The non-linear effect of Airbnb activity on house prices and the role of an area's income level in the relationship

(1)

The Non-Linear Effect of Airbnb Activity on House Prices and the Role of

an Area’s Income Level in the Relationship

Lukas Bruder, 11124202 January 2018, Amsterdam

Bachelor Thesis Economics and Business Specialization: Finance and Organization Supervisor: Ieva Sakalauskaite

(2)

Abstract

This paper investigates whether there is a non-linear effect of Airbnb activity on house prices and whether the possible positive effect depends on an area’s income level. A panel data analysis is conducted using a quadratic regression model. Hereby house prices are used as independent, and Airbnb activity as an explanatory variable. Key house price determinants are added as control variables, as well as year and district fixed effects. The sample consists of twelve districts from Amsterdam, each with annual data between 2005–2016. Unfortunately, two control variables have limited annual observations. For the second analysis, the initial sample is divided depending on the districts’ relative income level. The quadratic regression is run again with each dataset half in order to compare the results. The empirical results suggest a non-linear relationship turning from positive into negative, and an initially stronger positive impact in areas with a relatively lower income level of Airbnb activity on house prices. Further research with fewer limitations is warranted to validate the results.

Statement of originality

This document is written by student Lukas Bruder, who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

1. Introduction

Peer-to-peer markets is a phenomenon of the 21st_{century. Their primary function is to connect}

sellers with buyers, and enable them a safe and comfortable transaction (Einav, Farronato & Levin, 2016). Initially it was essential to start a business and grow its network in order to offer a good or service. Nowadays, online market platforms like eBay and Amazon enable small businesses, or even private sellers, to offer their products. Platforms that create peer-to-peer markets also build new industries.

The idea of home sharing was widespread by Airbnb. The company created an online community marketplace that connects travelers and homeowners. It enables the traveler to book a room or an entire accommodation from a host for a short-term stay. As Airbnb established a new possible source of income for homeowners and long-term renters, there is a reason to assume that home sharing affects the housing market. The limited amount of research that exists in this field has found a positive relationship between Airbnb activity and house prices (Van der Bijl, 2016; Sheppard & Udell, 2016), as well as rents (Horn & Merante, 2017), to be statistically significant. More detailed research of this relationship is needed.

The law of economics of diminishing marginal utility says that the utility obtained from consuming a good decreases with every unit. This is applicable to income. Therefore, Airbnb activity’s effect on house prices may change over time, and depending on Airbnb’s popularity. There is, furthermore, room to speculate that Airbnb’s effect of on the housing market differs depending on the wealth level of the area, for the new income opportunity may be more appreciated if the area’s income level is relatively low.

In order to widen the scientific knowledge in this field, this paper attempts to answer the following two research questions:

Is there a non-linear effect of Airbnb activity on house prices?

If there is a positive impact of Airbnb activity on house prices, does it depend on an area’s income level?

Two hypotheses are tested in this paper. Correlational panel data analyses are conducted using a quadratic regression model. The panel data consists of district-specific annual house prices, Airbnb activity estimates, and control variables. Unfortunately, the analysis suffers partially from omitted variable bias or a limited number of observations due to data shortage of the

(5)

control variables. The results of this study indicate that annual Airbnb activity within a district is associated with a positive impact on house prices, which decreases with every marginal Airbnb activity unit and becomes negative at a certain level. This suggests a non-linear relationship. The second analysis suggests that Airbnb activity has a stronger effect on house prices in areas with relatively less income. However, the results of the second analysis were not completely consistent and additionally indicate that the positive impact of Airbnb activity on house prices diminishes more intensely in districts with relatively lower income. This could be caused by research’s limitations. Nevertheless, the results found in this thesis contribute to the scientific knowledge in this field, and may additionally assist the City of Amsterdam on how to regulate Airbnb activity.

2. Literature review

2.1 Key house price determinants

Properties are special assets. Properties can be bought as investments, residential areas, or commercial spaces. Real estate investors speculate that their asset valuation will rise and provide income during the holding period, e.g. rental income subtracted from repair costs. A simple return calculation for real estate investors is (Brown & Matysiak, 2000, p. 210):

property investment return = selling price + periodical income − purchase price purchase price

(0)

A property purchase is therefore only a good financial investment when the price is expected to sufficiently increase in the future or if the rental income is adequately high whilst there are only limited periodical costs. The demand for properties should rise with the expected return. Muellbauer and Murphy (2008) explain that the mispricing arises due to excessive expectations, e.g. when the fundamental price determinants experience positive changes over a period of time, the market overestimates the future house values; once the positive shocks are missing and the market realizes that it is overheated, the price falls are inappropriately deep due to exaggerated negative forecasts.

Basic economic theory says that the price of a certain good is determined by the equilibrium of the quantity supplied and demanded. Muellbauer and Murphy (2008) apply this law on the housing market. Their model states that the house price depends on the supply, stock

(6)

demand factors. In order to determine a house price, it is essential to know the drivers of the demand for houses.

One of the demand fundamentals is income. A property is a rather expensive good, and, therefore, it requires a certain wealth level to purchase a house, instead of renting one. Subsequently, the demand for houses increases with the income level of a household. Several researchers (Muellbauer & Murphy, 2008; Chen & Cheng, 2017; Égert & Mihaljek, 2007; Black, Fraser & Hoesli, 2006; Can & Megbolugbe, 1997) share this assumption. Case and Shiller (2003) showed through empirical research that there is a positive relationship between income and house prices. Another argument for the relationship between house demand and income is that an increase in household income should motivate house owners to upgrade to better accommodation (Capozza, Hendershott, Mack, & Mayer, 2002).

The level of interest rate or mortgage rate further influences the demand for houses. This can be explained by the fact that many property investors take out loans in order to finance the purchase of a house, and the cheaper this loan is, the higher the demand for properties. Historical surveys have documented that the public expects house prices to increase further when interest rates are low (Muellbauer & Murphy, 2008). Lowering the interest rate is an economy-enhancing instrument, and is, therefore, commonly assumed to be a fundamental factor of house prices (Muellbauer & Murphy, 2008; Clayton, Ling & Naranjo, 2009; Égert & Mihaljek, 2007; Chen & Cheng, 2017; Haffner & de Vries, 2009; Black, Fraser & Hoesli, 2006).

A third key demand factor in the real estate market is the demographic factor, i.e. the population. As properties serve as accommodation, the demand for them rises with the number of people who require a living area in a certain region. Several types of empirical research have confirmed the significant positive effect of demographic factors, such as the population on the housing demand and house price (Muellbauer & Murphy, 2008; Égert & Mihaljek, 2007; Chen & Cheng, 2017; Haffner & de Vries, 2009; Capozza, Hendershott, Mack, & Mayer, 2002).

Besides the key factors, regional specific conditions also affect the demand for properties, which is the so-called spatial dependency. Spatial dependency can be separated into adjacency effects and neighborhood effects, of which the former describes the externalities of the absolute location, and the latter takes neighborhood characteristics such as public services, neighbors, environment, and noise into account (Can, 1992). Can and Megbolugbe (1997), as well as Bourassa, Cantoni, and Hoesli (2007) found the spatial dependency to be a significant factor for house prices. These findings appear logical, given that location externalities and a

(7)

property’s direct environment are factors that directly affect the property’s resident and, therefore, also its value.

2.2 The effect of Airbnb activity on house prices – past research

The company Airbnb was founded by Joe Gebbia, Brian Chesky, and Nathan Blecharczyk in 2008, and is based in San Francisco, California. Its concept is an online platform through which homeowners can offer their whole properties, apartments, or single rooms—mainly for short-term stays—to others in exchange for payment. Long-short-term renters may also offer their accommodation or a single room on Airbnb if their landlords allow them to do so. As a result, the company has created a new income source for landlords through short-term rentals, which did not exist in the same form before.

Today, Airbnb is active in over 65,000 cities, counts over three million accommodation listings on its website, and has connected approximately 200 million guests to hosts, which demonstrates its popularity and that its externalities must be increasing (Airbnb, 2017).

Besides the new income source for the hosts, Airbnb renters also raise city populations. Short-term visitors’ consumption also supports local businesses (Sheppard & Udell, 2016). Consequently, Airbnb must have a positive impact on the demand for houses, as it increases both city population and the return on property investments.

Short-term stays, however, also create negative externalities, such as noise, uncertainty in neighborhoods, and the overuse of publicly provided goods (Sheppard & Udell, 2016; Van der Bijl, 2016). As mentioned above, the demand for houses has a spatial dependency. Hence, these negative neighborhood effects caused by Airbnb should decrease property prices. Van der Bijl’s research (2016) indicates that Airbnb activity is positively related with the perceived nuisance level in neighborhoods.

The limited existing amount of past empirical research about the effect of Airbnb on house prices, which was conducted in Amsterdam and New York, has indicated that a positive relationship between Airbnb’s services and house prices (Van der Bijl, 2016; Sheppard & Udell, 2016). Related empirical research about the impact of Airbnb on rents, which was undertaken in Los Angeles and Boston, showed that Airbnb activity is associated with the rise of rental payments, which consequently enhances house values (Horn & Merante, 2017). All the directly related past research conducted correlational analyses, with single sale price or unit rent as a dependent variable, Airbnb activity as an explanatory variable, and house characteristics as control variables, whilst controlling for locational and time fixed effects.

(8)

However, the definition of Airbnb activity varied. Horn and Merante (2017) used the number of listings divided by the number of residences to recreate Airbnb activity, whilst Van der Bijl (2016) used the number of reviews is tens of thousands. Sheppard and Udell (2016) applied several measurements of Airbnb activity, such as reviews and listings. This research applies the Airbnb activity measure by Van der Bijl (2016), because reviews reflect actual bookings, whilst listings themselves do not display activities. Contrary to the introduced studies, this research conducts the regression analysis with panel data, takes the median house price per square meter of a district as a dependent variable, and controls for key house price determinants. This is done in order to study the effect of Airbnb on house prices observed across time, and broaden the scientific knowledge of the relationship. The usage of districts and their median house price per square meter filters out house characteristics. The addition of the key house price determinants may avoid omitted variable bias and improve the accuracy of the research. As there still may be individual factors influencing house prices during different time periods or at specific locations, this research uses fixed effect as well.

2.3 Diminishing marginal utility of income

Airbnb has not even existed for a decade yet, and the complaints against it are increasing with time and popularity. Big cities like New York, Berlin, and London have restricted their regulations towards Airbnb because non-homeowners are disadvantaged by the rising social cost and in order to fight against housing shortage, provide living space for natives in the city center (“New York deflates”, 2016).

In Amsterdam specifically, the city council set limits for Airbnb hosts in 2014, which were nonetheless later proven not to have been adhered to (“Three-quarters of Amsterdam”, 2016). Consequently, in 2016 Airbnb and the City of Amsterdam signed an agreement that obliges the community marketplace to delete listings that were booked for more than 60 days in a year already from its website. Moreover, Airbnb complied to build a ‘neighbor tool’ on its website, which enables Amsterdam residents to complain about noise or other negative incidents caused by short-term renters (“Amsterdam, Airbnb agree”, 2016).

This news shows that the public opinion towards Airbnb in Amsterdam might have worsened drastically within the last three years. This may be explained by the fact that the quantity of short-term stays rose immensely in the last few years. From 2015 to 2016, the number of nights in which Airbnb customers stayed in Amsterdam increased by 125%, and reached nearly 1.7 million (“Airbnb continues to”, 2017). Therefore, the impact of the negative

(9)

externalities may begin to equalize the positive effects or even outrun them. This would result in a non-linear effect by Airbnb on house prices.

The law of economics of diminishing marginal utility states that the utility derived from the consumption of an additional product, keeping the utilization of other goods constant, decreases with every unit. This may be applied to the relationship between Airbnb and housing prices. Homeowners who offer their properties on Airbnb may receive less satisfaction from the income generated by the short-term rentals with every marginal tenant. Layard, Mayraz, and Nickell (2008) researched the marginal utility of income. Their study confirmed the assumption that the utility gained from supplementary income decreases with income.

Taking this into account, the predicted non-linear effect of Airbnb on house prices may not only arise from the increasing significance of the nuisance, but also a decreasing valuation of the additional income source. Whilst during the rise of Airbnb the opportunity to receive extra income from hosting guests seemed very attractive, the utility received by the income source decreases after some time, or after reaching a certain wealth level. Therefore, willingness to pay a higher price for a property, because it can be partially financed through short-term stays, decreases.

Hypothesis 1: The effect of Airbnb activity on house prices is non-linear.

Income level differs per area. Hence, according to the law of diminishing marginal utility, additional income should generate lower marginal utility in an area with initial high income than in an area with initial low income. Frey and Stutzer found (as cited in Easterlin, 2005) that higher income increases utility more in developing countries than in wealthy countries. Based on this, there is reason to presume that the positive effect of Airbnb activity on house prices falls with an area’s income level.

Hypothesis 2: The positive impact of Airbnb activity on house prices decreases with an

area’s income level.

Testing the two hypotheses will add knowledge to the existing literature, for similar research about non-linear effects of Airbnb on the real estate market, or differences in the intensity of the effect of Airbnb in relation to the income level of an area, have not been studied yet.

(10)

3. Empirical analysis

3.1 Method and model

The aim of this thesis is to study whether there is a non-linear effect of Airbnb on house prices, and whether the positive impact of Airbnb is stronger for less expensive houses. The two hypotheses in this paper are tested through quantitative empirical research. This is done with a panel data analysis of annual district-level data. It is attempted to estimate a non-linear relationship by conducting a quadratic regression, to which control variables are gradually added. Furthermore, district-fixed effects are included to the regression, to control for unobservable district characteristics and year-fixed effects, to capture time trends of house prices. The following regression is estimated in order to study hypothesis 1:

𝑙𝑛𝐻𝑜𝑢𝑠𝑒𝑃𝑟𝑖𝑐𝑒_𝑖𝑡

= 𝛼 + 𝛽₁𝐴𝑖𝑟𝑏𝑛𝑏𝐴𝑐𝑡𝑖𝑣𝑖𝑡𝑦_𝑖𝑡+ 𝛽₂𝐴𝑖𝑟𝑏𝑛𝑏𝐴𝑐𝑡𝑖𝑣𝑖𝑡𝑦_𝑖𝑡2

+ 𝛽₃𝑙𝑛𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛_𝑖𝑡+ 𝛽₄ 𝑀𝑜𝑟𝑡𝑔𝑎𝑔𝑒𝑅𝑎𝑡𝑒_𝑡

+ 𝛽₅𝑙𝑛𝐻𝑜𝑢𝑠𝑖𝑛𝑔𝑆𝑢𝑝𝑝𝑙𝑦_𝑖𝑡+ 𝛽₆𝑙𝑛𝐼𝑛𝑐𝑜𝑚𝑒_𝑖𝑡+ 𝜃_𝑡+ 𝛿_𝑖 + 𝜀_𝑖𝑡 (1)

where 𝑙𝑛𝐻𝑜𝑢𝑠𝑒𝑃𝑟𝑖𝑐𝑒𝑠_𝑖𝑡 is the median house price per square meter of a district i during year t and 𝐴𝑖𝑟𝑏𝑛𝑏𝐴𝑐𝑡𝑖𝑣𝑖𝑡𝑦_𝑖𝑡 is the quantity of Airbnb service reviews divided by ten thousand for a given district i during year t.

The dependent variable is the house price, and is measured by the median price per square meter. As commonly done in other real estate research, the natural logarithm of the dependent house price variable is used in order to get the percentage change caused by the independent variable.

The independent variable, Airbnb activity, is measured by the number of reviews written (in tens of thousands). The square of the independent variable is added to the regression in order to control for the non-linear effect. Its coefficient, 𝛽₂, is the coefficient of interest in the first analysis, since a non-linear effect would be shown by its significance.

𝜃_𝑡represents the year fixed effects and 𝛿_𝑖stands for the district fixed effects. 𝛼 is the constant and 𝜀_𝑖𝑡 represents the error term.

𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛_𝑖𝑡, 𝑀𝑜𝑟𝑡𝑔𝑎𝑔𝑒𝑅𝑎𝑡𝑒_𝑡, 𝐻𝑜𝑢𝑠𝑖𝑛𝑔𝑆𝑢𝑝𝑝𝑙𝑦_𝑖𝑡 and 𝐼𝑛𝑐𝑜𝑚𝑒_𝑖𝑡 are the control variables of the regression. As explained in the literature review, population, mortgage rate, and income are the key factors of property demand, and housing supply is the supply of

(11)

properties. The logarithm of the population, housing supply, and income variable are used in the regression to find the percentage effect.

To test the second hypothesis, the original data set is divided into two separate groups, depending on the level of income per district. This results in a dataset containing the districts with a relatively higher income, and one with districts in which the income is relatively lower.

Equation (1) is run again independently per dataset, and therefore reveals, through the comparison of the resulting 𝛽₁ per dataset, whether the positive impact of Airbnb activity differs depending on income level.

3.2 Sample selection and data

The Gemeente Amsterdam divides the city in 23 areas, which are visualized in Figure 1. In this thesis, 22 of the areas are consolidated into twelve districts, which are the entities of the panel data. The selection was based on the availability of district-specific information from the house price data. Westpoort is excluded from the research, as it is an industrial/harbor area with almost no residence. The composition of the districts is shown in table 1.

(12)

Table 1: Composition of the twelve districts of Amsterdam used in this thesis District

Number

District Name Areas

1 Centrum Centrum Oost, Centrum West

2 Westerpark Westerpark

3 Oud-West and De Baarsjes Oud-West/De Baarsjes

4 Zeeburg Indische Buurt, Ijburg - Zeeburgereiland

5 Bos en Lommer Bos en Lommer

6 Amsterdam-Noord Oud Noord, Noord Oost, Noord West

7 Geuzenveld - Slotermeer Geuzenveld - Slotermeer

8 Osdorp Osdorp, De Aker Nieuw Sloten

9 Slotervaart and Overtoomse Veld Slotervaart, Overtoomse Veld

10 Zuidoost Bijlmer Centrum, Bijlmer Oost,

Gaspterdam Driemond

11 Oost and Watergraafsmeer Oud Oost, Watergraafsmeer

12 Zuideramstel De Pijp, Oud Zuid, Buitenveldert Zuidas

The house price data for the dependent variable used in this thesis was provided by the Dutch association of realtors (Nederlandse Vereniging van Makelaars en Taxateurs in onroerende goederen NVM). They supplied yearly data of the transactions counts, median selling price, and median price per square meter for all properties sold (intermediate houses, so-called corner or end houses, two-under-one roof houses, and apartments) in the twelve districts of Amsterdam, outlined in Table 1, between 2005–2016. The NVM observed 91,701 transactions of properties in Amsterdam during the analyzed period.

The data for the Airbnb activity variable was provided by InsideAirbnb.com. Inside Airbnb is a project under the direction of Murray Cox, who provides Airbnb data to the public for research purposes. The data is gathered through web scraps from the publicly available Airbnb website. Therefore, the data by Inside Airbnb is sourced from the Airbnb site itself, but analyzed, purified, and pooled by Inside Airbnb. The website offers a review dataset that consists of all creation dates and the corresponding listing IDs of the reviews written in Amsterdam. In this study, the reviews conducted from 2009 until 2016 are used. Inside Airbnb

(13)

additionally provides a dataset with detailed information about every Airbnb listing in Amsterdam, such as the listing ID, room type, price, district, and coordinates. Matching the two datasets enables one to determine how many reviews per year were written in a district, and thus makes for the desired annual district-level data. Of the 238,353 reviews, 238,302 were taken into account in the regression, given that no location could be allocated for 51 of them.

The annual mortgage rate data was handed over from De Nederlandse Bank. It is the only variable in the analysis that does not vary per district.

The population data used in this analysis is district specific, and annually measured between 2005–2016. Only the number of people who were officially registered in Amsterdam at the beginning of a year were taken into account. Therefore, it does not include short-term users of Airbnb. The data was provided by OIS (Onderzoek, Informatie en Statistiek). OIS Amsterdam is a database containing district specific annual data from Amsterdam about over 500 variables in the categories, e.g. activities, population, living, sport, income, and many more. The data is gathered through surveys or sourced by the municipality of Amsterdam or the Central Bureau of Statistics.

The housing supply measure in this study was estimated by the number of registered addresses as of every 1st_{of January. OIS Amsterdam supply district-specific data annually.}

Unfortunately, they only provide data from 2011 on.

The data of the income variable used in this analysis is annual and at district-level. It was estimated by the average gross household income, deducted by insurance and tax costs. The data was supplied by OIS. Their income data collection is regretfully three years delayed, and, as a result, the income measure of this study is only provided between 2005–2014.

The data shortage of housing supply and income limits this panel analysis to only be able to control for these key factors from 2011 until 2014.

In order to conduct the second analysis, the districts were divided depending on their income level. This was done by calculating the median income of every district during the period of 2005–2014 (2014 is the most recent district specific income measure of OIS), and dividing the twelve neighborhood sections into two relatively lower and higher income groups. Table 2 shows the median annual household income of the districts in descending order and divides them into two new datasets.

(14)

Table 2: Division of initial dataset depending on income level District

Number

District Name Median annual average

household income 2005–2014 (in Euros)

Dataset H (Districts with relatively higher income):

12 Zuideramstel 35,500

4 Zeeburg 34,197

11 Oost and Watergraafsmeer 33,900

1 Centrum 33,600

8 Osdorp 32,722

9 Slotervaart and Overtoomse Veld 29,350

Dataset L (Districts with relatively lower income):

3 Oud-West and De Baarsjes 28,650

6 Amsterdam-Noord 28,050 2 Westerpark 26,950 7 Geuzenveld - Slotermeer 26,450 10 Zuidoost 26,200 5 Bos en Lommer 25,650 3.3 Descriptive statistics

Table 3 provides descriptive statistics of the variables. It shows that the median house price of all districts between 2005 and 2016 is spread from €1,703 to €5,558. The mean median house price per square meter of the districts with relatively higher income is higher than in districts with relatively lower income. Airbnb activity varies from zero to 3.55 for all districts between 2005 and 2016. On average, there is more Airbnb activity in dataset H than in dataset L.

(15)

Figure 2 visualizes the overall increase in house prices in Amsterdam between 2005 and 2016. The linear trend lines illustrate the rise in house prices since the introduction of Airbnb in Amsterdam. The median house price per square meter increased by 26.69 % since the short-term rental company started in the Netherlands until 2016. However, from 2011 until 2013, the

Table 3: Descriptive statistics of the original dataset, dataset H (Districts with relatively

higher income) and dataset L (districts with relatively lower income)

Variables Mean Median Standard Deviation

Min Max Observations

House price (2005 – 2016): overall for: original dataset dataset H dataset L 2,980.566 3,168.958 2,792.174 2,936.500 3,223.500 2,468.00 825.929 800.682 812.926 1,703.000 1,950.000 1,703.000 5,558.000 5,558.000 4,932.000 N = 144 (n = 12 T = 12) N = 72 (n = 6 T = 12) N = 72 (n = 6 T = 12) Airbnb activity (2005 – 2016): overall for: original dataset dataset H dataset L 0.165 0.210 0.121 0.001 0.002 0.000 0.461 0.569 0.317 0.000 0.000 0.000 3.550 3.550 2.045 N = 144 (n = 12 T = 12) N = 72 (n = 6 T = 12) N = 72 (n = 6 T = 12) Population (2005 – 2016): overall for: original dataset dataset H dataset L 64,831.720 71,845.540 57,817.890 62,759.500 62,759.500 54,995.500 28,840.670 32,473.510 22,825.010 30,078.000 30,605.000 30,078.000 143,258.000 142,258.000 92,917.000 N = 144 (n = 12 T = 12) N = 72 (n = 6 T = 12) N = 72 (n = 6 T = 12) Income (2005 – 2014): overall for: original dataset dataset H dataset L 29,645.130 32,756.930 26,533.330 28,900.000 33,283.462 26,500.000 4,030.199 2,982.067 2,042.736 21,400.000 25,400.000 21,400.000 39,400.000 39,400.000 30,003.330 N = 120 (n = 12 T = 10) N = 60 (n = 6 T = 10) N = 60 (n = 6 T = 10) Housing supply (2011 – 2016): overall for: original dataset dataset H dataset L 34,449.830 39,766.970 29,132.690 30,923.000 30,923.000 29,471.000 17,242.290 20,750.670 10,692.870 15,017.000 15,017.000 15,330.000 79,395.000 79,395.000 41,878.000 N = 72 (n = 12 T = 6) N = 36 (n = 6 T = 6) N = 36 (n = 6 T = 6) Mortgage rate (2005 – 2016): overall for: original dataset dataset H dataset L 0.041 0.041 0.041 0.0433 0.0433 0.0433 0.008 0.008 0.008 0.026 0.026 0.026 0.053 0.053 0.053 N = 144 (n = 12 T = 12) N = 72 (n = 6 T = 12) N = 72 (n = 6 T = 12)

(16)

Figure 2: Median house prices total Amsterdam

In total, 238,353 Airbnb reviews were written in Amsterdam during the observed period between 2005–2016. Figure 3 shows the growth of popularity of Airbnb in Amsterdam between 2009–2016 by means of the reviews, which at least duplicated with every year. Over 340 reviews were composed per day in Amsterdam in 2016. Table 4 shows the mean daily Airbnb reviews per year for every district. It shows that the Centrum, Oud-West and De Baarsjes, and Zuideramstel districts, which are the most central, have the highest Airbnb activity. This is explained by the fact that tourists want to stay close to the Centrum whilst visiting Amsterdam.

Figure 3: Number of Airbnb reviews written in Amsterdam

0 20000 40000 60000 80000 100000 120000 140000 2009 2010 2011 2012 2013 2014 2015 2016

(17)

Table 4: Mean daily Airbnb Reviews per year

District Number and Name 2009 2010 2011 2012 2013 2014 2015 2016

1 Centrum - 0.07 1.03 3.85 11.38 24.86 52.44 97.25 2 Westerpark - 0.04 0.27 0.68 2.37 5.91 13.21 24.22 3 Oud-West and De Baarsjes - 0.00 0.40 2.06 5.81 14.37 31.69 56.02 4 Zeeburg 0.03 0.08 0.42 0.62 1.65 4.32 9.85 19.17 5 Bos en Lommer - 0.04 0.39 0.72 1.51 3.78 8.25 16.35 6 Amsterdam-Noord - - 0.00 0.08 1.23 4.33 10.66 20.72 7 Geuzenveld - Slotermeer - - 0.02 0.12 0.25 0.54 1.18 3.72 8 Osdorp - - 0.02 0.02 0.40 0.92 3.14 7.22 9 Slotervaart and Overtoomse Veld - - 0.23 0.22 0.42 1.75 3.14 6.23 10 Zuidoost - - - 0.01 0.01 0.34 1.74 5.99 11 Oost and Watergraafsmeer - - 0.07 0.34 1.71 4.56 13.46 25.39 12 Zuideramstel - 0.01 0.47 1.58 5.36 15.36 32.90 61.92 Total 0.03 0.25 3.32 10.30 32.11 81.03 181.65 344.19

Table 9 in the appendix shows that in each of the twelve districts, the population, income, and housing supply increased since the launch of Airbnb.

4. Results

Table 5 shows the results of the regressions run in order to test the first hypothesis. It shows that all estimated coefficients of Airbnb activity have a positive sign and are significant. The coefficient of the quadratic Airbnb activity variable is negative, significant, and smaller than the coefficient of Airbnb activity for every regression. This confirms hypothesis 1, given that a non-linear relationship between house prices and Airbnb activity is indicated. In column (5), where population, mortgage rate, and fixed effects were added to the regression, and robust

(18)

standard errors used, the Airbnb activity coefficient is 0.186, with a significance level of 1%, and the coefficient of Airbnb activity2_{is -0.04 with at a significance level of 5%.}

This indicates, as shown through equation (2) and Figure 4, that the first ten thousand reviews of any given year in a district are expected to increase the median house price per square meter by 18.6% on average. However, if 23,250 reviews were conducted in that year already, a unit increase in Airbnb activity has, on average, no effect on house prices. At any higher Airbnb activity level within 12 months, an increase in reviews is expected to decrease house prices.

𝑑𝐻𝑜𝑢𝑠𝑒𝑃𝑟𝑖𝑐𝑒_𝑖𝑡

𝑑𝐴𝑖𝑟𝑏𝑛𝑏𝐴𝑐𝑡𝑖𝑣𝑖𝑡𝑦_𝑖𝑡 = 𝛽1− 2 × 𝛽2 × 𝐴𝑖𝑟𝑏𝑛𝑏𝐴𝑐𝑡𝑖𝑣𝑖𝑡𝑦𝑖𝑡

(2)

Figure 4: Expected percentage change of house prices by a unit increase in Airbnb activity

In column (6) - (8) of Table 5 the number of observations is only 48, but all control variables are added to the regression. Here, the negative coefficient of Airbnb activity2_{is substantially}

higher, whilst the coefficient estimate of Airbnb activity is only slightly higher. This could be caused by the unexpected significant positive estimate of the mortgage rate coefficient between 2011 and 2014, and the positive coefficients of housing supply and income. It indicates that the effect of Airbnb activity, on average, turns negative at a lower 12 month Airbnb activity level already, which is approximately 7,157 reviews. However, the overall R2_{is only 0.067.}

Without the two control variables, but with 144 observations, overall R2_{is 0.218.} -0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3 3.3 3.6 3.9 4.2 4.5 4.8 ex pect ed % ef fect o n h ous e pri ces by a o ne uni t i ncr eas e in Ai rbnb a ct ivi ty

(19)

Table 6 displays the results for the districts with a relatively higher income level. Its estimates of the coefficient of Airbnb activity are positive in every regression, and negative for the quadratic form. Nearly all Airbnb activity coefficients are significant. Table 7 shows the regression results based on the dataset of the districts with relatively lower income. The coefficient outcomes for Airbnb activity are positive and significant, except for the last two regressions, which reveal negative and insignificant coefficients.

Comparing the results of column (5) in Table 6 and 7 the first unit increase in Airbnb activity in 12 months is, on average, associated with an increase of 33.9% (significant at 5% level) in house prices in districts with relatively lower income, whilst in districts with relatively higher income a 10.9% price increase is indicated (significant at 10% level). Therefore, houses in areas with a lower income level have a 23 percentage-point higher expected increase in price. However, the same regression also shows that the quadratic Airbnb activity coefficient in Table 6 is negative, and five times higher than the one in Table 7 (both are significant at 10% level). This is visualized in Figure 5. Hypothesis 2 is confirmed to a certain extent, given that the positive income of Airbnb activity is initially stronger in districts with relatively lower income. Though, according to this result, the positive effect of Airbnb activity on house prices also diminishes more significantly in districts with a relatively lower income level, and becomes negative at a lower level of Airbnb activity than in districts with relatively higher income. This could be caused by omitted variable bias.

The results for areas with less income become insignificant when all key variables are added to the regression. In areas with a higher income, the regression revealed significant results for both Airbnb activity variables. Here, the marginal effect of Airbnb activity on house prices has a turn from positive to negative impact at 6,307 annual reviews, whilst, as indicated in Figure 5, without the income and housing variable, this is expected to happen if 25,952 annual reviews are exceeded.

(20)

Figure 5: Expected percentage change of house prices by a unit increase in Airbnb activity

for areas with relatively higher and relatively lower income

The results of the second analysis differ compared to the first analysis. In the regressions that do not include housing supply and income, the estimated significant coefficients for Airbnb activity are, compared to the first analysis, lower when dataset H is used and higher when dataset L is used. This supports hypothesis 2. However, when the two additional control variables and fixed effects are added to the regression the estimates for Airbnb activity using dataset H are significant and higher than the baseline results. As mentioned above, under these settings the Airbnb activity coefficient estimates become insignificant when the data of the districts with relatively lower income is used. The change in results may be caused by the data shortage of housing supply and income. Thus, hypothesis 2 is not fully confirmed.

-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3 3.3 3.6 3.9 4.2 4.5 4.8 ex pec ted % ef fec t on hous e pri ce s by a o ne uni t incr eas e i n A irbnb a ct ivi ty

initial state of Airbnb activity

Lower Income Higher Income

(21)

Table 5: Regression conducted to test hypothesis 1 Ln(House Price) 2005- 2016 Ln(House Price) 2005- 2016 Ln(House Price) 2005- 2016 Ln(House Price) 2005- 2016 Ln(House Price) 2005- 2016 Ln(House Price) 2011- 2014 Ln(House Price) 2011- 2014 Ln(House Price) 2011- 2014 Airbnb activity 0186*** (0.031) 0.186*** (0.053) 0.428*** (0.052) 0.186*** (0.031) 0.186*** (0.054) 0.429*** (0.159) 0.219** (0.082) 0.219** (0.072) Airbnb activity2 -0.040*** (0.009) -0.040** (0.013) -0.089*** (0.017) -0.040*** (0.009) -0.040** (0.013) -0.317* (0.194) -0.153* (0.076) -0.153** (0.058) Ln(Population) 0.092 (0.072) 0.021 (0.100) 0.021 (0.171) -1.787*** (0.194) 0.480 (0.404) 0.480 (0.468) Mortgage rate 6.024*** (1.284) -21.904*** (2.377) -21.904*** (6.031) 9.306*** (2.225) 11.422*** (2.288) 11.422*** (2.313) Ln(Housing supply) 1.575*** (0.180) 0.145 (0.350) 0.145 (0.434) Ln(Income) 0.863*** (0.206) 0.580 (0.383) 0.580 (0.385) Constant 7.807*** (0.012) 7.807*** (0.022) 6.651*** (0.804) 8.402*** (1.142) 8.402*** (2.006) 2.010 (1.995) -5.295 (5.208) -5.295 (7.163) Fixed Effects

Yes Yes No Yes Yes No Yes Yes

Time Fixed Effects

Robust No Yes No No Yes No No Yes

Groups 12 12 12 12 12 12 12 12 Observations 144 144 144 144 144 48 48 48 𝑹𝟐 overall between within 0.216 0.716 0.854 0.216 0.716 0.854 0.267 0.263 0.532 0.218 0.410 0.854 0.218 0.410 0.854 0.926 0.940 0.383 0.067 0.060 0.902 0.067 0.060 0.902

The numbers horizontal to the variables without parentheses are the coefficients. The standard errors of the coefficients are noted in brackets below them.

(22)

Table 6: Regression conducted for hypothesis 2, dataset H (Districts with relatively higher income) Ln(House Price) 2005- 2016 Ln(House Price) 2005- 2016 Ln(House Price) 2005- 2016 Ln(House Price) 2005- 2016 Ln(House Price) 2005- 2016 Ln(House Price) 2011- 2014 Ln(House Price) 2011- 2014 Ln(House Price) 2011-2014 Airbnb activity 0.116*** (0.032) 0.116** (0.041) 0.312*** (0.058) 0.109*** (0.032) 0.109* (0.048) 0.260 (0.266) 0.251** (0.110) 0.251*** (0.055) Airbnb activity2 -0.022** (0.009) -0.022* (0.009) -0.055*** (0.017) -0.021** (0.009) -0.021* (0.010) -0.186 (0.261) -0.199* (0.092) -0.199*** (0.045) Ln(Population) 0.299*** (0.097) -0.117 (0.099) -0.117 (0.137) -2.181*** (0.167) -0.384 (0.839) -0.384 (0.644) Mortgage rate 6.364*** (1.641) -26.655*** (3.031) -26.655** (7.596) 14.695*** (4.143) 6.448 (4.026) 6.448* (2.942) Ln(Housing supply) 1.793*** (0.143) 0.320 (0.856) 0.320 (0.682) Ln(Income) 2.065*** (0.275) 0.223 (0.616) 0.223 (0.370) Constant 7.873*** (0.014) 7.873*** (0.023) 4.407*** (1.100) 10.158*** (1.164) 10.158*** (1.749) -8.666** (2.551) 6.352 (8.430) 6.352 (6.962) Fixed Effects

(23)

Table 7: Regression conducted for hypothesis 2, dataset L (Districts with relatively lower income) Ln(House Price) 2005- 2016 Ln(House Price) 2005- 2016 Ln(House Price) 2005- 2016 Ln(House Price) 2005- 2016 Ln(House Price) 2005- 2016 Ln(House Price) 2011- 2014 Ln(House Price) 2011- 2014 Ln(House Price) 2011- 2014 Airbnb activity 0.378*** (0.080) 0.378** (0.116) 0.755*** (0.114) 0.339*** (0.072) 0.339** (0.124) 0.847* (0.514) -0.025 (0.166) -0.025 (0.085) Airbnb activity2 -0.118*** (0.037) -0.118* (0.049) -0.236*** (0.059) -0.105*** (0.034) -0.105* (0.049) -0.543 (0.835) 0.169 (0.232) 0.169 (0.095) Ln(Population) -0.249*** (0.093) 1.462*** (0.396) 1.462* (0.668) -1.416*** (0.155) 1.239** (0.554) 1.239** (0.407) Mortgage rate 7.607*** (1.027) -7.962* (3.978) -7.962 (6.534) 18.857*** (4.176) 15.073*** (2.297) 15.073*** (1.245) Ln(Housing supply) 1.131*** (0.175) -0.039 (0.364) -0.039 (0.281) Ln(Income) 2.001*** (0.523) 1.206** (0.496) 1.206*** (0.216) Constant 7.742*** (0.020) 7.742*** (0.041) 10.231*** (0.346) -7.833* (4.388) -7.833* (7.302) -9.527* (5.117) -18.172** (6.506) -18.172*** (2.677) Fixed Effects

(24)

5. Limitations

This research is limited for the following reasons: It is based on its definition of Airbnb activity, which assumes that the rate of reviews written per customer is the same for every district. The duration of the stays is not taken into account, and neither is the number of guests per review. In addition, even though location and time fixed effects were added to the regression, there can of course still be other shifts in variables that affect the house prices in a certain area. Finally, the limited number of observations of the annual income and housing supply control variable is a severe limitation of this study. One may consider increasing the number of observations by narrowing the district size.

6. Conclusion and discussion

This study was conducted in order to extent the research of the effect of Airbnb on house prices. Two hypotheses, one claiming a non-linear relationship and one stating that the positive income of Airbnb on house prices decreases with the income level of an area, were tested through correlational panel data analyses.

Consistent with prior research (Van der Bijl, 2016; Sheppard & Udell, 2016), the results of the first analysis suggest an initial positive impact of Airbnb activity on house prices. However, this research additionally found an indication for a non-linear effect of Airbnb on property values. The added quadratic Airbnb activity variable shows that the expected positive impact of Airbnb on house prices decreases with every marginal unit of Airbnb activity, and eventually becomes negative. This is in line with the prediction that the positive externalities of Airbnb first raise house prices, but are outweighed by the negative externalities when the frequency of the Airbnb services increases. This research does not reveal whether the associated non-linear relationship of Airbnb and house prices is caused by a marginal decrease in the utility of the extra income source, or whether the perception of the negative externalities increases. However, the findings that the satisfaction received through income declines with income (Layard, Mayraz, & Nickell, 2008) and that Airbnb activity is positively related to nuisance (Van der Bijl, 2016) suggests that both are significant causes. Additional research is warranted to find out whether the result is externally valid.

The results addressing hypothesis 2, which stated that the positive impact of Airbnb on housing prices decreases with the income level of an area, are less clear to interpret. Both analyses, the one with districts with relatively lower income and the one with relatively higher

(25)

income, indicate a non-linear effect of Airbnb on house prices. Once again, for both groups the results show that Airbnb is first associated with a house price increase, but after a turning point is reached, the effect becomes negative. All directly comparable significant outcomes for Airbnb activity, resulting from the samples of districts with relatively lower income and relatively higher income, suggested that, in areas with a lower income level, the initial effect of Airbnb on house prices is stronger than in areas with a higher level. This supports hypothesis 2.

However, not all coefficients were significant. Running the regression with all control variables and fixed effects, and using the sample of districts with relatively lower income revealed insignificant estimates of Airbnb activity. They indicate that Airbnb first lowers and then increases house prices. This may be caused by the limited number of observations due to the data shortage and the fall of median house prices per square meter in Amsterdam between 2011 and 2013.

It is surprising that the significant coefficients resulting from the second analysis indicate that the negative effect of Airbnb activity on property prices is stronger in less-wealthy areas. This would imply that, at first, the positive externalities of Airbnb, one of which is the additional income, are more valued in districts with relatively less income, but that with the increasing popularity of Airbnb, these districts are also stronger affected by the negative impacts than in relatively wealthier locations. In addition, it is surprising that the indicated amount of Airbnb activity, at which the effect on house prices are expected to turn negative, is lower for districts with less income compared to wealthier districts. The unexpected results are not logically explainable, based on the existing literature about the marginal utility of income (Frey & Stutzer as cited in Easterlin, 2005; Layard, Mayraz, & Nickell, 2008). It seems reasonable to assume that the inconsistency of the findings is caused by the low number of observations and omitted variable bias.

Thus, the findings cannot fully confirm hypothesis 2, but indicate that with less limited research it could be validated. Nevertheless, the study shows that the positive impact of Airbnb activity on house prices is dependent on an area’s income level. Further research in the field, with fewer limitations, is needed to get clear results. Additionally, it would be interesting to study how the negative impact of Airbnb on house prices depends on the income level of an area.

(26)

7. Policy implications

The finding of the non-linear effect of Airbnb on house prices of this study is valuable in practice. It shows that with the increasing popularity of Airbnb, property owners may possibly also be disadvantaged (besides the population without the possibility to rent out their homes), as their assets lose value at a certain level of Airbnb activity. This could indicate that the rise of the popularity of Airbnb may decrease autonomously in the future. Homeowners may reduce, or even stop, offering their accommodation to short-term tenants in order to prevent their property value from declining, or because they do not get any utility from it anymore. Furthermore, more landlords may forbid their long-term renters from offering their accommodation or a room on Airbnb.

The city of Amsterdam could take the finding about the non-linearity of the relationship of Airbnb and house prices into account and adapt it to their regulations for Airbnb. Currently, the number of nights that a listing can be offered per year is restricted to 60. Considering that districts have 15,000 to 80,000 housing units, which (if not privately restricted) could be offered as accommodation on Airbnb, the turning point of maximal 2.325 Airbnb activities per year (23,250 reviews) may be exceeded in many districts in the future. This applies especially, if further, less-limited research indicates that this turning point is at a lower Airbnb activity level.

In 2016, only the Centrum district of Amsterdam exceeded the estimated critical Airbnb level of 23,250 reviews per year. However, this may change if the rise in Airbnb’s popularity continues. Table 10 in the appendix shows a forecast of the median growth rate per district for 2017 and 2018, which is estimated through the method of least squares. Table 11 (Appendix) displays the predicted number of reviews written per district depending on estimated growth rates. The estimation indicates that in 2017 Airbnb activity has negative effects on house prices in five districts of Amsterdam already.

The social cost that would be accompanied by an Airbnb activity level over the turning point should be prevented. Therefore, the City of Amsterdam should consider limiting the number of short-term visitors per year within its districts. If this is not legally feasible, they could obligate the Airbnb hosts to register at the applying district council to have more control over hosts and short-term stays.

(27)

8. Bibliography

Airbnb (2017), ‘About Us’, Accessed via https://www.airbnb.com/about/about-us. URL: https://www.airbnb.com/about/about-us

Airbnb continues to grow in Amsterdam, hits 1.7 million nights in 2016. (2017, May 2).

DutchNews.nl. Retrieved from http://www.dutchnews.nl

Amsterdam, Airbnb agree new deal to stop illegal rents. (2016, December 1). DutchNews.nl. Retrieved from http://www.dutchnews.nl

Black, A., Fraser, P., & Hoesli, M. (2006). House prices, fundamentals and bubbles. Journal

of Business Finance & Accounting, 33(9‐10), 1535-1555.

Bourassa, S. C., Cantoni, E., & Hoesli, M. (2007). Spatial dependence, housing submarkets, and house price prediction. The Journal of Real Estate Finance and Economics, 35(2), 143-160.

Brown, G. R. & Matysiak, G. A. (2000). REAL ESTATE INVESTMENT: A capital market

approach. Harlow: Financial Times Prentice Hall

Can, A. (1992). Specification and estimation of hedonic housing price models. Regional

science and urban economics, 22(3), 453-474.

Can, A., & Megbolugbe, I. (1997). Spatial dependence and house price index

construction. The Journal of Real Estate Finance and Economics, 14(1), 203-222.

Capozza, D. R., Hendershott, P. H., Mack, C., & Mayer, C. J. (2002). Determinants of real

house price dynamics (NBER Working Paper No. 9262). Retrieved from National

Bureau of Economic Research website: http://www.nber.org/papers/w9262

Case, K. E., & Shiller, R. J. (2003). Is there a bubble in the housing market? Brookings

(28)

Chen, N. K., & Cheng, H. L. (2017). House price to income ratio and fundamentals: Evidence on long‐horizon forecastability. Pacific Economic Review, 22(3), 293-311.

Clayton, J., Ling, D. & Naranjo, A. (2009). Commercial Real Estate Valuation:

Fundamentals Versus Investor Sentiment. The Journal of Real Estate Finance and

Economics, 38(1), 5-37.

Easterlin, R. A. (2005). Diminishing marginal utility of income? Caveat emptor. Social

Indicators Research, 70(3), 243-255.

Égert, B., & Mihaljek, D. (2007). Determinants of house prices in central and eastern Europe. Comparative economic studies, 49(3), 367-388.

Einav, L., Farronato, C., & Levin, J. (2016). Peer-to-peer markets. Annual Review of

Economics, 8, 615-635.

Haffner, M., & De Vries, P. (2009, February). Dutch house price fundamentals. In Australian

Tax Research Foundation Housing and Tax Symposium (Vol. 11).

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

Layard, R., Mayraz, G., & Nickell, S. (2008). The marginal utility of income. Journal of Public

Economics, 92(8), 1846-1857.

Muellbauer, J. & Murphy, A. (2008). Housing markets and the economy: the assessment.

Oxford Review of Economic Policy, 24(1), 1-33.

New York deflates Airbnb. (2016, October 27). The Economist. Retrieved from https://www.economist.com

(29)

Sheppard, S. & Udell A. (2016). Do Airbnb properties affect house prices? (Working Paper No. 3). Retrieved from Williams College Department of Economics website: http://web.williams.edu/Economics/wp/SheppardUdellAirbnbAffectHousePrices.pdf

Three-quarters of Amsterdam Airbnb rentals ‘break council rules’. (2016, March 21).

DutchNews.nl. Retrieved from http://www.dutchnews.nl

Van der Bijl, V. (2016). The effect of Airbnb on house prices in Amsterdam A study of the

side effects of a disruptive start-up in the new sharing economy. (Unpublished master’s

(30)

9. Appendix

Table 8: Data sources, description and use

Source Description & Use

NVM (Nederlandse Vereniging van Makelaars en Taxateurs in onroerende goederen)

The NVM gathers property sales data from Amsterdam and supplied the median sales price and median price per square meter of 15 districts in Amsterdam between 2005– 2016.

InsideAirbnb Murray Cox, head of InsideAirbnb, supplies

data of Airbnb services such as number of listings, pricing, a record of all past reviews and their corresponding listings and location of each listing through crawls conducted on the official Airbnb website (Data set used in this thesis is from April 2017).

OIS Amsterdam (Onderzoek, Informatie en Statistiek)

The OIS reports annually relevant

information about Amsterdam, separated in 22 districts, as population, average

household income and housing supply. Population is measured by the number of registered inhabitants of district i at the 1st

of January. Annually and district specific available between 2005–2016.

Average household income is estimated by the average gross household income minus insurance and tax costs. Annually annually and district specific available between 2005–2013.

Housing supply is estimated by the amount of registered addressed at the 1st_{of January.}

(31)

Annually and district specific available between 2011–2016.

De Nederlandse Bank De Nederlandse Bank provided the Dutch mortgage rate between 2005–2016.

Table 9: percentage increase in population, housing supply and income

Districts Population 2016 Population increase since 2009 in % Housing Supply 2016 Housing Supply increase since 2011 in % Income 2014 Income increase since 2009 in % Centrum 86,499 6.4% 54,217 2.4% 36,700 8.3% Westerpark 36,116 5.8% 21,330 0.6% 30,000 11.5% Oud-West and De Baarsjes 39,424 1.7% 72,783 11.3% 30,800 6.9% Zeeburg 64,108 21.7% 29,015 3.7% 38,068 12.0% Bos en Lommer 35,065 15.2% 17,457 13.9% 28,000 9.4% Amsterdam-Noord 92,917 7.2% 41,878 4.1% 30,200 8.2% Geuzenveld - Slotermeer 44,740 9.8% 19,013 1.7% 27,500 1.5% Osdorp 38,560 18.5% 17,722 3.4% 28,800 2.1% Slotervaart and Overtoomse Veld 37,632 19.4% 16,806 10,8% 31,000 5.1% Zuidoost 86,057 6.9% 39,828 5.9% 27,100 2.3% Oost and Watergraafsmeer 68,313 14.3% 36,633 10.6% 36,900 8.5% Zuideramstel 143,258 8.4% 79,395 2.6% 39,400 12.6%

(32)

Table 10: Median growth rate of Airbnb activity per district

2013 2014 2015 2016 2017 2018

2,955 2,524 2,234 1,944 1,583* 1,293*

*indicates that the growth rate was forecasted through Excel, using the method of least squares

Table 11: Two year forecast of number of reviews per district

2016 2017 2018

Centrum 35496 56196 72679

Westerpark 8839 13993 18098

Oud-West and De Baarsjes 20449 32374 41870

Zeeburg 6996 11076 14324

Bos en Lommer 5968 9448 12220

Amsterdam-Noord 7562 11972 15483

Geuzenveld - Slotermeer 1359 2152 2783

Osdorp 2637 4175 5399

Slotervaart and Overtoomse Veld 2273 3599 4654

Zuidoost 2186 3461 4476

Oost and Watergraafsmeer 9266 14670 18972

Zuideramstel 22599 35778 46272

Bolt numbers signal an exceedance of the estimated turning point (23,250 reviews) that indicates a negative effect of Airbnb activity on house prices

The non-linear effect of Airbnb activity on house prices and the role of an area's income level in the relationship