Incumbent price setting behaviour facing
increased competition:
A worldwide investigation of hotel prices after the introduction of Airbnb
M.J. Boom
11095490 Supervisor: R. Sloof University of Amsterdam, NL MSc. Managerial Economics & Strategy
June 2016
Abstract
An increase in competition has a decreasing effect on prices. We test this elementary theoretical prediction in a market characterized by vertical product differentiation. The rise of the shared economy platform Airbnb introduced new competition in the hospitality market, a market that is segmented in quality by the star rating system. Although Airbnb mainly enters the lower quality segments, we find that their entry affects prices in the higher quality segments. Surprisingly, the negative impact of Airbnb supply on hotel prices is even larger for higher quality hotels. As higher quality hotels have larger profit margins (perhaps due to lower initial competition levels), they are better able to reduce their prices in response to the increased (indirect) competition from the other segments. This reveals that there exists competition both within and between quality segments in markets that are characterized by vertical product differentiation.
Keywords: vertical product differentiation, competition, hospitality industry, Airbnb, price setting J.E.L. classification: D22, D41, L11, L25, L83,
STATEMENT OF ORIGINALITY
This document is written by Marc Boom who declares to take full responsibility for the contents of this document.
“I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.”
I. INTRODUCTION
Imagine you walk into a store intending to buy a memory stick. The shop sells only two kinds of memory sticks. The sticks are from the same brand and the only difference is that one of the memory sticks has a double amount of storage. Surprisingly, both sticks are sold for the same price. Now, every rational person would buy the memory stick with the largest memory. This example illustrates one of the dimensions by which companies can differentiate their products from their competitors. If two products are sold for the same price, but differ in quality, then people buy the product of the highest quality. If two products are exactly equal in quality (including design), but different in price, then people buy the cheapest one (Sutton, 1986; Shaked and Sutton, 1987). Suppliers therefore face the trade-off between asking low-prices (and low quality) and providing high quality (and high prices) in order to attract customers. They search for a profit maximizing price-quality combination that attracts a certain amount of customers at a certain cost of quality. Differentiating your product by choosing a quality level is called vertical product differentiation.1 Whenever there are products of similar quality, firms
can compete by setting a lower price. In a stable market, firms determine their quality-price level and an equilibrium is found. Yet, whenever incumbent firms decide to change their quality-price level, for instance because new suppliers join the competitive playing field, the equilibrium is distorted and a new balance needs to be found.
In this paper we investigate one dimension of this dynamic product differentiation process in a market where the price-quality equilibrium of the incumbent firms is distorted. More specifically, we look at the price-setting behaviour of incumbent firms, in a market that is characterized by increased competition due to the entry of competitors. We are particularly interested in the change in prices when quality levels of incumbents are fixed in the short run. In order to do so we measure the impact of the entry of Airbnb (that facilitates the rental of private apartments and rooms) on the price setting behaviour of hotels per star rating segment. A unique characteristic of the hospitality industry is the globally recognized quality segmentation in the form of a star rating system. This makes the hospitality industry suitable for an investigation on vertical product differentiation. Each country has a national star rating system that assigns a number of stars (from 2 to 5) to hotels. The determination of the star rating depends on a variety of (quality and facility) conditions that a hotel should satisfy in order to
1 Obviously, products can also differentiate by means of marketing, brand name, design, et cetera. This kind of differentiation, horizontal product differentiation (Hotelling, 1929), is often reflected by taste and hardly by quality.
qualify for a certain quality level. In a framework of vertical product differentiation firms can compete by both quality and price. Yet, quality levels for hotels are rather fixed after they are initially determined at the moment a hotel enters the market. In order to satisfy the conditions for a certain star rating, quite substantial investments are often needed (e.g. in order to qualify for a four star rating, having an indoor swimming pool is often a requirement). These investments are usually not feasible (in the short run) once a hotel entered the market. Therefore, in this research we assume that quality is fixed and that hotels within quality segments only compete by price. This assumption is further discussed in section II.
Moreover, we assume that hotel supply is also fixed in the short term. Apart from large initial capital requirements at the initial entry stage, the scarcity of suitable locations (in metropolitan areas) causes a significant barrier to enter the market (Cheng, 2013). Due to these major initial investments, hotels will also not easily exit the market. Taking the above mentioned into account then, when there are no major shocks in hotel demand (e.g. Olympic Games), existing hotels are able to cope with small fluctuations in the number of tourists over time, and hotel supply will not face large adjustments (Cheng, 2013).
Despite of the relatively fixed supply, the hospitality branch is viewed as being characterized by high competition (Cheng, 2013) and has historically been among the sectors with the lowest profit margins.2 Yet, currently the hospitality branch is globally performing fairly well, with increasing revenues and profit margins.3 Arguably this is partly due to the increasing mobility and thus tourism worldwide. The number of international tourist arrivals has increased with 4,4% over 2015 towards a total of 1.184 million.4 This increase in tourism is not expected to slow down in the coming years5. Yet, these tourists have become more demanding. The introduction of websites like Booking.com and TripAdvisor significantly increased the transparency in the hotel industry. Previous experiences of other customers are easily shared online. Tourists are thus better able to compare hotels on their prices, quality, location and evaluations (Briggs, Sutherland and Drummond, 2007). Meanwhile, quality expectations of tourists are globally increasing, as they demand more extensive services from hotels (Freeman, R and Glazer K., 2015). Due to these high quality expectations and the increased transparency,
2 Subjective statement from Kristin Rohlfs consultant of the Hospitality Research Group of PKF Consulting 3 Margins by Sector (US). January 2016. Database from NY Stern University
4 International tourist arrivals up 4% reach a record 1.2 billion in 2015. January 2016. United Nations World Tourism Organization 5 United Nations World Tourism Organization (UNWTO) recalls in their annual “Tourism Highlights” report of 2015 that global tourism is expected to grow by 3,3 per year (until 2030).
hotels are maximizing their effort in order to realise the customer’s (high) expectations (Briggs et al, 2007).
To summarize. The hospitality industry is characterized by vertical product differentiation and has strict boundaries of these quality segments. In the short term, hotel supply and their quality levels are fixed. Hotels are facing high levels of competition, and thus rather low margins. Partly because of the internet, the market is currently highly transparent. In order to stay competitive within their quality segment, hotels need to meet the customers’ (high) service expectations. Differentiating by providing higher quality than expected is fairly tough and jumping a star staring is impossible in the short run. Hence, we assume that hotels within their quality segment are solely competing on price. Over the past few years, hotels have faced increased competition from the online rental platform Airbnb.6 We wonder whether hotels react to this new competitive challenge by adjusting their prices.
Airbnb is an internet platform that connect hosts with spare rooms with people that are looking for a place to stay. Airbnb is one of the best known examples of a business that introduces a decentralized peer-to-peer market or sharing economy model. Established as a small start-up in 2008, it has grown to a ten-billion-dollar firm enabling 80 million bookings per year. Airbnb offers a substitute for hotel rooms and thus can be seen as a serious competitor in the hospitality industry. Airbnb mainly enters the low quality segment and offers generally lower prices than hotels (as Table 1 illustrates). This sharing economy platform has received a lot of attention by the media and regional governments. This is perhaps not surprising, as Airbnb is believed to have a major disruptive impact on the current hospitality market and the economy in general. For instance, according to a report of the Hotel Association of New York City, Airbnb had a negative impact of 2.1 billion on the lodging market and broader economy of New York city during 2014.7 Furthermore, Airbnb is believed to increase the house and rental prices in large cities. For this reason Berlin even decided to ban Airbnb.8 On the other hand, Airbnb provides tourists with an extra opportunity for an overnight stay (that is generally more affordable than a hotel) on central locations in the city (Guttentag, 2015). Moreover, Airbnb offers homeowners the opportunity to generate extra income by renting their house or spare room(s). Mortgage
6 Airbnb has shown an exponential growth in the number of apartments and bookings over the past few years, e.g. the “Financieel Dagblad” wrote that in January 2016 there were in Amsterdam 500% more Airbnb bookings than a year earlier (07-06-2016).
7 “Airbnb and Impacts on the New York City Lodging Market and Economy” by HVS Consulting & Valuation (13-10-2015) 8 http://www.independent.co.uk/news/world/europe/Airbnb-rentals-berlin-germany-tourist-ban-fines-restricting-to-protect-affordable-housing-a7008891.html (02-15-2016)
loaners even provide higher mortgages to house buyers in Amsterdam, because of this opportunity to rent your house on Airbnb. This again, is also believed to have an increasing impact on the residential real estate prices.9 Very strong conclusions are often drawn, but the actual impact is rarely adequately investigated. Apart from a working paper from Zervas, Proserpio and Byers (2016) - who found that Airbnb supply has a negative impact on local hotel revenues - academic empirical research is rather scarce. Thus, aside of the economic relevance concerning the impact of increased competition on price setting behaviour, this research is socially highly relevant.
In this research we investigate how the increased competition from Airbnb relates to the change in hotel prices per quality level. We assemble the number of available Airbnb listings (supply) for 53 cities around the world in the period from 2008 until 2015. By the use of a fixed effects (panel data) analysis we estimate the relation between Airbnb supply and the development of hotel prices per quality segment. Although Airbnb supply does not seem to be related with average hotel prices, it does seem to be related with prices in the different quality segments. Interestingly, we find that this relation not necessarily holds for the prices in the lower quality hotels (where Airbnb mainly enters), but mainly for the prices in medium and high quality hotels. As profit margins seem to increase with quality, we argue that hotels in the medium and
9http://fd.nl/economie-politiek/1149391/ton-hogere-hypotheek-dankzij-inkomsten-uit-Airbnb (25-04-2016) TABLE 1
Average prices for hotels and Airbnb for a selection of cities
Average overnight prices during 2015
Hotel prices per segment Average overnight prices
Low-Quality Med-Quality High-Quality Hotel Airbnb Airbnb discount
Amsterdam £ 73 £ 93 £ 154 £ 105 £ 60 43% Barcelona £ 51 £ 87 £ 188 £ 97 £ 30 69% Berlin £ 46 £ 63 £ 104 £ 76 £ 30 60% London £ 61 £ 101 £ 208 £ 78 £ 44 44% New York £ 100 £ 159 £ 286 £ 177 £ 77 57% Paris £ 64 £ 103 £ 246 £ 111 £ 44 61% Sydney £ 35 £ 85 £ 150 £ 114 £ 81 29% Washington £ 78 £ 129 £ 255 £ 133 £ 67 49% Average £ 64 £ 109 £ 208 £ 116 £ 60 48%
Notes: Table shows average overnight prices for 2 persons for a hotel and an Airbnb rental in 2015. The Airbnb prices (which are originally measured in USD) are transformed to GBP by the average exchange rate in 2015. This table is for illustrative use and the 8
cities are exemplary. The average numbers in the last row are over all cities that are included in the research. The Airbnb discount is calculated as the percentage that an Airbnb rental costs less than an average hotel room {1 - P(Airbnb)/P(hotel)}.
high quality segments are better able to decrease their prices in response to increased (indirect) competition.
The remaining of this paper is structured as follows. First, we discuss the related literature, lay out the conceptual framework and develop our hypotheses (Section II). Then, we discuss the different data sources we used and necessary manipulations we made (Section III). Section IV discusses our empirical results based on panel data techniques. In section V we discuss the robustness checks and the limitations of our analysis. Finally, in section VI we conclude our paper
II. FRAMEWORK & HYPOTHESES
For this research we use a conceptual framework with multiple quality levels. We investigate the relation between an increase in competition due to market entry and the price development of the incumbent firms within these quality levels. The incumbent firms (hotels) are segmented in low, medium and high quality segments. Supply from the entrant (Airbnb) is segmented accordingly by two different approaches which we discuss below. Besides the price development within a quality segment due to market entry within this same segment, we are also interested in the spill over effects to other quality segments. In particular, does entry in the low quality segment also influence the prices in the higher quality segments? This is highly relevant for the case of Airbnb, that mainly introduces additional competition in (or below) the low quality (low price) segment.10 Theory predicts that whenever competition levels in a segment increases, firms react by adjusting their price or their quality, in order to find their new optimal price-quality level. When additional competition in a particular segment decreases average prices in this segment, between segment competition ensures spillover effects and will also cause a reduction in prices of the other segments.
Competing by the quality of a product, is called vertical product differentiation. By vertically differentiating your product, in other words choosing a certain quality, firms are able to increase their profitability (Shaked and Sutton, 1982). In a market with few firms, firms can relax the
10 Appendix 1 can be seen as suggestive evidence for the assumption that Airbnb mainly enters the low quality (low price segment). The table shows that Airbnb overnights can be bought (without exception) for substantial lower prices than a hotel overnight stay. On average the price for an Airbnb rental is approximately half of the price of a hotel room. This suggests that Airbnb mainly enters the low price (and low quality) segment.
price competition by choosing a different quality than their competitors. When the quality differences are small, price competition becomes more important; therefore, quality differences are generally rather substantial. In more competitive markets, product differentiation is used to segmented the market, in such a fashion that there exists enough quality difference between the segments to relax price competition between segments (Smith, 1956). These segments function as a sort of separate (but smaller) markets in which firms face less fierce price competition, which enables firms to ask a larger price premium. Hence, it follows that firms in competitive markets (typically characterized by low profit margins) are likely to engage in vertical product differentiation in order to enhance their profits (Shaked and Sutton, 1987)11. A critical assumption for the existence of vertical product differentiation is the existence of demand heterogeneity; or differently, the presence of different kinds of consumers. Although all consumers should agree on the ranking of the products in terms of quality, not all consumers are able (or think it is worth) to pay a higher price for a higher quality. Obviously, this depends on preferences and available budgets, i.e. consumer with higher incomes are more likely to buy high quality (high price) products (Shaked and Sutton, 1982).
We apply this conceptual framework to the hospitality market. The hospitality market is well in line with the characteristics of a perfectly competitive market12 that engages in vertical product differentiation. Hotels can differentiate themselves by quality and compete within this quality segment on price. Quality segments for hotels are easily identified by a well-known quality qualification, namely star ratings. These star ratings are monitored by national institutions and hence provide a proper measure of the relative quality within a country (Abrate, Fraquelli and Viglia, 2012). This quality ranking is uniformly accepted by customers and most of them will choose a high-quality hotel above a low-quality hotel when offered for the same price. Nevertheless, there exist large price differences between the different quality segments, which attracts different kind of customers. Recently the hospitality industry faced an increase in competition due to the entrance of Airbnb, mainly in the low-quality segment. We are interested in the price setting behaviour of the incumbent firms (the hotels) in the different quality segments due to this increase in competition.
11 This is under the assumption that quality improvement can be realised by the expenditure of fixed costs and a sufficiently small increase in the variable costs.
12 A perfect competitive market is characterized by 1) many suppliers and customers, 2) perfect information, 3) standardized products and 4) free exit and entry in the long run (Knight, 1921). The hospitality industry is characterized by assumptions 1, 2 and 4, whereas the perfect information assumption is just recently satisfied due to the existence of websites as Booking.com. Assumption 3 is partly satisfied, since the hospitality industry is characterized by vertical product differentiation, and the products within the quality segments are rather standardized.
Our research focusses on a period of 8 years, meaning that we investigate the market in the short run. As the hospitality market is characterized by large initial (sunk) costs, there should exist no free exit and entry in the short run (Stigler, 1968). For this reason, we assume that hotel supply is fixed. Moreover, because quality enhancements also require large investments, we assume that quality levels are determined at initial entry and also fixed in the short run. Hence, hotel supply is also fixed within the quality segments. Elementary economic theory predicts that an increase in competition will have a negative effect on prices. In line with this, we expect that average hotel prices have suffered from Airbnb’s entrance. Furthermore, we expect that the entrance of Airbnb (mainly in the low segment) will have a negative impact on the prices in the low quality segment, even more in the medium quality segment and will not have an (significant) impact on the prices in the high quality segment. In the remainder of this section we elaborate on the fixed quality assumption and on the expected price developments within quality segments.
II.1 - Fixed quality assumption
In this research we use the star rating system for the quality measurement of hotels. As it rarely occurs that hotels change star rating, we assume that quality levels in this industry are rather fixed. The assumption that quality levels are stable, regardless of the increase in competition, seems applicable for the hospitality market in the short run.
First, one could intuitively argue that an increase in star rating would rarely occur. Although the precise determinants of star ratings differ per country, the requirements are in general largely in line. The additional requirements for a jump in star rating are quite substantial. For a two-star hotel to become a three-star hotel, one needs to provide a 24h reception, offer cleaning services, provide a bathroom for every room and run a restaurant. A jump to a four-star hotel would imply that one needs to have a spa, a sauna, an indoor swimming pool and porters. These additional requirements demand major investments, a large increase in provided services (plus corresponding costs) and often a complete change of business model. Besides, considering that existing hotels are bounded by the available space in their building, is its often physically impossible to extend their hotels with these services and facilities. For this reason, it is plausible assume that hotels choose their quality level at entry and do not change this once they entered.
Secondly, we can justify the fixed quality assumption by a more theoretical approach. Generally, quality of an industry increases with competition, but only whenever the average fixed costs of quality improvement (a single investment cost that does not vary with quantity) decreases with the increase in competition. This decrease in average fixed costs requires the possibility of knowledge spillovers of quality enhancement (Banker, Khosla and Sinha, 1998). Yet, a quality increase (jump in segment) for a hotel relies more heavily on (physical) investments, rather than on the assembling of additional industry knowledge and know-how. Therefore, it seems unlikely that the beneficial knowledge spillovers of a quality investment outweigh the corresponding costs for a hotel to realize an increase in star rating. This suggests that an increase in competition due to the entry of Airbnb is not likely to enhance the quality level in the hospitality industry.
Another theoretical justification for the fixed quality assumption comes from the division in quality improvements by fixed and variable costs. Whenever quality enhancements can be realized by an increase in variable cost (as in restaurants), then an increase in market size (in terms of potential demand)13 increases the average quality. Whenever quality improvements come from fixed costs investments (as with newspapers), average quality only increases with market size in relatively small markets (Berry and Waldfogel, 2010). The hospitality industry seems an interesting compromise between these markets. An increase in quality that realises an increase in star rating often requires physical investments and thus fixed costs. Thus, as fixed cost quality enhancements are only feasible in relatively small markets - which we assume is not the case for the markets that Airbnb enters (mainly cities) - we should not expect large quality enhancements by the incumbent hotels due an increase in market size. Within these segments though, hotels can differentiate themselves by providing additional or better services, hence increasing quality within the boundaries of their star segment. This kind of quality improvement is realized by variable cost and thus predicted to increase by market size. Hotels in relatively large markets (cities), facing increased demand, are predicted to enhance their services, but not to engage in large investments and thus not to realise a jump in quality level. Nevertheless, because of the increased transparency (due to the evaluation platforms) of the hospitality market, it is highly important for hotels to match the expectations of the customers.
13 In the article of Berry and Waldfogel (2010) potential demand is estimated by population. For our research it would be more useful to look at tourist arrivals (which are the foremost hotel customers). Yet, as this research only focuses on major cities, we can assume that these are large markets, compared to smaller cities or the suburban areas.
Therefore, hotels are maximizing their efforts in order to provide the customers with the promised service levels (Briggs et all, 2007).
Figure 1 shows vertical differentiating possibilities in the hotel industry. Hotels enter the market in a certain segment. Once entered, these hotels cannot gain an additional star as variable costs quality investments are not sufficient to meet the requirements and the required fixed cost quality investment is not realistic. That is, qualifying for an additional star requires such a substantial amount of effort and investment that it would hardly occur. On the other hand, hotels are trying to fulfil the high expectations of the customers (based on the star rating) as much as they can, in order to receive positive evaluations on the evaluation websites.
II.2 - Price developments (within quality segments)
In this section we will elaborate upon the hypothesises regarding the price developments (within quality segments) due to the increased completion (due to Airbnb). This subsection is divided in three categories. First, we hypothesize on the impact of Airbnb supply on the average hotel prices. Secondly, we draw our hypotheses on this impact on prices in the different hotel quality segments. Lastly, we further identify the direct and indirect effect on prices by segmenting Airbnb supply.
II.2.a Total Airbnb supply on hotel prices
Assuming that quality is fixed (as explained above) and that all hotels within these star segments are maximizing their effort in order to receive good evaluations, we assume that their quality levels are relatively comparable. Ergo, we assume that hotels with the same number of assigned stars can only differentiate themselves by setting their price, formally known as Bertrand price competition (Bertrand, 1883).14 As economic theory predicts, hotels react to increased competition by lowering their prices (Abrate et al, 2012). For this reason, we hypothesize that Airbnb supply has a negative effect on the average hotel prices (where we average over quality segments).
H1. Overall average hotel prices are negatively related to Airbnb supply
II.2.b Total Airbnb supply on hotel prices per quality segment
For a deeper analysis we segment the hotels in quality levels. In this research we investigate the average prices of three segments, created by the use of the star rating system. The first segment, low-end hotels (PL), contains the average prices of 2 and 3 star hotels. The second segment, medium-quality hotels (PM), contains the average prices of 4 star hotels. The last segment, high-end hotels (PH), contains the average prices of 5 star hotels.
Although we presume average prices to decrease, we expect that PH is hardly impacted, because prices of high-quality hotels (5 star hotels) are less responsive to chances in competition levels (Tran, 2011). Previous empirical studies have found that prices of high quality hotels that were subject to demand shocks were hardly impacted, in contrast to low and medium quality hotels (Abrate et al, 2012). Although this paper investigates a supply, instead of a demand shock, the relation is expected to be similar. What happens to prices when the number of consumers per supplier adjusts? Apparently, prices of high quality hotels are less impacted by competition. This is also confirmed in a study that found that discount percentages for high-end hotels are significantly less impacted by increased competition levels, than those of low and medium quality hotels (Becerra, Santaló and Rosario Silva, 2013).
14 According to the stated assumptions, hotels within a quality segment can only engage in Bertrand price competition. Because firms within these categories are providing similar levels of quality and service, they are assumed to be rather homogenous. For this reason, consumers should more or less a similar price for hotels within a quality category. This implies that the within-segment price variance should be much smaller than the between-segment price variance. Unfortunately, we do not have the data to formally test this, but for further research it would be an interesting hypothesis to check empirically.
Furthermore, it should be noted that high quality hotels focus on different customers than those that Airbnb attracts. The most eminent argument to book an Airbnb accommodation instead of an hotel is the lower price (Guttentag, 2015). Customers that book a high quality hotel care much less about price (Tran, 2011) and more about provided services and facilities, something that Airbnb hardly can compete with. As Table 1 suggests, Airbnb mainly enters the low and (perhaps) medium quality segment, when compared by price. The high-quality segment can therefore only be impacted indirectly, i.e. decreased prices in the medium quality segment could have a reducing effect on high segment prices. Given that high-end hotel prices are less responsive to competition we expect that this indirect effect is not sufficiently strong. We thus do not expect that that Airbnb has a significant impact on PH.
H2: Hotel prices of the high-end hotels (PH) are not affected by Airbnb
Prices of hotels in the low and medium segment, on the other hand, are responsive to competition levels (Abrate et al, 2012) and thus expected to decrease with an increase in competition (Becerra et al, 2013). Furthermore, as hotels in these quality segments face both direct competition (from Airbnb) and indirect competition (via the other segments), we expect that the prices the low quality hotels (PL) and medium quality hotels (PM) are negatively related to Airbnb supply.
H3. Hotel prices of low-end (PL) and medium-quality (PM) hotels are negatively related to Airbnb supply
Moreover, we expect that PM is more strongly negatively associated with Airbnb supply than PL, because the margins of these hotels provide a broader range for adjustment. Standard economic theory suggests that firms are able to change prices above marginal costs in vertical differentiated markets. Products with the highest quality have the largest margin above marginal cost (Martin, 2010). In markets that are characterized by vertical product differentiation, profit margins are ranked in the same order as quality levels (Donnenfeld and Weber, 1991). In this theory it is assumed that quality enhancements are mostly realized by investments (fixed costs) and less dependent on variable costs (only rise slowly with quality). Hence, firms with the better quality product can choose for the strategy to set the price equal to the lower quality product, gain a larger market share and thus realize a larger profit. As the hospitality industry is characterized by this vertical product differentiation, we expect that high quality hotels have
larger profit margins than medium quality hotels, which accordingly have larger margins than low quality hotels. In other words, margins of hotels are likely to be positively related with star rating.
That quality levels in the hospitality industry are predominantly obtained by investments, further strengthens this statement. Since hotels like to make an overall profit, they will include these investments in their profit and loss statements by an annual depreciation. In order to stay profitable, hotels will set their price just above the total average costs (sum of average fixed and average variable costs). Nevertheless, as these (fixed) investment costs can be seen as sunk costs, the main short term priority is to keep the price above the variable cost. This in turn suggests that hotels will try to keep their prices above the total average costs, but when their quantity is decreasing (e.g. due to additional competition from Airbnb) they are able to respond by lowering their prices towards their average variable (or marginal) cost. Because of the high initial investments, high quality hotels will have a larger average fixed costs share and are thus better able to respond to additional competition by lowering their prices.
For this reason, we expect that higher quality hotels are better able to change their prices in changing market situations. This is indeed what Abrate et al (2012) found. They found that when the demand for hotel rooms increases, the price differential between the star ratings increases. In other words, when there are more consumers per supplier, the price differential increases. Additional demand makes it possible for hotels to increase their prices towards (or above) their average total costs, which are relatively higher for higher quality hotels; thus increasing the price differential15. When supply increases (e.g. by Airbnb), the number of consumers per supplier decreases and the opposite relation should hold. In order to stay competitive, firms should lower their prices. Since higher quality hotels have larger margins, they are able to lower their prices to a larger extent and the price differential between star rating segments will decrease.
Interestingly, these margins (calculated as the price margin above marginal costs, divided by the price) are frequently referred to as the Lerner index. This index is a measure for the market power of a firm; in other words, a measure of competitiveness (Lerner, 1934). When there is a
15 This increase in price differential causes quality downshifting. As higher quality products have the possibility to ask a larger premium above the marginal cost, some consumers will buy a worse quality product than they would have done if the prices were set at marginal costs. This loss in consumer surplus is larger than the gain in producer surplus and therefore causes substantial welfare losses for an economy (Bresnahan, 1981).
lot of competition, firms generally have low market power and thus small margins. In a market with low competition levels, firms have high market power and are able to set higher margins. This smoothly links to our example of the hospitality market. We just argued that hotels within lower quality segments have lower profit margins. According to the Lerner index this implies that hotels within this segment should have less market power, because they face more competition. Table 2 illustrates the distribution of hotels over the different quality segments. This distribution suggests indeed that there is the most competition in the lower quality segment, less in the medium quality segment and the least in the high quality segment.16 Which is intuitive, as due to the high initial investment costs, it is much harder to enter the market in the high quality segment, than in the low quality segment.
TABLE 2
Distribution of hotel quality segments for 8 cities Number of hotels
Low-Quality Med-Quality High-Quality
Amsterdam 192 85 24 Barcelona 244 172 34 Berlin 492 173 38 London 406 408 183 New York 423 212 95 Paris 272 205 70 Sydney 142 168 54 Washington 347 140 21 Average % 55% 34% 11%
Notes: Table shows for 8 cities the number of hotels per quality segment according to the hotel reservation website Booking.com. The average % is calculated by the average share of the segments over these 8 cities. This table is for illustrative use and the 8 cities are exemplary.
As lower quality hotels face more initial competition and have therefore lower profit margins, the introduction of Airbnb will perhaps lower the prices, but only slightly. Medium quality hotels, on the other hand, have lower initial competition levels and thus higher profit margins. Additional competition in a market with low competition reduces the market power more severely and will therefore have a larger impact on the hotel prices in this segment. Yet, as
16 Note that this table only provides suggestive evidence for the competition levels within these segments. We should be aware of three limitations. First, the table only shows the competition levels for 8 cities and is just for illustrative use. Secondly, the table shows the number of hotels in a segment, instead of the number of hotel rooms. As higher quality hotels have typically larger hotel rooms (Briggs et al., 2007), these numbers are perhaps somewhat exaggerated. Lastly, competition levels are not only determined by the number of competitors, but also by the number of consumers. When there exists much more demand for low quality hotels, then a higher share of hotels in this segment does not necessarily imply a higher competition level.
mentioned before, we do have reasons (less price responsive to competition levels and no direct competition) to expect that the high-end hotels are not significantly impacted by the entrance of Airbnb. Therefore, we expect that an increase in competition from Airbnb has a larger impact on the prices in the medium-quality hotel segment, than those of the low-quality hotel segment.
H4. Airbnb supply has a larger negative impact on medium-quality hotels prices (PM), than on low-end hotel prices (PL)
II.2.c Segmented Airbnb supply on hotel prices per quality segment
In order to get a deeper understanding of the impact of Airbnb on the different hotel segments, it is helpful to segment the Airbnb supply in such a fashion that the segments can be linked to the hotel star rankings. In doing so, we should keep in mind that Airbnb apartments do not provide a perfect substitute for hotels. Although Airbnb offers a range of outstanding and very luxurious accommodations, they all lack the additional services that hotels provide. On the other hand, Airbnb generally provides accommodations for lower prices than hotels. Hence, Airbnb is likely to enhance competition mostly for hotels targeting the low-end customers. For a segmentation of Airbnb supply that matches the hotel segmentation, we use two approaches. First, we create three segments by the sort of rentals; a shared room (AirbnbS); a private room (AirbnbP) and an entire apartment (AirbnbA). These are respectively proxies for low quality Airbnb supply, medium quality Airbnb supply and high quality Airbnb supply. Secondly, we use a more controversial approach by segmenting Airbnb supply on their own price17, with price boundaries related to the hotel prices in that particular city. Considering that lower prices are an often mentioned motivation to book an Airbnb rental above a hotel (Guttentag, 2015),
we segmented supply based on discounted hotel prices. The low Airbnb segment (AirbnbL)
contains all apartments with a price lower that 90% of the low segment hotel rates (PL). The
medium segment (AirbnbM) consists of all apartments with a price between 90% of the low
segment hotel rates and 90% of the medium segment hotel rates. The high segment (AirbnbH) includes all apartments with a higher price than the before mentioned. Although creating segments with hotel-prices is perhaps not the ideal approach, prices for Airbnb are arguably the best predictor of quality (apart from individual photo assessment on the website). Besides, it seems that hotel prices are a good estimator for star ratings (Abrate et al, 2012; Bull, 1994;
17 As hotel rooms are mostly for 2 persons, the Airbnb price is also measured by price per night for two persons. (e.g. the price for a four person Airbnb apartment is divided by two).
Israeli, 2002) and therefore perhaps a useful way to segment Airbnb apartments in similar groups as the hotels.
In line with earlier hypotheses, we expect that the supply of an Airbnb segment has a negative impact on the prices of the corresponding hotel segment, i.e. the entry of medium quality Airbnb rentals (AirbnbM) will lower the prices in the medium quality hotel segment (PM).
H5. The supply of an Airbnb segment (AirbnbQ) has a negative impact on the prices in the corresponding hotel segment (PQ)
Nevertheless, quality segments also compete with each other. Firms choose their optimal price-quality combination such that their profits are maximized. Consumers trade-off price-quality against price and choose that quality level where marginal utility (of quality improvements) equal the marginal costs (of acquiring a higher quality product). Assuming that quality is fixed, a decrease in the price of a product in a particular segment will enhance the utility that this product provides to the consumer. Thus, a decrease in the prices within a segment will shift consumers from the other segments towards the segment with the lower prices. Hotels in these other segments see their profits decrease, due to the fewer customers that are willing to pay for the quality of their products. When the profit margins tolerate it, these hotels will as a response lower their prices to find the new profit maximizing price. Accordingly, we expect that the supply of a particular Airbnb segment, also has a negative indirect impact on the prices of the adjacent hotel segments, i.e. the entry of medium quality Airbnb rentals (AirbnbM) will also lower the prices in the low (PL) and high (PH) quality hotel segments. We expect that the indirect effect of the non-adjacent hotel segments is not substantial enough to be significant, i.e. the entry of low quality Airbnb rentals (AirbnbL) on the prices in the high quality hotel segments (PH)
H6. The supply of an Airbnb (AirbnbQ) segment also has a negative (indirect) impact on the prices of the adjacent hotel segments (PQ-1and PQ+1)
III. DATA
For this research we assembled data from several different data sources. In this section we provide an overview of the sources of this data, report on some minor modifications that we made and give a first informational insight of the variables. This section contains three subsections: 1) hotel prices, 2) Airbnb supply and 3) control variables. An overview of all these variables is presented in Table 4 at the end of this section.
III.1 - Hotel prices
For the hotel prices (per star rating) we used the Hotel Price Index reports (HPI) from Hotels.com. Starting in 2004, this HPI-report is launched every six months and contains the aggregated hotel prices of all the bookings made and actually paid on hotels.com (including tax and fees). The HPI report is globally seen as the industry standard for hotel prices, perhaps the most comprehensive benchmark available and broadly used as a reference tool by the media, financial analysts, hoteliers, investors and academics. Combining all UK HPI-reports available yields us with hotel price data per star rating segment for the years from 2008 to 2015 for 53 cities worldwide. Obviously, we correct the hotel prices for inflation rates as reported by the Worldbank.
In order to compare the hotel prices, regardless of regional economic fluctuations, the report is launched for several countries and reports the hotel prices that are globally paid by inhabitants of that particular country. For this research we use the UK-reports, the numbers thus reflect the actual prices paid for hotels (worldwide) by UK inhabitants. In order to control for possible
£90 £95 £100 £105 £110 £115 2008 2009 2010 2011 2012 2013 2014 2015 A ve ra ge hot el pr ic es FIGURE 2
Average hotel prices over the years
Average hotel prices of 53 cities over the year 2008-2015 as reported by the Hotel Price Index. The prices reflect the nominal prices paid by UK inhabitants in hotels all around the
world. The real hotel prices are constructed by correcting for inflation. Nominal hotel price
currency effects, we include the relative purchasing power parity of the destination country towards the UK.18
Figure 2 gives an impression how prices evolved over time. Interestingly, for a long period industry reports and the media have commonly agreed on the fact that Airbnb did not have a significant impact on the hospitality industry yet.19 They argued that due to the increasing tourist numbers and thus demand for hotel rooms, hotels were able to keep up their prices and performances. Now, the first opposing sounds are heard that Airbnb currently starts to have an impact on the hospitality industry, including a negative impact on the hotel prices (which were assumed to be stable before).20 Yet, a simple deeper look to the price data in this research shows that hotel prices over the past years have not been so stable as previously assumed. Industry reports often only mention the nominal hotel room prices. But when these prices are corrected for local inflation the worldwide average hotel room prices have actually already been decreasing from 2011 onwards (see figure 2 and table 3).
TABLE 3 Average hotel prices
All hotels Low Quality Hotel Medium Quality Hotel High Quality Hotel
2008 £ 106 £ 113 £ 121 £ 163 2009 £ 100 £ 105 £ 111 £ 163 2010 £ 99 £ 100 £ 113 £ 174 2011 £ 103 £ 101 £ 113 £ 175 2012 £ 102 £ 95 £ 110 £ 177 2013 £ 101 £ 94 £ 109 £ 182 2014 £ 98 £ 89 £ 101 £ 171 2015 £ 95 £ 88 £ 104 £ 174
N 52 cities 52 cities 53 cities 52 cities
Notes: Table shows the average hotel prices paid per hotel segment for 53 cities. The hotel prices are from the HPI-reports from Hotels.com and are the actual prices paid by UK inhabitants. The prices are corrected for inflation. The low quality segment contains hotels with 2/3 stars, the medium quality segment contains hotels with 4 stars and the high quality with 5 stars.
18 Because the data reflects the prices paid by UK inhabitants, there might exist a bias due to currency effects. A decrease of a foreign currency towards the GBP would thus decrease the hotel prices in that particular country. To control for these currency effects we include a control variable containing the ratio of PPP (purchase parity power) of the foreign (destination) country divided by the PPP of the UK. This should correct for the currency (dis)advantages that UK inhabitants would face in these countries.
19 Recent articles in i.e. the “NRC Handelsblad” (23-02-2016) and industry reports like the Rabo-bank industry report (23-02-2016) argued that Airbnb does not have a significant impact on hotel performance. Even the Vice President of the Americal Hotel and Lodging
Association argued in a personal email that “The United States just hit its 72 straight month of increased RevPAR which would indicate that
the U.S. hotel industry has improving revenue per available room alongside the presence of Airbnb.” (18-04-2016) Also the CEO of Airbnb,
Nathan Blecharczyk, argued on the Startup-Fest in Amsterdam that Airbnb does not cannibalize the hospitality industry (24-05-2016). 20 E.g. New York Post (18-04-2016) argued that New York hotel prices have decreased in 2015 due to Airbnb. Even the Rabo-bank industry report (11-06-2016) of 4 months later lowered their tone and argued that Airbnb started to become a serious competitor.
III.2 - Airbnb supply
Because the historical Airbnb supply (the number of Airbnb apartments) in a particular city is not publicly available, we try to estimate the supply with a proxy. By the use of a web scraping method, we assemble all data of every currently available apartment of the 53 cities (matched to the HPI-data) from the Airbnb website. This provided us in total with almost 400 thousand unique observations. Every unique observation represents a particular Airbnb apartment/room, that contains information about the number of guests, price, type of room, location and registration date of the host. Using these registration dates we applied backwards reasoning and approximate the number of apartments for each city (e.g. the number of Airbnb hosts in New York that registered before 1 January 2013). Due to the exponential growth of Airbnb growth, cumulative supply is strongly correlated with instantaneous supply (Zervas, Proserpio and Byers, 2016). See Appendix 1 for a further explanation about this approach. Figure 4 shows the average Airbnb listings (of 53 cities) and the standard deviations over the years. It is clear that Airbnb has experienced a quite substantial growth over the recent years, but that this differs a lot between cities. Paris has for instance more than 36.000 listings, while Mumbai has less then 400 listings in 2015.
III.3 - Control variables
In order to disentangle the actual relation of Airbnb supply in a particular city on the price setting behaviour of hotels, we control for industry specific trends (the number tourists, costs
Mean Mean - Std. Dev. Mean + Std. Dev. -2.000 0 2.000 4.000 6.000 8.000 10.000 12.000 14.000 2009 2010 2011 2012 2013 2014 2015 A ve ra ge num be r of A ir bnb li st ing s FIGURE 3
Average Airbnb listings over the years
Graph shows the average number of Airbnb rentals as calculated by the backward reasoning method. The average number of Airbnb rentals was
of traveling), some macroeconomic trends (GDP, unemployment) and a demographic trend (population density). Because the analysis is performed on city-level and city-level data is rather scarce, we were not able to perfectly match the data of the control variables with the dependent variable (hotel prices). Hence, including all control variables reduces completeness of the dataset from 53 cities towards 35 cities.
As hotel prices are largely driven by the demand for a place to sleep, we proxy the demand for overstays by the number of tourist arrivals in a city. We use two estimates. The first estimate is the number of international overnight visitors per city as estimated in the annual “Global Destination Cities Index” reports of Mastercard (2010-2015). The second estimate is the number of international arrivals from the Euromonitor International bi-annual report “Top 100 city destinations” (2008-2014). The number of arrivals of these two estimates were largely in line, but not perfectly equal. Hence, for the overlapping years we used the average of the two measures. This data matches 47 of the cities in the database. Interestingly, Figure 3 shows (apart from 2009) a permanent increase in the average number of tourist arrivals over the past years. Suggesting that the demand for hotels has been increasing worldwide.
Hotel prices might also be driven by local price conditions. Price changes of particular goods/services for hotel customers might differ from the regular inflation rates. For this reason, we included the Corporate Travel Index (sum of hotel, transport and diner) to account for this.
3.000.000 4.000.000 5.000.000 6.000.000 7.000.000 2008 2009 2010 2011 2012 2013 2014 2015 N um be r of tour is t a rr iva ls FIGURE 4
Average number of annual tourist arrivals
Average annual tourist arrivals of 47 cities over the years 2008-2015. Averages of the Global Destinations Index (Mastercard, 2010-2015) and the Top 100 Global City
Obviously, hotel prices were excluded from this index. This data matches to 48 cities in our database.
As macroeconomic variables we included unemployment rates and real GDP (per capita). Unemployment rates are used as a proxy for the economic condition of a country, which might have an impact on the average hotel performance. According to their performance hotels might decide to change their prices (either up - or down), depending on their operational strategy. The population density of a city is included to control for the demand for squared meters in a city. Real GDP (per capita) is used as a proxy for domestic hotel demand. As the real GDP (per capita) is corrected for inflation, an increase implies that the average inhabitant of a city will be able to purchase more domestically. Since hotel bookings can be seen as a luxury good, we might expect that domestic demand for hotels (perhaps especially for high-quality hotels) increases, with real GDP and thus has an upward effect on hotel prices. We also include population density as a proxy for the demand for housing in a city. Whenever population density is strongly increasing, the demand for housing increases and simultaneously the rent and house prices (prices per square meter). Increasing prices for a living-accommodation might be positively related to hotel prices. These macroeconomic variables where collected from the OECD Metropolitan Area database. Unemployment rates and population level included data for the years (2008-2014). The OECD data on real GDP (2008-2013), was complemented with GDP growth rate data (2013-2014) from the “Brookings analysis of data from Oxford Economics, Moody’s Analytics, and U.S. Census Bureau”. Because there is just a limited set of these macroeconomic variables on city-level available, we were able to match this data for 40 cities in our dataset. Furthermore, we also collected country level macro-economic data (Real GPD per capita and unemployment rates) of all the 53 cities. This data was assembled from the “World Economic Outlook” (April 2016) report from International Monetary Fund. This data is particularly helpful as this country macroeconomic data is available for all 53 cities (for all the years).
TABLE 4
Variable overview
Variable name Data description Data source Notes Nr. of cities Mean Std. Dev
Hotel price variables
AvP Average hotel price (2 persons) Hotel Price Index reports Average price of all hotelrooms booked 52 80,3 49,5 P_low Price for low quality hotels (2 persons) Hotel Price Index reports Constructed by the average of 2 and 3 star hotel prices 52 37,3 33,3 P_med Price for med quality hotels (2 persons) Hotel Price Index reports Contains only 4 star hotels 53 62 41,1 P_high Price for high quality hotels (2 persons) Hotel Price Index reports Contains only 5 star hotels 52 105,3 74 RealAvp Real average hotel price (2 persons) Hotel Price Index reports Corrected for inflation 52 75,1 49,4 RealP_low Real price for low quality hotels (2 persons) Hotel Price Index reports Corrected for inflation 52 35,3 32,7 RealP_med Real price for med quality hotels (2 persons) Hotel Price Index reports Corrected for inflation 53 58 40,7 RealP_high Real price for high quality hotels (2 persons) Hotel Price Index reports Corrected for inflation 52 97,7 71,6
Airbnb supply variables
Airbnb Total number of Airbnb listings per city (* 1.000) Webscrape Constructed from registration date host 53 1,942 4,012 Airbnb_s Number of Airbnb apartment listings per city (* 1.000) Webscrape Constructed from registration date host 53 0,043 0,087 Airbnb_p Number of Airbnb private room listings per city (* 1.000) Webscrape Constructed from registration date host 53 0,623 1,389 Airbnb_a Number of Airbnb shared room listings per city (* 1.000) Webscrape Constructed from registration date host 53 1,34 2,807 Airbnb_low Number of Airbnb listings in low price segment (* 1.000) Webscrape All listings with a price below 90% of P_low (hotels) 49 1,419 3,336 Airbnb_med Number of Airbnb listings in medium price segment (* 1.000) Webscrape All listings in between 49 0,511 0,943 Airbnb_high Number of Airbnb listings in high price segment Webscrape All listings with a price above 90% of P_med (hotels) 48 0,17 0,385
Control variables
CPI Consumer Price Index (Inflation) OECD Used for inflation corrections 53 100,4 22,5 PPP_corrected Relative Purchasing Power Parity IMF PPP hotel country / PPP UK 53 169,4 847,4 L_GDP Real GDP per capita per city (Local) OECD & Brookings Analyses 2015 values are missing 40 27955,5 27317,9
L_Unemp Unemployment rates per city (Local) OECD 2015 values are missing 40 4,3 4,6
PopulationDensity Population per squared km per city OECD 2015 values are constructed with growth rates 40 412 724,5 N_GDP Real GDP per capita per country (National) IMF Few 2015 values predicted by IMF 53 35082 21492 N_Umemp Unemployment rates per country (National) IMF Few 2015 values predicted by IMF 49 7,2 4,3 TouristArrivals Number of tourist arrivals per city (* 1.000.000) Euromonitor & Mastercard reports Average arrival numbers of the two reports 47 3477 3814 CTI Average restaurant price in Corporate Travel Index Business Travel News reports - 48 115,8 57,9
Complete analyses Regression analyses (excluding control variables) - - 52/53 - -
Complete analyses Regression analyses (including control variables) - - 35 - -
IV. RESULTS
In a panel data setup, we investigate the relation of Airbnb supply (Airbnbit) on hotelprices (Pit).
To disentangle the actual relational effect from other external influences, we include entity fixed effects ("#), time trends ($%) and several control variables (Wk,it), where i indicates the
index for cities, t for years and k for the control variables. We structure the discussion of the results along the lines of the hypotheses spelled out in Section II. Firstly, we discuss the impact of Airbnb on the average hotel prices. Secondly, we further elaborate on the effect of Airbnb on the hotel prices per quality segment. Lastly, we further identify the direct and indirect impact on prices per hotel quality segment by segmenting Airbnb supply.
IV.1 - Total Airbnb supply on average hotel prices
Hypothesis H1 predicts that an increase in competition (due to the entry of Airbnb) will have a negative effect on the average prices of the incumbent firms (the hotels). To examine this straight forward hypothesis, we use regression equation (1) for our analyses, We predict that &' (coefficient of Airbnb on hotel prices) is negative, taking into account a time trend, entity fixed effects and control variables.
(#% = &* + &',-./0/#% + 1232,#%+ "# + $%+ 5#% 1) Table 5 shows the results from this first analysis (including 52 cities), where a time trend, city fixed effects, country clusters and a HAC-robust adjustment21 are sequentially included. Column (5) reveals that average hotel prices are in direction indeed negatively (-0,020) related to Airbnb, this result is however insignificant. Therefore, this analysis does not directly provide us with evidence that additional competition (due to entry from Airbnb) is negatively related to hotel prices.
In the columns 2-8 of Table 6 several control variables are included. The variables are separately included to adequately observe what the impact of the control variable on hotel price is, and how the inclusion of this variable changes the direction, magnitude or significance of the relation of Airbnb supply on hotel prices.
TABLE 5 Average hotel prices
(1) (2) (3) (4) (5)
RealAvP RealAvP RealAvP RealAvP RealAvP
Airbnb -0.092 -0.13 -0,2 -0.20 -0.20
(0.17) (0.22) (0,22) (0.19) (0.19)
year 0.079 0,17 0.17 0.17
(0.33) -0,33 (1.15) (1.15)
time trend no yes yes yes yes
city fixed effects no no yes yes yes
country cluster no no no yes yes
HAC no no no no yes
obs. 352 352 352 352 352
N of cities 52 52 52 52 52
Notes: Table shows the regression results of Airbnb supply (per 1000) on average hotel prices in a panel data analyses including 52 cities. Column 1 shows the regular panel regression analyses. In Column 2 a time trend is included. Column 3 includes also time fixed effects. Column 4 also includes country clusters. Column 5 includes a correction for heteroskedacity and autocorrelation. Standard errors in parentheses;* p<0.05, ** p<0.01, *** p<0.001
The inclusion of the PPP-correction control variable (column 2) corrects for possible currency (dis-)advantages that might bias the results. This does not substantially change either the direction, magnitude or significance level of the coefficient of the variable of interest (Airbnb). In column (3) the number of tourist arrivals is included, which substantially reduces the negative coefficient of the Airbnb coefficient. Yet, the coefficient of TouristArrivals itself is insignificant. The inclusion of the Corporate Travel Index (CTI) in column (4) even changes the direction of the Airbnb regression coefficient, although it remains insignificant. The coefficient of the CTI is positive and significant, suggesting that high travels costs relate to high average hotel prices. In columns (5) we control for population density (PopulationDensity), which makes the coefficient of Airbnb on hotel prices more negative and significant. The coefficient of Population Density itself is insignificant. In column (6) real GDP per capita (N_GDP) is included, which changes the direction of the Airbnb coefficient, but the coefficient remains insignificant. The estimate of N_GDP itself is positive and significant. As GDP per capita proxies the domestic hotel demand, this suggests that an increase in GDP per capita (of 1000 pound), increases hotel demand and sequentially increases the hotel prices (with £1.50). Unemployment rates (N_Unemp) are included in column (7) and make the Airbnb coefficient, more negative and significant. The estimate for N_GDP is significant and negatively related to the hotel prices. The unemployment rate proxies the state of the economy, which has a negative effect. This indicates that better economic conditions (reduced unemployment rates) increases hotel prices. Although in most regressions negative relation between Airbnb supply and hotel prices is shown, only in the regressions with the sole inclusion of PopulationDensity or Unemployment as control variable, this relation becomes significant. Nevertheless, most
variables seem to either affect the significance or magnitude of the Airbnb coefficient or to have a substantial impact on hotel prices themselves (and thus increasing the precision of the regression). For this reason, it is appropriate to include all variables in the regression. Which we did column (8). Yet, the regression coefficient Airbnb, though slightly negative, remains insignificant. Hence, we cannot confirm the negative relation of Airbnb on hotel prices and thus reject hypothesis H1.
TABLE 6 Average hotel prices
(1) (2) (3) (4) (5) (6) (7) (8)
RealAvP RealAvP RealAvP RealAvP RealAvP RealAvP RealAvP RealAvP
Airbnb -0.20 -0.17 -0.042 0.026 -0.29* 0.17 -0.26* -0.043 (0.19) (0.19) (0.27) (0.21) (0.13) (0.21) (0.11) (0.17) year 0.17 0.19 0.27 0.78 1.71 -0.66 0.54 0.50 (1.15) (1.15) (1.37) (1.29) (1.73) (0.85) (1.00) (1.19) PPPcorrection -0.084*** -0.88 -1.02 -0.77 -0.083*** -0.084*** -0.46 (0.0065) (0.92) (0.90) (0.72) (0.0045) (0.0048) (0.73) TouristArrivals -0.63 -1.08 (1.56) (1.42) CTI 0.11* -0.027 (0.045) (0.025) PopulationDensity -0.13 -0.00095 (0.11) (0.068) N_GDP 0.0015*** 0.0015*** (0.00027) (0.00022) N_Unemp -1.93** -0.78* (0.65) (0.28) obs. 352 352 307 328 273 352 334 228 N. of cities 52 52 47 48 40 52 49 35 Notes: Table shows the regression results of Airbnb supply (per 1000) on average hotel prices in a panel data analyses including 53 cities. All regressions include a time trend, city fixed effects, country clusters and are corrected for heteroskedacity and autocorrelation. Standard errors in parentheses;* p<0.05, ** p<0.01, *** p<0.001
IV.2 - Total Airbnb supply on hotel prices per quality segment
In the further analyses we assess the impact of Airbnb on the different hotel segments. We hypothesized that Airbnb has a negative impact on the hotel prices in the medium and low segment (hypothesis H3) and no significant impact on those in the high segment (hypothesis H2). Therefore, for the regression equations (2, 3 and 4) below we respectively expect that β2
< 0, that β3 < 0 and that β4 = 0. Besides, as stated in hypothesis H4 we expect that PM is more
negatively associated with Airbnb supply than PL, suggesting that β
(#%6 = & 7,8#%+ 1232,#%+ "# + $%+ 5#% 2) (#%9 = & :,8#%+ 1232,#%+ "# + $%+ 5#% 3) (#%; = & <,8#% + 1232,#%+ "# + $%+ 5#% 4)
The results in Table 7 show the regression estimates for the different hotel segments, where column 1-3 show respectively the results for the low, medium and high quality segment. Columns 4-6 show the same results, but now with inclusion of the PPP correction control variable. In column 7-9 the remaining control variables are included.
We hypothesized that additional competition is negatively related to incumbent prices. Since hotels in the low and medium quality segment face direct competition from Airbnb, we expect this is especially the case for these hotels. As hotels in the high quality segment are less likely to face direct competition and are generally less price responsive to competition, we do not expect a significant impact. The results in column 1-3 of Table 7 show a negative Airbnb coefficient for all hotel quality segments, but this coefficient is only significant for the medium quality hotel segment. This results partly confirm hypothesis H3 and hypothesis H2. It seems that prices of the hotels in the low quality segment did not decrease due to increased completion from Airbnb. Arguably, this is because the low profit margins on prices simply do not let them. Prices of medium quality hotels are negatively with Airbnb supply at a 5% significance level. In particular, an increase with 1000 Airbnb rentals in a city decreases hotel prices in the medium quality segment with £0.85.
To illustrate the magnitude of this result, we take Amsterdam as an example. In 2015 there were more than 2500 new Airbnb listings in Amsterdam. This would relate to a decrease in the average hotel price of more than 2 pounds (2,5 * £0,85). The average medium quality hotel price in Amsterdam in 2015 was about £93; the increase in Airbnb supply in 2015 is thus related to a price decrease in the medium quality hotel segment of approximately 2,3%. Or differently speaking; in 2015 there were approximately 10.000 Airbnb rentals in Amsterdam. If these Airbnb rentals were not present hotel prices for medium quality hotels could have been approximately £8,50 (almost 10%) higher.
TABLE 7 Hotel prices per segment
without control variables with the PPP correction with control variables
(1) (2) (3) (4) (5) (6) (7) (8) (9)
RealP_low RealP_med RealP_high RealP_low RealP_med RealP_high RealP_low RealP_med RealP_high
Airbnb -0.33 -0.85* -0.58 -0.32 -0.83* -0.57 -0.22 -0.81* -1.43*** (0.18) (0.38) (0.63) (0.18) (0.38) (0.63) (0.13) (0.28) (0.35) year -0.34 0.056 0.30 -0.33 0.13 0.35 0.25 0.38 0.74 (0.86) (1.19) (1.28) (0.86) (1.18) (1.29) (0.75) (0.46) (0.69) PPPcorrection -0.035*** -0.046*** -0.021*** 0.090 0.019 0.037 (0.0034) (0.0043) (0.0057) (0.46) (0.42) (0.82) TouristArrivals -1.29 -2.81*** -0.38 (0.94) (0.44) (2.24) CTI 0.0051 -0.022 -0.047 (0.023) (0.034) (0.091) PopulationDensity 0.00039 0.13** 0.30** (0.034) (0.032) (0.089) N_GDP 0.00058** 0.0010*** 0.0027** (0.00020) (0.00014) (0.00072) N_Unemp -0.71* -1.26* -1.13 (0.27) (0.47) (0.64) obs. 325 369 326 325 369 326 211 232 199 N of cities 52 53 52 52 53 52 36 36 35 Notes: Table shows the regression results of Airbnb supply (per 1000) on average hotel prices in a panel data analyses including 53 cities. All regressions include a time trend, city fixed effects, country clusters and are corrected for heteroskedacity and autocorrelation. Standard errors in parentheses;* p<0.05, ** p<0.01, *** p<0.001
