The Rise of the Sharing Economy:

∗
Estimating the Impact of Airbnb on the Hotel Industry

Georgios Zervas Davide Proserpio

Questrom School of Business Marshall School of Business

Boston University University of Southern California

John W. Byers

Computer Science Department

Boston University

Last revised: November 18, 2016

First draft: December 14, 2013

Abstract
Peer-to-peer markets, collectively known as the sharing economy, have emerged as
alternative suppliers of goods and services traditionally provided by long-established in-
dustries. We explore the economic impact of the sharing economy on incumbent rms
by studying the case of Airbnb, a prominent platform for short-term accommodations.
We analyze Airbnb's entry into the state of Texas, and quantify its impact on the Texas
hotel industry over the subsequent decade. We estimate that in Austin, where Airbnb
supply is highest, the causal impact on hotel revenue is in the 8-10% range; more-
over, the impact is non-uniform, with lower-priced hotels and those hotels not catering
to business travelers being the most aected. The impact manifests itself primarily
through less aggressive hotel room pricing, an impact that benets all consumers, not
just participants in the sharing economy. The price response is especially pronounced
during periods of peak demand, such as SXSW, and is due to a dierentiating feature
of peer-to-peer platforms  enabling instantaneous supply to scale to meet demand.

∗
The authors thank the participants and organizers of SCECR '13, WISE '13, Platform Strategy '14,
ACM EC '15 and the seminar participants at Telefonica I+D Research Barcelona, Technicolor Research
Paris, Simon Business School, and Microsoft Research New England for their helpful feedback. We thank
Smith Travel Research (STR) for sharing data with us. We are also indebted to Flavio Esposito for motivating
us to investigate Airbnb and for his contributions to our earlier research on the topic.

1
1 Introduction

The emergence of peer-to-peer platforms, collectively known as the sharing economy, has

enabled individuals to collaboratively make use of under-utilized inventory via fee-based

sharing. Consumers have so far enthusiastically adopted the services oered by rms such

as Airbnb, Uber, Lyft, and TaskRabbit. The rapid growth of peer-to-peer platforms has

arguably been enabled by two key factors: technology innovations and supply-side exi-

bility. Technology innovations have streamlined the process of market entry for suppliers,

have facilitated searchable listings for consumers, and have kept transaction overheads low.

Supply-side exibility is another hallmark of these platforms: Uber drivers can add or re-

move themselves from the available supply of drivers with a swipe on an app, and similarly

other suppliers can readily list and de-list the selection of goods or services they have on

oer.

In our work, we focus on the impacts that these peer-to-peer platforms have on incumbent

rms, specically focusing on the case of Airbnb, a provider of travel accommodation and

a pioneer of the sharing economy. With Airbnb having served over 50 million guests since
1 2
it was founded in 2008, and a market capitalization eclipsing $30 billion, we hypothesize

that Airbnb has a measurable and quantiable impact on hotel revenue in aected areas.

Our hypothesis is that some stays with Airbnb serve as a substitute for certain hotel stays,

thereby impacting hotel revenue, and that this impact is dierentiated: by geographic region,

by hotel market segment, and by season. Incumbent rms, despite both facing higher xed

costs and oering less personalized products than peer-to-peer platforms, have only recently

started to take competition from platforms like Airbnb as a serious threat. For example,

hotel executives have publicly issued largely dismissive statements regarding competitors

like Airbnb, arguing that these peer-to-peer platforms are either a niche market or that they

target complementary market segments from that targeted by hotel chains. Interestingly,

Airbnb appears to also espouse this latter view: according to Airbnb, in many cities, over
3
70% of Airbnb properties are outside the main hotel districts, suggesting complementarity

of their oerings.

In this paper we provide empirical evidence to this debate by studying the dierentiated

impact of Airbnb's entry in the Texas hotel market on hotel room revenue. Our study ex-

plores the relationship between Airbnb and hotels in the state of Texas by estimating monthly

1 See http://blog.airbnb.com/wp-content/uploads/2015/09/Airbnb-Summer-Travel-Report-
1.pdf.
2 See http://www.wsj.com/articles/airbnb-raises-850-million-at-30-billion-valuation-
1474569670.
3 See http://blog.airbnb.com/economic-impact-airbnb/.

2
hotel room revenue as a function of Airbnb entry in the market. Using data we collected

from Airbnb, monthly hotel room revenue from approximately 3,000 hotels in Texas dating

back to 2003, and several other auxiliary datasets to compile controls (described in detail

in Ÿ 2 of the paper), we quantify the extent to which Airbnb's entry to the accommodation

market has negatively impacted hotel room revenue.

To identify the causal impact of Airbnb on hotel revenue, we employ a dierence in dif-

ferences (DD) empirical strategy. Specically, due to the signicant variability in both the

temporal rate and the spatial density of Airbnb adoption in Texas, as well as the geographic

specicity of both our hotel and Airbnb datasets, we are able to treat Airbnb market en-

try as a variable intervention in space and time against the hotel room revenue data. Our

DD strategy identies the Airbnb treatment eect by comparing dierences in revenue for

hotels in cities aected by Airbnb before and after Airbnb's entry, against a baseline of dif-

ferences in revenue for hotels in cities unaected by Airbnb over the same period of time.

To perform the analysis, we regress against two measures of Airbnb supply: a cumulative

measure that denes supply as all listings appearing prior to a given date in a given city,

and an instantaneous measure that denes supply as those Airbnb listings active within a

short (e.g., 3-month) time period. In all our specications, we include a rich set of controls

that vary by location and over time: population, wages, unemployment, total hotel room

supply in each market, each hotel's own capacity over time, airport passenger counts, and

the TripAdvisor ratings for each hotel as a proxy for quality. In addition to these measured

covariates, we include city-specic trends, and city-month dummies to account for seasonal

variation in demand accross dierent market. Using our preferred cumulative specication,

we nd that, in Texas, each additional 10% increase in the size of the Airbnb market resulted

in a 0.39% decrease in hotel room revenue, with similar, but somewhat smaller estimated im-

pacts using the instantaneous supply measure. These eects are primarily driven by Austin,

where Airbnb inventory has grown extremely rapidly over the past few years, resulting in an

estimated revenue impact of 8-10% for the most vulnerable hotels in Austin.

We next investigate the market response to Airbnb entry in Ÿ 3.6, and study the mecha-

nisms whereby aected hotels might react to Airbnb's market entry both in the short-term

and in the long-term. In the short-term, likely responses could take the form of a price

response or an occupancy response. Using hotel industry performance metrics as dependent

variables, we nd a small decrease in occupancy rate and a signicant decrease in hotel room

prices. Notably, such a price response benets all consumers, not just participants in the

sharing economy. With respect to longer-term responses, such as diminished investment or

hotel entry and exit, we do not nd evidence of an eect yet, consistent with evidence we

present showing that the timescale of such a response would occur with a multi-year lag.

3
 Our next set of results, presented in Ÿ 4, develops a more nuanced understanding of the

mechanisms behind Airbnb's impact on hotel room revenue by unpacking the eects to study

the dierentiated impacts that Airbnb has had across hotels, cities, and time. First, given

the nature of rentals on Airbnb today, which typically provide fewer amenities and services

than many hotels, we expect those hotels providing more dierentiated services to be less

aected. We examine three such cases in Ÿ 4.1: high-end hotels, chain hotels, and hotels

catering to business travelers, each of which provide amenities that a typical Airbnb host does

not. First, after segmenting hotels in ve industry-standard price tiers (Budget, Economy,

Midprice, Upscale, and Luxury) we nd the impact of Airbnb is gradually magnied as we

move down the price tiers. Then, through a similar analysis, using conference and meeting

room space as a proxy for the extent to which a hotel caters to business travel, we nd that

the impact of Airbnb also falls disproportionately on those hotels lacking conference facilities.

Finally, we examine Airbnb's dierential impact on chain hotels versus independent hotels,

and conrm our expectation that chain hotels will be less aected than independents, for

reasons ranging from larger marketing budgets and stronger brands, to providing predictably

consistent service.

In our nal main result, we study the impact that Airbnb has during periods of peak

demand, leveraging our instantaneous measure of supply. Use of this measure enables us rst

to conrm that there are signicant seasonal uctuations in city-level Airbnb supply that

are correlated with periods of peak demand in those cities. In Ÿ 4.2, we study the impacts

that Airbnb has exerted, year-over-year, during the highly popular SXSW festival in Austin,

and during the Texas State Fair in Dallas. Our nding is that Airbnb's ability to exibly

scale instantaneous supply in response to seasonal demand has signicantly limited hotels'

pricing power during periods of peak demand. Indeed, we argue that accommodating surges

in demand through exible scaling of supply is a dening feature of the sharing economy,

and we interpret our result as evidence of the power of this capability, which appears dicult

for incumbent rms like hotels to directly counteract.

Finally, we mention several robustness checks that we conduct to support a causal in-

terpretation of our estimates, detailed in Ÿ 5. First, we show that the basic set of controls

included in our DD specication (i.e., hotel xed eects and time trends) explain approxi-

mately 88% of the variation in Airbnb supply, whereas time-varying observables that could

potentially drive hotel revenue have almost no additional explanatory power. Second, we

check whether Airbnb adoption is driven by hotel performance, which would be a case of our

confusing cause and eect. To the contrary, we nd that a wide range of pre-Airbnb demo-

graphic and market characteristics  including, for example, hotel room prices, occupancy

rates, and hotel room supply per city, which are all signicant predictors of post-Airbnb

4
hotel room revenue  are not correlated with the patterns of Airbnb adoption we see in

our data. Third, we dene a measure of competing Airbnb supply at a per-hotel granular-

ity, accounting for the geographic distance between the hotel and Airbnb inventory. This

distance-based analysis shows a magnied negative impact from Airbnb on hotels as prox-

imity between hotels and Airbnb inventory increases. Fourth, we show that our results are

robust to alternative measures of Airbnb supply. Finally, in a separate analysis, we combine

DD with coarsened exact matching (Iacus et al., 2012). Specically, we match each treated

hotel aected by Airbnb to a control hotel belonging to the same price-tier and sharing the

same aliation, discarding hotels that remain unmatched (e.g., an upscale Hilton in Austin

where Airbnb adoption is high, and an upscale Hilton in Dallas where Airbnb penetration

is low.) We nd that our CEM estimate is similar to our main analysis. Taken together,

these robustness checks provide signicant support for the assumptions underlying our DD

analysis.

We conclude in Ÿ 6 by discussing managerial and policy implications related to the rapid

growth of Airbnb specically and the sharing economy more broadly.

1.1 Related Work

Relatively few papers have yet studied competition betweeen peer-to-peer markets and in-

cumbent rms oering similar goods or services. In one line of recent work, (Einav et al.,

2016) discuss the design and regulation of peer-to-peer markets, and provide theoretical

predictions of the eects of competition from these markets on incumbent rms. A key

prediction they make, that is borne out in our data, is that peer-to-peer markets can reduce

price variability by exibly scaling supply to accomodate increased demand. As for empirical

work, a handful of studies have examined the adoption and eects of car-sharing; for example,

two studies have used survey analysis methods to nd that car-sharing is associated with

signicant decreases in miles traveled, gasoline consumption, and car ownership (Cervero

et al., 2007; Martin et al., 2010). In the domain of accommodation sharing, we nd a large

number of opinion pieces in the popular press and on blogs, but little in the way of academic

literature. Our closest comparison point is a set of short studies, commissioned by Airbnb,

which claim that the Airbnb business model is complementary to the hotel industry, but

primarily focus on arguing for and quantifying the substantial net economic benet to cities
4
that Airbnb travelers provide. While our work is related to these studies, we apply a more

sophisticated identication strategy, methodology, and segmentation analysis, resulting in

conclusions that are both dierent and more nuanced. Notably, recent analyses have con-

4 See: https://www.airbnb.com/economic-impact/

5
rmed our initial ndings in Texas in other markets; for example, Credit Suisse analysts

used STR data to estimate that in New York City, January 2015 revenue per hotel room
5
was impacted by 18.6%, year-over-year. .

Our work contributes to the growing literature on multi-sided platform competition, as

Airbnb exemplies a two-sided platform. Much of this literature establishes the economic

theory of two-sided markets, for example through structural models that establish theories of

price structure and usage (Rochet and Tirole, 2003; Rysman, 2009; Weyl, 2010), and models

which connect innovations in product design to network eects (Parker and Van Alstyne,

2005). Other work, more closely related to our own, contributes empirical results to the liter-

ature that seek to explain the behavior of rms and individuals in two-sided markets (Jin and

Rysman, 2012), including the role of multihoming (Landsman and Stremersch, 2011), mod-

eling response to regulation (Carbó Valverde et al., 2010), and understanding the supply-side

labor market (Hall and Krueger, 2015). Our work, in contrast to these, empirically studies

a setting where a peer-to-peer market oers a substitute for consumer services supplied by

traditional rms.

It is in this latter context that our work contributes to the literature on substitution

between peer-to-peer markets and incumbent rms, as markets like Airbnb can be viewed as

providing enabling technology that facilitates suppliers of niche inventory to exibly bring

their products to market. In contrast to traditional markets, Airbnb provides suciently low

cost of revenue for individuals to protably list remnant inventory online; moreover, Airbnb

provides enhanced reach by reducing consumer search costs (Bakos, 1997). As such, our

study can be viewed as investigating the consequences of an online platform lowering the

barrier to entry for suppliers. Related work has studied similar examples in other domains.

For example, a number of recent studies have focused on the impact of Craigslist  a website

featuring free online classied ads  on the newspaper industry (Kroft and Pope, 2014;

Seamans and Zhu, 2013).

Finally, our work contributes to the literature studying the impact of external shocks

on the tourism and the hospitality industry. Much of the prior work though, has centered

on demand shocks. For example, O'Connor et al. (2008) study the impact of terrorism on

tourism in Ireland; Baker and Coulter (2007) estimate the impact of the 2002 and 2005

terrorist attacks in Bali on the islands' vendors. Similarly, Kosová and Enz (2012) examine

the adverse eects of the 9/11 attack and the 2008 nancial crisis on hotel performance.

5 See New York City hotel rooms are getting cheaper thanks to Airbnb at http://qz.com/341292

6
2 Data and the Airbnb Platform

For our study, we collect and combine data from various sources including the Airbnb website,

the Texas Comptroller Oce, Smith Travel Research (STR), county demographics from

the U.S. Census Bureau, airport passenger counts from the U.S. Bureau of Transportation

Statistics (BTS), the Current Population Survey (CPS) from the U.S. Bureau of Labor

Statistics (BLS), and hotel reviews from TripAdvisor.

2.1 The Airbnb Platform

Much of the data used in our study is collected directly from the Airbnb website. Airbnb

describes itself as a trusted community marketplace for people to list, discover, and book

unique accommodations around the world, and exemplies a peer-to-peer marketplace in

the sharing economy. Prospective hosts list their spare rooms or apartments on the Airbnb

platform, establish their own nightly, weekly or monthly price, and oer accommodation to

guests. Airbnb derives revenue from both guests and hosts for this service: guests pay a

9 − 12% service fee for each reservation they make, depending on the length of their stay,

and hosts pay a 3% service fee to cover the cost of processing payments. Since its launch in

2008, the Airbnb online marketplace has experienced very rapid growth, with more than two

million properties worldwide and over 50 million guests that used the service by September
6
2015.

Airbnb's business model currently operates with minimal regulatory controls in most

locations, and as a result, hosts and guests both have incentives to use signalling mechanisms

to build trust and maximize the likelihood of a successful booking. To reinforce this behavior,

Airbnb has built an online reputation system that enables and encourages participants to

rate and review each completed stay. Guests use star ratings to rate features of their stay,

e.g., cleanliness, location, and communication, while both guests and hosts are encouraged
to post public reviews of each stay on the platform.

2.2 Airbnb Listings Data

To estimate the extent of Airbnb's market entry, we collected consumer-facing information

from airbnb.com on the complete set of users who had listed their properties in the state

of Texas for rental on Airbnb. We refer to these users as hosts, and their properties as
their listings. Each host is associated with a set of attributes including a photo, a personal
6 See http://blog.airbnb.com/wp-content/uploads/2015/09/Airbnb-Summer-Travel-Report-
1.pdf.

7
statement, their listings, guest reviews of their properties, and Airbnb-certied contact in-

formation. Similarly, each listing displays attributes including location, price, a brief textual

description, photos, capacity, availability, check-in and check-out times, cleaning fees, and

security deposits. Figures 2 and 3 display a typical Airbnb listing, and a typical Airbnb user

prole, respectively. Our collected dataset contains detailed information on 10, 555 distinct

hosts and 13, 935 distinct listings spanning a period from 2008 to August 2014.
To conduct our analysis, we must choose an appropriate level of geographic aggregation.

Here, our data is suitably granular (with location accuracy to roughly 100 meters) to permit

analysis at many dierent scales. Our preferred specication employs city-level granularity,

and is driven by the observation that a city is the largest geographic unit within which

we reasonably expect to see signicant substitution patterns between hotels and Airbnb

properties. However, distance-based measures also arguably have operational validity. We

discuss these along with our other modeling decisions and robustness checks.

Another central element of our analysis is to accurately quantify Airbnb supply; however,

this cannot be directly inferred from available data, and is thus a highly nuanced modeling

decision. Indeed, inferring instantaneous Airbnb supply is a challenging task even for Airbnb

itself due to stale vacancies, i.e., Airbnb listings that appear to be part of available supply
only because the hosts neglected to update the availability status of those listings. By

analyzing proprietary Airbnb data, Fradkin (2014) nd that between 21% and 32% of guest

requests are rejected due to this eect.

Despite imperfect information, we do have substantial data with which to construct

proxies for supply, namely the date that hosts became Airbnb members, and the date for

each review of each property. Signicantly, Fradkin et al. (2014) report that 67% of Airbnb

guests left a review about their stay across their large dataset. For market entry, we can

estimate the (unobservable) entry date of individual listings either by using the date their

owners became Airbnb members or by the date of the rst review. Similarly, we can construct

proxies for both cumulative and instantaneous supply by leveraging the review histories we

compile. We detail and justify our approach in Section 3.2.

2.3 Hotel Data: Revenue, Prices, and Occupancy Rates

The main dependent variable we use in our analysis is monthly hotel room revenue, which

we obtained from public records furnished by the Texas Comptroller of Public Accounts,
7
in their capacity as auditors of state tax collection. In addition to monthly hotel room

revenue, the dataset includes basic information including hotel name, address, and capacity.

7 Available at http://aixtcp.cpa.state.tx.us/hotel/hotel_qtr_all_srch.php

8
The raw dataset spans the period between Jan. 2003 and Aug. 2014.

Interestingly, according to Texas law, a hotel is considered to be any building in which

8
members of the public rent sleeping accommodations for $15 or more per day. For this

reason, revenue from Airbnb properties (as well as various other vacation rental options)

whose owners are in compliance with the Texas tax code is also reported in this dataset.

This is evident from Figure 4, which plots the number of unique tax-paying properties in

Austin broken down by capacity, i.e., maximum occupancy. We conjecture that the rapid
increase in low capacity properties starting in 2008 is related to Airbnb's entry into the

Texas market at the same time. To exclude non-hotel properties from our analysis of impact

on hotels, we cross-reference the Texas Comptroller dataset with the U.S. hotel census data

provided to us by STR. The STR census includes all U.S. hotels and contains a rich attribute

set for each hotel, including its opening date, price segment, capacity, operation type (chain

vs. independent), and geographic location. In total, the STR dataset contains information

on 3, 747 hotels in Texas metropolitan areas. After linking the STR census dataset with the

Texas tax dataset, we obtain high-condence matches for a panel of 3, 619 properties (96%

of STR hotels, which account for over 95% of the revenue in our data).

Airbnb can aect hotel room revenue through lower occupancy rates, decreased hotel

room prices, or a combination of these two factors, conventionally reported within the hotel

and hospitality industry as RevPAR (revenue per available room), which is the product of

average room price and occupancy. Because the data we obtained from the Comptroller's

oce does not report either occupancy rates or hotel room prices, we obtain additional data

on these quantities for a subset of Texas hotels from STR. The room price (also referred to

as average daily rate, or ADR in the industry) and occupancy rate data from STR covers a

subset of 2, 584 hotels in Texas who chose to report this information to STR over the same

time period (Jan. 2003 to Aug. 2014).

2.4 Auxiliary data sources

We assemble a set of control variables derived from publicly available sources. First, for

each hotel we collect its entire TripAdvisor review history  a total of 424,583 reviews. We

then use TripAdvisor star ratings to control for changes in hotel quality over our observation

period. Second, we collect passenger arrival data for all Texas airports from the BTS. We

then associate each city in Texas with its nearest airport, and use the passenger data to

control for changes in tourism demand over time that are unrelated to Airbnb. The data

is a monthly panel of passenger counts, in which we exclude passengers connecting through

8 See http://www.window.state.tx.us/taxinfo/hotel/faqhotel.html.

9
Texas airports. Third, we obtain monthly unemployment and wage data at the MSA level

from the BLS at bls.gov. Unemployment statistics are updated monthly, while the wage

data, which comes from the Occupational Employment Statistics Survey, is updated once a

year. Finally, we obtain demographic information at the county level from the U.S. Census

Bureau at census.gov.

3 Quantifying Airbnb's Impact in Texas

3.1 Empirical Strategy

Airbnb has seen widely varying degrees of traction within dierent local, regional and in-

ternational markets, both with respect to initial market entry and the rate at which it has

been adopted within markets. For example, consider Figure 1, which depicts the current

extent of market penetration both of Airbnb properties and hotels within the state of Texas

(top panels), and within the county encompassing the state capital, Austin (bottom panels).

Unlike hotels, which have coverage throughout the state, and pockets of local density, such as

in downtown Austin, Airbnb has spotty coverage at best throughout the state, but broader

coverage across metro areas, including suburbs and exurbs. Table 1 reveals that patterns

of Airbnb adoption, over the past eight years in the ten most populous cities in Texas, are

themselves diverse, with several cities experiencing early adoption and rapid growth, while

others experienced minimal Airbnb adoption. Our empirical strategy exploits this variability

to identify the impact of Airbnb's rise on hotel room revenue using a dierence in dierences

(DD) identication strategy. Specically, we estimate Airbnb's impact on hotel room rev-

enue by comparing changes in hotel room revenue before and after Airbnb enters a specic

city, against a baseline of changes in hotel room revenue in cities with no Airbnb presence

over the same period of time.

The key identication assumption we have to make to support a causal interpretation of

this DD estimate is that there are no unobserved, time-varying, city-specic factors that are

correlated with both Airbnb entry and hotel room revenue, resulting in endogeneity. Stated
dierently, we assume that unobserved factors that could potentially jointly aect both

Airbnb adoption and hotel room revenue do not systematically vary both between dierent
cities and over time. For instance, the following unobserved factors are accounted for in our
estimate and do not bias our estimates: 1) city-specic time-invariant dierences in adoption

rates (e.g., consumers in Austin overall being more likely to adopt Airbnb than consumers

in Dallas); 2) factors that vary arbitrarily over time but do not vary across cities (e.g., a

generally increasing awareness of Airbnb shared across all consumers in Texas over time),

10
and, 3) city-specic trends, which allow for unobserved confounders that vary both between

cities and over time according to a pre-specied functional form (linear or quadratic).

Our DD specication takes the following form:

0
log Hotel Revenueikt =β log Airbnb Supplykt + Xikt γ (1)

+ hi + τt + Cityk × Montht + ikt .

The dependent variable is the log of monthly room revenue of hotel i in city k at time t. Our

model includes hotel xed eects hi , and time (year-month) xed eects τt . To implement

the DD strategy, we dene treated hotels to be those hotels in cities with an Airbnb presence,

and non-treated hotel to be those hotels in cities with no Airbnb presence. The rst dierence

is taken using the hotel xed eects, which allow for time-invariant dierences in hotel room

revenue between treated hotels and non-treated hotels. The second dierence in our DD

specication is taken over time using year-month xed eects τt which allow for unobserved

time-varying revenue dierences that are common across dierent cities. The coecient of

interest is β, which has the usual DD interpretation: it is an estimate of the percentage

change in hotel room revenue in treated (Airbnb-adopting) cities subsequent to Airbnb's

entry compared against a baseline of changes in hotel room revenue over the same time

period in untreated (non-adopting) cities. We interpret a statistically signicant negative

coecient on Airbnb supply as indicating that Airbnb listings lead to Airbnb bookings that

substitute for hotel stays and impact hotel room revenue. We interpret a coecient that is

not statistically signicantly dierent from zero as indicating that Airbnb listings having no

eect on hotels. We interpret a positive coecient, though implausible, as indicating that

Airbnb listings benet hotels. Next, we elaborate several measures of Airbnb supply that

we employ in Equation 1, and the various economic impacts each measure can identify.

3.2 Modeling Airbnb Supply

Our rst approach uses a cumulative measure of Airbnb supply, quantied at the granularity

of individual cities: for a given city and date, we count the number of distinct listings that

have cumulatively appeared on Airbnb in that city prior to that date. We approximate

the unobservable entry date of individual listings by using the displayed date their owners

became Airbnb members. By construction, a weakness of the cumulative measure of Airbnb

supply is that it ignores listing exit, which we do not observe in our data. Therefore, our

estimate of Airbnb's impact will be consistent if the unobserved fraction of active Airbnb
listings is not endogenously correlated with cumulative listing supply and hotel revenue.

To demonstrate when the (observed) cumulative supply and (unobserved) actual monthly

11
supply yield the same consistent estimate, we relate cumulative supply to actual supply

through a set of (unobserved) multipliers fkt ∈ [0, 1] such that Actual Airbnb Supplykt =
fkt × Cum. Airbnb Supplykt . Here, fkt is the fraction of Airbnb listings that entered the

market prior to time t and are still actively in the market at time t. Because we work

with a log-log specication, fkt becomes an unobserved quantity that enters the error term

additively. Therefore, only residual variation in fkt after controlling for observables, xed

eects, and trends that is correlated with residual cumulative supply, will cause bias.

Our second approach employs an instantaneous proxy measure of actual Airbnb supply.

To build an instantaneous measure, we exploit the fact that Airbnb requires guests who wish

to submit a review to do so within 14 days of a stay and reports the check-out date (with

monthly precision) in each review, thus listings that receive a review must be on the market

at that time. Moreover, the incidence of reviewing is high: Fradkin et al. (2014) report that

67% of Airbnb stays in their large dataset resulted in a review. Taken together, these two

facts indicate that a time-series of Airbnb reviews reects time-varying supply. For each

Airbnb listing in our data, we observe its entire historical record of reviews, which includes

reviews for the listing, as well as reviews for each guest (by the host). Using the review

dataset, we apply the following heuristic to determine when each Airbnb listing was active:

when an Airbnb listing enters the market we assume that it remains active for m months,

which we refer to as the listing's time-to-live (TTL); whenever a listing is reviewed, its TTL

is extended by m months from the date of the review; if a listing exceeds its TTL, it exits

the market; nally, listings become active again after exiting the market if they receive a

new review.

The main advantage of the instantaneous supply measure is that it can capture a key

dierentiating feature of Airbnb, its ability to scale supply. This measure has both descriptive

value and allows us to conrm that our results are not driven by our choice of a cumulative

supply measure. A limitation of the instantaneous supply measure, arising from the way

we construct it, is that it may underestimate Airbnb inventory in low season. During low

season Airbnb listings face lower demand, which in turn leads to fewer reviews. Therefore,

during low season, some listings that are available may receive zero reviews and thus be

misclassied as unavailable.

Figure 6 compares the cumulative and instantaneous Airbnb supply measures for the four

biggest cities in our data. We see that our instantaneous Airbnb supply measure uctuates

signicantly over time, dierentiating it from our cumulative supply measure. Moreover, its

pattern of variation over time correlates with periods when we would expect Airbnb supply

to be highest, such as March in Austin, when the SXSW festival takes place.

A nal issue that pertains to both measures of Airbnb supply that we have to deal with is

12
that the unit of analysis is hotel monthly room revenue, but the treatment, Airbnb adoption,

occurs at the city level. This mismatch in the level at which we measure our dependent

variable compared to the treatment variable can result in understating the standard error

of the estimate of Airbnb's impact, because it is likely that hotel room revenue is serially

correlated over time within a city. We correct for this mismatch by clustering standard errors

at the city level, which lets us account for possible serial correlation in hotel room revenue.

In doing so, we follow the standard practice in the literature for analyzing panel data in

a DD setting (Bertrand et al., 2004; Donald and Lang, 2007). We report standard errors

clustered at the city level for all subsequent regressions.

3.3 Incorporating Controls: Hotel Supply & Quality, Demand Shifters,

and Demographics
An initial identication challenge we face is that increased demand for accommodation is

likely correlated with increases in both Airbnb supply and hotel room supply. Concretely, it

is plausible that over our decade-long observation period, hotel rms have been strategically

developing new properties in areas of anticipated high demand. As high demand could

also correlate with increased Airbnb adoption, this pattern of competition could bias our

estimation, because city-specic increases in hotel room supply could drive per-hotel room

revenue down, and this eect could be misattributed to increased Airbnb adoption. To guard

against this concern, we construct a control variable Hotel Room Supply−ikt , which measures
the total supply of hotel rooms in the same city as hotel i (but excluding hotel i itself, thus the

−i in the subscript), for each time t. To construct this variable, we rely on the same monthly
panel of tax reports provided by the Comptroller as, in addition to revenue, taxpayers have to

report the capacity of their properties with each ling. Therefore, Hotel Room Supply−ikt
captures changes in competitors' total room supply over time including changes resulting

from hotels expanding or shrinking, and entering or exiting the market. This control, which

we also incorporate in Xikt , allows for increases in the supply of hotel rooms provided by

competitors to impact the room revenue of each hotel in our data, much as we hypothesize

an increase in Airbnb rooms does. In addition, we control for hotel i's own capacity and

quality over time, both of which may change, for instance, following renovations. We derive

hotel capacity from the tax data, and we use TripAdvisor ratings as a proxy for quality.

Second, as we explained earlier, our DD estimate will be biased if there exist unobserved

factors that vary across cities and over time, and which jointly inuence Airbnb entry and

hotel room revenue, most notably demand for accommodation. This type of bias likely works

against nding a negative Airbnb eect: both Airbnb supply and hotel revenue should

13
respond positively to shifts in accommodation demand, which implies that if we omit a

control for demand, then Airbnb supply will absorb its eect and become biased upwards.

We use three types of controls to account for variation in accommodation demand across

dierent cities. First, we include quadratic city-specic trends as a control in Xikt . The

inclusion of these trends relaxes the DD assumption of no cross-city time-varying unobserv-

ables that are correlated with both Airbnb supply and hotel revenue. A concern with the

inclusion of city-specic time-trends is that they can be confounded with hotels' response

to Airbnb (Wolfers, 2006). Fortunately, our dataset covers a long pre-Airbnb period from

2003 to 2008, allowing us to estimate these trends on a large sample of pre-treatment ob-

servations. Second, we include city-month (e.g., Austin-March) xed eects to control for

dierences in seasonal demand patterns across the dierent cities. For instance, March in

Austin is especially popular due to the SXSW festival. The city-month xed eects control

for such seasonal dierences. Finally, we associate each city in our data with the nearest

airport and use the (log of ) the number of passengers disembarking at that airport as their

nal destination as a control.

A further issue relates to the unobserved incentives of consumers who choose to list their

homes on Airbnb. For example, Airbnb touts the help it provides to struggling or unemployed
9
homeowners in paying their mortgage. . Conceivably, an increase in the unemployment rate

could simultaneously drive Airbnb adoption and independently cause demand for hotels to

soften. Therefore, failure to control for cross-city dierences in the demographics could

potentially bias our estimation. In this case, the bias likely works in favor of nding a

negative Airbnb impact. To address this concern, we incorporate unemployment rate, the

median annual wage, and population as a controls in Xitk .

3.4 Identication checks

Before proceeding with estimation, we conduct a series of identication checks to assess

whether our proposed empirical strategy is capable of recovering Airbnb's causal impact on

hotel room revenue. Our DD identication strategy relies on randomness in Airbnb adoption

with respect to unobserved city-specic time-varying factors (ikt ) that are also correlated

with changes in hotel room revenue (conditional on the control variables we include). As

with any study relying on observational data, there is no conclusive test of this assumption.

However, we can exploit the richness of our data to check if this assumption is likely to hold

in practice. Similar to Akerman et al. (2013), we perform two checks that support the basis

for our key identication assumption.

9 See How Airbnb helps users save their homes, August 2012, http://finance.fortune.cnn.com/2012/
08/16/Airbnb-foreclosure/

14
 First, we show that most variation in Airbnb adoption is explained by regressing (the log

of ) Airbnb supply on time-invariant city-specic factors, time xed eects, and city-specic

trends  all of which are part of the DD model. These factors explain 88% of the variation

in Airbnb adoption, suggesting that our modeling assumption has a sound basis in practice.

Next, we repeat this regression with the addition of city-specic time-varying observables

that could potentially be correlated with hotel room revenue: population, unemployment

rate, and employment in the accommodation sector. The inclusion of these factors does not

increase the explanatory power of the regression.

Second, we perform a randomization check by testing whether pre-treatment city char-

acteristics predict future Airbnb supply, where the time of treatment is taken to be 2008,

when Airbnb entered the Texas market. The idea behind this test is that assuming Airbnb

adoption is exogenous (with respect to hotel revenue), it should not be correlated with pre-

treatment factors. To perform this identication check, for each city, we compute its most

recent pre-treatment (2007) population, unemployment rate, employment in the accommo-

dation sector, hotel room supply, hotel room prices, and hotel occupancy rates. We then

interact these pre-determined factors (Zk,2007 ) with a vector of post-treatment year-month

xed eects (τt ), and regress them on Airbnb supply. Concretely, with the units of analysis

being post-2007 city-months, we estimate:

log Airbnb Supplykt = Cityk + (τt × Zk,2007 )0 θ + ekt . (2)

Each coecient in the vector of coecients θ is interpreted as a correlation between a spe-

cic pre-treatment characteristic and Airbnb adoption in each post-treatment period (from

January 2008 onwards). Figure 5 presents the estimated coecients θ for each characteris-

tic together with their 95% condence intervals. The only signicant association we nd is

between pre-Airbnb population and subsequent Airbnb adoption, and, for this reason, we

include population as a control in all our specications. Visually, there also appears to be

a weak correlation with pre-Airbnb unemployment rate, further justifying the inclusion of

county-level unemployment rates as a control in Equation 1. Beyond these associations, we

nd no other discernible trend in the remaining coecients (whose 95% condence intervals

always include the zero point, or, no eect). It is especially reassuring that the pre-treatment

hotel industry structure  as captured by hotel room supply, occupancy rates, room prices,

and accommodation sector employment in 2007  do not predict Airbnb supply from 2008

onwards.

Here, we have shown that various factors potentially aecting hotel room revenue, in-

cluding demographic trends, as well as the structure and performance of the hospitality

15
industry across dierent cities, are not correlated with local patterns of Airbnb adoption.

These checks increase our condence that the identication assumptions needed to estimate

Airbnb's causal impact on hotel room revenue hold in our data.

3.5 Results and Economic Signicance

We report the results of estimating Equation 1 using our cumulative Airbnb supply measure

and incorporating the control variables discussed in Section 3.3 in the rst column of Table 3.

We estimate the coecient β = −0.039 (p < 0.01), or equivalently, a 10% increase in Airbnb
listings is associated with a statistically signicant 0.39% decrease in monthly hotel room

revenue. The estimated coecients for the controls have the signs and magnitudes one

would expect (e.g., increased hotel room supply and unemployment are both associated

with decreased hotel room revenue), although we note that our estimate for β without any

controls (not presented) is comparable (β = −0.035, p < 0.01). As stated earlier, we interpret

a negative coecient β as indicative of some Airbnb stays substituting for hotel stays in cities

with an established Airbnb presence.

Our estimates are sensitive to the functional form of the city-specic time trends. Table 4

compares models without city-specic trends, and with city-specic trends of increasing or-

der. Without time trends, we estimate a positive (but insignicant) eect, whereas once

trends are included, our estimate becomes negative and signicant. To explain this observa-

tion, we hypothesize that city-specic demand trends drive both hotel revenue and Airbnb

supply. Therefore, when we omit city-specic trends from the model, Airbnb supply stands

in for the omitted trend and becomes biased upwards. In other words, increasing Airbnb

supply is a sign of increasing demand for accommodation. This analysis guides the func-

tional form we choose to control for city-specic trends: since moving from quadratic to

cubic trends our estimates remain unchanged, we settle for the simpler quadratic form.

The economic signicance of this estimate is best understood in the context of Airbnb's

growth. For instance, in Austin, the city in Texas with the highest Airbnb penetration,

we estimate that the impact of Airbnb over the past 5 years is roughly 10% of hotel room

revenue (the calculation is based on an increase in cumulative Airbnb supply from approx-

imately 450 listings in 2010 to over 8,500 listings in 2014, yielding a revenue impact of

1 − (8, 500/450)−0.039 ). Considering the high xed costs associated with operating a hotel,

this gure could represent a signicant fraction of hotel prots.

An alternative way to assess the economic signicance of Airbnb is through a direct

comparison of the eects of Airbnb and hotel room supply on hotel revenue. By interpreting

the coecient of log Hotel Room Supply in the rst column of Table 3, we nd that a 10%

16
increase in the supply of hotel rooms in Texas is associated with a roughly 1.6% decrease

in Texas hotel room revenue, as compared with the smaller 0.39% decrease associated with

a 10% increase in Airbnb supply. It makes intuitive sense that increasing Airbnb supply

has a smaller impact than increasing hotel room supply, as we do not expect all Airbnb

stays to substitute for a hotel room stay. Nevertheless, the two eects are surprisingly

comparable in size: an increase in Airbnb supply has one-fourth the negative revenue impact

of a corresponding increase in hotel room supply. Taken at face value, this suggests that

incremental Texas Airbnb inventory does weakly substitute for incremental hotel inventory.

And, although the impact of additional Airbnb supply is not as large, the signicantly higher

costs associated with increasing hotel room supply implies that hotels are less likely to be

able to expand inventory as rapidly, an issue we return to shortly.

Next, we estimate Equation 1 using our instantaneous Airbnb supply measure. We

present these results in the second and third columns of Table 3. In the second column we

use a TTL of 3 months, while in the third, we use a TTL of 6 months. In both cases, we

obtain negative and signicant estimates, though the 3-month TTL estimate for β is smaller

than the 6-month TTL estimate (−0.025 vs. −0.035, p < 0.01 for both estimates). Our

conclusion is that regressing on either a cumulative or an instantaneous measure of Airbnb

supply captures a signicant eect on hotel revenues due to Airbnb.

This analysis reveals an additional insight regarding Airbnb's economic impact: the sig-

nicant uctuation in instantaneous Airbnb supply suggests that Airbnb's impact on hotel

revenue will vary signicantly over time too. For instance, in our data, we estimate that

instantaneous Airbnb supply during SXSW has historically been approximately 60% higher

than the rest of the year. In turn, this suggests that Airbnb impact on hotel revenue is

approximately 1.5 percentage points larger during SXSW (calculated as log(1.6) ∗ 0.035).
Variation in instantaneous supply is not the only reason why Airbnb's impact could be

more pronounced during SXSW or during other large events. Perhaps it is the case that

Airbnb might be especially appealing to SXSW participants, but has little or no appeal to

travelers the rest of the year. However, we nd that this is not the case: when we censor

SXSW from our data, the elasticity that we estimate is unchanged (β = −0.039, p < 0.05)
using our cumulative supply measure. This result suggests that Airbnb's impact is not solely

due to idiosyncratic preferences of the SXSW demographic.

In summary, we see evidence that Airbnb's impact in Texas is observable through the lens

of both cumulative and instantaneous supply measures. We further see that while its impact

is most strongly concentrated in Austin, and has maximum impact correlated with periods

of peak demand, the impacts are present year-round. Using the instantaneous measure, we

attributed seasonal variation in impact to a feature that is unique to the sharing economy,

17
supply exibility. We later rene this top-level analysis to analyze how the economic impacts

are dierentiated across types of hotels in Section 4.1 and further unpack the eects of supply

exibility on the peak pricing power of hotels in Section 4.2.

3.6 Hotels' Responses to Airbnb: Price, Occupancy, Entry & Exit

So far, we have measured Airbnb's impact in terms of hotel revenue. Now we turn to the na-

ture of responses by incumbent hotels to Airbnb market entry. In the short term, hotels could

plausibly respond to Airbnb market entry through a price response, an occupancy response,

or both. In the long run, Airbnb could cause hotel investments to change course, ultimately

impacting market entry and exit. All of these impacts can be investigated naturally by

measuring alternative dependent variables, other than revenue.

Recall that hotel room revenue is the product of two quantities: average occupancy rate

within a given time period, and average daily room price (ADR) during that same period

of time. A hotel that exerts no response to a supply shock would exhibit a reduction in

occupancy, whereas an active manager could alternatively maintain occupancy levels via a

price response. A notable dierence between the two responses is that the latter response,

reduced prices, is a net benet for all consumers seeking accommodations, whether they use

Airbnb or not.

To estimate these eects, we re-estimate the DD specication in Equation 1, substituting

the dependent variable rst with occupancy rate, and then with the log of ADR, and retaining

the controls. Similar to the room revenue analysis, these two quantities vary by hotel and

by month. The price and occupancy dataset that we use masks individual hotel identities,

therefore, we cannot link it with the TripAdvisor data on a hotel-by-hotel basis. Instead,

we control for changes in hotel quality at the city-level using the average hotel rating and

fraction of reviewed hotels in each city. We report these results in Table 5. As reported

in the rst column of this table, we nd a small and weakly signicant (p < 0.1) negative

connection between increased Airbnb listings and occupancy rate. (Note that, in contrast

to our other dependent variables, occupancy rate is already expressed as a percentage and

therefore we do not log-transform it. Therefore, the coecient of this regression has a

level-log interpretation.) In the second column, we regress against ADR, and we nd that

a 10% increase in Airbnb supply is associated with a statistically signicant (p < 0.01)
price decrease of 0.19%. This suggests that aected hotels actively respond by lowering

their prices. Note that this behavior is consistent with basic hotel revenue management
10
practices, where hotels set prices accordingly to the level of occupancy rates observed. To

10 Indeed, the hospitality industry has high xed costs and low marginal costs, and therefore the thinking
is that it's better to put a head in a bed  at a low price  than not at all.

18
understand the economic signicance of these results we can repeat the same calculation

performed in Section 3.5, which suggests that in Austin, Airbnb negatively impacted hotel

prices by roughly 6%.

Both the price and occupancy eects we investigated above constitute immediate re-

sponses to Airbnb. In the longer run, Airbnb may also aect hotels' entry, exit, and invest-

ment decisions. To better understand the decision-making process and timetables of hotel

development, we assembled a proprietary dataset (from STR) that records all currently on-

going Texas hotel projects, including both new construction and renovations (we do not

have access to the historical record of completed renovations.) STR records the dates that

projects enter their various phases of development. Using this dataset, we computed the av-

erage time it takes to transition from one phase to the next, which we diagram in Figure 7.

The average estimated time between pre-planning and projected opening is approximately 4

years, although there exists signicant variation depending on the project type. Therefore,

hotel projects that were completed or were ongoing during our observation period were likely

conceived before Airbnb became a concern for the hotel industry. Indeed, basic Poisson re-

gressions of hotel entry and exit against Airbnb supply, which we do not report, yielded

no correlation. As Airbnb continues to become more established, and hotels have time to

incorporate Airbnb in their investment strategies, studying the nature of hotels' longer-term

response will be worth revisiting.

4 Variation of Impact Across Hotels and Across Time

4.1 Which hotels are most aected and why?

We have provided evidence that Airbnb has a negative impact on hotel room revenue in

Texas, treating hotels as a homogeneous set. In this section, we investigate various mech-

anisms through which Airbnb could exhibit heterogeneous impacts across dierent types of
hotels, and provide supporting empirical evidence. To motivate this analysis, we observe

that while Airbnb can surely sometimes provide a suitable alternative to hotels, one can

hardly expect it to be a perfect substitute for all travel needs. As Airbnb has its roots in

casual stays, including those involving shared accommodations, we expect it to be a more

attractive option for travelers on a budget. Conversely, business travelers and vacationers

who frequent high-end hotels are examples of consumer groups we argue are less likely to

substitute a hotel stay with an Airbnb stay. Business travelers, in particular, are often less

price-sensitive, as they are typically reimbursed for their travel; moreover, they also make use

of business-related hotel amenities not typically provided by Airbnb properties. Following

19
this logic, we further isolate the impact of Airbnb on hotel room revenue by partitioning

hotels in three dierent ways, each dividing hotels into segments that we expect to be less

vulnerable to Airbnb's entry and other segments that we expect to be more vulnerable, then

estimating the additional interaction eects in our original DD specication. In our rst

partition, we segment hotels by price tier, following the STR hotel census, which divides

hotels into ve tiers: Budget, Economy, Midprice, Upscale, Luxury. In our second partition,

we dierentiate hotels by their customer base: those that target business travelers versus

those that do not. Finally, we consider the dierentiated impact on chain hotels versus

independents.

To estimate heterogeneous treatment eects, we estimate a new specication that adds

an interaction eect between hotel types and Airbnb supply to the DD specication in

Equation 1:

log Hotel Revenueikt =β1 log Airbnb Supplykt (3)

+ β2 log Airbnb Supplykt × Hotel Typei

0
+ Xikt γ + αi + τt + ikt .

The coecients of interest are β2 , which capture the dierential impact of Airbnb on the

various segmentations by hotel type that we investigate. For our rst segmentation, we

dene Hotel Typei as a categorical variable identifying each one of the hotel price segments

used by STR. In the second and third segmentations, we dene Hotel Typei to be a binary

indicator: whether or not hotel i has conference or meeting space, and whether or not it is

a chain, respectively.

The results of these analyses appear in Table 7. We start with price segmentation,

presented in the rst column. We estimate Equation 3, interacting hotel price segments

with Airbnb supply. Here, we use Luxury hotels as a reference category least unaected

by Airbnb, motivated by the observation that these hotels are least comparable to Airbnb

based on average room price and also by their amenities (e.g., pools, conference rooms,

concierge). We nd the negative impact of Airbnb increasing as we step down price tiers, with

statistically signicant interaction coecient estimates at the 1% level for each of the three

lowest tiers (Midprice, Economy, and Budget). In contrast, we nd only a small negative and

insignicant eect for the Upscale and Luxury segment (the latter being the reference level,

and hence being captured by the main eect). From a managerial standpoint, this result

has direct import: even though lower-end hotels in Texas account for a disproportionately

small amount of room revenue as compared with upmarket hotels, they nevertheless bear the

20
brunt of the impact of the market entry of Airbnb. Our evidence suggests that consumers

are increasingly substituting Airbnb stays for lower-end hotels in Texas, possibly identifying

the former as oering better value at a similar price point. While this increased competition

aords consumers greater choice, it also places lower-end hotels in regions with high Airbnb

penetration at greater risk.

In the second column of Table 7, we report the results of the segmentation of hotels

catering to business travelers. We use those hotels having conference and meeting space as the

reference category. The estimated coecient β2 for the interaction between Airbnb supply

and the indicator variable denoting absence of meeting space is negative and statistically

signicant (−0.015, p < .01), suggesting that hotels lacking business facilities are more

aected by Airbnb. These results are consistent with our prior segmentation as well as

with Airbnb's marketing strategy to date, which has primarily targeted vacation travel. We

do note though that, seeing a growth opportunity in the business travel segment, Airbnb
11
recently launched an initiative to attract more business travelers. An interesting open

question going forward is the extent to which business travel will continue to dierentiate

the impact of Airbnb on hotels.

The third distinction that we explore, which relates primarily to hotel operation, is

between chain hotels (including franchises) and independent hotels. Unlike independent

hotels, chain hotels allocate large marketing budgets to advertising, brand building, guest

loyalty programs, and other tactics which should make them less vulnerable to competition.

In addition, many chains provide a more predictable standard of service, which further

dierentiate them from both Airbnb and independent hotels. We present this analysis in

the third column of Table 7, using chain hotels as a reference level. The overall eect due to

Airbnb remains negative and statistically signicant (−0.038, p < .01), suggesting that hotels
of both operation structures were aected. However, the estimated interaction coecient for

independent hotels (−0.008, p < .05) is also negative and statistically signicant, suggesting

that Airbnb has indeed had a slightly larger impact on independent hotels.

Overall, we nd that independent hotels, hotels that do not cater to business travelers,

and lower-end hotels are all more heavily aected by Airbnb than our respective reference

categories, hotels without these characteristics. While these results help us better understand

the most vulnerable hotel segments, and are certainly of importance to hoteliers, they also

serve as robustness checks to our primary nding, in that the heterogeneous substitution

eects they reveal align with the eects we hypothesized based on the value proposition to

consumers that Airbnb oers.

11 See http://bits.blogs.nytimes.com/2014/07/28/airbnb-expands-into-business-travel/.

21
4.2 Airbnb and peak pricing power of hotels
Our analysis so far has focused on quantifying the extent to which Airbnb supply substitutes

for hotel room supply and its dierentiated impact across various hotels segments. But

we now show that Airbnb supply is more than just a partial substitute for low-end hotel

supply, by proposing and empirically evaluating mechanisms whereby changes in Airbnb

supply exhibit fundamental dierences from changes in hotel room supply. In particular,

we investigate the ability of Airbnb suppliers to exhibit a more exible response to peak

seasonal demand, and in so doing, crimp operating margins of hotel operators during these

peak periods.

During localized periods of peak demand, it is well understood that hotels can respond
12
by raising prices , but they cannot materially increase supply, due to high xed costs of

new inventory. In contrast, many of the micro-entrepreneurs providing Airbnb supply can

elect to take inventory on and o the market on very short time scales and with near-zero

cost. Thus, the aggregate decisions of Airbnb providers comprise both a price response and

a supply response. Our subsequent analysis is therefore motivated by the hypothesis that,

during localized periods of peak demand, regions with exible Airbnb supply serve to more

eectively absorb high seasonal demand than regions in which Airbnb is not present. If

the hypothesis is operative, the managerial implication is that the hotel industry's ability

to command high rents during peak periods, which we will refer to as their peak pricing
power, has become diminished in regions where Airbnb has actively entered the market, as
compared with other locales where Airbnb is less prevalent.

To motivate our denition of peak pricing power, consider that city-specic travel patterns

are highly seasonal, and many periods of peak demand predictably recur with an annual

frequency. Therefore, for each hotel-year in our data, we will refer to peak demand months

as the high season, and the remaining months as the low season. For each hotel i, we will
denote high season prices during year y by pH
i,y and low season prices by pLi,y . Given these

two quantities, we will dene hotel i's peak pricing power as:

Pi,y = log pH L
i,y − log pi,y , (4)

which can be interpreted as the percentage increase in prices during high season compared to
13
low season. Because we are interested in understanding changes in  rather than absolute
levels of  hotel pricing power as Airbnb adoption grows, the quantity we analyze is the rst

12 For example, see evidence of surge pricing coinciding with the annual shareholders' meeting of Berkshire
Hathaway in Omaha, Buett's revenge, The Economist, 1/9/16.
13 The percentage interpretation is most accurate for smaller values of this dierence.

22
dierence of peak pricing power:

∆Pi,y = (log pH L H L
i,y − log pi,y ) − (log pi,y−1 − log pi,y−1 ), (5)

which can be interpreted as the year over year change in a hotel's ability to increase prices

during high season. Rearranging terms of Equation 5 gives us the more convenient form:

∆Pi,y = (log pH H L L
i,y − log pi,y−1 ) − (log pi,y − log pi,y−1 )

= ∆ log pH L
i,y − ∆ log pi,y , (6)

which is the dierence between year-over-year changes in high season prices and low season

prices. Intuitively, double dierencing allows us to adjust changes in high season pricing

(likely related to exible scaling of Airbnb supply) using low season changes in pricing (likely

unrelated to exible Airbnb supply) as a baseline. For instance, if year-over-year percentage

price changes are equal during high and low season, it is unlikely that they are jointly driven

by Airbnb hosts exibly scaling supply to accommodate peak demand during specic months

of the year; hence, in this case, ∆Pi,y will be estimated to be zero.

To study changes in peak pricing power of hotels in our dataset, we considered the impact

of two large events that take place annually in Texas: the South by Southwest (SXSW)

festival in Austin in March, and the Texas State Fair (TSF) in Dallas in October. Both

events draw a very large number of out-of-town visitors, and have a substantial impact on

the bottom line of area hotels as a result. Both events have also grown in popularity in the

past decade, but with the much smaller SXSW festival growing more rapidly in percentage

terms. Figure 10 displays attendance for SXSW Interactive, which together with SXSW

Film and SXSW Music, are the major components of SXSW. March and October represent

the peak months for demand of hotels in Austin and Dallas respectively, measured both in

terms of occupancy and ADR (average daily room rate). In both cases, ADR and occupancy

range between 8-15% above the corresponding values for the rest of the year, consistently

over the past decade. However, Airbnb has grown much faster in Austin than it has in Dallas,

suggesting that if Airbnb aects peak pricing power, this eect will be more pronounced in

Austin.

We begin our analysis by visualizing changes in peak pricing power. Motivated by our

previous results, where we found that Airbnb has a stronger impact on lower-end hotels,

we segment hotels by price category and consider year-over-year changes in pricing power

for high season versus all other months combined. Following Equation 6, for each hotel, we

compute year-over-year changes in high and low season prices (i.e., ∆ log pH
i,y and ∆ log pLi,y .)
Figure 8 displays the annual average of these quantities in Dallas for the period 2010-2014.

23
The gap between the solid line (changes in high season prices) and the dashed line (changes

in low season prices) can be interpreted as the year-over-year change in hotel pricing power

during periods of peak demand. Visually, we see little discernible dierence between the

two lines, with the gap between them always close to zero. This suggests that the pricing

power of hotels in Dallas during the State Fair has not changed signicantly compared to

the remainder of the year.

Next, we consider Austin. With the very rapid growth in SXSW, one could naturally

conjecture that the rate at which peak pricing power grows would outstrip that of non-peak

periods. Consider the data plotted in Figure 9, where we depict the year-over-year percentage

changes in SXSW prices for March (solid line), in comparison to changes in prices during the

remaining months of the year (dashed line). During the initial period, roughly 2010-2012,

visual evidence suggests the hotel pricing power for SXSW increased faster than during the

rest of the year, consistent with rapid growth in SXSW. In the second half of the period,

2012-2014, a new phenomenon is at work. The gap between high and low season price

changes starts to narrow, as hotels lose the ability to exert the same pricing power, despite

the continued growth of SXSW. This eect is especially pronounced for lower end hotels,

as our previous results would predict. Overall, these visualizations are consistent with an

explanation of exible Airbnb supply coming online during SXSW to accommodate peak

demand, thereby crimping the peak pricing power of lower-end hotels specically.

As a nal step in understanding the statistical signicance of the eect we visualized, we

estimate a descriptive model of changes in peak pricing power. The dependent variable we

analyze is the seasonal price dierence for each hotel i and year-month t, which is dened

as follows:

O12 log pi,t = log pi,t − log pi,t−12 , (7)

where OD is the seasonal dierence operator of order D. As before, the interpretation of

this quantity is the percentage change in prices for hotel i compared to prices during the

same month one year ago. Unlike our visualization, where we lumped all low-season months

together, here we separately dierence each month in our data. The model we estimate takes

the following triple-dierences form:

O12 log pi,t =β1 Austini + β2 Marcht + Yeart

+ β3 Austini × Marcht + β4 Austini × Yeart + β5 Marcht × Yeart
+ β6 Austini × Marcht × Yeart + i,t , (8)

where Marcht is a dummy for March hotel-months, the Yeart are year xed eects, and

Austini is an indicator for hotels in Austin. In addition to these explicit controls, seasonal

24
dierencing wipes out both hotel xed eects, as well as hotel-month-specic linear trends in

year-over-year prices changes (e.g., a specic hotel increasing March prices by 5% every year,

April prices by 2% every year and so on.) The coecients of interest are contained in the

vector β6 , and they can be interpreted as changes in SXSW pricing power. Intuitively, the

model estimates March-specic changes in pricing power in Austin and then adjusts these

estimates for a) March-specic changes in pricing power outside Austin and b) non-March-

specic changes in Austin. Figure 11 displays the coecients β6 and their associated 95%

condence intervals. Our conclusions here mirror our earlier observations: SXSW pricing

power has signicantly declined as Airbnb popularity grew, despite the fact the SXSW

attendance has continued to steadily grow over time.

Unlike our earlier analyses, the results in this section are descriptive. When jointly

interpreted with our causal estimates of Airbnb on hotel revenue, they paint a picture of

Airbnb reducing hotel pricing power during periods of peak demand, consistent with our

hypothesis that the exible provisioning of inventory to accommodate peak demand is a

distinguishing feature of the sharing economy. Better understanding this phenomenon with

both more sophisticated modeling and data spanning more large events is future work.

In closing, we compare and contrast our observations with another sharing economy

study that observes exible supply entering the Uber market during peak periods (Hall

et al., 2015). In this work, researchers study the eectiveness of surge pricing on Uber,

whereby drivers are incented to drive at peak times through higher payment multipliers.

The study reports that the surge pricing mechanism is eective, and leads to reduced wait

times during periods of peak demand, comparable to levels seen in low-demand periods. In

comparison, our work shows that a similar incentive drives Airbnb suppliers to scale room

supply during periods of peak demand, when they can command higher rents. We witness

this eect indirectly, through decreased peak pricing power of hotels in high season. While

Uber directly incentivizes increased supply through central setting of price multipliers, a

similar eect arises in Airbnb without direct control, but instead through the collective,

decentralized decision-making of its suppliers. Interestingly, Airbnb is moving towards a

14
variable pricing model, where it dynamically adjusts listing prices in response to demand.

5 Robustness checks

We perform three checks to reinforce the causal interpretation of our DD estimate: a distance-

sensitive denition of the Airbnb supply variable; a specication test using an alternative

14 See Airbnb shakes up pricing model to meet surging demand at http://www.ft.com/cms/s/0/

bc875c4c-88ee-11e5-9f8c-a8d619fa707c.html. We thank Avi Goldfarb for pointing out this connection.

25
functional form of Airbnb supply; and a matching method, which we use as a more stringent

alternative in dening (otherwise similar) treated and untreated properties.

5.1 Distance-based market denition

In our analyses so far, we have assumed that travelers may substitute between hotels and
15
Airbnb properties within the same city irrespective of the distance between properties.

While this seems reasonable for smaller cities in our dataset, it is likely a less reliable ap-

proximation of how travelers form consideration sets when visiting sprawling cities like Hous-

ton. To test the sensitivity our results to narrower market denitions, we next analyze a

proximity-based Airbnb supply measure. Specically, for each hotel in our data, we measure

the cumulative number of Airbnb rooms within a xed radius at any given point in time.

This measure allows dierent hotels in the same city to face dierent levels of competition

by Airbnb. To be consistent with this market denition, we also dene hotel competition in

the same way: for each hotel, we measure the number of hotel rooms within the same xed

radial distance.

Using a hotel-specic market denition introduces a new type of endogeneity concern that

we need to address. For any given hotel, increased competition by nearby Airbnb properties

or hotels is likely correlated with increased demand for that hotel. In other words, even within

the same city, new hotel rooms and Airbnb properties are more likely to be located near hotels

that are facing growing demand by travelers. The city-specic trends we previously included

do not allow for correlation between local measures of Airbnb and within-city hotel revenue

variation. Therefore, we estimate a model that includes hotel-specic quadratic trends. This

model, known as the correlated random trends, or random growths, model (Wooldridge, 2005;
Murtazashvili and Wooldridge, 2008; Wooldridge, 2009), allows for correlation between the

hotel-specic trend component and time-varying observables. The specic model we use

takes the following form:

0
log Hotel Revenueikt =β log Local Airbnb Supplyikt + Xikt γ (9)

+ hi + ai1 t + ai2 t2
+ τt + Cityk × Montht + ikt .

Following standard practice (see, e.g., Wooldridge (2005)), we eliminate the hotel-specic
16
quadratic trends by second-dierencing our data. Second-dierencing requires the sacrice

15 Hollenbeck (2014), who analyzes the same Texas data, also uses city-level markets to model competition
between hotels.
16 To see this, note that rst-dierencing transforms the linear trend to a constant, and the quadratic trend

26
of the rst two monthly observations of each hotel. Our decade-long panel is sucient to

comfortably accommodate this transformation. The nal model we estimate is:

0
∆2 (log Hotel Revenueikt ) =β∆2 (log Local Airbnb Supplyikt ) + ∆2 (Xikt )γ (10)

+ ai2 + τt + Cityk × Montht + ikt ,

where ∆2 is the second-dierence operator. Note that dierencing also eliminates the hotel

xed eect hi . The model can be estimated using the within transformation to eliminate the

hotel-specic intercepts ai2 . We continue to cluster errors at the city-level.

Table 6 displays our results. In the rst column, we display our results using a radius of

1 mile around each hotel. We estimate a signicant Airbnb eect with magnitude −0.032
(p < .05), similar to our prior estimates. In the second column, we experiment with a larger

radius of 5 miles. Our estimate is again signicant, however, it is now smaller in magnitude

(−0.025, p < .05). One natural interpretation for the dierence between these two estimates

is that the greater the distance between Airbnb listings and hotels within a city, the less likely

travelers are to substitute between the two. Overall, our results support our prior hypothesis

that Airbnb is directly aecting hotel revenue, while producing the additional insight that

this impact is sensitive to the distance between hotels and Airbnb listings within a city.

5.2 Alternative functional form for Airbnb supply

The second robustness check we perform guards against a functional specication concern in

Equation 1: regressing the log of Airbnb supply on the log of hotel room revenue implicitly

assumes a constant elasticity relationship between the two quantities. While this might be a

reasonable assumption in data with limited variation in Airbnb supply, the constant elasticity

assumption is likely violated in our setting, as it is implausible that doubling Airbnb supply

from 1 to 2 units will have the same eect on hotel room revenue as doubling Airbnb supply

from 100 to 200 units. To ensure that our results are not driven by this modeling choice, we

model Airbnb supply non-parametrically using a categorical variable, which takes on one of

the following (roughly log-binned) values: 0 Airbnb units, 1-99 Airbnb units, 100-999 Airbnb

units, 1000+ Airbnb units. Specically, we estimate:

log Hotel Revenueikt =β1 I(Airbnb Supply 1-99)kt + β2 I(Airbnb Supply 100-999)kt (11)

0
+ β3 I(Airbnb Supply 1000+)kt + hi + τt + Xikt γ + ikt ,

to a linear one: ai1 t + ai2 t2 − ai1 (t − 1) − ai2 (t − 1)2 = ai1 + ai2 (2t − 1). Taking a second dierence, we arrive
at ai1 + ai2 (2t − 1) − ai1 − ai2 (2(t − 1) − 1) = 2ai2 , which has ai2 as a hotel-specic intercept.

27
where the I(.) are dummy indicators for the corresponding ranges of Airbnb supply.

This model allows for the eect of Airbnb to vary depending on the level of Airbnb list-

ings present in each city during a given period. In addition, it provides easier to interpret

estimates compared to the log-log estimates of Equation 1. In this model, each of three

estimated coecients associated with the three levels of the categorical Airbnb supply vari-

able we use represents a percentage change in hotel revenue. We estimate this model by

replacing Airbnb supply with this new categorical variable in Equation 1 using zero Airbnb

units as the reference level. We present our results for cumulative Airbnb supply in the

rst column of Table 8. These estimates provide directly interpretable estimates of Airbnb's

economic impact. We nd that increasing levels of Airbnb penetration have proportionally

larger impacts on hotel room revenue, as we would expect. For example, at Airbnb adoption

rates exceeding 1000 rooms, the estimate (−0.085, p < .05), indicates (since we are now

working with a log-level specication) an average impact of 8.5% on hotel room revenue.

These estimates are in line with our previous estimates in Section 3.5.

It is reassuring that we nd no statistically signicant eect at low levels of Airbnb

supply. In fact, this model claries that Austin is the primary driver of the Airbnb eect

we estimate  no other city in our data had more than 1000 (cumulative) Airbnb listings

during our observation period and therefore the 8.5% decrease in revenue we estimate for

Airbnb penetration at this level is driven by Austin. Indeed, we also nd that our estimate

between 100-999 listings is also primarily driven by Austin  deleting Austin from the data

and re-estimating the model gives a negative but statistically insignicant impact. Even

though a few cities in our data saw Airbnb adoption at this level, those cities were larger,

had more hotel rooms, and did so late in our observation period (see Table 1). Therefore,

it would be rather surprising had Airbnb exerted a statistically signicant impact on hotel

revenues outside of Austin. Does this result suggest that the Airbnb eect is specic to

Austin? We think this is unlikely for two reasons. First, Airbnb is popular in many other

destinations outside of Texas both nationally and globally, and we do not have reason to

believe the incumbent hotel industry is better defended from Airbnb in those other cities.

Second, our earlier ndings in Ÿ 4.2 indicate that hotels facing seasonal tourism demand have
17
a structural susceptibility to Airbnb that is universal, not local to Austin.

17 A recent report by CBRE Hotels' Americas Research ranks Austin as the 13th most vulnerable city to
Airbnb in the US (NYC ranks rst), taking into account both the ratio of Airbnb units to hotel rooms in
each market, as well as hotel room and Airbnb prices. See http://cbrepkfcprod.blob.core.windows.
net/downloads/store/12Samples/An_Analysis_of_Airbnb_in_the_United_States.pdf.

28
5.3 A matching estimate using CEM
Since Airbnb adoption is clearly not random by design, to provide evidence in support of

the DD identication assumptions, we showed that observed pre-treatment demographic and

market characteristics do not correlate with the patterns of Airbnb adoption we observe in

our data, which is what we would expect with exogenous Airbnb entry. Here, we combine

DD with matching to further limit the potential for unobserved confounders biasing our es-
timates. To explain the matching approach, rst recall our source of identication: roughly

speaking, for each treated hotel, i.e., a hotel aected by Airbnb competition, our DD
analysis constructs a counterfactual outcome using a set of untreated hotels, i.e., hotels
unaected by Airbnb. The intuition behind matching is that the more similar treated and

untreated hotels are in their observed characteristics, the less likely they are to dier in

unobserved ways, including bias-inducing factors. Matching methods aim to reduce endo-

geneity concerns by ensuring comparability between treated and untreated units (Heckman

and Navarro-Lozano, 2004). While various matching methods exist, here we use the Coars-

ened Exact Matching (CEM) procedure (Iacus et al., 2012), because it is intuitive and works

well with categorical data (like most hotel characteristics).

CEM takes places in two steps. First, hotels are stratied based on observed charac-

teristics; we use price segment (Budget to Luxury), operation (independent or chain), and

hotel chain aliation (e.g., Hilton, or Marriott), if any. After this rst step, each stratum

contains hotels that are identical on the basis of these characteristics. For instance, a single

stratum contains all Upscale Marriott hotels, some of which are eventually treated and some

of which are not. In a setting with a binary treatment indicator, it is clear which units are

eventually treated. In our case, where treatment intensity varies, we make this distinction by

dening hotels in cities which see no Airbnb penetration by the end of our observation period

as untreated, and the remaining hotels as treated. One could argue that this denition of

treatment is too permissive; while we do not present these results for brevity, we found our

CEM analysis to be robust to alternative denitions of treated units, such as hotels in cities

that eventually have at least 100 Airbnb listings. In the second step of CEM, we discard

strata containing only treated or untreated hotels, and re-normalize weights of observations

in the remaining strata to place equal weight on treated and untreated units in each stra-
18
tum. Applying CEM to our data leaves us with 1, 946 hotels. Finally, we re-estimate the

DD specication in Equation 1 on the subset of matched hotels using the CEM weights.

Conceptually, DD on the CEM sample estimates a treatment eect within each stratum of

18 CEM entails a trade-o between matching granularity, and the number of discarded observations. We
chose our matching criteria to strike a reasonable balance between ensuring units within each stratum are
similar, and discarding too many observations. Our results our robust to alternate matching criteria.

29
comparable treated and untreated hotels, then averages these treatment eects to arrive at

a nal estimate. We report this estimate in the second column of Table 8. We nd that the

eect of Airbnb on hotel room revenue is robust to CEM, attaining a magnitude β = −0.043,
p < .01) that is highly comparable to our original estimate, β = −0.039, reported in column

1 of Table 3.

6 Discussion and conclusions

The sharing economy has recently emerged as a viable alternative to fullling a variety of

consumer needs, ranging from prepared meals to cars to overnight accommodations, that

were previously provided primarily by rms rather than entrepreneurial individuals. As

the size of the sharing economy has grown, so has the magnitude of its economic impacts.

Our work is among the rst to provide empirical evidence that the sharing economy is

signicantly changing consumption patterns, as opposed to generating purely incremental

economic activity. Focusing on the case of Airbnb, a pioneer in shared accommodations,

we estimate that its entry into the Texas market has had a quantiable negative impact

on local hotel room revenue. The substitution patterns we observe strongly suggest that

Airbnb provides a viable, but imperfect, alternative for certain traditional types of overnight

accommodation. Our analyses pinpoint lower-end hotels, and hotels not catering to business

travelers, as those that are most vulnerable to increased competition from rentals enabled

by rms like Airbnb. Moreover, our work gives evidence that Airbnb supply is dierentiated

from hotel supply, as evidenced both by Airbnb supply-side exibility and carrying through

to the impact on hotel peak pricing power.

Our work has some limitations which could be addressed in future work. First, one must

recognize that our ndings are representative of the state of Texas; directly generalizing

them to other markets may not be appropriate given the varying of dynamics of supply

and demand for accommodation across dierent regional markets. Additional studies which

model the impact of Airbnb across these markets could be a useful contribution. A second

limitation of work is that we analyze properties listed only on Airbnb, but not properties

available through related vacation rental platforms like HomeAway and VRBO. We do not

believe that our results are signicantly aected by these competitors, since these rms pri-

marily serve the smaller vacation rental market; moreover, they have not experienced the

extremely rapid growth of Airbnb. Nevertheless, one could investigate the impact of all of

these rms in aggregate, or individually. A nal limitation of our study pertains to the

precise characterization of hotels' response: here we have analyzed two metrics, price and

occupancy rate, that managers can invoke as a response in the short-term. On longer time

30
scales, hotels have other ways of responding to Airbnb, including alterations to their invest-

ment schedules, to their entry and exit decisions, and to their marketing campaigns. New

promotions, advertising campaigns, and even re-positioning to provide more personalized

Airbnb-like services are all options. Work that either informs or interprets the shape of the

response by hotels in the longer run will address interesting open questions.

Our results have direct implications for hotels, travelers, and policy makers. For hotel

managers, the competition their rms face from peer-to-peer platforms has several unique

features that dierentiate it from competition with other rms. First, the Airbnb platform

has near zero marginal cost, in that a new room can be incrementally added to (or removed

from) the platform with negligible overhead. Because of this, Airbnb can scale supply in a

near frictionless manner to meet demand, even on short timescales. By contrast, increasing

hotel room supply involves buildout, causing signicant marginal costs for hotel chains.

As we have shown, this unique feature of Airbnb has already signicantly aected hotel's

pricing power during periods of peak demand. Second, Airbnb oers a much wider range

of products and services than hotels: Airbnb users can rent anything from an apartment

to a yurt. More importantly, because Airbnb leverages existing housing inventory, it can

potentially expand supply wherever houses and apartment buildings already exist. This

is in contrast to hotels, which must be built at locations in accordance with local zoning

requirements. Therefore, competition by Airbnb is potentially harder for incumbents to

adapt to, compared to competition by other hotel rms.

Turning to consumers, we show that hotels in areas where Airbnb has an established

presence have responded to increased competition by lowering their prices, which harms

their revenues, but benets travelers, even those who do not use Airbnb. In addition to

reduced prices, consumers also benet from increased variety provided through peer-to-peer

platforms. Furthermore, consumers on the supply side benet through additional income

generated by providing goods and services via peer-to-peer platforms.

Finally, our results have implications for policy makers. Municipal revenues rely in part

on tax receipts from well-regulated industries such as hotels and taxicabs. With demand

shifting away from these incumbent rms, and to the extent that regulation and taxation

of peer-to-peer platforms proves to be more challenging, the bottom line of cities with an

established Airbnb presence could be hurt in the short run. Of course, peer-to-peer platforms

can also bring about increased demand, which would provide direct benet to cities, making

the net impact on cities harder to measure. Quantifying the net impact of peer-to-peer

platforms remains an interesting direction for future research.

Returning to the thesis that the sharing economy has the potential to transformatively

increase social welfare, as evangelized by Botsman (2012) and others, we assert that a large

31
population of individuals worldwide have indeed beneted from Airbnb: not only hosts that

derive incremental income by renting properties through Airbnb, and guests who select an

Airbnb rental as an alternative to a hotel stay, but also those consumers who benet from

lower prices and increased competition in the accommodation industry. More broadly, one

can weigh the positive change the sharing economy can bring about, not only by providing

imperfect substitutes for existing products, but also, through an application of Say's Law,

by generating demand that did not previously exist, through the supply of new products and

services. Harkening back to arguments Airbnb has made, supply of inexpensive accommo-

dations can increase travel and tourism spend overall, and thus, the sharing economy could

be a net producer of new jobs. However, these positives must be evaluated against various

costs, including those estimated in this paper.

References

A. Akerman, I. Gaarder, and M. Mogstad. The Skill Complementarity of Broadband Internet.

Technical report, IZA Discussion Paper, 2013.

K. Baker and A. Coulter. Terrorism and Tourism: The Vulnerability of Beach Vendors'

Livelihoods in Bali. Journal of Sustainable Tourism, 15(3):249266, 2007.

J. Y. Bakos. Reducing buyer search costs: Implications for Electronic Marketplaces. Man-
agement Science, 43(12):16761692, 1997.

M. Bertrand, E. Duo, and S. Mullainathan. How Much Should We Trust Dierences-In-

Dierences Estimates? The Quarterly Journal of Economics, 119(1):249275, 2004.

http://www.ted.
R. Botsman. The Currency of the New Economy is Trust. [Video le].

com/talks/rachel_botsman_the_currency_of_the_new_economy_is_trust.html,
2012. [Online].

S. Carbó Valverde, S. Chakravorti, and F. Rodriguez Fernandez. Regulating Two-Sided

Markets: An Empirical Investigation. Technical report, Working Paper, Federal Reserve

Bank of Chicago, 2010.

R. Cervero, A. Golub, and B. Nee. City CarShare: Longer-Term Travel Demand and Car

Ownership Impacts. Transportation Research Record: Journal of the Transportation Re-

search Board, 1992(1):7080, 2007.

S. G. Donald and K. Lang. Inference with Dierence-in-Dierences and Other Panel Data.

The Review of Economics and Statistics, 89(2):221233, 2007.

32
L. Einav, C. Farronato, and J. D. Levin. Peer-to-peer markets. Annual Review of Economics,
8(1), 2016.

A. Fradkin. Search frictions and the design of online marketplaces. Technical report, Working

Paper, 2014.

A. Fradkin, E. Grewal, D. Holtz, and M. Pearson. Reporting Bias and Reciprocity in Online

Reviews: Evidence From Field Experiments on Airbnb. 2014. Cited with permission.

J. Hall, C. Kendrick, and C. Nosko. The Eects of Uber's Surge Pricing: A Case Study.

Technical report, Working paper. Link, 2015.

J. V. Hall and A. B. Krueger. An Analysis of the Labor Market for Uber's Driver-Partners

in the United States. http://arks.princeton.edu/ark:/88435/dsp010z708z67d, 2015.

J. Heckman and S. Navarro-Lozano. Using Matching, Instrumental Variables and Control

Functions to Estimate Economic Choice Models. Review of Economics and Statistics, 86

(1):3057, 2004.

B. Hollenbeck. The spread of horizontal chains: Eciency or market power? Technical

report, Working paper, UCLA, 2014.

S. M. Iacus, G. King, and G. Porro. Causal Inference without Balance Checking: Coarsened

Exact Matching. Political analysis, 20(1):124, 2012.

G. Z. Jin and M. Rysman. Platform Pricing at Sports Card Conventions. Technical report,

National Bureau of Economic Research, 2012.

R. Kosová and C. A. Enz. The Terrorist Attacks of 9/11 and the Financial Crisis of 2008

The Impact of External Shocks on US Hotel Performance. Cornell Hospitality Quarterly,

53(4):308325, 2012.

K. Kroft and D. G. Pope. Does Online Search Crowd Out Traditional Search and Improve

Matching Eciency? Evidence from Craigslist. Journal of Labor Economics, 32(2):259

303, 2014.

V. Landsman and S. Stremersch. Multihoming in Two-Sided Markets: An Empirical Inquiry

in the Video Game Console Industry. Journal of Marketing, 75(6):3954, 2011.

E. Martin, S. A. Shaheen, and J. Lidicker. Impact of Carsharing on Household Vehicle

Holdings. Transportation Research Record: Journal of the Transportation Research Board,

2143(1):150158, 2010.

33
I. Murtazashvili and J. M. Wooldridge. Fixed eects instrumental variables estimation in

correlated random coecient panel data models. Journal of Econometrics, 142(1):539552,

2008.

N. O'Connor, M. R. Staord, and G. Gallagher. The Impact of Global Terrorism on Ireland's

Tourism Industry: an Industry Perspective. Tourism and Hospitality Research, 8(4):351

363, 2008.

G. G. Parker and M. W. Van Alstyne. Two-Sided Network Eects: A Theory of Information

Product Design. Management Science, 51(10):14941504, 2005.

J.-C. Rochet and J. Tirole. Platform Competition in Two-Sided Markets. Journal of the
European Economic Association, 1(4):9901029, 2003.

M. Rysman. The Economics of Two-Sided Markets. Journal of Economic Perspectives, 23

(3):12543, 2009.

R. Seamans and F. Zhu. Responses to Entry in Multi-Sided Markets: The Impact of

Craigslist on Local Newspapers. Management Science, 60(2):476493, 2013.

E. G. Weyl. A Price Theory of Multi-sided Platforms. American Economic Review, 100(4):

164272, 2010.

J. Wolfers. Did Unilateral Divorce Laws Raise Divorce Rates? A Reconciliation and New

Results. The American Economic Review, 96(5):18021820, 2006.

J. M. Wooldridge. Fixed-eects and related estimators for correlated random-coecient and

treatment-eect panel data models. Review of Economics and Statistics, 87(2):385390,

2005.

J. M. Wooldridge. Correlated random eects models with unbalanced panels. Manuscript

(version July 2009) Michigan State University, 2009.

34
Table 1: Airbnb's spatial and temporal penetration. Cumulative counts of Airbnb listings
per year in the ten most populous Texas cities.

San Fort Corpus

Houston Antonio Dallas Austin Worth El Paso Arlington Christi Plano Laredo
(Pop.) 2.16M 1.38M 1.24M 0.84M 0.78M 0.67M 0.38M 0.31M 0.27M 0.24M

2008 1 9 0 25 0 0 0 0 0 0

2009 6 13 7 146 2 0 1 0 0 0

2010 39 22 23 468 10 0 3 0 1 0

2011 169 72 109 1862 34 3 19 7 5 1

2012 425 171 271 5158 68 8 27 24 20 1

2013 695 271 422 7489 93 23 36 49 33 1

2014 891 346 526 8575 114 31 52 60 44 2

Table 2: City-months that experience positive Airbnb entry

2008 2009 2010 2011 2012 2013 2014

Jan. 0 3 11 39 70 100 121

Feb. 0 3 11 40 74 104 123

Mar. 1 4 12 46 77 107 126

Apr. 1 6 13 49 79 109 127

May 1 6 16 49 84 109 128

Jun. 1 6 17 49 86 111 128

Jul. 1 6 23 58 89 112 130

Aug. 2 6 26 60 90 114 −
Sep. 2 10 28 65 93 117 −
Oct. 2 10 29 65 96 118 −
Nov. 2 11 30 66 96 120 −
Dec. 3 11 34 67 98 120 −

35
Table 3: Dierence-in-dierences estimates of the impact of Airbnb
on hotel room revenue using dierent measures of Airbnb supply.

(1) (2) (3)

Revenue Revenue Revenue

log Cum. Airbnb Supply −0.039***

(−4.40)
log Inst. Airbnb Supply (TTL 3 mo.) −0.025***
(−2.82)
log Inst. Airbnb Supply (TTL 6 mo.) −0.035***
(−3.92)
log Hotel Room Supply −0.157*** −0.154*** −0.156***
(−6.25) (−6.12) (−6.21)
log Capacity 0.034 0.034 0.034
(1.50) (1.50) (1.50)
log Median Annual Wage −0.212 −0.364 −0.290
(−0.60) (−1.01) (−0.82)
Unemployment Rate −0.060*** −0.058*** −0.058***
(−4.48) (−3.98) (−4.10)
log Population 0.049 0.061 0.030
(0.33) (0.42) (0.21)
log Airline Passengers 0.150*** 0.138*** 0.148***
(3.24) (2.94) (3.23)
Is Reviewed −0.057*** −0.056*** −0.057***
(−3.03) (−2.94) (−2.98)
TripAdvisor Star-Rating 0.031*** 0.031*** 0.031***
(6.93) (6.81) (6.86)
N 294383 294383 294383
2
Within R 0.013 0.011 0.012

Note: The dependent variable is log Hotel Revenueikt . Cluster-robust t-

statistics (at the city level) are shown in parentheses. All specications in-
clude hotel xed eects, year-month xed eects, city-month xed eects,
and a city-specic quadratic time trend.

Signicance levels: * p<0.1, ** p<0.05, *** p<0.01.

Table 4: Dierence-in-dierences estimates of the im-

pact of Airbnb on hotel room revenue using city-specic
trends of increasing order.

(1) (2) (3) (4)

No trends Linear Quadratic Cubic

log Cum. Airbnb Supply 0.009 −0.025** −0.039*** −0.039***

(1.26) (−2.48) (−4.40) (−3.29)
N 294383 294383 294383 294383
Adj. Within R2 0.024 0.012 0.013 0.011

Signicance levels: * p<0.1, ** p<0.05, *** p<0.01.

36
Table 5: Dierence-in-dierences estimates of the impact
of Airbnb on hotel occupancy rates and prices.

(1) (2)
Occupancy rate Room price

log Cum. Airbnb Supply −0.005* −0.019***

(−1.66) (−2.84)
log Hotel Room Supply −0.132*** −0.060***
(−8.36) (−4.26)
log Capacity 0.075*** −0.007
(5.98) (−0.43)
log Median Annual Wage −0.263 −0.050
(−1.65) (−0.26)
Unemployment Rate −0.025*** −0.016**
(−4.50) (−2.47)
log Population −0.004 0.140**
(−0.09) (2.02)
log Airline Passengers 0.012 0.044**
(0.80) (2.22)
Is Reviewed −0.060 −0.129**
(−1.34) (−2.33)
TripAdvisor Star-Rating 0.002 0.008**
(0.86) (2.59)
N 264172 264172
2
Within R 0.018 0.012

Note: The dependent variable is Occupancy rateikt in column

1 and log Hotel Room Priceikt in column 2. Cluster-robust t-
statistics (at the city level) are shown in parentheses. All speci-
cations include hotel xed eects, year-month xed eects, city-
month xed eects, and a city-specic quadratic time trend.

Signicance levels: * p<0.1, ** p<0.05, *** p<0.01.

37
 Table 6: Local measures of Airbnb and hotel room supply.

(1) (2)
Within 1 mile Within 5 miles

∆2 (log Local Cum. Airbnb Supply) −0.032** −0.025**

(−1.97) (−2.15)
∆2 (log Local Hotel Room Supply) 0.006 0.016
(0.95) (0.69)
∆2 (log Hotel Room Supply) −0.005 −0.004
(−0.19) (−0.18)
∆2 (log Capacity) −0.026 −0.026
(−1.18) (−1.18)
∆2 (log Median Annual Wage) −0.131 −0.129
(−0.38) (−0.37)
∆2 (Unemployment Rate) −0.017* −0.017*
(−1.88) (−1.86)
∆2 (log Population) −0.156 −0.155
(−0.89) (−0.89)
∆2 (log Airline Passengers) 0.174*** 0.176***
(5.49) (5.57)
∆2 (Is Reviewed) 0.043*** 0.043***
(2.65) (2.67)
∆2 (TripAdvisor Star-Rating) −0.007* −0.007*
(−1.66) (−1.67)
N 285187 285187
2
Within R 0.0011 0.0011

Note: The dependent variable is ∆2 (log Hotel Revenueikt ), where ∆2 is the

second dierence operator. Cluster-robust t-statistics (at the city level) are
shown in parentheses. All specications include hotel xed eects, year-
month xed eects, city-month xed eects, and a city-specic quadratic
time trend.
Signicance levels: * p<0.1, ** p<0.05, *** p<0.01.

38
Table 7: Dierence-in-dierences estimates of heterogeneity in Airbnb's impact on hotel room
revenue.

(1) (2) (3)

Price segment Meeting space Operation

log Cum. Airbnb Supply −0.016 −0.033*** −0.038***

(−1.61) (−3.58) (−4.23)
Price segment × log Cum. Airbnb Supply (ref. Luxury)
Budget −0.039***
(−5.39)
Economy −0.031***
(−6.02)
Midprice −0.020***
(−5.20)
Upscale −0.007
(−1.45)
w/o Meeting Space × log Cum. Airbnb Supply −0.015***
(−4.28)
Independent × log Cum. Airbnb Supply −0.008**
(−2.53)
log Hotel Room Supply −0.158*** −0.158*** −0.158***
(−6.26) (−6.27) (−6.26)
log Capacity 0.034 0.035 0.033
(1.49) (1.53) (1.50)
log Median Annual Wage −0.225 −0.219 −0.215
(−0.64) (−0.62) (−0.61)
Unemployment Rate −0.060*** −0.060*** −0.060***
(−4.46) (−4.46) (−4.47)
log Population 0.086 0.058 0.047
(0.63) (0.39) (0.31)
log Airline Passengers 0.151*** 0.150*** 0.150***
(3.28) (3.26) (3.24)
Is Reviewed −0.032** −0.047*** −0.056***
(−2.12) (−2.64) (−2.97)
TripAdvisor Star-Rating 0.026*** 0.029*** 0.031***
(7.15) (6.94) (7.00)
N 294383 294383 294383
2
Within R 0.018 0.014 0.013

Note: The dependent variable is log Hotel Revenueikt . Cluster-robust t-statistics (at the city level) are shown
in parentheses. All specications include hotel xed eects, year-month xed eects, city-month xed eects,
and a city-specic quadratic time trend.

Signicance levels: * p<0.1, ** p<0.05, *** p<0.01.

39
Table 8: Robustness checks: the rst column tests an alter-
native functional form for cumulative Airbnb supply; the
second column estimates Airbnb's impact using a CEM-
matched subset of hotels.

(1) (2)
Revenue Revenue

log Cum. Airbnb Supply −0.043***

(−4.25)
Cum. Airbnb Supply (ref. 0)
1-99 listings −0.020
(−1.40)
100-999 listings −0.063**
(−2.05)
1000+ listings −0.085**
(−2.16)
log Hotel Room Supply −0.152*** −0.151***
(−6.06) (−5.70)
log Capacity 0.034 0.075**
(1.50) (2.40)
log Median Annual Wage −0.432 −0.246
(−1.15) (−0.66)
Unemployment Rate −0.059*** −0.055***
(−3.87) (−3.55)
log Population 0.128 0.152
(0.78) (1.12)
log Airline Passengers 0.127*** 0.165***
(2.65) (3.52)
Is Reviewed −0.056*** −0.050***
(−2.93) (−2.72)
TripAdvisor Star-Rating 0.031*** 0.034***
(6.79) (6.40)
N 294383 188818
2
Within R 0.011 0.015
CEM Sample No Yes

Note: The dependent variable is log Hotel Revenueikt . All specica-

tions include hotel xed eects, year-month xed eects, city-month
xed eects, and a city-specic quadratic time trend.

Signicance levels: * p<0.1, ** p<0.05, *** p<0.01.

40
 ● ●
● ● ●
●
●

● Airbnb ●
Hotels ●
●

● ●
●●●
●
●● ●●
●
●
● ● ●
●●
● ●●●
● ●
● ●
● ●● ● ●
●
●●
● ●●
●
● ●●
●
● ●
●
● ●
● ● ●
● ●
●
● ● ● ●
●
●
● ●
● ●
● ●●●
●
● ● ● ● ●
●
●●●
● ●●●
● ●●● ● ●● ●
●
●
● ● ● ●●
●
●
●
●
●● ● ● ●● ●
●
● ● ● ● ●●
●
● ● ● ● ● ● ● ● ●
●●
●●● ●● ● ● ●
● ● ● ● ●●● ● ●
●●● ● ●● ● ●●●
● ●
● ●●●●●●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●●
● ●
● ● ● ●● ●●●
●● ●
●●
●
●
●●
●
●
●
●
●
●
●
● ●● ●● ● ● ●● ●
●
●●●
● ●
●● ● ● ●●●
●●●
● ●●●
●
●●●
●●●
●●●
●●
●
●
●
●
●●
●●
●
●
●
●
●
●
●●
●
●●●● ● ● ●● ● ● ● ●● ● ●● ●●
●
● ●●●
●
●
●
●●
●●●●
●
●
●●
●
●
●●●
●
●●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●● ●●
●
●
● ● ●●
●
● ●●
●
●●
●●●●
● ● ●● ● ● ● ● ● ●●●
● ●●
●●
●
●●
●●
●
●●
●●● ●●●● ● ●● ●
●
●●
●
●●●
●●
●
●●
●
●
●●
●
●
●
●●● ●● ●
●
● ●
●
●●
●
●●
●●
●
●
●
●
●●
●●
●
●
●
●
●
●●
●●
●●
●
●● ●
● ● ●
Texas ● ●
● ● ●●● ● ● ● ● ●● ●●
●
●● ● ●● ●●● ●● ●●●● ● ●●● ●●● ●
● ●
● ● ● ●
● ●●●● ●● ● ● ●● ●
●
● ●
● ● ● ●
● ●
●
● ●●
● ● ●
● ●
● ● ●
● ●●
●
● ●● ●● ●
●● ●●
● ● ● ● ● ● ●● ● ● ●
● ● ● ●●●
●
● ●
● ● ●● ●
● ● ● ● ●
●●
●● ● ●● ● ●● ●
●●
● ● ●● ●● ●
●
●●
●
●●
● ● ●
●
● ●● ● ● ●● ●
● ● ● ● ● ●
● ● ●
●● ●● ●● ● ●
●● ● ● ● ● ●
●
●
●
●
● ● ● ● ●●
● ● ●
● ●
● ●
● ● ●● ● ●
● ●● ●
● ● ● ● ●●● ● ●● ●
●
● ● ●●
●
● ● ● ●
● ●
● ● ●
●
● ●
●● ●
●
●●●
●●
●
● ● ●●
●
● ● ● ●
● ●
● ●● ● ●
● ●●● ●● ● ●● ●● ●● ● ●● ●●● ● ●
● ●
● ● ●
● ● ●●● ● ●
●
●● ● ●
●●
● ● ● ● ● ● ●●
● ●
● ● ●
●
●●
●
●●
●
●
●
●●
●
●
● ● ●
●● ● ●● ●●● ●●●
●●
●
●●● ● ●
●● ●
●●
●
●
●●
●
●
●●
●
●
●
●●
●
●
●
●●●● ● ●●●● ●
● ●
● ●
●●
●●● ● ● ● ● ●● ● ●●
● ●
● ●● ● ●●●●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●●
●●
●
●
●
●
●●
●
●●
●●●
●
●
●●
●
●
●
●
●
●
●
●●
●●
●
●
●●● ● ● ● ● ● ● ●●
●
● ●
●
●●
●
●
●
●
●
●
●
● ● ●
●
●
●● ● ●●
●
●
●●
●●
●
●
●
●
●●
●
●●
●
●●●
●
●
●●●
●● ● ● ● ●
●
● ●
●● ● ● ●● ●●
● ●
● ● ● ●● ●●
●
● ● ●● ● ●●●●●● ● ●● ● ● ●●
●●● ● ●●●●● ●● ● ●●● ● ●
● ● ● ●●●
● ●
● ●● ●● ●●
●●● ● ● ●
● ● ●
●
●●
●
●
●●
●●
●●●● ●
●● ● ●
●
●
●
●●●
●
●
● ●●●
● ●
● ●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●●
●
●
● ●
●
●
●
● ●● ●
● ●●
●●
● ●●●●
●● ●●●●●
●
●
●
●
●●●● ● ●
● ●●
● ●●●
● ●●● ●
●
●●
●
●
●●●
●
●
●
●●
●●
●
●
●●
●●
●
●●
●●
●●
●●●●
●
●●●● ●
● ●●
● ● ●●
●
●●●
●●
●
●●
●
● ●
● ●
● ●●
● ● ●●●● ● ●
●
●●
●●
●
●
●
●●
●
●●
●●
●●●
● ●
● ●● ●● ●
●
●
●
●
●
●
●
●
●
●
●●
●●●
●
●●
● ●
● ●●
●●
●●●●
●●● ●
● ●
● ● ● ●
●
●
●●
●
●
●
●
●
●
●●
●●
● ● ●● ● ● ● ● ●
●
●●
●●
●
●●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
● ●●●
●●
●●
●
●
●
●
●●●
● ● ●●
●●
●● ●
●
●● ●
●
●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●●
●● ●
●●●
● ●
● ● ● ● ●
●
● ●●
●●
●
● ● ●●
●●●
● ● ●● ●●
● ●● ● ●
● ●● ●● ● ●
●
● ● ● ●
●
● ●●●
●
●
● ● ●●
●
●
●
●
● ● ● ● ● ●
● ●
● ● ● ● ●
● ●● ● ●
● ●
●● ● ●
● ●
●
●
● ●● ● ● ● ●●●
●
●● ●
● ●
● ●● ●●● ●●●●●
●●
● ●
●●
● ●
● ●
●
● ● ●
●●
●
● ● ● ●
●
●●
●
●●●
● ● ●
●●
●
●
●
● ●
●●●● ●●● ●● ●●●● ●
● ●●
●● ● ●● ●● ●●●
● ●
●
● ●● ● ● ●
●
● ●
● ●
● ●
0 200 400 km ●
● ●
● ●
●
●●●
●
●
● ● ● ●●
●● ●
●
●
●
●
●●
●●
●
●●●
●●●
●●
●
●●●●●
●
●●
●
● ●
●●
●
●
●●
●

● ●
●● ●●
● ●
●
● ●
●●
● ●
●● ●
● ● ●● ● ●
● ●●
●
● ● ● ● ●
●●● ● ● ● ●
● ● ●
● ● ●●●
●●
●
● ●● ●
●● ●●
● ● ● ●
●
●●
● ● ● ● ●● ● ● ●
● ● ● ● ●
● ●●
●● ●● ● ● ●
● ●
●● ● ● ● ●●
● ● ●
●
●● ●
Travis County

●● ● ●●● ● ●● ●
●
● ● ●
● ● ● ●● ● ●●
● ● ●●●●
●
● ●
● ● ● ●● ●
● ● ● ●●●●
●● ● ●●●
● ●
●● ● ●● ●
●● ●● ● ● ●●●● ●● ●● ● ● ●●● ●●
●
● ●●
●●
●
● ●●
● ●
●
●
● ● ●●
●●●
● ●●● ●●
●●
● ●●●●●
●●
●●
●●
●●●●●●●●● ●
● ●● ●
●
●●
● ● ● ● ●●● ● ● ● ● ●● ●
●● ● ●
●●
●
●
● ● ● ●●●●
● ●● ●●●●●●● ●
●● ●●
●●
●●
●● ● ● ●
●●
●
●
●● ●
●●● ● ●● ●●● ●
●
●●● ●
●●● ●
●●
● ● ● ● ● ●●
● ● ●
●● ● ●●
●● ●● ●
●●●
●
●●
●
●
● ●●● ●● ●
●● ●● ●● ●
●
● ●● ●
● ●● ● ●
● ●
●●
● ● ●
●
● ●●
● ●
●●● ●●
● ●●●● ●
●●●
●●
●●
● ●
●
●
●●
●
●
●●
●
●
●
●
●
●
●●
●
●
● ●● ●●
●● ● ● ●●●
●
●
●
●
● ●
● ● ●● ●● ● ●●
●●●
●●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●●● ● ● ●●
●
●
●●●● ● ●
●
● ●●
● ● ●
●●
● ●●
●
●●
●
●●●
●
●
●●
●
●
●
●●
●●
●
●
●●
●●
●
●
●
●●
●
●
●●
●
●● ● ●●●
● ● ●
●
● ●
●●●●●●●
●
●
●●
●
●●
●
●●
●
● ●
●●
●●
●●● ●
●●●● ●●●●●●
●
●●
● ●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
● ●●
●● ●●
● ● ● ●
●
● ●● ●●●●●●
●
●
●●
● ●
●
●
●
●
●●●
●
●
●
●●
●●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●●
●
●
●
●●
●
●●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●●●
●●
●
●
●●
●
●●●
● ● ● ●●
●●●
●
● ● ●●
●●●
●
●●
●
●●
●
●●●
●
●●
●
●●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●●●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
● ●
●
●●
●
●
●
●●
●●
●●
●●
●
●●
●
●●●
●●●
●● ●●
● ● ●
●
●
●
●
●
●
●
● ● ●● ●●
● ● ●●● ●
●●
●●
●
●●
●
●
●
●●
●
●
●●
●
●
●●
●
●●
●
●●
●
●
●●●
●
●●
●
●●
●
●●
●
●
●●
●●
●
●●
●
●
●●
●
●●●
●● ●
●●● ● ●
●●●
● ●● ●
● ●●
●
●● ●
●● ●
●
●
●●
●
●
●●
●●●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●●
●
●
●●
●
●●
●●
●
●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●
●
●●●● ●●
●● ● ●●●
●
● ●● ●●
● ●●
●●● ●●
●●
●●
●
●● ●
●●●
●●●●
●
●●●
●
●
●●
●
●
●●●
●
● ●●●●
●
●
●
●
●
●●
●
●● ●●
●
●
● ●● ●
● ●
●
●● ● ●●●
●
●
●●
●
●●
●
● ●
●●
●
●
●●●
●
●●●
●
●
●
●
●
●●
●●●
●
●
●●
●●
● ●
●●●
● ●● ●
● ● ●● ●●● ●●
● ●●
●
●
●●
●
●
●
●
●●
●
●
●
●
●
●●
●●
●
●●
●
●●
●
●
●
●
●
●
●
●
●●
●
● ●
●
●
● ●●
●● ● ● ●● ●
● ●●●●
● ●
●● ●●●● ●● ● ● ●
● ● ● ●● ●●●● ● ●●● ●●●● ●
● ●●●
● ●● ●●●●● ● ●
●●●
●● ●●
●●
●
●●●
● ●●●● ●●
●
● ●
● ● ● ●●● ●●
●
●● ● ● ● ●
● ● ●
● ● ●● ● ●
● ● ● ●● ● ●
●●● ● ● ● ● ● ● ● ●
●●
●● ●
●
● ●●● ● ● ●● ●
● ●
● ●
●
●● ●●
● ● ●
● ●●
● ●● ● ● ●
● ● ● ●
●● ●● ● ●
●● ●
● ● ●
● ●● ● ● ● ●
● ●● ● ● ●
●
●
0 10 20 30 km ●
● ●●● ● ● ●
● ●
● ●●
●
●● ●
●●●●
● ● ●
● ●● ● ● ● ● ●● ●
● ● ● ● ●
● ● ● ●
●
●● ●
● ●
●●

Figure 1: Geographical distribution of hotels and Airbnb listings in the state of Texas (top)
and in Travis County, TX (bottom) in 2013.

41
Figure 2: A typical Airbnb listing.

42
Figure 3: A typical Airbnb user prole.

43
 1000 Acc. sector empl. (log) Hotel occupancy rate
Capacity 0.4
0.04
Fewer than 5 rooms 0.0
0.00
At least 5 rooms −0.4
750 −0.04
Tax paying properties

−0.8
Hotel room price (log) Hotel room supply (log)
0.25 0.2
500 0.1
0.00
0.0
−0.25 −0.1

Population (log) Unemployment rate

250
0.3
0.1
0.2
0.1 0.0
0.0 −0.1
0
2003 2005 2007 2009 2011 2013 2008 2010 2012 2014 2008 2010 2012 2014

Figure 4: Annual counts of Austin properties Figure 5: Correlation between Airbnb sup-
that pay hotel occupancy tax, broken down ply and pre-Airbnb (year 2007) city charac-
by capacity. teristics, with 95% condence intervals.

Austin Dallas Houston San Antonio

8000

Cumulative
6000
4000
Airbnb supply

2000
0
SXSW

Instantaneous
2000

1000

0
2004 2009 2014 2004 2009 2014 2004 2009 2014 2004 2009 2014

Figure 6: Cumulative vs. instantaneous Airbnb. The seasonal peaks of the Austin instanta-
neous supply curve correspond to SXSW.

Pre 529 days 100 days Final 228 days Entered 542 days Projected
Planning
planning planning construction opening

Figure 7: Average time between various stages in the hotel pipeline construction.

44
 Budget Economy Midprice Upscale Luxury
YoY change in prices

20%

10%

0%

−10%
2010 2015 2010 2015 2010 2015 2010 2015 2010 2015

State Fair of Texas Rest of the year

Figure 8: Year-over-year changes in Dallas hotel prices broken down by hotel price level.
The solid line displays changes during the State Fair of Texas (October) while the dashed
line displays changes for the rest of the year.

Budget Economy Midprice Upscale Luxury

YoY change in prices

20%

10%

0%

−10%
2010 2015 2010 2015 2010 2015 2010 2015 2010 2015

SXSW Rest of the year

Figure 9: Year-over-year changes in Austin hotel prices broken down by hotel price level.
The solid line displays changes during SXSW (March) while the dashed line displays changes
for the rest of the year.
SXSW hotel pricing power

10%
30000
SXSWi attendance

YoY changes in

5%
20000
0%

10000 −5%

0 −10%
1998 2002 2006 2010 2014 2007 2009 2011 2013 2015

Figure 10: SXSWi attendance Figure 11: Percentage changes in year-over-

year Austin hotel peak pricing power (SXSW
vs. rest of the year).

45