baarsjes and Oud-Zuid. This limitation of the dataset is discussed in section 6 more elaborately.
The third database used for this thesis is the house price database, which is made available by the NVM. This database covers 95 percent of all the housing transactions in Amsterdam for the period 2009-2016. The dataset provides a variety of variables such as the transaction price, transaction date and house characteristics including house size, house type and construction year. Because the exact transaction price and the size in square meters are known, the price per square meter can be calculated. After correcting for incomplete
Table 3
Descriptive statistics: housing transactions in Amsterdam.
Mean Std. dev. Min Max
Transaction price (€) 302,928 212,052 48,500 2,500,000 Price per m2 (€) 3,442 1,175 818 7,500 Size in m2 87.400 41.511 25.000 350.000 Rooms 3.333 1.338 1.000 17.000 Apartment 0.873 Terraced 0.113 Semi-detached 0.007 Detached 0.007 Parking 0.143
Maintenance quality – good 0.899
Construction year < 1910 0.142 Construction year 1910 – 1944 0.301 Construction year 1945 – 1959 0.052 Construction year 1960 - 1970 0.099 Construction year 1971 – 1980 0.039 Construction year 1981 – 1990 0.125 Construction year 1991 – 2000 0.097 Construction year ≥ 2000 0.146 Transaction year 2013 2.348 2009 2016 Number of observations 39,181
observations, the dataset contains 39,669 observations of housing transactions in Amsterdam. A scatter plot analysis shows that the dataset contains 488 potential outliers in transaction price, house size and price per square meter. These observations are deleted from the sample and account for 1.2 percent of the total sample. More specifically, all the observations with a transaction price above 2.5 million Euros (103 observations) are dropped. In addition, the dataset provides information about whether or not home size is checked. To get reliable results, all 50 observations that contain a house size that is not checked are dropped. Furthermore, all the observations with house size above 350 square meters (123 observations) are dropped. At last, all the observations with a price per square meter above 7,500 Euros (212 observations) are dropped.
Table 3 displays the descriptive statistics for the housing transaction database after correcting for outliers. The average transaction price for a house in the period 2009-2016 in Amsterdam is just over 300,000 Euros and the average house size is about 87 square meters. Price per square meter is the variable of interest. However, also control variables are reported in the descriptive statistics. These control variables include dummies for the house types, parking availability and construction year scopes. As table 3 reports, 87.3 percent of the housing transaction in the period 2009-2016 concern apartments and only 14.3 percent include parking availability. The dummy for maintenance quality equals 1 if the maintenance of both the inside and outside is at least good (i.e. 7 on a scale of 1 to 10). 90 percent of the housing sold in the sample period has a good maintenance quality.
The housing transaction database is used to account for neighborhood fixed effects on the hotel industry. The price per square meter per neighborhood is included in the model discussed in section 4 as a control variable. The database includes four-digit zip codes for each observation. Using the four-digit zip code the house is linked to the neighborhood it is located in and the average price per square meter per neighborhood is calculated. Using level six zip code indicators to account for neighborhood fixed effects would result in more precise estimates. However, level six zip code indicators are not provided in the database. This limitation is also discussed in section 6.
(1)
(2)
4. Methodology
This thesis focuses on the effect of Airbnb on the hotel industry in Amsterdam. To answer the research question what the effect of Airbnb is on the hotel industry in Amsterdam, the analysis uses a time series regression including the main independent variable of interest and control variables. First the effect of Airbnb on the hotel industry is measured at the neighborhood level. Amsterdam consists of 22 neighborhoods in 7 different districts. Therefore, also the effect of Airbnb on the hotel industry in Amsterdam is measured at the district level. More specifically, the following equations (1) to (6) are regressed at both the neighborhood and district level.
The dependent variable is the price per night of a hotel room and the price per night of an Airbnb listing is the main independent variable of interest. The following basic time series regression model is used:
𝑃𝐻!" = 𝛼!+ 𝛽!𝑃𝐴!"!!+ 𝜀!"
where αi is the constant in neighborhood i and PHit is the average price per night of a hotel
room at time t in neighborhood i. PAit-1 measures the average price per night of Airbnb
listings in Amsterdam at time t-1 in neighborhood i. According to Ye et al. (2009), the hotel industry is affected by the number of online reviews. On the other hand, Yang et al. (2016) find that hotel class and online quality-signaling factors affect hotel prices. Therefore, the following model also includes the average hotel star rating:
𝑃𝐻!" = 𝛼!+ 𝛽!𝑃𝐴!"!!+ 𝛽!𝑋!!+ 𝜀!"
where Xit represents a vector of control variables for each neighborhood in Amsterdam
including the average hotel star rating. The vector also includes the number of tourists (in millions) in Amsterdam at time t. However, the number of online reviews is not included in the model. Appendix 1 shows that the number of tourists in Amsterdam is highly correlated with the number of online reviews. To avoid multicollinearity, the number of online reviews is not included.
Each neighborhood in Amsterdam has its own specific characteristics affecting the hotel industry. The neighborhood specific effects are accounted for in the constant of the
(3)
(4)
(5)
(6) model. However, the average house price per square meter per neighborhood is included as an independent variable. This variable is also included in the vector Xit in equation (2). The next
model also accounts for time fixed effects to control for time trends in the hotel room price:
𝑃𝐻!" = 𝛼! + 𝛽!𝑃𝐴!"!!+ 𝛽!𝑋
!+ 𝜑!!+ 𝜀!"
where φit represents the time fixed effects.
Since Airbnb can affect the hotel industry through lower occupancy rates, decreased hotel room prices, or a combination of these two components, the dependent variable PHit is
replaced by RevPARit. As already mentioned in section 3, RevPAR is the revenue per
available room, which is the product of average hotel room price and occupancy rate. By replacing the dependent variable, the basic model looks as follows:
𝑅𝑒𝑣𝑃𝐴𝑅!" = 𝛼! + 𝛽!𝑃𝐴!"!!+ 𝜀!"
where αi is the constant in neighborhood i and RevPARit is the revenue per available room at
time t in neighborhood i. PAit-1 measures the average price per night of Airbnb listings in
Amsterdam at time t-1 in neighborhood i. Since the revenue per available room is the product of average hotel price and occupancy rate, online quality-signaling factors and hotel class can affect the dependent variable. For this reason the average hotel star rating is added to equation (4). The following model also includes this control variable:
𝑃𝐻!" = 𝛼!+ 𝛽!𝑃𝐴!"!!+ 𝛽!𝑋
!"+ 𝜀!"
where Xit represents a vector of control variables for each neighborhood in Amsterdam at
time t, including the average hotel star rating. The vector also includes the number of tourists in Amsterdam at time t. To account for the neighborhood fixed effects, the average house price per square meter per neighborhood is included as a control variable. This variable is also included in the vector Xit in equation (5). The next model also accounts for time fixed effects
to control for time trends in the hotel room price:
𝑅𝑒𝑣𝑃𝐴𝑅!" = 𝛼!+ 𝛽!𝑃𝐴!"!!+ 𝛽!𝑋
!+ 𝜑!"+ 𝜀!"
5. Results
This section describes the results of the regression estimates of the effect of Airbnb on the hotel industry in Amsterdam. First, the regression results for the effect of Airbnb on the hotel room prices are discussed, followed by the regression results for the effect of Airbnb on the revenue per available room. In the latter case, the dependent variable hotel room price per night is replaced by the revenue per available room. These regression estimates show that Airbnb not only affects the hotel industry in Amsterdam through decreased hotel room prices, but also through lower occupancy rates. The variables of both mentioned regression results are measured at the neighborhood level. Amsterdam consists of 22 neighborhoods in 7 different districts. In the last part of this section the effect of Airbnb on the hotel industry in Amsterdam is discusses at the district level.
5.1. Effect of Airbnb on hotel room prices
In this subsection the main results of the regression of the effect of Airbnb on the hotel room prices are presented. In addition, the regression results displayed in table 4 are discussed more detailed. As already mentioned in section 4, first the dependent variable is the average hotel room price and the main independent variable of interest is the price per night of an Airbnb listing. Column (3) in table 4 shows the regression estimates of equation (3) in section 4. The coefficient of the price per night of an Airbnb listing (PAit-1) indicates that, on average, the
hotel room price per night decreases by 0.08 Euros per every one Euro increase in the price per night of an Airbnb listing in the previous year. This is the effect of Airbnb after controlling for both neighborhood and time fixed effects. However, the estimated coefficient is not significant.
Table 4 displays the baseline regression results of the effect of Airbnb on hotel room prices in Amsterdam. The variables are measured at the neighborhood level. The coefficients in columns (1) to (3) represent the regression estimates for equations (1) to (3) in section 4. In column (1), the coefficient estimate of equation (1) is shown. This is the basic regression model including only the main independent variable of interest, the price per night of an Airbnb listing in Amsterdam, and the dependent variable, the price per night of a hotel room. The coefficient of the price per night of an Airbnb listing indicates that, on average, the hotel room price per night increases by 0.46 Euros per every one Euro increase in the price per night of an Airbnb listing in the previous year. The estimated coefficient is significant at the
1% level. However, the adjusted R-squared is 0.2022, meaning that the model only predicts 20.22 percent of the variance in hotel room price.
In column (2) the control variables are added. This increases the adjusted R-squared to 0.7537, meaning a better fit of the model. The control variables include the average hotel star rating and the total number of tourists per year in Amsterdam. Also the average house price per square meter is added to the model to control for neighborhood specific differences in quality. The estimated coefficient of the price per night of an Airbnb listing becomes negative, indicating a decrease of 0.09 Euros in the price per night of a hotel room per every one Euro increase in the price per night of Airbnb listing in the previous year. However, this coefficient estimate is not significant. Also the average hotel star rating and the house price per square meter do not significantly affect the hotel room price. On the other hand, the coefficient estimate of the number of tourists in Amsterdam is significant at the 5% significance level. This implies that an extra million tourists in Amsterdam per year leads to a 9.69 Euros increase in hotel room price per night.
To account for time trends in the average hotel room price in Amsterdam, year fixed effects are added to the model in column (3). Adding these fixed effects leads to a higher
Table 4
Baseline regression results: the effect of Airbnb on hotel room price at the neighborhood level.
(1) (2) (3)
Basic Control variables Year fixed effects
Price Airbnb 0.4571*** -0.0934 -0.0785
(0.0935) (0.1291) (0.1434)
Average hotel star rating -0.7963 1.9077
(8.3835) (9.0946)
Number of tourists 9.6947** 21.9621***
(4.9231) (6.3783)
Price per m2 0.0073 0.0004
(0.0072) (0.0.120)
Year fixed effects No No Yes
Neighborhood fixed effects No Yes Yes
Number of observations 80 80 80
Adjusted R2 0.2022 0.7537 0.7942
Notes: The dependent variable is the average hotel room price. The variables are measured at the neighborhood
level. Standard errors are clustered at the PC4 level and in parentheses. *, **, *** indicate 10%, 5%, 1% significance, respectively.
adjusted R-squared of 0.7942. The effect of the price per night of an Airbnb listing on the hotel room price per night decreases to 0.08 Euros, but is still not significant. In addition, the average hotel star rating and the average house price per square meter still do not significantly affect the hotel room price in Amsterdam. The estimated coefficient of the number of tourists significantly affects the hotel room price at the 1% significance level. The coefficient indicates that an extra million tourists in Amsterdam per year lead to a 21.96 Euros increase in hotel room price per night. This is expected since an increase in the number of tourists leads to more demand of hotel rooms.
Table 4 shows the estimated regression results of the effect of Airbnb on the hotel room price per night at the neighborhood level. After controlling for neighborhood and year fixed effects in column (3), the effect of Airbnb on the hotel room price per night in Amsterdam is not significant. This is not in line with Zervas et al. (2016) who find that Airbnb negatively impacts hotel revenue in Texas.
5.2. Effect of Airbnb on revenue per available room
This subsection presents the main results of the regression estimates of the effect of Airbnb on the revenue per available room. In table 5 the regression results are displayed. Section 4 already discussed that the dependent variable hotel room price is replaced by revenue per available room. Airbnb can affect the hotel industry through decreased hotel room prices, lower occupancy rates, or a combination of these two factors. For this reason, the dependent variable is replaced by revenue per available room, which is the product of average hotel room price and occupancy rate. The main independent variable of interest is still the price per night of an Airbnb listing. Column (3) in table 5 presents the regression estimates of equation (4) in section 4. The coefficient of the price per night of an Airbnb listing indicates that, on average, the revenue per available hotel room decreases by 0.12 Euros per every one Euro increase in the price per night of an Airbnb listing in the previous year. This is the effect of Airbnb after controlling for neighborhood and time fixed effects. However, the estimated coefficient is not significant.
Table 5 shows the baseline regression results of the effect of Airbnb on the revenue per available hotel room in Amsterdam. The variables are measured at the neighborhood level. The coefficients in columns (1) to (3) represent the regression estimates for equations (4) and (6) in section 4. In column (1), the coefficient estimate of the basic regression model is shown. This basic model only includes the main independent variable of interest, the price
per night of an Airbnb listing, and the dependent variable, the revenue per available room. The coefficient of the price per night of an Airbnb listing indicates that, on average, the revenue per available hotel room increases by 0.47 Euros per every one Euro increase in the price per night of an Airbnb listing in the previous year. The estimated coefficient is significant at the 1% level. However, the adjusted R-squared is 0.2095, meaning that the model only predicts 20.95 percent of the variance in revenue per available hotel room.
In column (2) the control variables average hotel star rating and the total number of tourist per year in Amsterdam are added. In addition, the average house price per square meter is included in the model as a control variable. This increases the adjusted R-squared to 0.9005, meaning that the model predicts 90.05 percent of the variance in revenue per available hotel room. The estimated coefficient of the price per night of an Airbnb listing becomes negative, indicating a 0.11 Euros decrease in the revenue per available hotel room per every one Euro increase in the price per night of an Airbnb listing in the previous year. However, this coefficient estimate is not significant. In addition, the house price per square meter does not significantly affect the revenue per available hotel room in Amsterdam. On the other hand, the coefficient estimates of the average hotel star rating and the number of tourists
Table 5
Baseline regression results: the effect of Airbnb on revenue per available room at the neighborhood level.
(1) (2) (3)
Basic Control variables Fixed effects
Price Airbnb 0.4734*** -0.1059 -0.1190
(0.1126) (0.0941) (0.0799)
Average hotel star rating 29.5695*** 29.9215***
(5.9432) (5.7092)
Number of tourists 10.9627*** 21.8558***
(4.0270) (4.3832)
Price per m2 0.0035 -0.0033
(0.0049) (0.0083)
Year fixed effects No No Yes
Neighborhood fixed effects No Yes Yes
Number of observations 78 78 78
Adjusted R2 0.2095 0.9005 0.9271
Notes: The dependent variable is the revenue per available hotel room. The variables are measured at the
neighborhood level. Standard errors are clustered at the PC4 level and in parentheses. *, **, *** indicate 10%, 5%, 1% significance, respectively.
in Amsterdam per year are significant at the 1% level. More specifically, an extra million tourists in Amsterdam per year causes the revenue per available room to increase by 10.96 Euros. An additional hotel star rating, on average, indicates a 29.57 Euros higher revenue per available hotel room in Amsterdam.
To account for time trends in the revenue per available hotel room in Amsterdam, year fixed effects are included in the model in column (3). Adding these fixed effects increases the adjusted R-squared to 0.9271, meaning a better fit of the model. The effect of the price per night of an Airbnb listing is 0.12 Euros. However, this estimated coefficient is not significant. Also the average house price per square meter does not significantly impact the revenue per available hotel room in Amsterdam. On the other hand, both the estimated coefficients of the average hotel star rating and the number of tourists are significant at the 1% significance level. An additional hotel star rating, on average, indicates a 29.92 Euros higher revenue per available room, and an extra million tourists in Amsterdam per year leads to an increase of 21.86 Euros in the revenue per available room.
Table 5 displays the baseline regression results of the effect of Airbnb on the revenue per available hotel room in Amsterdam at the neighborhood level. After controlling for neighborhood and year fixed effects in column (3), the effect of Airbnb on the revenue per available room is not significant. In section 5.1 it s already mentioned that the effect of Airbnb on the hotel room price per night is not significant at the neighborhood level. For this reason the effect of Airbnb on the hotel industry in Amsterdam is investigated at the district level in the following section.
5.3. Effect of Airbnb at the district level
Amsterdam consists of 22 neighborhoods in 7 different districts. The hotel room database provided by the RIS Amsterdam covers the hotel room prices, occupancy rates and star category for each neighborhood. However, the database is not complete since there are missing observations. In section 6 this limitation is discussed more elaborately. Most of the observations occur in the neighborhoods Centrum-Oost, Centrum-West, Oud-West/De Baarsjes and Oud-Zuid. Furthermore, table 2 shows that for the neighborhoods Noord-West and Bijlmer-Oost there are no observations of the hotel room price and occupancy rate in the period 2009-2016. Therefore, in this section the effect of Airbnb on the hotel industry in Amsterdam is investigated at the district level. To do so, the dependent and independent variables are clustered at the 7 districts level. Appendix 2 shows that the average house price
per square meter is highly correlated with the other independent variables. To avoid multicollinearity, the house price per square meter is not included in the regression.
In table 6 the regression results are shown. The dependent variable is the average hotel room price, and the main independent variable of interest is the price per night of an Airbnb listing in the previous year. In contrast to the regression results displayed in table 4, the variables in table 6 are measured at the district level. Column (3) in table 6 shows the regression estimates of equation (3) in section 4. The coefficient of the price per night of an Airbnb listing indicates that, on average, the hotel room price per night decreases by 0.45 Euros per every one Euro increase in the price per night of an Airbnb listing in the previous year. This is the effect of Airbnb after controlling for district and time fixed effects. The estimated coefficient is significant at the 1% level.
Table 6 displays the baseline regression results of the effect of Airbnb on the hotel room prices in Amsterdam. The variables are measured at the district level. The coefficients in columns (1) to (3) represent the regression estimates for equations (1) to (3) in section 4. In column (1), the coefficient estimate of equation (1) is shown. This basic regression model only includes the main independent variable of interest and the dependent variable. The coefficient of the price per night of an Airbnb listing indicates that, on average, the hotel room price per night increases by 0.33 Euros per every one Euro increase in the price per night of
Table 6
Baseline regression results: the effect of Airbnb on hotel room price at the district level.
(1) (2) (3)
Basic Control variables Year fixed effects
Price Airbnb 0.3304*** -0.4650*** -0.4516***
(0.1273) (0.1385) (0.1337)
Average hotel star rating -3.970 -3.1150
(12.9801) (11.6722)
Number of tourists 9.6118*** 17.2274***
(3.6088) (3.7536)
Year fixed effects No No Yes
Neighborhood fixed effects No Yes Yes
Number of observations 41 41 41
Adjusted R2 0.1951 0.7135 0.8064
Notes: The dependent variable is the average hotel room price. The variables are measured at the district level.
Standard errors are clustered at the PC4 level and in parentheses. *, **, *** indicate 10%, 5%, 1% significance, respectively.
an Airbnb listing in the previous year. This estimated coefficient is significant at the 1% level. However, the adjusted R-squared is 0.1951, implying that only 19.51 percent of the variance in hotel room price per night is predicted by the model.
In column (2) the control variables are included in the regression. This leads to an increase of the adjusted R-squared to 0.7135, meaning a better fit of the model. The control variables include the average hotel star rating and the number of tourists in Amsterdam per year. The estimated coefficient of the price per night of an Airbnb listing is still significant at the 1% level, which is in contrast to the regression results displayed in table 4. The negative coefficient indicates, on average, a decrease in hotel room price of 0.47 Euros per every one Euro increase in the price per night of an Airbnb listing in the previous year. Also the estimated coefficient of the number of tourists is significant at the 1% level. This coefficient indicates that an extra million tourists in Amsterdam per year increases the hotel room price by 9.61 Euros per night. On the other hand, the average hotel star rating does not significantly affect the hotel room price.
To account for time trends in the average hotel room price per night in Amsterdam, year fixed effects are included in the model in column (3). Adding these fixed effects leads to a better fit of the model. The effect of the price per night of an Airbnb listing is still significant at the 1% level. One Euro increase in the price per night of an Airbnb listing in the previous year leads to a decrease of 0.45 Euros in the price per night of a hotel room. Also the number of tourists in Amsterdam has a significant impact on the price per night of a hotel room. An extra million tourists per year increases the hotel room price per night by 17.23 Euros. After including the year fixed effects, the average hotel star rating still does not significantly impact the average hotel room price.
For the research on the effect of Airbnb on the revenue per available hotel room at the district level, the dependent variable hotel room price is replace by revenue per available hotel room. This is in line with the research on the effect of Airbnb on the hotel industry at the neighborhood level discussed in section 5.2. In table 7 the regression results are displayed. The main independent variable of interest is the price per night of an Airbnb listing in the previous year. In contrast to the regression results showed in table 5, the variables in table 7 are measured at the district level. Column (3) in table 7 shows the regression estimates of equation (6) in section 4. The coefficient of the price per night of an Airbnb listing indicates that, on average, the revenue per available hotel room decreases by 0.19 Euros per every one Euro increase in the average price per night of an Airbnb listing in the previous year. This is
