The effect of the growth of AirBnB on house prices in Amsterdam, London and Berlin

1

The effect of the growth of AirBnB on house

prices in Amsterdam, London and Berlin

Bachelor thesis

Economics & Business Student: Thijmen Sangers Studentnumber: 10734872 Supervisor: L.M. Treuren

2

Statement of originality

This document is written by Thijmen Sangers who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Abstract

This study focuses on the effect of the growth of AirBnB on house prices in Amsterdam, London and Berlin. Previous studies mainly focused on residential housing and the hotel industry. Besides, little research about AirBnB is done in Europe. From previous studies it is clear that AirBnB grows rapidly and that its growth has an effect on rental prices and the hotel industry. This study focuses on house prices, which are considered to be sticky, therefore estimations are done considering that the house prices react with a lag of one quarter, one year or two years. The results imply that house prices react with a lag of one year. Making use of a fixed effects regression model we find that an increase in the relative growth of one percentage point of the total offered nights through AirBnB results in an +0.0072841 increase in the percentage points change in the House Price Index. Although the result is significant, the effect seems to be weak. Only if local total offered nights increased by huge numbers, the effect on house prices would not be negliglible. Besides, we find that tax rates and tourism expenditures are significant explanatory factors of house prices, but also these effects seem to be weak.

3

Table of contents: Page:

1. Background and literature review (4-10)

1.1 Background (4-6)

1.2 Literature review (6-10)

2. Methodology (10-16)

2.1 Assumptions (10-13)

2.2 Regression model (13-15)

2.3 The variables (15-16)

3. Descriptive statistics (17)

4.Results and discussion (18-21)

4.1 Results (18-20)

4.2 Discussion (20-21)

5. Conclusion (21)

6. References (22)

4

1. Background and literature review

1.1 Background

Peer-to-Peer markets have grown over the last years by the name ‘sharing economy’. Individuals can become, mostly temporarily, suppliers of goods and services. Proponents of the sharing economy suppose that economic efficiency is improved because of full use of capacity. Underutilized products can be optimally used by sharing products. Enormous growth of companies like Peerby, Uber and AirBnB reflect the growth of peer-to-peer markets. Opponents of the sharing economy criticize the concept by arguing that much of the growth comes from a lack of regulations (Barron, K. et al, 2017). Hotels for instance, face more regulations than AirBnB, hotels have to pay occupancy taxes where AirBnB hosts do not. AirBnB is a well-known company which is based on the principles of the sharing economy, a company which gives people the opportunity to become active on the residential housing market. AirBnB has created a website where buyers and sellers are easily connected. Low search and transaction costs are important for consumers, AirBnB managed to find a way to provide in this. AirBnB gives suppliers the opportunity to set their prices themselves unlike for example Uber, where prices are market based.

This study originated from a need to investigate how the growth of AirBnB affects house prices. Previous studies imply a relation between AirBnB and the residential housing market, it is likely that a relationship between AirBnB and house prices also exists.

Recent studies show that rental prices and the hotel industry are affected by the growth of AirBnB. For example Barron, K. et al (2017) found that a 10% increase in listings on AirBnB resulted in a 0.42% increase in rental prices and Horne and Merante (2017) found that a one standard deviation increase in the density of AirBnB listings leads to a 0.4% increase in rental prices. It is likely that AirBnB also affects property prices, however there are little studies about the effect of AirBnB on house prices. People take into account potential additional income if they rent out their homes through AirBnB, therefore they are willing to pay more when they buy properties. This may lead to higher bids and causes higher prices on the housing market. Not everyone is happy with the development of AirBnB.

Recently there has been a lot of commotion considering AirBnB. Governments are struggling with regulations and rules since AirBnB does not face these aspects where hotels are strictly attached to them. The competition in this market seems to become unfair. AirBnB acts like a broker and does not own any properties. AirBnB was not the first company to act as an intermediary between short-term consumers and producers in the residential housing market, but became the biggest during last years. The company minimizes the costs for participants and gives people the opportunity to become flexible suppliers. By 2016 there

5 were about 2.000.000 accommodations in 34.000 cities across the world signed up for AirBnB, right now there are about 3.000.000 accommodations in 65.000 cities spread over 191 countries all over the world (Sheppard, S., & Udell, A., 2016). From this fact, it is clear that AirBnB is still growing. Another way to show the tremendous growth of AirBnB is by its market capitalization, in 2016 AirBnB was valued about 25 billion dollars, where Marriott International Inc. was valued at 17.9 billion dollars. To put it in perspective, Marriott International Inc. owned more than 4.000 hotels, so these numbers actually tell that AirBnB is worth more than Marriott International Inc. (Kokalitcheva, 2015). Some people say AirBnB is the most successful example of the sharing economy considering these numbers (Martin, 2016).

Most research about this subject focuses on the effect of AirBnB on rental prices or the hotel industry, therefore investigating the effect on house prices contributes to the discussion about AirBnB. There is evidence that the growth of AirBnB causes house prices to rise in certain places, especially in the United States of America. Other subjects that were studied were subjects that have to deal with tourism specifically, like negative externalities of higher tourism rates due to the growth of AirBnB. However, there seems not to be much research about the effect on house prices, especially not in Europe. Therefore there is need to investigate the effect of AirBnB on house prices in Amsterdam, London and Berlin. House prices in these cities are likely to be affected by the growth of AirBnB. The main question is; ‘What is the effect of the growth of AirBnB on house prices in Amsterdam, London and Berlin?’. Since AirBnB seems to keep growing, this study will investigate the effect of the growth of offered total nights on AirBnB on house prices. First all relevant data is gained after which several different models can be regressed in Stata. The preferred method is the fixed effects regression so that city specific effects can be accounted for, further explanation follows in the methodology section There have been done 10 regressions, 3 fixed effects regressions with a considered lag in the reaction of house prices for one quarter, one year and two years. There have been done 3 fixed effects regressions with time dummies considering a lag in the reaction of house prices for one quarter, one year and two years. Furthermore, 2 fixed effects regressions with real house price index values are done considering a lag in the reaction of house prices for one quarter and one year. One regression considering a lag in the reaction of house prices for one year has been done with House Price Index values and total nights offered through AirBnB in natural logarithms. Last, we did a regression with a considered lag in the reaction of house prices for one year omitting the variable tourism expenditures, because we want to see what impact this variable might have on the model. The effect of the growth of AirBnB on house prices is significant in the models with and without time dummies considering a lag in the reaction of house prices for one year. All other regressions are insignificant considering a significance level of 5%.

6 The models with a lag in the reaction of the house prices for one year possible fits the model best because house prices are too sticky to react with a lag of a quarter or too little sticky to react with a lag of two years. Although the effect with a lag in the reaction of one year is significant, the effect is weak.

In the next section relevant papers will be discussed. Outcomes of the papers will be compared and assumptions will be held against each other. The section after the literature review is the methodology section, this section is about the used methods and assumptions about the model. The descriptive statistics will give insight in trends we can derive from figures about house prices and AirBnB. The results and conclusion will be presented after the descriptive statistics. In the last part the reference list and the appendix are showed.

1.2 Literature review

This section is about relevant existing literature about AirBnB. Papers relatable and relevant for this study will be discussed and outcomes will be presented and explained. Assumptions made in different papers will be held against each other to see at which points studies differ. Studies from Sheppard and Udell (2016), Horn & Merante (2017), Barron, K. et al (2017), Yrigoy, I. (2016) and Jefferson-Jones, J. (2015) will be discussed.

There have been several studies about the impact of AirBnB in different cities in the past 4 years. AirBnB is shown to be a company which has a great impact on a variety of sectors. Property prices and rental prices are said to rise since the willingness-to-pay would be higher taking into account the potential increase in income (Sheppard, S., & Udell, A., 2016). Sheppard and Udell (2016) claimed that AirBnB could have a positive influence on the economy through new available income streams for flexible suppliers. On the other hand tourism is encouraged because reducing costs for renting a property in the short-term makes it more attractive for tourists to come over. They also claim that property prices are affected by the growth of AirBnB. Sheppard and Udell (2016) studied the effect of AirBnB on property values in New York City. They found that a doubling in the amount of AirBnB listings goes together with increases in property values between 6% and 11%, considering their hedonic model. Sheppard and Udell found an even bigger effect by using a difference-in difference approach, in that case properties that are related to AirBnB are considered to rise in value by 31%. All combined and based on averages and the assumption that everyone fully takes advantage of their potential income streams, house values are estimated to rise by 17.7%.

7 Figure 1: Transmission Mechanisms for the impact of AirBnB activity on housing prices (Sheppard an Udell, 2016)

Sheppard and Udell explain the impact on property values using the figure above. The figure explains the channels through which fluctuations in property values can occur considering AirBnB’s impact. The first impact is the potential additional income which leads to lower costs of owning/renting a home. This obviously leads to a rise in property values because costs go down, so there is more money available to buy or rent a property. The second impact is the increase in population, a growing population leads to a higher demand for space and therefore higher demand for properties, which leads to a rise in value. The third impact comes through the possibility that guests of AirBnB listings stimulate the local neighborhoods economy. Guests need to eat, drink and they take part in tourists trips. The total demand for goods and service thus rises, upward streaming demand leads to higher prices and higher income and therefore to a rise in property values. The last impact is the only negative impact on property values. Guests can be too noisy, leave a lot of garbage or can be part of criminal activities. The neighborhood’s quality will decrease by these factors and leads to a decrease in the property values.

Horn & Merante (2017) investigated the effect of the growth of AirBnB on rental prices in Boston. They found that a one standard deviation increase in the density of AirBnB listings leads to a 0.4% increase in local rents, total increase in local rents rate is not given in this study. Unfortunately, no distribution is given. In their study they warn for the fact that not all sources of bias can be ruled out. This study reports that demand for rental houses is outpacing supply and thus drives up rental prices. The concept of home-sharing has a great impact in these dynamics and is said to need further investigation. If growth of AirBnB in

8 Boston will continue at the same rates, AirBnB usage will likely double in the coming 3 years. Since they found that a one standard deviation increase in the density of AirBnB listings leads to a 0.4% increase in local rents, there seems to be need for a solution to prevent rental prices to explode. An important assumption made in this study is to relax the assumption that house supply is static in the short-run. Despite the fact that Horn and Merante (2017) take into account neighborhood effects, reverse causality and tract effects, they state that not all bias can be ruled out in their study. Another finding is that tax and regulation seem to need to play a bigger role in Boston since competition is becoming unfair. Further empirical and theoretical research therefore has to be done according to Horn and Merante (2017).

Barron, K. et al (2017) found that a 10% increase in listings on AirBnB resulted in a 0.42% increase in rental rates and a 0.76% increase in house prices. Interesting outcome of their study is the finding that in zip codes with a lot of owner-occupiers the effect of AirBnB is smaller. Owner-occupiers are not able to rent out your home, that is the theory behind the above finding. Heterogeneity is assumed across zip codes since their findings across different cities vary a lot. Owner-occupiers are only able to offer their home on AirBnB when they are not home, like when they are on holidays. 0.27% of annual rent growth and 0.49% in growth of annual house prices can be explained by the growth of AirBnB according to Barron et al. (2017). By which is meant that for example 0.49% is a part of the total growth of 1%. Total growth of rental prices and property prices are not given. They argue that the growth of rental rates and house prices occurs through 2 channels. Rental rates are increased through home-sharing because landlords switch from renting out their properties on the long-term market to renting them out on the short-term market. Apparently, people are more willingly to pay for short-term rental housing. The second channel explains that homeowners can rent out their homes and therefore are willing to pay more for a property since they have a potential additional income. Barron et al. (217) also address the finding by mentioning the study of Zervas et al. (2017), namely the entry of small suppliers in the market who compete traditional suppliers like the hotels. An important assumption in this study is that homeowners are more likely to enter the short-term rental market when they live in less touristy neighborhoods than homeowners who live in a touristy neighborhood. Percentage change in rental prices is the dependent variable, the independent variables are the amount of AirBnB listings and observed characteristics of the investigated zip code areas. This is comparable to this study, in which the effect on house prices will be estimated instead of rental prices. Instead of zip code areas, 3 different cities are investigated in this study. Another similarity is the fact that Barron et al. (2017) control for unobserved time-varying effects. An instrumental variable is added to the regression, this variable is the amount that ‘airbnb’ is searched via Google. The study of Barron et al. (2017) is in many

9 ways comparable to this study, except for the fact Barron et al. (2017) investigate the effect on rental prices instead of house price.

Some specific research has been done about the Spanish island Mallorca (Yrigoy, I. 2016). Also this study reports a causal relationship between the growing amount of listings on AirBnB and the rental prices. Rental prices are again showed to increase as a result of increased activity of AirBnB. The contrast of this study compared to the previous discussed studies lies in the claim from Yrigoy (2016) that potential additional income is no reason for people to offer their home on AirBnb. In contrast to Yrigoy (2016), other studies, Barron et al. (2017) and Sheppard and Uddel (2016), claimed that a specific reason of the growth of AirBnB is the potential increase in income. It seems there is a contradiction between these studies considering these claims. It is likely that both studies are right about their claims. A reason for the claim of Yrigoy (2016) to be true is that Mallorca is a very touristic place and therefore the demand for short-term rental is high. In studies of Barron et al. (2017) and Sheppard and Udell (2016) the finding that people list their homes on the AirBnB site because of the potential additional income is likely to be true because it is of course attractive to rent out your home if you do not need to be there yourself.

Jefferson-Jones, J. (2015) found that interest in staying in someone’s home rather than in a hotel has increased for both people that are on holiday as for business people. Jefferson-Jones (2015) suggests that the possibility to rent someone’s property can help struggling house-owners. The potential additional income for example, can be used to amortize mortgages. House-owners can be financially supported through renting out their home and are in this way protected against negative house market downturns.

All papers well-considered, rental prices and house prices seem to be positively connected to the growth of AirBnB. Most of them are based on empirical studies in the United States of America. This paper focuses on Europe since there is little research about the effect of AirBnB on economies in Europe. The cities Amsterdam, London, Berlin and Barcelona are big cities in the west of Europe which face high levels of tourism. As for AirBnB as a whole, also in this cities the company seems to have an effect on the housing market. House prices are rising in all 3 cities and it is not clear what part can be attributed to AirBnB.

As described above, there have been several studies about the effect of the growth of AirBnB on the housing market. Most of them found a positive relationship between the growth of AirBnB and the rental prices. Most common argument for this relationship is the potential additional income that can be earned by renting out a home on the short-term market. Tourism is another factor that plays a big part in the growth of AirBnB and therefore affects the house prices and rental prices. Sheppard and Udell (2016) show the effects on the property prices through a transmission mechanism model. Four impacts of AirBnB are

10 described, three of them are positive effects on the property prices, one is a negative effect regarding negative externalities. Horn and Merante (2017) investigated the effect of AirBnB in Boston on residential housing market. Their findings indicate that an increase of the density of AirBnB of 1 unit, leads to an increase of 0.4% in local rents. They argue that further research is needed and that there is still some bias in their research despite controlling for some effects. The study of Barron et al. (2017) is set up in the same way as this study. A regression line is set up with percentage change in rental rates as dependent variable, where in this study the dependent variable is the percentage change in house prices. Yrigoy (2016) investigated the effect of AirBnB in Mallorca and also found a positive effect of the growth of AirBnB on the short-term rental prices. Barron et al. (2017) and Sheppard and Udell (2016) argued that potential additional income is one of the motivators for people to rent out their home. Yrigoy (2016) shows another perspective by stating that this is not the case for the majority of the people on Mallorca. Jefferson-Jones. (2015) adds value to the discussion by arguing that potential additional income can be used to amortize mortgages.

2. Methodology section

2.1 Assumptions

In this section the methodology will be discussed. The model assumptions are presented and the estimated model will be showed. The method used to estimate the effect of the growth of AirBnB on the house prices is explained. Several variables that influence house prices are used in the regression. For each of them is explained why the variable affects the house prices.

This paper contains the outcome of an empirical, quantitative data analysis about the effect of the growth of AirBnB on the change in house prices. This is an empirical study based on several assumptions that are made based on theories about the housing market and the other variables. The first assumption is that house prices are somewhat sticky, that is one of the reasons quarterly data is used. The effect on the House Price Index is measured considering a lag in the reaction of house prices for one quarter, one year and two years. By testing for all three lags we can see which of the three better fits the model. A causal effect of the amount of AirBnB listings on the change in house prices is assumed, this will be tested against a 5% significance level. Further assumptions that have to be made are that each city shows its own characteristics, an observable part and an unobservable part. The observable part is reflected by the following variables; the mortgage rate, the rental price index, GDP growth, the tax rate, the employment rate and a tourism expenditures in

11 each city. These factors are assumed to have an effect on the change in house prices. These control variables try to cover most of the change in house prices. However, it is possible and obvious that not all factors are covered, but the variables that are considered to be the most important are recorded in the model. The unobservable part of the change in house prices is reflected by alpha. This part may contain cultural effects, regulations and policies which are different for each city. We have to control for this part since it may cause bias in the estimated effect. An important assumption that needs to be made is that the unobservable parts of each city may not be correlated with each other. In other words, the error term of a city may not be correlated with the unique characteristics (error term) of another city. This assumption needs to be made because otherwise fixed effects, the method used to estimate the change in house prices, cannot be used. The fixed-effects model controls for all time-invariant differences between the individuals, so the estimated coefficients of the fixed-effects models cannot be biased because of omitted time-invariant characteristics. Fixed effects is used to see what causes the changes in house prices in each city. Besides, another important assumption is that higher tourism levels are likely to positively affect the amount of AirBnB listings and the house prices directly. Through the effect on AirBnB growth, house prices are indirectly affected. Thus, the house prices are affected through two ways, directly, and indirectly through the effect on the growth of the amount of AirBnB listings. Higher tourism levels cause higher demand for short-term rentals. Through this effect on AirBnB, house prices can possibly be positively affected. For doing a fixed effects regression, 7 assumptions are needed to make certain the estimators are unbiased, consistent and efficient, this however is not likely to be the case.

- Assumption Fixed Effects 1. For each i, the model is yi,t = β1 * Xi,t1 … βk* Xi,tk +αi + εi,t where t = 1, …, T, indicates the time period and i indicates the city. The β’s are the parameters to estimate and αi is the unobserved effect. Almost every estimation is set up this way, linear in parameters. In this study we assume that this is the best possible way to describe reality. It is no unreasonable description. Although, figure 7 does not suggest a linear relationship between house prices and AirBnB. This may have several reasons, for example the small sample space. This assumption is likely to hold.

- Assumption Fixed Effects 2. We have a random sample from the cross section. Data is extracted from the Organization for Economic Cooperation and Development. We may assume that this source provides representative data. The sample of the three cities is carefully chosen, this study lacks external validity. Considering the results,

12 conclusions can only be made for Amsterdam, London and Berlin. This assumption is likely to hold

- Assumption Fixed Effects 3. Each explanatory variable changes over time (for at least some i), and no perfect linear relationships exist among the explanatory variables.

Every explanatory variable varies over time. Every independent variable shows changes in almost every time-period and they are not in perfect linear relationship with each other. Although, there might be some relationships between the independent variables, these will not be perfect linear. For example higher tourism levels can possibly increase the employment rate, but these are not perfectly correlated. The third assumption for the fixed effects holds.

- Assumption Fixed Effects 4. For each t, the expected value of the idiosyncratic error given the explanatory variables in all time periods and the unobserved effect is zero: Expectation (εit I Xi, αi) = 0. In other words, the expectations of the value of the error term is 0. This does not hold in this study since the independent variables cannot fully explain the variance in the change in the House Price Index, therefore the expectation of the error term is not equal to 0. The fixed effects estimator is thus likely to be biased. The definition of unbiasedness is that the expectation of the error term is 0. This assumption does not hold

Under the first 4 assumptions the fixed effects estimator is unbiased. The key for fixed effects estimators to be unbiased is the exogeneity assumption, assumption number 4. Exogeneity is when the explanatory variables are not correlated with the error term.

- Assumption Fixed Effects 5. Variance (εit I Xi, αi) = σε2, for all t = 1, …, T, where t indicates the time period.

The variance of the error term is equal to sigma, but sigma is not the same for all time-periods. The variance is thus dependent of the time-period. This is the reason why we use robust standard errors in the regression, to account for heteroscedastic error terms. Time dummies take away some of this effect.

- Assumption Fixed Effects 6. For all t ≠ s, the errors are uncorrelated (conditional on all independent variables and αi): Cov(εit, εis IXi, αi) = 0. This is known as serial correlation or autocorrelation, which means that past values impact present and future values. Figure 6 shows a plot with residuals versus fitted values. Uncorrelated

13 errors may be assumed if the points in the scatterplot are totally random. From figure 6 we can see that the points center around residuals value of zero and fitted values value of 0.005. This assumption is not likely to hold.

Under assumption 1 to 6, the fixed effects estimator is BLUE, or in words, the Best Linear Unbiased Estimator. There is one last assumption.

- Assumption Fixed Effects 7. Conditional on Xi and αi, the εit are identical and independent distributed. Each observation is drawn from the same distribution. Each observation is unrelated to any other observation, so we can conclude independency. This assumption is likely to hold for this model.

Not all assumptions hold for this model, but assumption 1, 2, 3, and 7 will likely hold. Though, the reason why fixed effects is used is because we have to take into account the differences between the 3 cities.

2.2 Regression model

The investigated time span is from the first quarter of 2003 till the second quarter of 2016. AirBnB was founded in 2008, so this way any trends from before the foundation of AirBnB can be found. All data is quarterly because not all available data was monthly, annual data is considered too rough and would not give a precise indication of the changes in house prices because a lot of events can take place during a year, making annual data invalid. All data is modified and adjusted in the way that the time periods of the variables matches each other. Since house prices are considered sticky, the effect on house prices is estimated accounting for a lag in the reaction of house prices for one quarter, one year or two years.

All these variables are likely to influence house prices, a statement about the significance and the strength of these effects can be made after the fixed effects regression. The main goal of this study is to investigate what the effect of the growth of AirBnB is on house prices. The regression formula is assumed to has the following form and Pi,t is assumed to has a causal relationship with Gi,t:

P

= α

+ β

*G

+

+ ε

: P

= α

+ β

*G

+ X

+ ε

14 The formula is shown for t-1, but t-4 and t-8 are also tested in this study. For example, t-1 denotes a lag in the reaction of house prices of one quarter. Let Pi,t-1 represent the House Price Index index for city i at the time t-1 and let Gi,t be the percentage growth of the amount of AirBnB listings in city i at time t. Xi,t contains the observed characteristics for each city i at time t. Each letter represents the following:

Pi,t = change in percentage points in house price index for city i at time t-1 αi = Y-intercept

Gi,t-1 = change in percentage points in offered total nights for city i at time t Ri.t-1 = change in percentage points in the mortgage rate for city i at time t Ri t-1 = change in percentage points in the rental price index for city i at time t Yi,t-1 = change in percentage points in GDP for city i at time t

Ti,t-1 = change in percentage points in the tax rate for city i at time t

Ei,t-1 =change in percentage points in the employment rate for city i at time t Oi,t-1 = change in percentage points in the tourism expenditures for city i at time t

ε

Xi,t represents the observable characteristics of each city i at time t and αi denotes the unobservable characteristics of each city i. If the intercept, thus the unobservable characteristics, are uncorrelated with Xi,t, it is possible to consistently estimate the beta or Xi,t. However, the error term is assumed to be correlated with the independent variables. This is needed since heterogeneity in characteristics across the 3 cities is considered. Fixed effects regression remove the effect of those time-invariant characteristics so we can assess the net effect of the predictors on the outcome variable. αi denotes the intercept for each city. Due to the characteristics of each city, as said in the part above, the intercept differs for each city.

The alpha in the moel cannot be directly observed. Strict exogeneity with respect to the error term is required. As said, exogeneity means in this case, that the independent variables are not correlated with the error term for each single time period. Because αi contains the unobservable part of the city characteristics, it cannot be directly controlled for. By using the so called ‘within transformation’, fixed effects eliminate αi . We have to control for heterogeneity in the error term, because we may not assume that error terms are homoscedastic (see assumption 6). Therefore a robust option will be used, so that heterogeneity is accounted for. Considering the previous part, it hopefully is clear that fixed effects is a valid method for estimating the change in house prices as result of a change in the amount of AirBnB listings in each city i at time t because unobserved characteristics for each city need to controlled.

15 The following part explains why and what the causal effect of the independent variables is on the dependent variable. In this paper the effect of the growth of AirBnB on the house prices in the cities Amsterdam, London, Berlin and Barcelona will be investigated and compared by assuming that national data is representative for each city. However, it’s common known that national values can differ from specific urban values. Unfortunately, there is no specific data available for every used variable for each city so that we have to relax the assumption that national values can differ from the values which are related to the cities themselves.

2.3 The variables

The growth of AirBnB is expressed in terms of total nights offered on AirBnB per quarter per city. All values of the variables are quarterly because house prices are considered to be sticky, so monthly data would not be useful. The needed data considering AirBnB is available on the websites https://console.bluemix.net, https://airdna.com and https://insideairbnb.com. Most of the used data will come from https://console.bluemix.net because the needed variable, total nights offered per quarter per city, is available to download on that website. There is a rise of total nights offered of which the effect on the house prices per city will be studied. There is chosen for an approach by total nights offered instead of for example revenue because the growth of AirBnB is likely to have an effect on the house prices through more supply, not through pure revenue. As said before, people might be willing to pay more for property then in the past, because now they can rent out there home via AirBnB, so it’s not the revenue they make, but the possibility to rent out their house, which is expressed in offered nights.

House prices are not only responding to the growth of AirBnB, but also to some other indicators. This study picked a few control variables to investigate the real effect of AirBnB on the house prices. Data for mortgage rates, rental price indices, GDP growth, tax rates, employment rates, tourism indices and Consumer Confidence indices are collected because each of these variables is expected and assumed to have an effect on the house prices and the growth of AirBnB. Furthermore, the house price indices from the first quarter of 2010 till the second quarter of 2017 for each country in which the city is located, are gained from the website of the Organisation for Economic Cooperation and Development even as the data for the mortgage rates, rental indices, GDP growth, tax rates, employment rates, tourism expenditures and the Consumer Confidence indices.

First, the mortgage rates are an important indicator of growing demand and supply in the housing market because if mortgage rates rise, house buyers have to pay more interest so the total pay sum is higher and vice versa. Quarterly mortgage growth rates are gained

16 through modifying monthly mortgage rates of each country. Rental price are gained in the same way. Higher rental prices influence the demand because when rental prices rise, buying a house becomes lucrative.

GDP stands for Gross Domestic Product by which is meant ‘the standard measure of the value of final goods and services produced by a country during a period minus the value of imports’. (OECD.stat, 2017) GDP growth rate is another indicator of house prices and growth of AirBnB since higher GDP reflects a higher amount of total produced goods and services. Higher GDP itself has a positive effect on the economy which leads to more expenditure. Consumer confidence has an impact on house prices and especially the demand for houses. When consumer confidence grows, people have more trust in the market. The increased trust can lead to the fact that people are faster willing to buy a new home. When people are more willing to buy, demand rises. Therefore consumer confidence can influence the housing market.

Taxes are another important role player in the housing market. Higher taxes lead to lower demand for properties since the total costs of owning/renting a property are increased. Tax on property values are derived from OECD.stat (2017) and is defined as “tax on the use, ownership or transfer of property”.

Employment rates can be a valuable measure of the growth of the economy. If employment rates rises, the economy is likely to move into a positive direction. Employment rates are again derived from OECD.stat (2017) and are calculated as ratio of the employed population to the working age population. Employment rates can be affected in many different ways. If for example the demand for AirBnB listings rises caused by higher tourism rates, total demand may rise and so will employment rates. If employment rates rise, total GDP rises, higher/more incomes affect the house prices to rise.

Touristic factors are likely to affect house prices in two directions. An indicator of tourism is the expenditure in each city by tourists, which is used in this study. Tourism leads to higher local demand and thus higher employment rates. As said in the previous paragraph, growing employment rates can lead to higher house prices through higher/more incomes. On the other side, tourism may lead to negative externalities like too much noise and waste on the street. If this is the case, people are less willingly to live in a neighborhood where these kind of things occur. This leads to lower house prices since demand decreases. The data is gained from OECD.stat ( 2017).

17

3. Descriptive statistics

In this paragraph the descriptive statistics will be presented. As can be seen in the tables below, not all variables have the same amount of observations. The amount of 30, 30 and 29 observations for AirBnB listings is due to a lack of available data. Tourism expenditures are only calculated by the OECD till the last quarter of 2015. All data is in percentage points, if you multiply the data points by 100% you get the percentage growth. Amsterdam shows a quarterly mean of 0.171646 percentage points growth of AirBnB, that is roughly 17.16%. In other words, the amount of AirBnB listings growths on average with 17.16% each quarter from Q1-2004 till Q2-2017. Important to realize is that AirBnB was founded in 2008, all data points before the foundation are equal to zero, this takes the average growth down. The amount of AirBnB listings growths on average with 24.64% each quarter in London and in Berlin the amount of AirBnB listings growths on average with 15.58%. Figure 2 shows the real House Price Index values over time for all three cities. Interesting to see is that the House Price Index values rise in London in 2005 and 2006, where this is not the case for Amsterdam and Berlin. 20050 represents the beginning of 2005 and 20075 represents the half of 2007 for example. In 2007 we see a tremendous drop in the house price index for London, which can be explained by an early impact of the financial crisis. Berlin does not seem to suffer from the financial crisis considering the graph. Amsterdam shows a decreasing line from July 2007 till July 2012, during the crisis, after which the house price index values rise again till present. London and Berlin also pick up the rising line since July 2012, this can indicate that the economy is growing and the financial crisis is past from then. Amsterdam however still underperforms in index numbers compared to London and Berlin.

Figure 3, 4 and 5 reports the increase of offered nights over time for the three cities. Every point represents the amount of offered nights that is added in the previous quarter. As you can see, the amounts of offered nights differ between the three cities, where London has the highest increase during the end of 2016 of about 550.000 offered nights. However, it is more import to look at the trends we can derive from the graphs. We see that after AirBnB’s foundation, the amount of offered nights does not grow rapidly, till 2011. Some explanation for the immediate high increase in 2011 might be that the financial crisis was past. Another explanation might be that AirBnB was simply unknown by customers until then. After people got to know about AirBnB they possibly made use of it. After 2012 the cities does not seem to produce a stable amount of added offered nights, but it keeps growing on average. Since these are all positive numbers we can again conclude that AirBnB is still growing and a lot of people decide to offer their home through AirBnB.

18

4. Results and discussion

In this section the results will be presented. Table 2 and 3 presents the outcomes of the fixed effects regressions. There have been done 10 regressions, a regressions with and without time dummies with a considered lag in the reaction of house prices for one quarter, one year and two years. The tables also contain a regression with time dummies for t-1 and t-4 with real House Price Index values to see if there might be a difference between the effect on real values or growth values. Furthermore, there has been done a regression with the change of the House Price Index and the growth of AirBnB in natural logarithms, this way we can say something about the elasticity; the coefficient represents the elasticity of the growth of AirBnB on the change of the House Price Index. In other words, if AirBnB growth increases with 1, the change of the House Price Index changes with the coefficient times %. The last fixed effects regression is a regression without time dummies for t-4 and omitting the variable tourism expenditures. First, the regressions without time dummies for t-4 shows the lowest p-values for almost all variables. The p-values for the growth of AirBnB seem to increase tremendously when tourism expenditures are omitted from the model. Tourism expenditures are likely to have a big impact on AirBnB and the other way around. Not all variables have turned out to be a good estimator for the change of the House Price Index, this may be due to a little amount of data points, measurement errors or it can be the case that the independent variable simply is no good estimator for the change of the House Price Index.

4.1 Results

The regressions without time dummies for t-1, t-4 and t-8 respectively show p-values of 0.255, 0.026 and 0.560. For a lag of 4 quarters in the reaction of house prices, the growth of AirBnB seems to have a positive, significant effect on the change of the House Price Index considering a significance level of 0.05. However, this effect is very small, the corresponding coefficient is 0.0072841. The effect thus is significant but very weak. The regressions with time dummies for t-1, t-4 and t-8 respectively show p-values of 0.175, 0.035 and 0.138. Again, only for t-4 the effect is positively significant, but all coefficients seem to be very small and therefore the effect is again weak. From this we can conclude that a lag in the reaction of house prices of one year fits the model best. The assumption we made, that house prices are sticky seems to be a fair one. Tax rates and tourism expenditures are significant at a level of 5% for t-1 and t-4 in the regressions with and without dummies where is not the case for t-8. An explanation for this is that house prices are not that sticky in the long term. Although these explanatory variables seem to be significant, the coefficients are small, thus

19 the effect is not really strong. If small changes in the growth of AirBnB, tax rates or tourism expenditures occur, the effect on house prices is almost negligible. Time dummies can explain how different time periods contribute to explain the variation in the dependent variable, this seems to work for t-8. The p-value with time dummies decreased to 0.138, where the p-value without time dummies for t-8 is 0.560. For t-8 the different quarters seem to explain a big part of the variation in the change of the House Price Index. This effect is less for t-1 and t-4. A possible explanation for time dummies to improve the model is that time dummies account for time trends like the financial crisis. The financial crisis affected for example house prices during that period, using time dummies in the model takes this into account.

The fixed effects regression for t-1 and t-4 with time dummies and real House Price Index values give p-values of 0.475 and 0.998, thus insignificant at a level of 0.05. The growth of AirBnB thus seem to have no effect on the real values instead of the percentage change of the House Price Index.

The regression with the growth of AirBnB and the change of the House Price Index expressed in logarithms gives a p-value of 0.073, significant at a 0.10 level but not at a 0.05 level. For doing this regression we picked the most significant estimation so far, for t-4 without time dummies, this way we hope to have gain a significant estimation with terms expressed in logarithms. As already said, the coefficient represents the elasticity, which is 0.176723. If total offered nights through AirBnB increases by 1 percentage point, the change of the change of the House Price Index equals +0 .176723%. Remind that an increase of 1 percentage point in total offered nights is equal to an increase of thousands of offered nights. Tourism expenditures seem to have a small, but significant impact on both house prices and AirBnB. Not only was the p-value of tourism expenditures significant at a 0.05 level most of the times, it also decreased the p-value of AirBnB enormously every time tourism expenditures was added in the model. There seems to be some collinearity between the two variables. Tourism expenditures are in theory likely to increase if AirBnB grows because it can indicate that more tourists come to the city and make use of AirBnB. The other way around, AirBnB is likely to grow if tourism expenditures increase since higher tourism expenditures can indicate more tourists, as already said. The same effect is likely to work on both variables. The effect can be seen from the p-values, the same regression model including the variable tourism expenditures gives a p-value of 0.026, where the same regression model omitting the variable tourism expenditures gives a p-value of 0.412.

The most significant estimation is the fixed effects regression without time dummies for t-4 with a p-value of 0.026. However, the best estimation is the estimation considering a lag in the reaction of house prices of one year with time dummies. This way trends over time like the financial crisis are accounted for. This is consistent with the leading assumption of

20 this study; house prices are sticky. The corresponding coefficient is 0.0072841. The interpretation behind this number is that if the growth of the amount of offered nights on AirBnB equals 1 percentage point, the change of the change of the House Price Index will

be +0.0072841%.

4.2 Discussion

This study gives new insights on the effects that AirBnB has in Amsterdam, London and Berlin. Most research that has been done is about the effect of AirBnB on rental prices and the hotel industry. Most of these studies are related to American markets. The effect of the growth of AirBnB in big cities in Europe was not well investigated so far. Also, the effect on house prices was not a main topic in previous studies about AirBnB. From this study it is clear that AirBnB has a significant effect on the house prices in the three cities. The effect of the growth of AirBnB on the House Price Index was significant in the fixed effects regression for t-4 without time dummies. This confirmed our assumptions that house prices are sticky, a change in the total offered nights through AirBnB at a specific time works through on the house prices one year later. Therefore, this study shows some significant evidence that the AirBnB positively affects house prices. It can be concluded that there is indeed an effect of AirBnB on house prices, but as showed in the results, the effect is very small. Only if the total offered nights through AirBnB would increase by big, unrealistic numbers, the house prices would be affected noticeable. Every small increase in total offered nights through AirBnB would cause a negligible effect on house prices. However some of the results were significant, but weak, we have to take into account some possible factors that may influence this study. First, the economy is stabilizing the last few years, house prices are rising, incomes increase and the economy as a whole is picking up. A good question might be if house prices were to grow at the current levels if AirBnB had not existed, although a positive significant effect is estimated. Second, the little amount of data might influence the outcome of the estimates. For good estimating, a big sample size is needed, the lack of this can influence the outcome of this study. Besides, not all control variables are showed to have a significant effect on the house prices. This can have several reasons, maybe the control variable just simply is no good estimator for the house prices, or national data that is used does not well represent the city characteristics. National GDP growth for example is taken in this study, where GDP growth on the country-side might differ a lot from GDP growth in big cities as Amsterdam, London and Berlin. There is a relatively high chance that this causes bias in the estimators. Further research needs to be done about this subject. Because AirBnB was founded in 2008, there are little data points available, these might cause bias in

21 the estimates. More factors that influence the house prices have to be controlled for. Since this study is not external valid, research about other cities in Europe is needed.

5. Conclusion

The problem this study attempted to address was the possible effect of the growth of AirBnB on house prices in Amsterdam, London and Berlin. The need to investigate this effect originated from the outcomes of previous studies about the effect of AirBnB on rental prices and the hotel industry. The hypothesis stated in the introduction was that AirBnB has positively affected house prices in Amsterdam, London and Berlin. As reported in the results and discussion, not all regressions turned out to be significant. Although some of the results were significant, most of them were very weak. Small changes in the explanatory variables caused negligible impacts on house prices in the three cities. However, the estimation with the highest significance, the fixed effects regression without time dummies with a considered lag of one year in the reaction of house prices, was significant with a p-value of 0.026. The coefficient turned out to be 0.0072841, which indicates a small effect. Since the outcome is significant, the hypothesis stated in the beginning of this paper is accepted. The conclusion is that AirBnB has positively affected house prices in Amsterdam, London and Berlin from their foundation in the third quarter in 2008 till the second quarter of 2016.

22

6. References

-Barron, K., Kung, E., & Proserpio, D. (2017). The Sharing Economy and Housing Affordability: Evidence from Airbnb.

-Horn, K., & Merante, M. (2017). Is home sharing driving up rents? Evidence from Airbnb in Boston. Journal of Housing Economics, 38, 14-24.

-Jefferson-Jones, J. (2015). Can Short-Term Rental Arrangements Increase Home Values?: A Case for Airbnb and Other Home Sharing Arrangements.

- Kokalitcheva. K. (2015,17 June) Here’s how Airbnb justifies its eye-popping $24 billion valuation. http://fortune.com/2015/06/17/airbnb-valuation-revenue/

-Litvak, J. (2016). Could Helpful Short-Term Rental Platforms be Negatively Affecting the Housing Market?.

-Martin, C. J. (2016). The sharing economy: A pathway to sustainability or a nightmarish form of neoliberal capitalism?. Ecological Economics, 121, 149-159.

-Sheppard, S., & Udell, A. (2016). Do Airbnb properties affect house prices (No. 2016-03). -Yrigoy, I. (2016). The impact of Airbnb in the urban arena: towards a tourism-led

gentrification. The case-study of Palma old quarter (Mallorca, Spain). Turismo y crisis, turismo colaborativo y ecoturismo. XV Coloquio de Geografía del Turismo, el Ocio y la Recreación de la AGE, 281-289.

-Zervas, G., Proserpio, D., & Byers, J. W. (2014). The rise of the sharing economy: Estimating the impact of Airbnb on the hotel industry. Journal of Marketing Research.

23

7. Appendix

Figure 2: Graph of the real House Price Index values of Amsterdam, London and Berlin.

Figure 3: increase of offered nights on AirBnB in Amsterdam over time

24

Figure 4: increase of offered nights on AirBnB in London over time

Figure 5: increase of offered nights on AirBnB in Berlin over time

2008 2012 2016

25

Table 1: summary table

Variable Obs Mean Std. Dev. Min Max

AirBnB growth 165 0.1912717 0.3992928 0 3.768865 HPI 165 0.0037288 0.0166209 -0.06252 0.043956 Mortgage rate 165 0.0341244 0.4226494 2.541666 3.397 GDP 165 0.3668234 0.7604281 -4.48549 2.047503 CCi 165 0.000363 0.0054776 -0.01857 0.016547 Rental index 165 0.005186 0.0034005 -0.00375 0.025314 Tax rate 165 0.0017949 0.0149072 -0.03096 0.042609 Employment rate 165 0.0014107 0.0043325 -0.02267 0.010579 Tourism expenditure 165 0.0744806 0.5499074 -2.1765 0.899862

26

Table 2: P-values and standard errors , the effect of the growth of AirBnB on the house prices in Amsterdam, London and Berlin.

t-1 t-4 t-1 t-4 t-1 time dummies Real House Price no time dummies time dummies Index

(1) (2) (3) (4) (5) AirBnB 0.255 (0.003) 0.026*** (0.003) 0.175* (0.005) 0.035*** (0.005) 0.475 (2.213) Mortgage rate 0.769 (0.003) 0.945 (0.003) 0.553 (0.004) 0.941 (0.004) 0.893 (2.012) GDP 0.896 (0.002) 0.113* (0.002) 0.931 (0.003) 0.903 (0.003) 0.336 (1.599) CCI 0.005*** (0.237) 0.643 (0.245) 0.121* (0.349) 0.694 (0.347) 0.050*** (168.233) Rental index 0.089** (0.376) 0.346 (0.388) 0.118* (0.498) 0.911 (0.495) 0.217 (242.351) Tax rate 0.000*** (0.081) 0.001*** (0.084) 0.009*** (0.110) 0.049*** (0.110) 0.007*** (52.929) Employment rate 0.658 (0.010) 0.840 (0.310) 0.997 (0.377) 0.937 (0.374) 0.125* (178.237) Tourism exp. 0.000*** (0.002) 0.000*** (0.002) 0.001*** (0.003) 0.010*** (0.003) 0.130* (0.037) Constant 0.018*** (0.002) 0.657 (0.003) 0.170* (0.009) 0.695 (0.009) 0.000*** (4.353)

* * *

**

*

27

Table 3: P-values and standard errors , the effect of the growth of AirBnB on the house prices in Amsterdam, London and Berlin.

t-4 t-8 t-8 t-4 t-4

time dummies Log HPI an AirBnB Omitting Tourism exp. Real House Price no time time no time no time

Index value dummies dummies dummies dummies

(6) (7) (8) (9) (10) AirBnB 0.998 (2.238) 0.560 (0.003) 0.138* (0.003 0.073** (0.050) 0.412 (0.004) Mortgage rate 0.768 (2.111) 0.320 (0.002) 0.105* (0.001) 0.383 (1.013) 0.805 (0.001) GDP 0.181* (1.678) 0.079** (0.001) 0.071** (0.002) 0.738 (0.729) 0.472 (0.002) CCI 0.492 (176.536) 0.212 (0.193) 0.211 (0.251) 0.214 (27.74 0.472 (0.215) Rental index 0.128* (254.321) 0.057** (0.125) 0.236 (0.234) 0.197* (93.641) 0.346 (0.266 Tax rate 0.243 (55.541) 0.577 (0.135) 0.831 (0.131) 0.877 (16.642 0.046 (0.066) Employment rate 0.145* (187.034) 0.865 (0.243) 0.341 (0.283) 0.529 (152.36) 0.390 (0.239) Tourism exp. 0.365 (0.039) 0.067** (0.002) 0.915 (0.005) 0.466 (0.128 - - Constant 0.000* * * (4.568) 0.078** (0.001) 0.571 (0.003) 0.020*** (0.746) 0.819 (0.002)

* * *

**

*

28

Figure 6: Scatterplot: Residuals versus fitted values

Figure 7: Scatterplot: House Price Index versus growth of AirBnB

-. 0 6 -. 0 4 -. 0 2 0 .0 2 .0 4 R e s id u a ls -.03 -.02 -.01 0 .01 .02 Fitted values -. 0 6 -. 0 4 -. 0 2 0 .0 2 .0 4 H P I 0 1 2 3 4 AIRBNBgrowth