• No results found

The relationship between Airbnb and non-business

N/A
N/A
Protected

Academic year: 2021

Share "The relationship between Airbnb and non-business"

Copied!
29
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Faculty of Economics and Business (FEB)

Bachelor thesis Economics and Finance

The relationship between Airbnb and non-business

overnight stays in the Netherlands

By Jens van de Pol, 10579060

Supervisor: Dr. Andras Kiss

January 29

th

, 2017 Amsterdam

Abstract

This thesis analyses if Airbnb attracts additional non-business overnight stays in the Netherlands and its provinces, which are grouped together in a low and high Airbnb listings group. It does so by using a DD regression technique. The main findings are that there is no evidence to support that Airbnb attracts additional non-business overnight stays for the Netherlands and the low group. But for the high group Airbnb does attract additional non-business overnight stays, its effect is however not big enough to compensate for the low group and make it significant for the Netherlands overall.

(2)

Verklaring eigen werk

Hierbij verklaar ik, Jens van de Pol, dat ik deze scriptie zelf geschreven heb

en dat ik de volledige verantwoordelijkheid op me neem voor de inhoud

ervan.

Ik bevestig dat de tekst en het werk dat in deze scriptie gepresenteerd wordt

origineel is en dat ik geen gebruik heb gemaakt van andere bronnen dan die

welke in de tekst en in de referenties worden genoemd.

De Faculteit Economie en Bedrijfskunde is alleen verantwoordelijk voor de

begeleiding tot het inleveren van de scriptie, niet voor de inhoud.

(3)

Tables of contents

I. Introduction ………. 3

II. Literature review ………...…. 4

III. Methodology ……….... 5 A Data ………... 5 B Estimation technique ………... 6 C Descriptive statistics ………. 8 IV. Results ……….... 10 A The Netherlands ……….. 10 B The provinces ……….. 10 C Robustness check ……… 14 V. Conclusion ………. 15 References ……… 17 Appendices ……….. 18

A.

The Netherlands ……….. 18

B.

The provinces ……….. 20

C.

Robustness check ……… 23

(4)

I. Introduction

Airbnb has shaken up the traditional tourism accommodation by providing a marketplace platform that facilitates rental accommodation between individuals, also known as a peer-to-peer market (P2P). Over the past few years Airbnb has experienced explosive growth (Guttentag, 2015). Airbnb had in late 2014 over one million listings on their site and in the summer of 2014 over 375,000 guests per night were accommodated by their hosts. In the beginning of 2015 Airbnb was valued at $20 billion, which is more than most established hotel brands (Fraiberger & Sundararajan, 2015).

Molenaar & de Groot state that the OIS (the municipal research agency) expect the amount of hotel overnight stays in Amsterdam to be above 14 million in 2016, which is a 9-percent increase to 2015. It also states that Amsterdam receives 4 million visitors who will spend the night via Airbnb and day visitors. This combined amount is expected to be 23 million in 2025 (2016). While there are complains, critique and claims on Airbnb by hotel organizations, for example in France (de Jong, 2016), according to the OIS the amount of hotel stays in Amsterdam will be bigger than the year before. So it seems like both the hotel overnight stays as well as the stays facilitated by Airbnb are increasing. Which would mean that Airbnb attracts additional customers instead of just taking them from the hotels.

The aim of this thesis is to evaluate if Airbnb attracts additional non-business overnight stays in the Netherlands. So the research question is; Does Airbnb attracts additional non-business

overnights stays in the Netherlands? In this process I will analyse the Netherlands and a group of provinces with a large number of Airbnb listings and a group with a smaller number of listings. To investigate this effect, the number of total overnight stays and business stays are collected and used to calculate non-business overnight stays. All data used comes from the Central Bureau for Statistic (CBS). The CBS collects, edits and publishes statistical data about the Netherlands and is controlled by the Dutch government. To investigate the relationship between Airbnb and non-business overnight stays, I use a regression model with overnight stays as dependent variable, a dummy variable for the time Airbnb entered the market as independent variable and months and dummy variables for non-business stays and control variables for low and high season months.

An answer to this research question, positive, neutral or negative is of great relevance for the economy of countries it may concern. Choi et al. (2015) say that when people plan a trip, they will consider many things like transportation, food and entertainments. So if Airbnb attracts additional non-business overnight stays those sectors could be significantly influenced and generate more income for countries and businesses.

The next section is the literature review, here research on Airbnb and economic effects in recent years will be discussed. In the section thereafter the methodology, the data, regression models and statistic descriptive will be discussed. Then the section with results will come, that consists of

(5)

presenting, evaluating and a robustness check of the results. And finally in the section conclusion the conclusion of the research will be presented and recommendations for further research will be made.

II. Literature review

Because there has not been done any research like the one in this thesis before, this section presents related studies about Airbnb. In this section I will discus 5 studies, the first 4 are very similar, this is due to their difference-in-differences (DD) research method which they have based on the first study. The 4 studies all have Airbnb supply as independent variable and hotel revenues as dependent variable, only 2 studies have hotel revenue per room and the other 2 have just hotel revenue. The first study estimated the impact of Airbnb on the hotel room revenue in Texas. The second study researches the impact of Airbnb on hotel revenue in the Nordics. The third one researches the relationship

between hotel revenue and Airbnb listings in South Korea. The fourth one studies the impact of Airbnb on the hotel revenue in the Netherlands. And the last one focuses on the effect of Airbnb on the hotel occupancy rates in San Francisco and Chicago.

Zervas, Proserpio & Byers (2014) use, as mentioned above, a DD empirical strategy to identify the Airbnb impact by comparing differences in revenue for hotels in cities affected by Airbnb before and after the entry of Airbnb against the difference in revenue for hotels in unaffected cities over the same period. In their regression the most important variables they use are monthly hotel room revenue and Airbnb supply. They find a negative coefficient (-0.039). Zervas et al. (2014) state that a 1-percent increase in Airbnb supply leads to a 0.05-percent decrease in monthly hotel revenue in Texas. Further they state that Airbnb is a viable, but an imperfect, alternative for traditional types of accommodations.

Neeser (2015) did research in Norway, Finland and Sweden. In his DD technique Neeser uses average revenue per hotel room and the total number of Airbnb listings as his main variables. In his research he finds no significant effect. In his conclusion Neeser (2015) states that Airbnb doesn’t have a significant effect on hotel revenue per available room, but Airbnb did contribute to a significant reduction in the average price of hotel rooms in the Nordics.

Choi et al. (2015) investigated the effect of Airbnb in South Korea. For this they use panel data of Seoul, Busan and Jeju which are the cities with the largest hotel revenue, they are selected to represent South Korea. Further the hotels are classified in 5 types; luxury, upscale, midscale, economy and budget. Choi et al. use hotel revenue and Airbnb listings as their main variables. The result of their regression model concludes that there are no significant coefficients for any of the 5 classes in South Korea. For the cities individually Seoul has only a significant coefficient for the budget class (-0.10), Busan shows significant coefficients for upscale and midscale classes (-0.07 and 0.09) and Jeju has no significant coefficients. Due to too small and meaningless, in economic sense, coefficients Choi et al. (2016) conclude that Airbnb listing’s has no effect on the hotel revenue in South Korea.

(6)

Hooijer (2016) investigate the impact of Airbnb in the Netherlands between 2005 until the second quarter of 2015. In her research Hooijer runs 3 regression models, with in each regression the main variables hotel revenue and number of Airbnb listings. The first regression is only with the independent and dependent variables, where she finds a significant positive effect (0.0125). In the second regression the control variable is added; here Hooijer finds again a significant positive effect (0.0166). And in the last regression with 2 control variables she discovers a significant negative effect (-0.0312). Hooijer (2016) bases her conclusion on the last regression and states that Airbnb listings do have a negative impact on the hotel revenue, 1-percent change in the number of Airbnb listings results in a -0.0312-percent change of hotel revenue in the Netherlands.

Goree (2016) investigate the effect of Airbnb on hotel occupancy rates in San Francisco and Chicago, She chose these cities because Airbnb was founded in San Francisco and because Chicago is a major U.S. city where there was less scepticism against Airbnb than in New York. The research is conducted on reviewing existing literature about the effects of Airbnb on the hotel industry and conducting an empirical analysis of hotel occupancy and economic data. In Goree’s regressions the hotel occupancy rate in San Francisco and Chicago are the dependent variables, GDP of the U.S., a dummy variable indication post and after-recession and an interaction between GDP and the dummy are the independent variables. Her results are that in San Francisco there is no significant effect and in Chicago the coefficient is only significant at a significance level of 10-percent (-24.119). Goree (2016) concludes that from the period 2008 to 2014 in San Francisco Airbnb has no significant effect on the hotel occupancy rates, while in Chicago there is.

Although the setup of the previous studies is relatively similar there are different outcomes. In Texas, the Netherlands, San Francisco and Chicago there is a significant effect concluded, while in South Korea and the Nordics there is no significant effect detectable.

III. Methodology

This section is composed of three subsections. First I will describe how the data used is collected. Secondly I discus the estimation technique used to investigate if Airbnb attracts additional overnight stays. And as third the descriptive statistics are presented.

A. Data

To perform the research in this thesis five datasets are used. All five datasets are from the CBS. The first dataset consists of total monthly overnight stays and monthly business overnight stays in the Netherlands over the period January 2000 until December 2012 (time period 1), this set contains 156 observations of total overnight stays and 72 observations of business overnight stays, so in total of 228 observations. The second dataset provides the same information over the same time period but divided

(7)

over the different provinces of the Netherlands and contains 156 observations of total overnight stays and 72 of business overnight stays per province except for Flevoland that has 150 observations of total overnight stays, so this dataset has a total of 2730 observations. The third dataset provides the same data as the first but over the period January 2012 until September 2016 (time period 2) and contains a total of 114 observations. The fourth dataset provides the same data as the second but over time period 2 and contains a total of 1368 observations. And the last dataset provides quarterly GDP values from January 2000 until September 2016 and contains 67 observations. All datasets are in absolute values times thousand except for the quarterly GDP, which is in absolute value times billion.

The two datasets containing the overnight stay values for the Netherlands and the two for the provinces are merged together to become one dataset providing monthly overnight stays for the Netherlands and the provinces over the time period January 2000 until September 2016.

The drawback of these datasets is that the sets of the two different time periods are based on two different researches, so they deviate from each other. This can be found in the values of the overlapping months in 2012 contained by both datasets. How this issue and others are resolved I will discuss in the subsection descriptive statistics.

B. Estimation techniques

Zervas et al. state that business travellers make use of business-related hotel amenities not typically provided by Airbnb properties (2014). And when you think about it, this seems very logical. Because business travellers sometimes check in late due to arriving with a late flight or delay; in a hotel this is not a problem because there is always someone at the reception to help them while this is not always the case with Airbnb where the hosts sometimes have to come from their home or are already sleeping. There is also the eating facilities in the hotel that make it more attractive to business travellers, like a restaurant which prevents the traveller from having to do groceries and cook when coming back late from work and the breakfast a hotel provides makes it possible for the guest to take a quick breakfast and go to work. Most business hotels also have a room or area available for a meeting. The services hotels provide like room service and daily room cleaning makes them also more

appealing to business travellers. Where with hotels the cleaning is included in the price, with Airbnb you are expected to clean the room/apartment yourself or you have to pay extra for cleaning costs. All this is why I made the assumption that business travellers are not influenced by the entry of Airbnb.

Due to this assumption I use a DD model between business and non-business overnight stays (Angrist & Pischke, 2015) to investigate if Airbnb attracts additional non-business overnight stays in the Netherlands. The DD technique enables me to estimate the different effects of the business

overnight stays and the non-business ones and the effect of Airbnb on the non-business overnight stays after it entered the market. I applied this DD model on the Netherlands and on two groups of

(8)

The Hague and Rotterdam (2015). So by making a high Airbnb listings group and a low one it is possible to evaluate the difference, possibly caused, more specifically. The monthly sum of overnight stays of the provinces Noord-Holland, Zuid-Holland and Utrecht are grouped together in the high Airbnb listings group and the remaining provinces are grouped together in the low Airbnb listings group.

The DD model I use takes the following form:

𝑆𝑡𝑗 = 𝛼0+ 𝛾𝑚𝑡+ 𝛿𝐺𝑗+ 𝜗𝑚𝑡𝐺𝑗+ 𝛽𝑚𝑡𝐸𝑡𝐺𝑗+ 𝜂𝑡𝑗+ 𝜀𝑡𝑗 (1) Graph 1 shows the essence of this model. The dependent variable is the number of overnight stays in month t of group j. The dependent variable is dependent of a constant, month t which is 0 at January 2008 the time Airbnb became active in the Netherlands (Airdna, 2017), Group j which is a dummy variable with value 0 for business stays and 1 for non-business stays, E as another dummy variable for when Airbnb became active with value 0 if t<0 and 1 if t≥0, a combination between month and E and a combination of the month, group and E which provides the critical coefficient 𝛽 on which this paper makes its conclusion. I make my conclusion based on this coefficient because it represents the effect of the non-business overnight stays for the months from the point Airbnb became active in the Dutch market. Further represents 𝜂𝑡𝑗 a group of control dummy variables, which account for monthly changes per group and there is an error term in this model. The control variables take low and high monthly overnight stay changes into account. The dummy variables groups months together in groups A,B,C and D for the business overnight stays and groups months together in groups A,B and C for the non-business overnight stays. Group A represents low season months, D high season months and the other group(s) is/are respectively in between. The months are distributed based on the interval between the smallest and biggest mean and the 25-percent range each group represents of this interval (see Appendix A table 1 and Appendix B table 2 and 3). I run two regressions, one without control variables and one with.

(9)

Based on the enormous growth Airbnb has made both in overnight stays facilitated and in revenue, the small effects found in some of the studies discussed and the expected growth of hotel stays in Amsterdam, it seems impossible that Airbnb growth is made at the expenses of hotels, because of this a positive 𝛽 is expected.

Based on the DD model explained and on what is expected for 𝛽, the hypothesis takes the following form:

𝐻0: 𝛽 = 0 𝐻1: 𝛽 > 0

If 𝛽 does significantly deviate from 0 in a positive way than 𝐻0 gets rejected and based on this research Airbnb does attract additional non-business overnight stays in the Netherlands. For this a critical assumption has to be made that if 𝛽 is significantly deviating from 0 this is due to Airbnb.

C. Descriptive statistics

Before the descriptive statistics will be presented, first I will discuss how the data is

manipulated to make them useable. These data still contain problems that have to be dealt with before it can be regressed.

- Two different datasets

The datasets of the two time periods come from two different researches. To deal with this and to make the two datasets fluent the mean of the 2012 monthly differences for the total and business overnight stays are tested at a 95-percent interval to evaluate if they significantly deviate from 0 (period 2-period 1) (see Appendix A tables 4 and 5). If they do, the sum of the differences are

calculated as a percentage of the 2012 sum of the monthly values of period 2 (because these values are used instead of period 1’s) 𝑠𝑢𝑚 𝑚𝑜𝑛𝑡ℎ𝑙𝑦 𝑑𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑐𝑒𝑠 2012

𝑠𝑢𝑚 𝑚𝑜𝑛𝑡ℎ𝑙𝑦 𝑣𝑎𝑙𝑢𝑒𝑠 2012 𝑜𝑓 𝑝𝑒𝑟𝑖𝑜𝑑 2= 𝜌, 𝜌 is for total 0.094 and business 0.0061. The next step is making the data fluent through the calculation; (1 − 𝜌) ∗ 𝑑𝑎𝑡𝑎 𝑝𝑒𝑟𝑖𝑜𝑑 2. The same process is done with data of the provinces (see Appendix B tables 6 and 7).

- Missing data for the Netherlands

Unfortunately the dataset of period 1 is missing the business overnight stays from January 2000 until December 2006. Without these months of data there is only one year of data available before the entering of Airbnb on the Dutch market, which is not enough for running a reliable regression model, so a regression model has to be used to “predict” the missing data. But because the GDP is only provided in quarterly data the average percentage of the month’s business stays contribution to their

(10)

corresponding quarterly business stays has to be calculated in order to attribute that percentage of the outcome of the prediction model to the corresponding months. These percentages multiplied by the forecast values provide the missing data.

So first the monthly business overnight stays from 2007 until 2016 are divided by their corresponding quarterly value; 𝑚𝑜𝑛𝑡ℎ𝑖 𝑏𝑢𝑠𝑖𝑛𝑒𝑠𝑠 𝑜𝑣𝑒𝑟𝑛𝑖𝑔ℎ𝑡 𝑠𝑡𝑎𝑦

𝑞𝑢𝑎𝑟𝑡𝑒𝑟𝑖 𝑏𝑢𝑠𝑖𝑛𝑒𝑠𝑠 𝑜𝑣𝑒𝑟𝑛𝑖𝑔ℎ𝑡 𝑠𝑡𝑎𝑦= 𝜑𝑖 and of these values the means are taken the of each month (see Appendix A table 8).

Secondly the OLS regression model is estimated with as dependent variable the quarterly business overnight stay and independent variable the GDP of the Netherlands over the time period 2007 until 2016 with t identifying the quarters. This model is:

𝑄𝑢𝑎𝑟𝑡𝑒𝑟𝑙𝑦𝑏𝑢𝑠𝑖𝑛𝑒𝑠𝑠𝑠𝑡𝑎𝑦𝑠𝑡 = 𝛼0+ 𝛽𝐺𝐷𝑃𝑡+ 𝜀𝑡 (3) (see Appendix A table 9). From this regression is the following forecast model made:

𝑄𝐵𝑆̂ = 2537.378 + 4.8284𝐺𝐷𝑃𝑡 𝑡.

Then to predict the missing monthly data the quarterly GDP from the period 2000 until 2006 is put through the model and multiplied by the corresponding 𝜑 mean; 𝑄𝐵𝑆̂ ∗ 𝜑𝑚𝑒𝑎𝑛.

- Missing data for the provinces

The CBS has no GDP for the provinces individually, so to predict the missing data an average percentage of the monthly business overnight stays to the total month overnight stays per provinces has to be calculated over the period 2007 until 2016. And then these values are multiplied to the monthly business overnight stays of the period 2000 until 2006 calculated above.

So first the average contribution in percentage of a provinces to the total business overnight stays per month over the period January 2007 until September 2016 are calculated;

𝑚𝑒𝑎𝑛 (𝑏𝑢𝑠𝑖𝑛𝑒𝑠𝑠 𝑜𝑣𝑒𝑟𝑛𝑖𝑔ℎ𝑡 𝑠𝑡𝑎𝑦𝑠 𝑜𝑓 𝑚𝑜𝑛𝑡ℎ𝑖 𝑜𝑓 𝑝𝑟𝑜𝑣𝑖𝑑𝑎𝑛𝑐𝑒𝑗

𝑡𝑜𝑡𝑎𝑙 𝑏𝑢𝑠𝑖𝑛𝑒𝑠𝑠 𝑜𝑣𝑒𝑟𝑛𝑖𝑔ℎ𝑡 𝑠𝑡𝑎𝑦𝑠 𝑜𝑓 𝑚𝑜𝑛𝑡ℎ𝑖 ) = 𝜔𝑖𝑗, where i represents the month and j the

provinces (see Appendix B table 10).

Then 𝜔𝑖𝑗 multiplied by the predicted business overnight stays of the Netherlands of 𝑚𝑜𝑛𝑡ℎ𝑖 provides the missing data.

- Making the datasets useful for this research

The final step is making the datasets useful for this research. For the dataset of the Netherlands this means calculating the monthly non-business overnight stays by subtracting the monthly business overnight stays with the total monthly overnight stays this dataset now contains 402 observations (see Appendix A table 11 and for QDP see table 12).

For the provinces first the monthly non-business overnight stays are calculated the same way as the Netherlands only then per province. Then the provinces Noord-Holland, Zuid-Holland and

(11)

Utrecht (high Airbnb listings group) are grouped together in a High Airbnb Listings Group (HALG and the remanding ones are grouped in a Low Airbnb Listings Group (LALG). The HALG has 390 observations and the LALG has 390 observations this is due to the 6 months (equivalent to 12

observations) of missing observations in the province Flevoland and in order to avoid to get false data all of these monthly data has to be deleted for the LALG provinces (see Appendix B tables 13 and 14).

IV. Results

In this section I discuss the results if Airbnb attracts additional non-business overnight stays in the Netherlands. I do so by describing the results of the estimated coefficients provided by two DD regressions and then the interpretations for these coefficients results are discussed. First the Netherlands is discussed, secondly the two groups and finally the robustness check.

A. The Netherlands

Because model (1) is a linear-linear regression it should be interpreted as follows (if 𝛾, 𝛽 and 𝜗 significantly deviate from 0); one unit change in the independent variable m results in a 𝛾 + 𝛽 + 𝜗 change in the expected value of S for the non-business group and only a 𝜗 change of S for the business group (Stock & Watson, 2015). But because 𝛽 is the critical coefficient in this paper we are only interested in its effect on S. So based on this model we expect while time passes and m increases per month with one unit that S increases with 𝛽 (for non-business group), we expect this because while time moves away from the entry time of Airbnb more people will offer their room, apartment or house on Airbnb because it becomes more popular what will attract more non-business people.

If 𝛽 is significantly larger than 0 then it is possible Airbnb effects transportation, restaurants, bars, clubs, museums and local registrations by attracting additional non-business overnight stays.

Table 15 shows the results of the estimated coefficients. In this table you can see that in both regressions 𝛽 doesn’t significantly deviate from 0. Further is notable that by adding the control variables to the regression the R-squared value increases drastically.

What do these findings mean, it means that against expectations for the Netherlands overall 𝐻0 cannot be rejected which indicates that based on this research there is no evidence to support that Airbnb attracts additional non-business overnight stays in the Netherlands.

B. The provinces

Table 16 shows the results of the estimated coefficients of HALG and table 17 of LALG. In table 16 you can see that 𝛽 significantly deviates from 0 at a 1-percent level in regression 2. In the

(12)

other regression 𝛽 doesn’t significantly deviate. Further it is again notable that the R-squared values drastically increase due to the control variables.

These findings mean that based on this research Airbnb attracts additional non-business overnight stays in the HALG. So in provinces with a high supply of Airbnb listings Airbnb becomes more used over time and attracts more non-business stays.

Table 17 shows that for both regressions 𝛽 doesn’t significantly deviate from 0 for the LALG. And again for this group increased the R-squared values drastically for the regression with control variables.

These findings mean that, based on this research, Airbnb doesn’t attract additional non-business overnight stays in the LALG.

Table 15

Difference-in-Differences analysis the Netherlands

Overnight stays Variable (1) (2) Month 0.9421*** 0.9039*** (0.195) (0.111) Group 4377.296*** 1599.798*** (462.317) (123.071) Month*Group -0.8637 -0.5199 (9.729) (2.757) Month*E*Group 10.604 7.3045 (16.495) (4.613)

Dummy Business A/B 116.5915***

(19.234)

Dummy Business C 225.534***

(18.140)

Dummy Business D 347.4698***

(18.415)

Dummy Non-B A/B 3607.079***

(132.886)

Dummy Non-B D 9278.469***

(241.101)

Constant 1229.492*** 1018.47***

(11.894) (13.614)

Regression type OLS OLS

Observations 402 402

R-squared 0.4896 0.9588 Notes: Each OLS regression accounts for robust std errors. Regression 2 is with control variables *** significant at 1-percent level, ** significant at 5-percent level, * significant at 10-percent level

(13)

Table 16

Difference-in-Differences analysis HALG

Overnight stays Variable (1) (2) Month 0.7265*** 0.7026*** (0.124) (0.069) Group 891.8349*** 254.6788*** (115.910) (34.232) Month*Group 0.5622 0.6629 (2.411) (0.670) Month*E*Group 4.7851 3.9510*** (4.145) (1.112)

Dummy Business A/B 93.7486***

(10.603)

Dummy Business C 146.7224***

(11.264)

Dummy Business D 228.3195***

(11.481)

Dummy Non-B A/B 956.3583***

(33.245)

Dummy Non-B D 2321.435***

(61.102)

Constant 747.2072*** 614.3599***

7.216 (8.906)

Regression type OLS OLS

Observations 402 402

R-squared 0.4418 0.9542 Notes: Each OLS regression accounts for robust std errors. Regression 2 is with control variables *** significant at 1-percent level, ** significant at 5-percent level, * significant at 10-percent level

(14)

Table 17

Difference-in-Differences analysis LALG

Overnight stays

Variable (1) (2)

Month 0.1912** 0.2243***

(0.079) (0.049)

Group 3416.729*** 1607.207***

(349.114) (109.810)

Month*Group -6.6775 -2.6616

(7.699) (2.562)

Month*E*Group 12.27 5.499

(12.673) (3.981)

Dummy Business A/B 52.600***

(12.231)

Dummy Business C 91.6669***

(8.302)

Dummy Business D 134.3376***

(8.536)

Dummy Non-B A/B 2589.419***

(125.846)

Dummy Non-B D 6637.029***

(195.006)

Constant 484.586*** 410.5952***

(5.225) (5.057)

Regression type OLS OLS

Observations 390 390

R-squared 0.5185 0.9509

Notes: Each OLS regression accounts for robust std errors. Regression 2 is with control variables *** significant at 1-percent level, ** significant at 5-percent level, * significant at 10-percent level

(15)

C. Robustness check

The results in sections IV.A and IV.B are based on a linear-linear regression to see if these results would change when we account for a non-linear relation I run the two regressions in a log-linear form: 𝑙𝑜𝑔𝑆𝑡𝑗 = 𝛼0+ 𝛾𝑚𝑡+ 𝛿𝐺𝑗+ 𝜗𝑚𝑡𝐺𝑗+ 𝛽𝑚𝑡𝑡𝐺𝑗+ 𝜂𝑡𝑗+ 𝜀𝑡𝑗

(see Appendix C tables 18,19 and 20). The tables show now that 𝛽 doesn’t significantly deviate from 0 in the HALG, which indicates that Airbnb has no effect in the HALG. However because the p-value of the 𝛽 coefficient from regression 2 is 0.101, so this coefficient isn’t a significant deviation from 0 at a 10-percent level by 0.1-percent and the R-squared is lower than in the linear regression this change is not big enough to question the findings in part A and B. It does however trigger me do another

robustness check.

The results in sections IV.A and IV.B are based on the estimated predicted value of the quarterly GDP of the business overnight stays. To see if the results differ when another dependent variable is used to predict the business overnight stays, the whole research is repeated but this time with vacancies as dependent variable for the regression model (see Appendix C table 21) to predict the business overnight stays for the period Jan 2000 until Dec 2006. This changes stays values for the business group and the non-business group.

These results for the Netherlands can be seen in Appendix C table 22. Table 22 shows that now the monthly coefficients don’t significantly differ from 0 anymore, further there aren’t any significant changes; 𝛽 still doesn’t significantly deviate from 0 for the Netherlands.

The results for the HALG are in Appendix C table 23. These results show the same changes for the month coefficient as with the Netherlands and there are some changes in the month*group

coefficient, so became this coefficient significantly deviating from 0 at a 10-percent level in model 2 and at a 1-percent level for regression 2. Further is 𝛽 still significantly deviating from 0 at a 1-percent level in regression 2.

The LALG results are in Appendix C table 21. These results show some odd changes for the month coefficients, these coefficients are now significantly deviating from 0 at a 5-percent level but this time negatively for regression 1 and at a 1-percent level for regression 2. Further is 𝛽 still not deviating from 0 in both regressions.

Although some changes have occurred due to the change in predicting variable nothing has changed for the 𝛽 coefficient in a significant way, so the same conclusion can be drawn from this research.

(16)

V. Conclusion

This paper analyses if Airbnb attracts additional non-business overnight stays in the

Netherlands. It does so by analysing the difference in business overnight stays, which are assumed not being influenced by entry of Airbnb and non-business overnight stays by applying a difference-indifferences technique. All the datasets come from the CBS. Besides analysing the Netherlands overall this paper also analyses a high listings group of provinces and a low one in order to be able to more specifically evaluate the effect Airbnb has in these groups.

The main conclusion of this thesis is that Airbnb doesn’t attract additional non-business

overnight stays in the Netherlands, which is against expectations. However when analysing the results of the two groups a logical explanation is found why the results of the Netherlands are against

expectations. The results for the HALG show a significant effect, which means Airbnb does attract additional non-business overnight stays in provinces where Airbnb is highly present. The results of the LALG showed no significant effect, so in provinces where Airbnb is less present Airbnb doesn’t attract additional non-business overnight stays, which is logical because Airbnb can’t cause an significant effect if it has a small presence. So the additional stays Airbnb attracts in the three high Airbnb listings provinces isn’t big enough to compensate for the nine low Airbnb listings provinces and doesn’t make its effect significant for the whole Netherlands.

The findings for the HALG could be a reason why in some of the previous discussed studies there are no significant effect found on if Airbnb hurts hotel revenue, because it could be that in those studies the analysed area is a high Airbnb listings area and attracts a new kind of customer who wouldn’t have come if Airbnb wasn’t present in that area.

For further research it would be interesting to see if indeed these additional tourists are a new kind of customers and solely interested in staying in an Airbnb accommodation. For research done in other countries if Airbnb does attract additional non-business overnight stays one complete dataset with no years of missing data is recommended in order not to have to predict these missing data and making the research more reliable.

Looking at the research done, the biggest problem is the data. The amount and kind of data manipulations done makes this research less precise and reliable.

Firstly by making the datasets of the two periods fluent by multiplying (1 − 𝜌) with the data of period 2 the dataset became more precise than it was, but less then it would have been if it was one dataset from 2000 until 2016. Secondly by using the quarterly GDP as dependent variable to predict the missing business data while its regression estimate wasn’t significantly deviating from 0, because I predicted 12 months * 7 years = 84 months of data for the Netherlands, and this is 84/402= 0.209 so 20.9-percent of the data I used is predicted. And afterwards multiplying the outcome of the prediction model with 𝜑𝑚𝑒𝑎𝑛 made the data less precise and so the analysis less reliable than it would have been if the business data weren’t missing. Thirdly filling the missing business data for the provinces with

(17)

multiplying the forecast outcome for the Netherlands with 𝜔𝑖𝑗 made the data less precise than it would have been if the business data weren’t missing. And the years of data are 8 years before the entering and almost 8 years after the entering of Airbnb. But when you account for that it takes time for Airbnb to take off and grow, than I had less then 8 years of data to estimate an effect over.

All these manipulations made the dataset used for this thesis less precise and reliable than it could have been if all the data needed would have been available, but they were necessary to be able to perform the DD regression analysis.

(18)

References

Airdna. (2017). : https://www.airdna.co/city/nl/amsterdam

Angrist, J. D., & Pischke, J.-S. (2015). Mastering metrics: the path from cause to effect. Woodstock, Oxfordshire, United Kingdom: Princeton University Press.

Choi, K.-H., Jung, J., Ryu, S., Kim, S.-D., & Yoon, S.-M. (2015, October). The relationship between Airbnb and the hotel revenue: In the case of Korea. Indian Journal of Science and Technology. de Jong, H. (2016, 6 24). Franse hotelorganisatie klaagt Airbnb aan. Financieel Dagblad.

Fraiberger, S., & Sundararajan, A. (2015). Peer-to-peer rental markets in sharing economy. NYU Stern School of Business, research paper.

Goree, K. (2016). Battle of the beds: The economic impact of Airbnb on the hotel industry in Chicago and San Francisco. Scripps College, Bachelor thesis.

Guttentag, D. (2015). Airnbn; disruptive innovation and the rise of an informal tourism accommodation sector. Current issues in tourism , 18 (12), 1192-1217.

Hooijer, P. (2016). The relationship between Airbnb and the hotel revenue: Evidence from the Netherlands. University of Amsterdam, Bachelor thesis.

Molenaar, C., & de Groot, G. (2016, 12 9). Toeristische drukte Amsterdam blijft toenemen. Financieel

Dagblad.

Neeser, D. (2015). Does Airbnb hurt hotel business: Evidence from the Nordic countries. University Carlos III in Madrid, Master thesis.

Stock, J. H., & Watson, M. W. (2015). Introduction to econometrics. Essex, England: Pearson. Thole, H. (2015, 8 26).:

https://www.businessinsider.nl/airbnb-nederland-amsterdam-verspreiding-appartement-582139/

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 in-press .

(19)

Appendices

A. Appendix A the Netherlands

Table 1

Month distribution dummy variable the Netherlands Group Business Non-Business

A Jan, Feb, Dec Jan, Feb, Mar, Nov, Dec

B Jul Apr, Mar, Jun, Sep, Oct

C Mar, Aug, Nov

D Apr, May, Jun, Sep, Oct Jul, Aug

Table 5

ρ

Total

Business

0.094

0.061

Table 8

The Netherlands

Month

φ

Jan

0.307

Feb

0.314

Mar

0.380

Apr

0.329

May

0.341

Jun

0.330

Jul

0.303

Aug

0.318

Sep

0.380

Oct

0.393

Nov

0.346

Dec

0.261

Table 4

Significant different from 0 NL Total Business

(20)

Table 9

Regression analysis

Quarterly business stays

Variable (3)

GDP 4.8284

(3.781)

Constant 2537.378**

(1053.187)

Regression type OLS

Observations 39

R-squared 0.0407

Note: Each OLS regression accounts for robust std errors

*** significant at 1-percent level, ** significant at 5-percent level, * significant at 10-percent level

Table 11

Descriptive statistics for the Netherlands

Variable Mean Standard deviation Minimum Maximum

Stays 3564.206 3356.16 880 14526.51

m 4 58.095 -96 104

G 0.5 0.501 0 1

m*G 2 41.128 -96 104

m*E*G 135.821 27.634 0 104

Dummy Business A/B 0.0423 0.201 0 1

Dummy Business C 0.1244 0.33 0 1

Dummy Business D 0.209 0.407 0 1

Dummy Non-B A/B 0.209 0.407 0 1

Dummy Non-B D 0.0846 0.279 0 1

Table 12

Descriptive statistics for the Netherlands Variable Mean Standard

deviation Minimum Maximum

(21)

B. Appendix B the provinces

Table 2

Month distribution dummy variable HALG Group Business Non-Business

A Jan, Feb, Dec Jan, Feb, Mar, Nov, Dec

B Jul Apr, Mar, Jun, Sep, Oct

C Mar, Jun, Aug, Nov

D Apr, May, Sep, Oct Jul, Aug

Table 3

Month distribution dummy variable LALG Group Business Non-Business

A Jan, Feb, Jul, Dec Jan, Feb, Mar, Apr Nov, Dec

B Aug Mar, Jun, Sep, Oct

C Mar, Apr, Nov

D May, Jun, Sep, Oct Jul, Aug

Table 6

Significant different from 0 Provinces Provinces Total Business

GR Yes No FL Yes No DR Yes No OV Yes No FL No No GL Yes No UT Yes No NH Yes Yes ZH Yes Yes ZE Yes No NB Yes Yes LM Yes No

(22)

Table 10 The provinces Month ω GR ω FR ω DR ω OV ω FL ω GL ω UT ω NH ω ZH ω ZE ω NB ω LM Jan 0.024 0.027 0.017 0.034 0.009 0.076 0.070 0.355 0.185 0.025 0.107 0.070 Feb 0.024 0.026 0.019 0.036 0.009 0.077 0.065 0.354 0.182 0.026 0.109 0.074 Mar 0.021 0.027 0.020 0.036 0.007 0.077 0.065 0.359 0.181 0.027 0.102 0.078 Apr 0.021 0.029 0.018 0.031 0.009 0.076 0.064 0.370 0.190 0.027 0.095 0.069 May 0.022 0.030 0.018 0.033 0.010 0.081 0.059 0.366 0.185 0.029 0.096 0.072 Jun 0.024 0.036 0.017 0.033 0.010 0.079 0.056 0.362 0.181 0.030 0.100 0.071 Jul 0.022 0.032 0.018 0.032 0.009 0.068 0.055 0.387 0.182 0.033 0.091 0.071 Aug 0.025 0.032 0.020 0.036 0.010 0.069 0.052 0.377 0.182 0.034 0.091 0.072 Sep 0.021 0.029 0.019 0.034 0.009 0.080 0.063 0.376 0.174 0.029 0.097 0.070 Oct 0.022 0.028 0.021 0.037 0.009 0.082 0.062 0.363 0.175 0.028 0.100 0.073 Nov 0.023 0.025 0.016 0.038 0.008 0.084 0.064 0.354 0.181 0.025 0.105 0.075 Dec 0.024 0.024 0.019 0.037 0.008 0.073 0.060 0.363 0.171 0.024 0.105 0.092 Table 7 Ρ

Provinces Total Business

GR 0.209 - FL 0.073 - DR 0.138 - OV 0.121 - FL - - GL 0.124 - UT 0.069 - NH 0.056 0.129 ZH 0.102 0.026 ZE 0.106 - NB 0.138 0.026 LM 0.046 -

(23)

Table 13

Descriptive statistics for the HALG

Variable Mean Standard deviation Minimum Maximum

Stays 1262.146 811.343 530 4.231.719

m 4 58.095 -96 104

G 0.5 0.501 0 1

m*G 2 41.128 -96 104

m*E*G 135.821 27.634 0 104

Dummy Business A/B 0.0423 0.201 0 1

Dummy Business C 0.1667 0.373 0 1

Dummy Business D 0.1667 0.373 0 1

Dummy Non-B A/B 0.209 0.407 0 1

Dummy Non-B D 0.0846 0.279 0 1

Table 14

Descriptive statistics for the LALG

Variable Mean Standard deviation Minimum Maximum Stays 2344.376 2590.266 3573.684 10786.57

m 4 58.095 -96 104

G 0.5 0.501 0 1

m*G 2 41.128 -96 104

m*E*G 135.821 27.634 0 104

Dummy Business A/B 0.0423 0.201 0 1

Dummy Business C 0.1244 0.33 0 1

Dummy Business D 0.1667 0.373 0 1

Dummy Non-B A/B 0.1667 0.373 0 1

(24)

C. Appendix C robustness check

Table 18

Difference-in-Differences analysis the Netherlands

logOvernight stays

Variable (1) (2)

Month 0.0007*** 0.0007***

(0.0002) (0.00009)

Group 0.0776*** 0.9872***

(0.076) (0.026)

Month*Group 0.00002 0.0001

(0.002) (0.0005)

Month*E*Group 0.0011 0.0006

(0.003) (0.0008)

Dummy Business A/B 0.1083***

(0.017)

Dummy Business C 0.2001***

(0.016)

Dummy Business D 0.2901***

(0.015)

Dummy Non-B A/B 0.8213***

(0.027)

Dummy Non-B D 1.4631***

(0.027)

Constant 7.1039*** 6.9238***

(0.010) (0.012)

Regression type OLS OLS

Observations 402 402

R-squared 0.7489 0.9783

Notes: Each OLS regression accounts for robust std errors. Regression 2 is with control variables *** significant at 1-percent level, ** significant at 5-percent level, * significant at 10-percent level

(25)

Table 19

Difference-in-Differences analysis HALG

logOvernight stays

Variable (1) (2)

Month 0.0009*** 0.0009***

(0.0002) (0.00009)

Group 0.6985*** 0.3977***

(0.066) (0.028)

Month*Group 0.0008 0.0009*

(0.001) (0.0005)

Month*E*Group 0.0017 0.0012

(0.002) (0.0007)

Dummy Business A/B 0.1424***

(0.016)

Dummy Business C 0.2130***

(0.016)

Dummy Business D 0.3133***

(0.016)

Dummy Non-B A/B 0.6979***

(0.024)

Dummy Non-B D 1.2366***

(0.030)

Constant 6.6053*** 6.4180***

(0.010) (0.013)

Regression type OLS OLS

Observations 402 402

R-squared 0.5458 0.9491

Notes: Each OLS regression accounts for robust std errors. Regression 2 is with control variables *** significant at 1-percent level, ** significant at 5-percent level, * significant at 10-percent level

(26)

Table 20

Difference-in-Differences analysis LALG

logOvernight stays Variable (1) (2) Month 0.0004** 0.0004*** (0.0002) (0.0001) Group 1.9262*** 1.5921*** (0.0834) (0.0366) Month*Group -0.0016 -0.0006 (0.002) (0.0007) Month*E*Group 0.0027 0.001 (0.003) (0.001)

Dummy Business A/B 0.1191***

(0.025)

Dummy Business C 0.2005***

(0.017)

Dummy Business D 0.28***

(0.017)

Dummy Non-B A/B 0.8166***

(0.036)

Dummy Non-B D 1.4409***

(0.036)

Constant 6.1719*** 6.0145***

(0.011) (0.012)

Regression type OLS OLS

Observations 390 390

R-squared 0.8396 0.9756 Notes: Each OLS regression accounts for robust std errors. Regression 2 is with control variables *** significant at 1-percent level, ** significant at 5-percent level, * significant at 10-percent level

(27)

Table 21

Regression analysis

Quarterly business stays

Variable (3)

Vacancies 2.9664**

(1.386)

Constant 3440.976***

(200.061)

Regression type OLS

Observations 39

R-squard 0.1055

Note: Each OLS regression accounts for robust std errors

*** significant at 1-percent level, ** significant at 5-percent level, * significant at 10-percent level

Table 22

Difference-in-Differences analysis the Netherlands

Overnight stays

Variable (1) (2)

Month -0.0701 -0.1094

(0.209) (0.128)

Group 4307.466 *** 1537.332***

(462.1) (123.453)

Month*Group 2.0931 2.4379

(9.735) (2.759)

Month*E*Group 8.8444 5.5453

(16.495) (4.615)

Dummy Business A/B 120.0296***

(20.473)

Dummy Business C 234.955***

(20.032)

Dummy Business D 355.5797***

(18.97)

Dummy Non-B A/B 3603.774***

(132.91)

Dummy Non-B D 9278.607***

(241.171)

Constant 1286.442*** 1069.4***

(12.285) (14.59)

Regression type OLS OLS

Observations 402 402

R-squared 0.479 0.9579

Notes: Each OLS regression accounts for robust std errors. Regression 2 is with control variables *** significant at 1-percent level, ** significant at 5-percent level, * significant at 10-percent level

(28)

Table 23

Difference-in-Differences analysis HALG

Overnight stays

Variable (1) (2)

Month 0.1112 0.0866

(0.134) (0.083)

Group 849.408*** 217.243***

(115.758) (34.676)

Month*Group 2.3598 2.4612***

(2.413) (0.672)

Month*E*Group 3.7149 2.8814***

(4.144) (1.112)

Dummy Business A/B 96.901***

(12.667)

Dummy Business C 151.7808***

(12.72)

Dummy Business D 234.198***

(12.578)

Dummy Non-B A/B 954.012***

(33.131)

Dummy Non-B D 2320.746***

(60.913)

Constant 781.822*** 645.0652***

(7.652) (9.887)

Regression type OLS OLS

Observations 402 402

R-squared 0.4195 0.9521

Notes: Each OLS regression accounts for robust std errors. Regression 2 is with control variables *** significant at 1-percent level, ** significant at 5-percent level, * significant at 10-percent level

(29)

Table 24

Difference-in-Differences analysis LALG

Overnight stays

Variable (1) (2)

Month -0.1987** -0.1645***

(0.083) (0.051)

Group 3389.653 *** 1583.337***

(349.075) (109.995)

Month*Group -5.522 -1.5083

(7.704) (2.567)

Month*E*Group 11.5769 4.807

(12.677) (3.987)

Dummy Business A/B 55.655***

(10.892)

Dummy Business C 95.1088***

(8.048)

Dummy Business D 138.7291***

(7.891)

Dummy Non-B A/B 2587.424***

(126.02)

Dummy Non-B D 6637.59***

(195.081)

Constant 506.612*** 429.9738***

(5.199) (4.871)

Regression type OLS OLS

Observations 390 390

R-squared 0.5131 0.9503

Notes: Each OLS regression accounts for robust std errors. Regression 2 is with control variables *** significant at 1-percent level, ** significant at 5-percent level, * significant at 10-percent level

Referenties

GERELATEERDE DOCUMENTEN

86 Similarly, sampling can be used to establish quality control in the clerical field, where it may be used by the internal audit function, as well as in the course

Zwaap T +31 (0)20 797 88 08 Datum 2 december 2014 Onze referentie ACP 50-1 ACP 50. Openbare vergadering

Derhalve kan worden geconcludeerd op basis van de resultaten zoals beschreven in de voorgaande paragrafen en deze paragraaf, dat beursgenoteerde klanten van de Big Four geen

Here, the slope of conditional direct effect appeared to be significant only for the high levels SN, in other words, only for the situations were procrastination was seen as

A research into how cultural factors influent the success of fashion bloggers from the Netherlands, the United Sates and Taiwan.. in Business Administration- Marketing Track

De Triodos Bank geeft medewerkers waardering in de vorm van ‘schouderklopjes’ aangezien zij van mening zijn dat financiële beloningen maar effect hebben op korte termijn. De

-General vs firm specific -Formal vs informal Employees’ -Performance -Turnover Employee commitment Organizational Climate − Opportunity to perform − Supervisor(s) support

I find that a large share of non-interest income does increase the insolvency risk for cooperative banks, but not for commercial and savings banks. The increase in insolvency risk