When Discounts Hurt Sales: The Case of Daily-Deal
Markets
Zike Cao, Kai-Lung Hui, and Hong Xu†
June 2015
Abstract
Online daily-deal websites such as Groupon.com and LivingSocial.com have proven
to be a novel price promotion tool for local merchants in geographic markets. Using
sales data that we collected from Groupon.com, we investigate the effectiveness of
this promotion tool by quantifying the impact of discount percentages on products
or services sales. We have conducted multiple instrumental variable (IV) estimations, robustness tests, validation and falsification exercises, and concluded that a
larger discount percentage leads to less product sales in daily-deal markets. Specifically, a 1% increase in a deal’s discount percentage would decrease its sales by
0.0195% per hour. This effect is more prominent when the deal is credence goods,
when it only accumulates few sales, or when the merchant is located at a wealthier or more educated city. Thus, the negative effect of discount percentages could
a result of consumers’ quality concerns; that is, a large discount percentage can
become a warning signal of product quality to consumers. Our results have important managerial implications for both local merchants who participate in daily-deal
websites and the daily-deal website owners.
†
Department of Information Systems, Business Statistics, and Operations Management, Hong Kong University of Science and Technology, Clearwater Bay, Kowloon, Hong Kong, zcaoaa@ust.hk, klhui@ust.hk,
hxu@ust.hk. This study is part of a PhD student’s dissertation.
1.
Introduction
Price promotions account for a significant share of the marketing mix budget in
many industries (Blattberg et al., 1995; Pauwels et al., 2002). Growing phenomenally
during the past several years, online daily-deal platforms have emerged as an innovative
form of price promotion that allows local merchants to advertise their products or services
to new customers through deep discounts. Daily-deal platforms are especially favorable
to small- and medium-sized local businesses with limited marketing budget, because they
do not pay upfront fees for running promotions but share a proportion (typically 50%) of
the resultant revenue with the daily-deal platforms.
The effectiveness of a deal hinges crucially on how many new customers can be
attracted by the discount. While discounts are often appealing to price-sensitive consumers, a large discount may also raise consumers’ quality concerns, and therefore, dissuade purchases (Erdem et al., 2008). Daily deal platforms are especially vulnerable to
such negative impact of discount due to the high product uncertainty associated with
online shopping (Overby and Jap, 2009). The spatial and temporal separation between
buyers and sellers further intensifies the information asymmetry between the transacting
parties in online markets (Ghose, 2009; Pavlou et al., 2007; Li et al., 2009). Moreover,
such quality concerns are exacerbated for experience and credence goods (Dimoka et al.,
2012), which represent the typical types of products and services offered on daily-deal
platforms.1
In this study, we investigate the impact of the depth of discount on consumers’ demand in daily-deal markets.2 By examining the demand side of a daily-deal platform, we
will be able to offer important guidance to both daily-deal platforms and the participating
1
Experience goods are products that can be evaluated only after consumption (Nelson, 1970); Credence
goods are products that are difficult or impossible to evaluate even after consumption (Darby and Karni,
1973).
2
Other decisions for a merchant to make include supplying quantity, duration of the deal, number of
days until expiration after purchase and number of options for customers to choose from.
1
merchants on how to effectively design a price promotion.
We have compiled a large panel data set from one of the leading firms in the dailydeal industry, Groupon.com, from January 8th, 2014 and March 31st, 2014. We have
collected 19,978 deals covering a wide range of categories, from 172 cities in the United
States and Canada. We recorded the sales information for each deal every hour, and
reached a total of 1,835,794 observations.
As a summary of the findings, we first discovered a significant and negative effect
of discount percentage on the sales that a deal can accumulate during each hour of its
duration. To our best knowledge, we are the first to document such an immediate negative
effect of discount on sales.
We further conducted two instrument variable (IV) estimation to address the potential endogeneity concern for discount percentage. Such concern may arise due to
unobservable and omitted variables (e.g., a merchant’s store location, environment, or
reputation). Our first IV estimation followed the spirit of the BLP method that is commonly used to address endogeneity of prices in differentiated products markets (Berry,
1994; Berry et al., 1995; Nevo, 2000; Berry and Haile, 2014). Our second IV estimation
used the average hourly wage (AHW) and the housing price index (HPI) for each deal’s
associated industry and local market as instruments.
The first IV test concluded that the negative effect of discount percentage on demand continue to hold in the Entertainment and Restaurants category. The second IV
estimation also showed that the negative effect of discount depth survived with a marginal
significance (p = 0.0998). Overall, these IV tests give us some confidence that the negative effect of discount is robust to the potential endogeneity concern caused by omitted
variables.
A reasonable interpretation of the negative effect of discount is that consumers have
increasing quality concerns for merchants offering deeper discounts. Several additional
tests buttressed the above explanation. We first found that the negative effect of discount
2
is more salient for credence goods than experience goods, since the former exhibit higher
quality uncertainty. Second, we found that when a deal has already sells a large number
of vouchers, it spurs less quality concern, possibly because popularity serves as a positive
signal of product quality (Bikhchandani et al., 1998; Moretti, 2011). Lastly, we found
that consumers in wealthier and more educated markets respond more negatively to deep
discount. In addition to support the underlying mechanism for the negative effect of discount depth, these findings also have immediate and specific implications for managerial
practice.
2.
Related Literature
This study is closely related to the literature on price as a signal of unobservable
product quality in economics and marketing. There is an extensive literature that have
established the role of price as a credibly signal to quality in both monopolistic (e.g.,
Bagwell and Riordan, 1991) and oligopolistic settings (e.g., Wolinsky, 1983; Daughety
and Reinganum, 2007, 2008; Janssen and Roy, 2010). The signaling effect of price arises
with the presence of informed consumers, who are only willing to patron the high-quality
sellers (who charge high price due to high production cost), and consequently, low-quality
sellers find it optimal to set a low price due to its low production cost (Wolinsky, 1983;
Bagwell and Riordan, 1991). In a more competitive market, even without informed
consumers, prices are used as a competition tool among the sellers, which may lead to
an equilibrium where high-quality products are sold at a higher price than low-quality
products (Daughety and Reinganum, 2007, 2008; Janssen and Roy, 2010).
Another prominent reason why price may function as a credible signal of quality is
the repeat-purchase mechanism (Tirole, 1988). Firms will maintain high quality because
they can command a sufficiently large price premium from rational consumers that makes
the discounted stream of rents on repeat sales outweigh immediate gains from cheating
(Klein and Leffler, 1981; Shapiro, 1983; Rao and Monroe, 1996; Orosel and Zauner, 2011).
Researchers from both Information Systems (IS) and economics have studied prod3
uct uncertainties in online markets. Overby and Jap (2009) discovered that consumers
are more likely to conduct transactions through electronic channels when the associated
quality uncertainty is low. Consumers’ quality concerns in online markets mainly arise
from the information asymmetry between buyers and sellers, which leads to both adverse
selection issues, where buyers have limited capability to assess the quality of different
sellers before purchase, and moral hazard issues where sellers intentionally misrepresent
their product quality to online buyers (Dewan and Hsu, 2004; Jin and Kato, 2006; Pavlou
et al., 2007; Ghose, 2009; Li et al., 2009; Dimoka et al., 2012).
Several studies in marketing have discovered that frequent price promotions or
“deals” may negatively affect brand sales in the long-run as consumers adjust their reference prices (Lattin and Bucklin, 1989), increase price and promotion sensitivity (Mela
et al., 1997), engage in stockpiling behavior (Mela et al., 1998), lower perception of brand
image (Jedidi et al., 1999; Erdem et al., 2008). Our results complemented this stream
with a novel perspective: price promotions can even have an immediate negative effect
because consumers hold increasing quality concerns for deeper discounts.
There have also been several empirical and theoretical studies on daily-deal websites
(see, e.g., Jing and Xie, 2011; Hu et al., 2013; Shivendu and Zhang, 2013; Li and Wu,
2013; Subramanian and Rao, 2014; Wu et al., 2015). To our best knowledge, we are
the first to empirically study the impact of discounting depth on deal sales in daily-deal
markets.
3.
3.1.
Background and Data
Overview of Daily-deal Websites and Groupon.com
Daily-deal websites represent a new e-commerce business model where merchants
sell vouchers to consumers at a discounted price. A daily-deal website publishes multiple deals from different categories every day, and each deal can last several days.
Groupon.com, established in November 2008, was the first daily-deal website to focus
4
on local businesses. By the second quarter of 2014, Groupon has operated in over 150
local markets in North America and over 100 markets in Europe, Asia and South America. The website has sold more than 600 million deals from over 650,000 merchants.3
Groupon thus holds a dominant position over its close rivals such as LivingSocial.com
and Woot.com (Foresee, 2012).4
Each deal on Groupon.com has its dedicated page to display all relevant deal information, such as descriptions of the deal and the merchant (in text and images), the
original price, the discounted price, the discount percentage, the number of vouchers sold,
terms of use, etc.. Some deals include the number of Facebook Fans and online reviews
about the merchant from third-party websites, such as Yelp, Yahoo!, Google, etc., on
their webpages.5
3.2.
6
Data
We collected data about 19,992 completed deals from 8 January 2014 to 31 March
2014. Using a proprietary software, we scraped down a list of attributes from each
deal’s Groupon webpage, including a deal’s original price, transaction price, discount
percentage, number of days until the deal voucher expires after purchase, the maximum
quantity allowed to be purchased per occasion, the number of different purchase options,
detailed terms of use specified by the merchant under the “Fine Print” section, and if
available, the number of Facebook fans, review quotes from third-party websites, as well
as the volume and valence of the reviews. We monitored each deal’s sales every hour,
using the Groupon Application Programming Interface (API). We also knew through
3
Source. https://www.grouponworks.com/benefits-of-advertising-with-groupon.
4
The field survey conducted by Foresee in 2012 showed that 60% of online visitors to the top 40
websites enrolled in at least one daily deal email program. Among the respondents, 52% used Groupon
while only 30% used LivingSocial and 9% Woot.
5
Please see Figure A.1 in the appendix for a screenshot of a typical deal page.
6
In the early days, merchants typically required a minimum number of purchases to “tip” a deal. Such
threshold requirement is no longer in use in Groupon, and our data shows that all merchants specify a
zero tipping point. See Wu et al. (2015) for an empirical analysis of the threshold effect on consumers’
communication behaviors.
5
Groupon API: (i) whether a deal webpage was created by Groupon or by the merchant
using Groupon’s deal-building tool; and (ii) whether or when a deal was sold out before
the end of its preset duration.
Our data has two limitations. First, when the number of vouchers sold is between
10 and 1,000, the API will round down the number to the closest multiple of 10, instead
of reporting the actual number.7 Second, when the number of vouchers sold is above
1,000, Groupon API will not report any update in sales before it reaches 5,000. The first
limitation may not change our result dramatically because our whole data are trimmed
by Groupon API in the same fashion. To address the second limitation, we dropped 14
deals with over 1,000 sales, resulting in 19,978 deals for the analysis.
Table 1 shows the percentage of deals from each of the 12 category. Each category
further contains multiple subcategories. See Table A.1 in the Appendix for a list of
categories and their subcategories.
** Table 1. Percentage of deals in each category **
Our sample covers 172 geographic markets defined at the level of cities or equivalents in the U.S. and Canada (152 American markets and 20 Canadian markets).8 To
supplement the Groupon data, we also compiled division-specific income and education
level data. We first matched each U.S. city with its corresponding county and then obtained from the 2010 U.S. Census (census.gov) the median household income and the
percentage of population aged 25 and above with a Bachelor degree or higher at the
county level.
We further obtained the national monthly industry-specific occupational employment and wage estimates for the 1st quarter of 2014 from the Bureau of Labor Statistics
(BLS, bls.gov)’s Current Employment Statistics (CES). We are then able to match each
7
For example, 89 is reported as 80 and 101 is reported as 100. Subramanian and Rao (2014) also had
a similar observation on this problem.
8
See Table A.2 in the Appendix for a complete list of the 172 local markets.
6
subcategory in our data with one (and only one) industry code in the BLS data.9 We also
gathered from the Federal Housing Finance Agency (FHFA, fhfa.gov) the U.S. monthly
house price index (HPI) in the nine census divisions for the 1st quarter of 2014. Each
deal in our data was matched with an HPI value based on the starting date of the deal
and the location of the merchant.
** Table 2. Summary statistics **
** Table 3. Correlations **
Figure 1 plots the empirical distribution of the focal decision variable — discount
percentage. Clearly, the discount depths in our data exhibit substantial variation, with
the mode of discount depth being 50%. It bears noting that the majority of the merchants
offered at least 30% discount off.
** Figure 1. Distribution of discount percentage in the data **
4.
Model and Results
We first supply an overview of the data in Figure 2. Panel (a) describes deals’
performance at different stages relative to the discount percentage, by plotting the sales
volume each deal achieved during the 30th, 40th, 50th, and 60th percentile hour of its
duration in each of the four graphs. The fitted trend in dotted line from all four graphs
demonstrate a clear downward slope. Panel (b) plots each deal’s total sales against the
corresponding discount percentage, and also yields a clear negative slope.10
** Figure 2. Deals’ average sales and discount percentages **
9
BLS uses the North American Industry Classification System (NAICS) in this data. For more details
on the implementation of NAICS in BLS, readers are referred to http://www.bls.gov/bls/naics.htm.
10
Note that, for both panels, we have dropped deals whose discount percentage is below the 5th percentile due to limited number of observations. In addition, in Panel (b) we have excluded deals with
duration below the 20th percentile or above the 80th percentile to avoid potential confounding influences
by a deal’s duration on its accumulated sales.
7
Our main econometric model is of the following specification:
ln Sijt = β1 ln discounti + β2 ln pricei + β3 ln CSij,t−1 + Xi α + γj + τt + ϵijt ,
(1)
where Sijt is the sales for deal i from market j during hour t; discounti is the discount
percentage that deal i offers off the original price; pricei is deal i’s transaction (discounted)
price; CSij,t−1 is deal i’s cumulative sales from hour 0 to t − 1; Xi is a set of control
attributes for deal i, included subcategory-fixed effects; γj represents city-specific effects;
and τt captures hour-specific effects. We specified all variables except dummies and the
average online review ratings in logarithms.11
We first fitted Equation (1) to the whole data by excluding the volume and average
rating of third-party online reviews in Xi , due to large number of missing values on
the two variables. As reported in Table 4, Column (1), the estimates, except that of
discount percentage, are largely consistent with our expectations. For example, consumers
generally have less demand for high-priced goods; Quantity sold up to the previous hour
is significantly and positively correlated with quantity sold in the current period. The
most surprising finding, and also the focus of this study, is that the coefficient of discount
percentage, −0.0195 (p < 0.01), is negative and very precisely estimated. Accordingly, a
1% increase in a deal’s discount percentage decreased its sales quantity by 0.0195% per
hour.
When we estimated Equation (1) using the subsample consisting of deals whose
webpages displayed third-party online review information (about one third), the coefficient of discount percentage, −0.03 (p < 0.01), has a even larger magnitude and is still
highly significant (Table 4, Column (2)). We also estimated Equation (1) separately in
each of the three major categories: Entertainment, Beauty and Spas, and Restaurants.
Column (3) - (5) in Table 4 report the estimates respectively. The results largely remain
11
As appropriate, we added one to the variable to avoid logarithms of zeroes.
8
consistent, except that in the Beauty and Spas category the negative effect of discount
percentage became insignificant, −0.0186 (p = 0.20).
** Table 4. The negative effect of discount depth **
We then conducted two IV estimations to address the potential endogeneity of discount decisions. Our first IV estimation borrowed from literature on identification of
demand in differentiated products markets (e.g., Berry, 1994; Berry et al., 1995; Nevo,
2000; Berry and Haile, 2014). So we focused in the Restaurant and Entertainment category, which each could be treated as a differentiated products market. We then defined a
focal deal’s competing products as those same-category deals that were offered by other
merchants located in the same city and that preceded or coincided with the focal deal
in opening time. We treated all other deal attributes except sales quantity, transaction
price, and discount percentage as exogenous characteristics, then instruments for deal i’s
discount percentage include (i) all exogenous variables of product i itself; and (ii) the
sum of exogenous variables of all product i’s competing products (Berry et al., 1995;
Hui, 2004). The intuition for these instruments can be obtained from a typical oligopoly
pricing model: products tend to have low markups and thus low prices (large discounts
in the Groupon setting) if they face good substitutes, and vice versa. Column 1 and
Column 2 in Table 5 reported this IV estimation in each of the two categories. The price
coefficients are negative and significant in both cases, and the values drop substantially
compared to OLS estimates in Table 4, which are consistent with previous findings (e.g.,
Berry et al., 1995; Hui, 2004). More important, discount depth still has a significant and
negative effect in both categories.
Our second IV test used the industry-specific average hourly wage (AHW), which
varied across time and subcategories, and monthly house price index (HPI), which varied
across time and across the nine U.S. census divisions, as instruments. We hope AHW
an HPI can to some extent reflect variations in merchants’ operation costs, which are
9
supposed to directly related to their discount decisions. The first-stage regression results
show that both AHW and HPI are negatively and significantly correlated with discount
percentage, confirming our intuition. More importantly, the second-stage regression results reported in Table 5, Column 3 show that there is still a marginally significant
negative effect of the discount percentage (−0.0609, p < 0.10).
4.1.
Underlying Mechanism
What explains this finding? We argue that consumers are quite concerned about
the product or service’s quality eventually delivered to them by the seller when they
purchase a deal on Groupon. Deep discounts exacerbate their such concerns or worries.
Consumers may believe high-quality sellers will generally choose lower discounts because
(i) high-quality products are probably more costly to produce, so lower discounts can save
more costs for high-quality sellers (Bagwell and Riordan, 1991), (ii) lower discounts are
more likely to guarantee that sellers will not cut quality (Klein and Leffler, 1981; Shapiro,
1983), and/or (iii) high-quality sellers might have a large base of informed customers who
know the sellers’ quality well (e.g., previous offline customers) and through whom product
information spreads. We next applied a battery of additional identification strategies that
scrutinized the heterogeneity of discount effect to support our arguments.
Experience vs. Credence Goods. We first estimated a model where the effect
of discount depth was allowed to vary depending on whether the product is classified
as a experience good (Nelson, 1970) or credence good (Darby and Karni, 1973). Unlike
experience goods whose utilities are known costlessly after consumption, credence goods’
qualities are expensive or impossible for consumers to judge even after purchase. Typical
examples of credence goods that have received much attention in the literature are medical
treatments, automotive repair services, home repair services and other kinds of expert
services (e.g., Darby and Karni, 1973; Emons, 1997; Dulleck and Kerschbamer, 2006).12
12
See Table A.1 in the appendix for the subcategories that are classified as credence goods (subcategories
that are italic and labeled with an asterisk).
10
If quality concern is indeed behind the negative discount effect we found, we should expect
that this negative effect is more severe for credence goods.
We added an interaction between discount depth and the dummy that indicates
whether a deal is credence goods in the regression:
ln Sijt = β1 ln discounti + β1C ln discounti × credencei + β2 ln pricei
(2)
+β3 ln CSij,t−1 + Xi α + γj + τt + ϵijt ,
Note that the main effect of credence is taken out by the subcategory-fixed effects and so
cannot be separately estimated. The estimates from this model are reported in Table 5,
Column 4. The coefficient of the interaction term credence×discount, β1C , has a negative
and significant estimate, −0.0488 (p < 0.05), implying that discounts significantly hurt
more severely the sales of credence goods. This is matched with our a priori expectation.
** Table 5. Identification tests **
Deal Popularity. Popularity is often an information source consumers resort to
in resolving quality uncertainties because a person might infer the value of the product
from the number of other people who have purchased the product (Bikhchandani et al.,
1998). This implies that when a deal’s sales volume is large enough, prospective consumers may have less worries about the product quality. Besides observational learning
by prospective consumers, larger existing buyer base may also increase the likelihood of
a deal receiving more buzz on social networking sites or physical social occasions, which
may also help reduce the concerns or uncertainties prospective consumers have about the
product quality (Moretti, 2011).
In line with this logic, we estimated a model of the following form:
ln Sijt = β1 ln discounti + β1P ln discounti × salesAboveT hresholdit + β2 ln pricei
+β3 ln CSij,t−1 + β4 salesAboveT hresholdit + Xi α + γj + τt + ϵijt ,
11
(3)
where salesAboveT hresholdit is a dummy to indicate whether deal i’s sales volume exceeds a certain threshold before hour t starts. For estimates from this specification reported in Table 5, Column 5, we chose the threshold to be 300.13 As our above arguments
predict, the estimate of β1P is positive and significant (0.0463, p < 0.01). Therefore, the
negative effect of discount percentage is significantly smaller in magnitude when a deal’s
sales volume exceeds 300.
Income and Education Effect. We also exploited variations in the average levels
of socioeconomic status (SES) across geographic regions. The two key SES indicators—
income and education—may systematically moderate how a consumer responds to deep
discount. High-income consumers may be generally less sensitive to prices and thus more
sensitive to quality. High-educated consumers may be more sensitive to quality when they
buy items on Groupon because they have better knowledge of the Groupon’s practices
and the tricks could be played by the merchants. As a results, discounts may seem less
alluring to high-income and high-educated consumers.
To test this, we estimated the following specification:
ln Sijt = β1 ln discounti + β1S ln discounti × SESj + β2 ln pricei
(4)
+β3 ln CSij,t−1 + Xi α + γj + τt + ϵijt ,
where SESj is city j’s income or education level. Note also that the main effect of SES
is taken out by the city-fixed effects.14 Estimates from this model are reported in Column
6 and 7 of Table 5. Our quality concern arguments would predict that discount effect is
more negative in geographic markets where consumers have higher average income and
education. We indeed found such patterns in our data.
13
The results are similar if changing the value of the threshold to 200 or 400.
14
Moreover, we also only used U.S. deals in this regression because we only obtained income and
education level data for U.S. cities.
12
4.2.
Falsification and Robustness Checks
To further validate our findings, we next did a falsification exercise using the data
on a sample of Groupon Goods deals we collected along with the main sample. Groupon
Goods is a online retailing channel launched in September 2011 and offers discounts on,
e.g., home furnishing products and electronics (Reuters, 2011). We expect consumers to
have less worries or uncertainties about products featured in Groupon Goods because (i)
products from Groupon Goods can be largely viewed as search goods; and (ii) consumers
deal directly with Groupon and Groupon gives consumers free return promises within
14 days of receipt.15 Therefore, offering discounts on products in this channel should
positively promote sales, at least not significantly hurt sales.
We obtained an analogous but substantially smaller (only 396 deals) sample of
Groupon Goods deals, because Groupon Goods channel featured much less deals compared to the channel for local merchants. We followed the same procedure in culling
the sales data as in the main sample, then we estimated a similar specification to Equation (1) based on this Groupon Goods sample (but Groupon Goods deals do not have
as many attributes as the deals offered by local merchants). The estimates, reported in
Table 4, Column 6, indicate that discount percentage have a significant positive coefficient, 0.236 (p < 0.01), which is in sharp contrast with the negative effect of discount
depth found before. This falsification test thus gives us more confidence in our quality
concern explanation for why deeper discounts offered by local merchants may lead to less
consumers to buy their products.
We also conducted several more robustness checks to examine the sensitivity of our
results. We first estimated a linear specification (Table 6, Column 1). We then estimated
a specification with the square term of lag cumulative sales (Table 6, Column 2) to allow
more flexible forms of social influence and diffusion effect. We also fitted Equation (1)
15
For more details on Groupon’s promise on this, see http://www.groupon.com/groupon-promise.
13
to a restricted data sample that discarded deals whose original price is below the 20th
or above the 80th percentile (Table 6, Column 3). This procedure could also help reduce
unobserved product heterogeneity that is not captured by the specification in Equation (1)
but related to product price. We also fitted Equation (1) to a restricted data sample that
discarded deals whose duration time is below the 20th or above the 80th percentile (Table
6, Column 4). Reassuringly, all the results largely remain consistent. One exception is
that when we dropped the lag cumulative sales (Table 6, Column 4), the discount depth’s
coefficient is not significant anymore but still negative signed.
** Table 6. Robustness checks**
Lastly, to see the implication of the limited number of lowly discounted deals, we
split the sample at the 5th percentile of discount percentage. We then fitted Equation
(1) separately to each subsample. The results, reported in Column 6 and 7 in Table
6 respectively, indicate that discount effect in the lowly-discounted subsample becomes
insignificant (although still has a negative sign), while the coefficient of discount estimated
from the highly-discounted subsample is still highly significant and its magnitude increases
substantially compared with the baseline estimate in Table 1, Column 1. These results
seem to suggest the negative influence of discount is probably mainly driven by deepdiscounted deals in our data and this finding may not apply when deals offers relatively
lower discounts.
5.
Implication and Conclusion
Previous research has generally agreed that “temporary retail price promotions
cause a significant short-term sales spike ... this result is fundamental to virtually all
research done in the area of promotions” (Blattberg et al., 1995, p. G123). On the other
hand, some researchers have found evidence of a long-term negative effect of price promotions, viz., a dip in sales after promotions in physical consumer package goods markets,
which could be because of downward-adjusted reference price, stockpiling behavior, re14
duced brand equity, etc. (Lattin and Bucklin, 1989; Mela et al., 1997, 1998; Jedidi et al.,
1999; Erdem et al., 2008). Contrasting and also complementing with prior research in
this stream, we suggest that the unique features of online daily-deal markets may render
price promotions even carry an immediate (short-term) negative influence on product
sales, viz., deterring consumer purchasing from the very beginning. This adds a novel
and important insight to marketing managers when they plan promotion campaigns on
experience/credence products, and/or in online distribution channels.
Evidently, the strategy of offering large discounts in the hope of catching consumers’
eyes may actually hurt merchants’ performance.16 We, therefore, would advise lower
discounts to merchants who are offering medium- to high-quality products or services.
For merchants who are offering credence goods, or who are located in a wealthier or moreeducated geographic market, such signaling effect is stronger and our advised strategy is
more applicable.17
For owners of daily-deal websites, it might be beneficial for them to impose policies
on the level of discounts allowed on their platforms. Lower discounts not only attract
more deal purchases but also generate higher revenue to be split between the merchants
and the platform owners.18
Moreover, it is crucial for the sites’ owners to know the heterogeneity of featured
merchants in terms of product quality and brand awareness whenever possible. Those
merchants that are of fairly good product quality but with poor brand awareness should
be the key clientele for daily-deal websites.19
16
http://www.geektime.com/2014/06/01/your-discounts-may-be-killing-off-your-sales/.
provides further empirical support to our suggestion that high discount hurt sales.
This article
17
But we are unable to conclude that the lower the discounts are, the better the sales performances
would be, as is suggested by our last robustness test. Answering this question probably requires more
careful designs in future research.
18
As noted before, a daily-deal website and each featured merchant have a revenue-sharing agreement
with fixed splitting ratio (typically 50/50).
19
Well-known merchants may have less incentives to participating in a daily-deal program in the first
place, or substitute daily-deal websites with other online and physical channels more easily, while recommending low-quality merchants probably would hurt the sites eventually.
15
References
Bagwell, Kyle, Michael H. Riordan. 1991. High and declining prices signal product quality. American
Economic Review 81(1) 224–239.
Berry, Steven T. 1994. Estimating discrete-choice models of product differentiation. RAND Journal of
Economics 25(2) 242–262.
Berry, Steven T., Philip A. Haile. 2014. Identification in differentiated products markets using market
level data. Econometrica 82(5) 1749–1797.
Berry, Steven T., James Levinsohn, Ariel Pakes. 1995. Automobile prices in market equilibrium. Econometrica 63(4) 841–890.
Bikhchandani, Sushil, David Hirshleifer, Ivo Welch. 1998. Learning from the behavior of others: Conformity, fads, and informational cascades. Journal of Economic Perspectives 12(3) 151–170.
Blattberg, Robert C., Richard Briesch, Edward J. Fox. 1995. How promotions work. Marketing Science
14(3) G122–G132.
Darby, Michael R., Edi Karni. 1973. Free competition and the optimal amount of fraud. Journal of Law
and Economics 16(1) 67–88.
Daughety, Andrew F., Jennifer F. Reinganum. 2007. Competition and confidentiality: Signaling quality
in a duopoly when there is universal private information. Games and Economic Behavior 58 94–120.
Daughety, Andrew F., Jennifer F. Reinganum. 2008. Imperfect competition and quality signalling. RAND
Journal of Economics 39(1) 163–183.
Dewan, Sanjeev, Vernon Hsu. 2004. Adverse selection in electronic markets: Evidence from online stamp
auctions. Journal of Industrial Economics 52(4) 497–516.
Dimoka, Angelika, Yili Hong, Paul A. Pavlou. 2012. On product uncertainty in online markets: Theory
and evidence. MIS Quarterly 36(2) 395–426.
Dulleck, Uwe, Rudolf Kerschbamer. 2006. On doctors, mechanics, and computer specialists: The economics of credence goods. Journal of Economic Literature 44(1) 5–42.
Emons, Winand. 1997. Credence goods and fraudulent experts. RAND Journal of Economics 28(1)
107–119.
Erdem, TÃijlin, Michael P. Keane, Baohong Sun. 2008. A dynamic model of brand choice when price
and advertising signal product quality. Marketing Science 27(6) 1111–1125.
Foresee. 2012. Daily deal websites and emails bring in new and existing customers for retailers. Foresee
Daily Deal Commentary (2012) .
Ghose, Anindya. 2009. Internet exchanges for used goods: An empirical analysis of trade patterns and
adverse selection. MIS Quarterly 33(2) 263–291.
Hu, Ming, Mengze Shi, Jiahua Wu. 2013. Simultaneous vs. sequential group-buying mechanisms. Management Science 59(12) 2805–2822.
Hui, Kai-Lung. 2004. Product variety under brand influence: An empirical investigation of personal
computer demand. Management Science 50(5) 686–700.
Janssen, Maarten C.W., Santanu Roy. 2010. Signaling quality through prices in an oligopoly. Games
and Economic Behavior 68 192–207.
Jedidi, Kamel, Carl F. Mela, Sunil Gupta. 1999. Managing advertising and promotion for long-run
profitability. Marketing Science 18(1) 1–22.
16
Jin, Ginger Zhe, Andrew Kato. 2006. Price, quality, and reputation: Evidence from an online field
experiment. RAND Journal of Economics 37(4) 983–1004.
Jing, Xiaoqing, Jinhong Xie. 2011. Group buying: A new mechanism for selling through social interactions. Management Science 57(8) 1354–1372.
Klein, Benjamin, Keith B. Leffler. 1981. The role of market forces in assuring contractual performance.
Journal of Political Economy 89(4) 615–641.
Lattin, James, Randolph Bucklin. 1989. Reference effects of price and promotion on brand choice behavior. Journal of Marketing Research 26(3) 299–310.
Li, Shibo, Kannan Srinivasan, Baohong Sun. 2009. Internet auction features as quality signals. Journal
of Marketing 73(1) 75–92.
Li, Xitong, Lynn Wu. 2013. Measuring effects of observational learning and social-network word-of-mouth
(wom) on the sales of daily-deal vouchers. System Sciences (HICSS), 2013 46th Hawaii International
Conference on. IEEE 2908–2917.
Mela, Carl F., Sunil Gupta, Donald R. Lehmann. 1997. The long-term impact of promotion and advertising on consumer brand choice. Journal of Marketing Research 34(2) 248–261.
Mela, Carl F., Kamel Jedidi, Douglas Bowman. 1998. The longterm impact of promotions on consumer
stockpiling behavior. Journal of Marketing Research 35(2) 250–262.
Moretti, Enrico. 2011. Social learning and peer effects in consumption: Evidence from movie sales.
Review of Economic Studies 78(1) 356–393.
Nelson, Phillip. 1970. Information and consumer behavior. Journal of Political Economy 78(2) 311–329.
Nevo, Aviv. 2000. A practitioner’s guide to estimation of random-coefficients logit models of demand.
Journal of Economics & Management Strategy 9(4) 513–548.
Orosel, Gerhard O., Klaus G. Zauner. 2011. Quality diversity and prices in markets for experience goods.
Journal of Economics & Management Strategy 20(3) 709–738.
Overby, Eric, Sandy Jap. 2009. Electronic and physical market channels: A multiyear investigation in a
market for products of uncertain quality. Management Science 55(6) 940–957.
Pauwels, Koen, Dominique M. Hanssens, S. Siddarth. 2002. The long-term effects of price promotions
on category incidence, brand choice, and purchase quantity. Journal of Marketing Research 39(4)
421–439.
Pavlou, Paul A., Huigang Liang, Yajiong Xue. 2007. Understanding and mitigating uncertainty in online
exchange relationships: A principal-agent perspective. MIS Quarterly 31(1) 105–136.
Rao, Akshay R., Kent B. Monroe. 1996. Causes and consequences of price premiums. Journal of Business
69(4) 511–535.
Reuters. 2011. Groupon gets into online retailing with groupon goods. Accessed March 23, 2015,
http://www.reuters.com/article/2011/09/28/us–groupon–idUSTRE78R4ZV20110928.
Shapiro, Carl. 1983. Premiums for high quality products as returns to reputations. Quarterly Journal of
Economics 98(4) 659–680.
Shivendu, Shivendu, Zhe Zhang. 2013. Economics of daily-deal websites: Advertising and sampling
effects. Proceedings of International Conference on Information Systems (ICIS) .
Subramanian, Upender, Ram C. Rao. 2014. An equilibrium analysis of daily deal strategies: When
should a daily deal website display deal sales? Working Paper. University of Texas at Dallas.
Tirole, Jean. 1988. The theory of industrial organization. MIT press, Cambridge, MA.
17
Wolinsky, Asher. 1983. Prices as signals of product quality. Review of Economic Studies 50(4) 647–658.
Wu, Jiahua, Mengze Shi, Ming Hu. 2015. Threshold effects in online group buying. Management Science
Forthcoming.
18
Figure 1. Distribution of discount percentage in the data
Figure 2. Deals’ average sales and discount percentages
(a)
(b)
Table 1. Percentages of deals in each category
Category
Freq.
Percent
Cum.
Automotive
271
1.36
1.36
Beauty & Spas
5,230
26.18
27.54
Education
691
3.46
30.99
Entertainment
6,524
32.66
63.65
Food & Drink
620
3.10
66.75
Health & Fitness
1,804
9.03
75.78
Home Services
226
1.13
76.91
Medical Treatment
558
2.79
79.71
Nightlife
60
0.30
80.01
Pets Services
90
0.45
80.46
Other Professional Services
734
3.67
84.13
Restaurants
3,170
15.87
100
Total
19,978
100
Table 2. Summary Statistics
Variables
Hourly sales
Lag cumulative sales
Transaction price
Original price
Discount percentage
Maximum purchaser allowed
Number of Facebook fans
Display review quotes (dummy)
Number of options
Proxy for use restrictions
Sold out finally (dummy)
Merchant created deal (dummy)
Days before expiration
Online deal (dummy)
Multi-regional deal (dummy)
Deal frequency
Duration
Review count
Average rating
Credence goods deal (dummy)
Median household income, 2008-2012
Bachelor's degree or higher,
percent of persons age 25+, 2008-2012
Average hourly wage
House price index
N
No. of deals
Unit
Mean
Std. dev.
1,835,794
19,978
1.24
5.77
1,835,794
19,978
76.33
149.32
1,835,794
19,978
USD
52.44
154.68
1,835,794
19,978
USD
142.74
364.00
1,835,794
19,978
0~100
55.55
14.34
1,835,794
19,978
5.49
21.16
1,835,794
19,978
190,781 1,589,734
1,835,794
19,978
0.015
0.12
1,835,794
19,978
2.00
1.40
1,835,794
19,978
424.70
133.33
1,835,794
19,978
0.027
0.16
1,835,794
19,978
0.026
0.16
1,835,794
19,978
212.78
120.91
1,835,794
19,978
0.012
0.11
1,835,794
19,978
0.33
0.47
1,835,794
19,978
1.10
0.35
1,835,794
19,978
Hour
102.64
55.76
624,924
6,692
531.61
1623.73
624,924
6,692
1~5
3.70
0.66
1,835,794
19,978
0.035
0.18
1,704,202
18,553
1,000 USD 54.77
10.60
Min
0
0
2
5
1
1
0
0
1
0
0
0
30
0
0
1
12
1
1
0
24.65
Max
990
990
6,440
12582
99
540
51,136,036
1
42
734.10
1
1
358
1
1
6
926
8778
5
1
90.75
1,704,202
1,704,202
1,704,202
8.50
12.30
177.10
58.10
38.59
262.10
18,553
18,553
18,553
0~100
USD
33.90
18.90
212.19
9.75
6.44
20.20
Table 3. Correlations
1. Hourly sales
2. Lag cumulative sales
3. Transaction price
4. Discount percentage
5. Maximum purchases
6. Facebook fans
7. Review quotes
8. Number of options
9. Use restriction proxy
10. Sold out finally
11. merchant-created
deal
12. Days before
expiration
13. online deal
14. Multi-regional deal
15. Deal frequency
16. Duration
17. Credence goods deal
1
1.00
0.31
-0.15
-0.08
0.05
0.11
0.02
-0.09
0.01
0.12
2
3
4
5
6
7
8
9
10
11
12
1.00
-0.46
-0.16
0.10
0.25
0.05
-0.23
0.04
0.19
1.00
0.06
-0.06
-0.14
-0.00
0.04
0.04
-0.05
1.00
-0.07
-0.12
-0.01
0.15
-0.05
-0.05
1.00
0.30
-0.00
-0.17
0.12
0.13
1.00
-0.02
-0.27
0.13
0.25
1.00
-0.00
-0.01
-0.01
1.00
0.03
-0.16
1.00
0.07
1.00
0.03
0.08
-0.01
-0.07
0.17
0.15
-0.01
-0.16
0.12
0.02
1.00
0.09
0.19
0.00
-0.19
0.36
0.35
0.01
-0.29
0.12
0.15
0.18
1.00
-0.02
0.04
-0.01
-0.05
-0.04
-0.05
0.08
-0.03
0.11
-0.12
0.08
-0.00
0.04
-0.01
0.14
0.04
-0.05
-0.00
-0.03
0.11
-0.01
0.37
0.03
0.09
-0.05
-0.06
0.42
0.02
0.11
-0.07
-0.01
-0.02
0.00
0.01
-0.01
-0.03
-0.21
0.01
-0.06
-0.02
0.02
0.07
0.03
0.05
0.01
-0.01
0.13
-0.01
-0.07
-0.03
-0.02
0.08
0.11
0.10
-0.03
-0.03
0.35
0.04
0.09
-0.06
13
14
15
1.00
0.02
-0.00
-0.02
0.15
1.00
0.02
0.11
-0.03
1.00
0.03
-0.02
16
17
1.00
-0.04 1.00
Table 4. Negative effect of discount depth
(1)
Whole Sample
price
discount
lag CS
days before expiration
merchant-created deal
facebook fans
has review quotes
sold out finally
duration
options
maximum purchases allowed
use restriction proxy
online deal
multi-regional deal
-0.0103**
(0.00406)
-0.0195***
(0.00699)
0.107***
(0.00617)
0.0328***
(0.00573)
0.0701***
(0.0206)
0.00493***
(0.000841)
0.0446**
(0.0176)
0.225***
(0.0281)
-0.238***
(0.0281)
-0.0168***
(0.00557)
0.000419
(0.00405)
0.00185
(0.00323)
-0.0404
(0.0581)
0.0103
(2)
Sub-sample
with Online Review
-0.0263***
(0.00785)
-0.0300***
(0.0111)
0.123***
(0.00416)
0.0536***
(0.00637)
0.0839***
(0.0234)
0.00396***
(0.00127)
0.0296*
(0.0172)
0.130***
(0.0228)
-0.352***
(0.0278)
-0.0113
(0.00735)
0.0167***
(0.00526)
-0.00415
(0.00677)
-0.0520
(0.0648)
0.00148
(3)
Entertainment
(4)
Beauty & Spas
(5)
Restaurants
-0.0132**
(0.00564)
-0.0190**
(0.00898)
0.129***
(0.00584)
0.0239***
(0.00709)
0.0688***
(0.0198)
0.00490***
(0.00153)
0.0184
(0.0194)
0.186***
(0.0352)
-0.296***
(0.0316)
-0.0103
(0.00947)
0.00517
(0.00554)
-0.00412
(0.00979)
-0.0602
(0.105)
0.0213*
-0.0141***
(0.00305)
-0.0186
(0.0141)
0.0444***
(0.00442)
0.0244***
(0.00610)
0.170***
(0.0351)
0.00457***
(0.000724)
0.0137
(0.0128)
0.0936***
(0.0327)
-0.0585***
(0.0144)
-0.0136*
(0.00690)
0.00950*
(0.00550)
0.00154
(0.00181)
0.00151
(0.0177)
0.0196**
-0.0289***
(0.00829)
-0.0267**
(0.0107)
0.135***
(0.00454)
0.0738***
(0.00609)
0.236**
(0.0832)
0.00699***
(0.00105)
0.0952***
(0.0301)
0.277***
(0.0609)
-0.329***
(0.0289)
-0.0162
(0.0148)
0.0109
(0.00867)
0.00522
(0.00793)
-0.0108
(0.0149)
-0.0217
(6)
Falsification:
Groupon Goods
-0.455***
(0.0265)
0.236***
(0.0906)
0.185***
(0.0140)
----0.0186
(0.163)
-0.544***
(0.0267)
-0.525***
(0.0476)
-0.0423
(0.0820)
----
review count
(0.00855)
-0.00897
(0.00931)
--
average rating
--
count × rating
--
division fixed effects
subcategory fixed effects
time fixed effects
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
-Yes
Observations
R-squared
1,835,794
0.159
624,924
0.176
620,317
0.178
469,746
0.066
287,932
0.160
17,692
0.810
deal frequency
(0.0145)
-0.0317**
(0.0132)
0.0239***
(0.00355)
0.00516
(0.00687)
-0.00101
(0.00334)
(0.0119)
-0.0322**
(0.0137)
--
(0.00759)
0.0111
(0.00804)
--
(0.0260)
0.0738***
(0.00609)
--
--
--
--
--
--
--
--
--
---
Notes. The dependent variable is log of sales within an hour. All independent variables except dummies and average rating are specified in logs. Robust standard
errors clustered by subcategory in parentheses (except column 6). Heteroskedastic robust standard errors used in column 6. *** p<0.01, ** p<0.05, * p<0.1.
Table 5. Identification tests
credence × discount
(1)
2SLS
(BLP
instruments;
Entertainment)
-0.0233***
(0.00608)
-0.101**
(0.0502)
--
(2)
2SLS
(BLP
instruments;
Restaurants)
-0.0737***
(0.00964)
-0.477***
(0.0897)
--
(3)
2SLS
(AHW and
HPI as
instruments)
-0.0133***
(0.00323)
-0.0609*
(0.0370)
--
salesAbove300 × discount
--
--
--
(4)
Experience
vs.
Credence
goods
-0.0104**
(0.00408)
-0.0192***
(0.00703)
-0.0488**
(0.0212)
--
salesAbove300
--
--
--
--
income × discount
--
--
--
education × discount
--
--
lag CS
0.129***
(0.00596)
0.0175**
(0.00795)
0.0698***
(0.0196)
0.00511***
(0.00197)
0.0190
(0.0195)
0.191***
0.138***
(0.00145)
0.0551***
(0.00515)
0.226***
(0.0850)
0.00649***
(0.000575)
0.100***
(0.0111)
0.253***
price
discount
days before expiration
merchant-created deal
facebook fans
has review quotes
sold out finally
(5)
Deal
popularity
(6)
Income
effect
(7)
Education
effect
-0.0156***
(0.00328)
-0.0219**
(0.00922)
--
-0.0102**
(0.00430)
-0.0156*
(0.00824)
--
-0.0100**
(0.00431)
-0.0137
(0.00852)
--
--
--
--
--
--
0.0463***
(0.0111)
0.182***
(0.0430)
--
--
--
--
--
-0.0589***
(0.0196)
--
0.108***
(0.00516)
0.0305***
(0.00292)
0.0666***
(0.0127)
0.00524***
(0.000689)
0.0520***
(0.00679)
0.237***
0.107***
(0.00615)
0.0329***
(0.00573)
0.0701***
(0.0206)
0.00491***
(0.000840)
0.0446**
(0.0176)
0.225***
0.0785***
(0.00456)
0.0286***
(0.00455)
0.0590***
(0.0194)
0.00443***
(0.000727)
0.0309*
(0.0158)
0.182***
0.108***
(0.00625)
0.0323***
(0.00554)
0.0664***
(0.0208)
0.00516***
(0.000875)
0.0523***
(0.0185)
0.234***
-0.0406***
(0.0128)
0.108***
(0.00625)
0.0322***
(0.00554)
0.0663***
(0.0208)
0.00520***
(0.000889)
0.0520***
(0.0185)
0.235***
(0.0358)
-0.297***
(0.0308)
-0.00803
(0.00920)
(0.0222)
-0.346***
(0.00714)
-0.0201***
(0.00528)
(0.0203)
-0.241***
(0.0236)
-0.0156***
(0.00357)
(0.0281)
-0.238***
(0.0280)
-0.0169***
(0.00558)
(0.0257)
-0.238***
(0.0273)
-0.0181***
(0.00480)
(0.0303)
-0.240***
(0.0291)
-0.0169***
(0.00556)
(0.0303)
-0.240***
(0.0290)
-0.0170***
(0.00557)
0.00227
0.00884***
0.000314
0.000490
0.00105
0.000376
0.000244
(0.00791)
-0.00718
(0.0100)
-0.0163
(0.0974)
0.0179
(0.0140)
-0.0307*
(0.0158)
(0.00338)
-0.00142
(0.00505)
--0.0633***
(0.0120)
-0.00492
(0.0120)
(0.00332)
0.00188
(0.00255)
-0.0199
(0.0258)
0.0102*
(0.00594)
-0.0109
(0.00725)
(0.00405)
0.00183
(0.00322)
-0.0400
(0.0581)
0.0104
(0.00856)
-0.00897
(0.00930)
(0.00352)
0.00289
(0.00276)
-0.0444
(0.0468)
0.00771
(0.00708)
-0.00114
(0.00792)
(0.00424)
0.00289
(0.00336)
-0.0320
(0.0668)
0.0110
(0.00865)
-0.0109
(0.00942)
(0.00425)
0.00297
(0.00337)
-0.0316
(0.0666)
0.0112
(0.00870)
-0.0110
(0.00948)
division fixed effects
subcategory fixed effects
time fixed effects
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Observations
R-squared
615,036
0.176
287,913
0.137
1,704,202
0.159
1,835,794
0.159
1,835,794
0.169
1,704,202
0.160
1,704,202
0.160
duration
options
maximum purchases
allowed
use restriction proxy
online deal
multi-regional deal
deal frequency
Notes. The dependent variable is log of sales within an hour. All independent variables except dummies are specified in logs. Robust standard errors
clustered by subcategory in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
Table 6. Robustness checks
price
discount
lag CS
lag CS squared
days before expiration
merchant-created deal
facebook fans
has review quotes
sold out finally
duration
options
maximum purchases
allowed
use restriction proxy
(1)
Linear
specification
(2)
Add lag CS
squared
(3)
Excluding
price
outliers
(4)
Excluding
duration
outliers
(5)
Excluding
lag CS
--
(6)
Sub-sample
with
discount <=
38.86%
-0.0393*
(0.0213)
-0.0167
(0.0162)
0.107***
(0.00938)
--
(7)
Sub-sample
with
discount >
38.86%
-0.0117***
(0.00412)
-0.0431***
(0.0165)
0.106***
(0.00619)
--
-0.000359*
(0.000189)
-0.441***
(0.127)
0.00839***
(0.000325)
--
-0.0338***
(0.00671)
-0.0278***
(0.0101)
0.105***
(0.00698)
--
-0.00622
(0.00392)
-0.0282**
(0.0114)
0.0934***
(0.00552)
--
-0.0796***
(0.00660)
-0.0159
(0.0147)
--
0.00114***
(0.000360)
0.112*
(0.0538)
2.98e-08***
(4.55e-09)
0.195
(0.136)
2.064***
(0.176)
-0.00601***
(0.00143)
-0.00871
(0.00918)
-0.0273***
(0.00233)
-0.0126***
(0.00322)
-0.0882***
(0.00603)
0.0313***
(0.000755)
0.0287***
(0.00834)
0.0628***
(0.00508)
0.00389***
(0.000543)
0.0278
(0.0200)
0.159***
(0.0139)
-0.242***
(0.0456)
-0.0161***
(0.00256)
0.0324***
(0.00572)
0.0820***
(0.0232)
0.00472***
(0.00103)
0.0453**
(0.0217)
0.233***
(0.0394)
-0.227***
(0.0359)
-0.0143**
(0.00613)
0.0278***
(0.00476)
-0.0164
(0.0198)
0.00405***
(0.000588)
0.0395**
(0.0185)
0.256***
(0.0585)
0.0229
(0.209)
-0.0214***
(0.00536)
0.0597***
(0.00759)
0.0981***
(0.0329)
0.00967***
(0.00140)
0.0829***
(0.0260)
0.375***
(0.0276)
-0.178***
(0.0255)
-0.0472***
(0.00815)
0.00801
(0.0245)
0.132***
(0.0477)
0.0119***
(0.00171)
0.183***
(0.0623)
0.156***
(0.0477)
-0.328***
(0.0444)
0.0174
(0.0315)
0.0342***
(0.00589)
0.0619***
(0.0202)
0.00470***
(0.000775)
0.0381**
(0.0173)
0.229***
(0.0291)
-0.232***
(0.0270)
-0.0194***
(0.00568)
5.56e-05
0.00188
0.00250
-0.00328
0.00155
0.00128
0.00191
(3.35e-05)
-0.000347**
(0.00171)
0.00372
(0.00404)
0.00224
(0.00426)
0.000227
(0.00541)
0.00956**
(0.0157)
-0.0674***
(0.00389)
0.00277
(0.000132)
0.0115
(0.254)
0.168***
(0.0373)
-0.0118
(0.0458)
(0.00252)
-0.0464*
(0.0221)
0.00598*
(0.00300)
-0.00332
(0.00665)
(0.00303)
-0.0606**
(0.0247)
0.0144
(0.00967)
-0.00600
(0.00936)
(0.00319)
-0.0214
(0.0400)
0.00459
(0.00694)
-0.00839
(0.00818)
(0.00429)
-0.120*
(0.0681)
0.0172
(0.0148)
-0.0213
(0.0144)
(0.0250)
0.0383
(0.102)
0.0586**
(0.0237)
-0.0684**
(0.0323)
(0.00302)
-0.0414
(0.0574)
0.00673
(0.00824)
-0.00619
(0.00844)
division fixed effects
subcategory fixed effects
time fixed effects
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Observations
R-squared
1,835,794
0.111
1,835,794
0.176
1,160,935
0.157
1,180,785
0.132
1,835,794
0.126
95,565
0.244
1,740,229
0.156
online deal
multi-regional deal
deal frequency
Notes. The dependent variable is log of sales within an hour. All independent variables except dummies are specified in logs. Robust standard errors
clustered by subcategory in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
Appendix
Figure A.1. Deal webpage on Groupon
Automotive Services:
Beauty and Spas:
Education:
Entertainment:
Food & Drinks:
Health & Fitness:
Home Services:
Medical Treatments:
Nightlife and Bars:
Table A.1. List of the subcategories in each category
Auot Glass Services, Auto Parts & Accessories, Auto Repair*, Car Dealers,
Car Wash & Detailing, Motorcycle Dealers, Oil Change, Parking, Stereo
Installation, Tires & Wheels (Total: 10)
Beauty Products, Body Wrap, Body-Contouring, Body Massage, Eyelash
Services, Facial Care, Foot Massage, Hair Salon, Laser Hair Removal*,
Makeup Artists, Men's Salon, Nail Salon, Oxygen Bar, Reiki, Salt Therapy,
Sauna, Skin-Tag Removal, Tanning Salon, Tattoo Removal, Teeth Whitening,
Vein Treatment, Waxing (Total: 22)
Acting Classes, Art Classes, Bartending Schools, Camera Techniques,
Computer Training, Cooking Classes, Cosmetology Schools, Dance Lessons,
Driving Lessons, Educational Services, Flight Instruction, Language Schools,
Makeup Class, Music Lessons, Paddleboard Lesson, Preschools, Private
Tutors, Specialty Schools, Speed Reading, Swimming Lessons, Training &
Vocational Schools, Wine Classes (Total: 22)
Alcohol Event, Amusement Park, Aquariums, Archery, Arts/Crafts/Hobbies,
Balloon Ride, Biking, Boat Tour, Boating, Botanical Garden, Bowling,
Brewery Tour, Casino, Circus, Comedy, Country Clubs, Creamery Tour,
Dance, Dinner Theater, Diving, Farm Tours, Film Festival, Fishing, Flight,
Food Tour, Gaming, Ghost Tour, Go-Kart, Golf, Historical Tour,
Home/Garden Show, Horse/Carriage Ride, Individual Speakers, Karaoke,
Kid's Activities, Laser Tag, Magic Show, Miniature Golf, Miscellaneous
Events, Miscellaneous Exhibition, Movie Tickets, Museum, Music Concert,
Mystery Date, Other Outdoor Adventure, Other Specialty Tour, Paintball,
Palace of Wax, Pool Party, Running Event, Segway Tour, Shooting,
Sightseeing Tour, Skating, Skiing, Skydiving, Speedway, Sporting Activity,
Sporting Event, Spring Jumping, Supercar Driving, Surfing, Symphony &
Orchestra, Talent Show, Theater & Plays, Train Tour, Water Park, Winery
Tour, Workshops and Seminars, Zipline Tour, Zoos (Total: 71)
Alcohol Store, Bagel Shops, Breweries, Butchers & Meat Shops, Candy
Stores, Cheese Shops, Chocolate Shops, Coffee & Tea Shops,
Cupcakes/Dessert/Bakery, Food Delivery Services, Grocery Stores, Health
Stores, Ice Cream & Frozen Yogurt, Juice Bars & Smoothies, Organic Food*,
Seafood Markets (Total: 16)
Badminton, Baseball, Bootcamp, Crossfit, Fitness Classes, Gyms & Fitness
Centers, Karate, Kickboxing, Martial Arts, Personal Training, Pilates, Rock
Climbing, Taekwondo, Tennis, Yoga (Total: 15)
Carpet Cleaning, Chimney Sweep, Gardeners, Gutter Cleaning, Handyman
Services*, Heating & Ventilation & Air Conditioning*, Home Cleaning, Home
Repair*, Interior Designers & Decorators*, Junk Removal, Lawn Care
Services, Movers, Painters, Pest & Animal Control, Pool Cleaners, Tree
Services, Window Washing (Total: 17)
Acupuncture*, Arthritis*, Chiropractic*, Craniosacral Therapy*, Dentists*,
Dermatology*, Detoxification*, Food Allergy*, Hearing aid*, Hormone
Therapy*, Hydrotherapy*, Hypnotherapy*, Laser Eye Surgery/Lasik*,
Medical Exam & Consultation*, Nail-Fungus Treatment*, Optometrists*,
Orthodontics*, Reflexology*, Stress Management* (Total: 19)
Cigar Bars, Dance Clubs, Gay Bars, Irish Pubs, Jazz & Blues Clubs, Lounges,
Music Venues, Night Clubs, Piano Bars, Pool Halls, Pubs/Sports Bars, Social
Clubs, Wine Bars (Total: 13)
Pets Services:
Restaurants:
Other Professional
Services:
Horse Services & Equipment, Pet Boarding/Pet Sitting*, Pet Groomers, Pet
Washing, Veterinarians* (Total: 5)
African, American, Asian, Breakfast & Brunch, Cafe & Tearoom, Caribbean,
Deli & Fast Food, European, French, Fusion Dishes, Hawaiian, Indian, Italian
Latin, Mediterranean, Middle Eastern, Pub Food, Seafood, Spanish, Specialty
Meal, Vegan & Health Food (Total: 21)
Accountants, Car Rental, Catering & Bartending Services, Digital Conversion,
Dry Cleaning & Laundry, Electronics Repair*, Event Planner, Magzine
Subscription, Photography, Printing & Copying Equipment & Services,
Resume Services, Self-Storage, Shoe Repair, Watch Repair* (Total: 14)
Notes. Subcategories that are classified into credence goods are italic and labeled with asterisk.
American
Cities:
Canadian
Cities:
Table A.2. List of the local geographic markets
Abilene, Akron-Canton, Albany Capital Region, Albany(GA), Albuquerque, AllentownReading, Amarillo, Anchorage, Ann Arbor, Appleton, Asheville, Athens(GA), Atlanta,
Augusta, Austin, Bakersfield, Baltimore, Baton Rouge, Billings, Birmingham, Boise,
Boston, Buffalo, Cedar Rapids-Iowa City, Central Jersey, Charleston, Charlotte,
Chattanooga, Chicago, Cincinnati, Cleveland, Colorado Springs, Columbia,
Columbia(MO), Columbus, Columbus(GA), Corpus Christi, Dallas, Dayton, Daytona
Beach, Denver, Des Moines, Detroit, El Paso, Erie, Eugene, Evansville, Fairfield County,
Fort Lauderdale, Fort Myers-Cape Coral, Fort Wayne, Fort Worth, Fresno, Gainesville,
Grand Rapids, Green Bay, Greenville, Hampton Roads, Harrisburg, Hartford, Honolulu,
Houston, Huntsville, Indianapolis, Inland Empire, Jackson, Jacksonville, Kalamazoo,
Kansas City, Knoxville, Lakeland, Lansing, Las Vegas, Lexington, Lincoln, Little Rock,
Long Island, Los Angeles, Louisville, Lubbock, Macon, Madison, Memphis, Miami,
Midland-Odessa, Milwaukee, Minneapolis-St. Paul, Mobile Baldwin County, Modesto,
Montgomery, Napa-Sonoma, Naples, Nashville, New Orleans, New York, North Jersey,
Ocala, Ogden, Oklahoma City, Omaha, Orange County, Orlando, Palm Beach, Pensacola,
Philadelphia, Phoenix, Piedmont Triad, Pittsburgh, Portland, Portland(ME), Providence,
Raleigh-Durham, Reno, Richmond, Rio Grande Valley, Roanoke, Rochester, Rockford,
Sacramento, Salem(OR), Salt Lake City, San Angelo, San Antonio, San Diego, San
Francisco, San Jose, Santa Barbara, Santa Cruz, Savannah-Hilton Head, Seattle,
Shreveport-Bossier, Sioux Falls, South Bend, Spokane Coeur D’Alene, Springfield(MA),
Springfield(MO), St. Louis, Stockton, Syracuse, Tallahassee, Tampa Bay Area, Toledo,
Topeka-Lawrence, Tucson, Tulsa, Ventura County, Washington DC, Westchester County,
Wichita, Wilmington-Newark, Worcester, Youngstown (Total: 152)
Abbotsford, Barrie, Calgary, Edmonton, Greater Toronto Area, Halifax, Kelowna,
Kingston, Kitchener-Waterloo, London, Ottawa, Regina, Saskatoon, St. John’s, St.
Catharines-Niagara, Sudbury, Vancouver, Victoria, Windsor, Winnipeg (Total: 20)