Mobile Payment Adoption: An Empirical Investigation on Alipay Yuqian Xu Gies College of Business, University of Illinois at Urbana-Champaign, Champaign, IL 61820, [email protected]Anindya Ghose Stern Business School, New York University, New York, NY 10012, [email protected]Binqing Xiao Industrial Engineering Department, Nanjing University, China, 210093, [email protected]The rapid development of mobile technology has introduced a new channel for consumer consumption, in addition to traditional online PC and offline (physical card) payment channels. In this paper, we investigate the determinants and outcomes of mobile channel adoption on consumer consumption behaviors. We utilize a unique data set from one of the largest banks in China, which contains the consumer credit card consumption from PC, offline, and mobile payment channels. The mobile payment channel under study here is from Alipay, which is now the world’s largest mobile payment platform. In our work, we find that both higher service demand and higher local penetration are associated with earlier mobile channel adoption. For the post-adoption behavior analysis, we show that the total transaction amount increases by around 2.4% after the Alipay adoption, and the total transaction frequency increases by around 23.5%. The relationship is even stronger for medium income consumers. Furthermore, we find that Alipay channel acts as a substitute for the offline channel, and as a complement for the PC payment channel. Both substitution and complementarity effects increase over time. Finally, we find that the increased credit card transaction activity and profitability are likely to be driven by hedonic shopping behavior with low value items. Our work aims to bring managerial implications for bank and retail managers on multi-channel management. Key words : Alipay, adoption, mobile payment, channel, PC, offline, online, credit card. 1 Electronic copy available at: https://ssrn.com/abstract=3270523
41
Embed
Mobile Payment Adoption: An Empirical Investigation on Alipay
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Mobile Payment Adoption:An Empirical Investigation on Alipay
Yuqian XuGies College of Business, University of Illinois at Urbana-Champaign, Champaign, IL 61820, [email protected]
Anindya GhoseStern Business School, New York University, New York, NY 10012, [email protected]
both pre- and post-adoption, subset analysis, and robustness checks. Section 7 summarizes
our results and outlines future research directions.
2. RELATED LITERATURE
In this section, we review three streams of literature that are closely related to our work,
namely, (i) PC channel adoption, (ii) features of mobile phone channel, and (iii) interde-
pendencies among different channels.
The first stream of work closely related to ours is on the adoption of the PC channel,
which is found to be associated with changes in service consumption, cost to serve, and
consumer profitability (see Hitt and Frei (2002) and Campbell and Frei (2010)). In par-
ticular, consumers who adopt PC channel are found to be more profitable although the
adoption will cause a reduction in short-term consumer profitability (Campbell and Frei
(2010)); moreover, consumers who adopt PC channel have a lower propensity to leave the
bank (Xue et al. (2011)). Recent work has started looking into the adoption of mobile
banking channel, and find that the mobile banking channel serves as a complement to
the PC banking channel (Liu et al. (2017)). However, so far limited work has been done
on exploring the adoption of mobile payment channel and the associated consequences
of mobile payment adoption. In this paper, we want to fill this void by investigating the
determinants and outcomes of mobile payment channel adoption on consumer consumption
behaviors.
Electronic copy available at: https://ssrn.com/abstract=3270523
Author:6 Article submitted to Management Science; manuscript no.
The second stream of literature closely related to our work is on the features of mobile
phone consumption. Prior work has looked into the effect of the screen size when analyzing
the impact of mobile phone on consumption decisions (see Chae and Kim (2004), Kim
et al. (2011), and Ghose et al. (2012)). Recent research has also focused on improving the
consumer experience of browsing the web on their smartphones (Adipat et al. (2011)).
Portability is also a key feature of mobile phones. Although PCs enable consumers to hold
a global perspective when shopping via the computer (Overby and Lee (2006)), the restric-
tion of portability offsets this benefit. Universality, as another essential feature, refers to
a channel’s ability to provide universal and stable service in space and time. Compared
to PCs, mobile channels overcome the constraints of places, which constantly support
consumers in activities such as information access or immediate transactions (Jung et al.
(2014)). However, the storage limitation of mobile devices constrains the volume of infor-
mation access and impairs the quality of PC services (Napoli and Obar (2014)). Thus,
determining the moderating features associated with mobile channel adoption is crucial
for bank and retail managers.
The third stream is the interdependencies among different channels, which has been
studied by a significant number of papers in the past few years. In general, findings in
these papers provide significant managerial insights into firms’ investment-strategies prob-
lem of whether to diversify their investments or concentrate on a single retail channel.
These papers have looked into the following problems: (1) the effects of substitution and
complementarity between PC and offline channels, (i.e., Brynjolfsson et al. (2009), Choi
et al. (2008), Ellison and Ellison (2006), Forman et al. (2009), Goolsbee (2001), and Prince
(2007)); (2) how offline sales channels could be affected by mobile ads based on location
or context (see Hui et al. (2013) and Molitor et al. (2016)); and (3) the interdependencies
Electronic copy available at: https://ssrn.com/abstract=3270523
Author:Article submitted to Management Science; manuscript no. 7
between the mobile and PC channels (see Bang et al. (2013) and Ghose and Han (2011)).
From these papers, we can see that a new channel will probably be established under the
condition that it has the ability to provide consumers with extra utility relative to what
they have received from the previous channels. For example, when the offline stores enter
the market, local consumers will shift away significantly from the previous PC channels
(Forman et al. (2009)). Despite the benefits of PC shopping, such as lower prices and a
wider area of product selection, the potential reasons why consumers turn to their local
offline stores may include lower traveling and shipping costs, as well as the ability to check
quality. Moreover, PC and offline channels can complement each other (Luo et al. (2013)),
and mobile and PC channels can substitute for and complement each other simultaneously,
depending on the product category (Bang et al. (2013)). Finally, the tablet channel acts
as a substitute for the PC channel and a complement for the smartphone channel (Xu
et al. (2016)). However, so far as we know, limited work has looked into the new mobile
payment channel and the intrinsic interdependencies that are related to mobile, PC, and
offline (physical card) payment channels. Thus, another goal of this paper is to explore
interdependencies among the three channels.
3. HYPOTHESES
A new channel will probably be established under the condition that it has the ability
to provide consumers with extra utility relative to what they have received from the pre-
vious channels. Random utility theory (McFadden (1974)) shows that consumers adopt
the product that provides them with the highest utility given the costs and benefits of
the product, and idiosyncratic consumer tastes. In our work, a consumer’s direct cost and
benefit are captured by the factor of service demand and the indirect cost and benefit are
mainly captured by the factor of local penetration. We discuss how these two factors are
Electronic copy available at: https://ssrn.com/abstract=3270523
Author:8 Article submitted to Management Science; manuscript no.
associated with consumer mobile channel adoption in Section 3.1. After the mobile channel
adoption, the usage of this new channel may change consumers consumption behaviors.
We analyze the consumer behavior changes from three perspectives, namely transaction
activities, consumer profitability, and channel substitution and complementarity. We study
these three aspects in details in Section 3.2.
3.1. Hypotheses: Mobile Channel Adoption
In this subsection, we discuss our two main hypotheses on pre-adoption behavior. We
consider two driven factors, namely service demand and local penetration.
Service Demand. Previous research on Internet payment adoption has shown that con-
sumers whose service-interaction demand is high would gain greater overall benefits from
any kind of service innovation, and thus they are more likely to adopt Internet payment
(see Lee and Lee (2001) and Xue et al. (2011)). Similarly, here those consumers who have
high service demand would also be more likely to adopt mobile payment. In our analysis,
following Xue et al. (2011), we use the transaction frequency to measure the consumers’
service demand. Moreover, previous work has shown that compared to PCs, mobile chan-
nels overcome the constraints of places due to channel ubiquity, and constantly support
consumers in activities such as information access or immediate transactions (Jung et al.
(2014)). This characteristic of channel ubiquity also enables users to enjoy videos to pass
the time and mitigate solitude when traveling (O’Hara et al. (2007)). Bang et al. (2013)
show that ubiquity is an important feature that measures the channel capability. In our
context, similar to Xu et al. (2016), we consider channel ubiquity as the channel’s abil-
ity to offer instant Internet access wherever and whenever a user wants. Intuitively, the
PC channel is constrained by Internet usage and hardware. However, the mobile channel
overcomes this limitation by offering ubiquitous Internet access (see Jung et al. (2014)
Electronic copy available at: https://ssrn.com/abstract=3270523
Author:Article submitted to Management Science; manuscript no. 9
and Venkatesh et al. (2003)). Therefore, we expect that the relationship between service
demand and earlier mobile channel adoption is stronger for consumers in need of channel
ubiquity, i.e., frequent travelers. We now summarize our first hypothesis as follows.
Hypothesis 1 (H1). Higher service demand is associated with earlier mobile channel
adoption; and the relationship is even stronger for frequent travelers due to channel ubiq-
uity.
Local Penetration. Previous product diffusion and network effects literature has shown
that the number of compatible products adopters may affect the product demand, see Bass
(1969) and Katz and Shapiro (1985). Moreover, the Internet payment adoption literature
has shown that local penetration is positively related to faster Internet payment adoption
(Xue et al. (2011)). In the mobile payment context, although consumers may not directly
interact with each other, following similar two reasons as Xue et al. (2011), we expect local
penetration plays an important role in mobile payment adoption. The first reason is the
local word-of-mouth effects found in may previous literature for other products such as PC
groceries, see Stavins (2001), Goolsbee and Klenow (2002), Forman et al. (2008), Choi et al.
(2010). The second reason is the indirect effects such as complementary investments made
by service providers who interact with mobile payment channel and want to expand service
payment channels, see Xue et al. (2011). Both explanations suggest that one may adopt
mobile payment channel faster with more prior adopters in a close area. In our analysis, we
use the number of existing Alipay adopters within the same zip code area as the adopter
under study to measure the level of local penetration. We specify our hypothesis as follows.
Hypothesis 2 (H2). Higher local penetration is associated with earlier mobile channel
adoption.
Electronic copy available at: https://ssrn.com/abstract=3270523
Author:10 Article submitted to Management Science; manuscript no.
3.2. Hypotheses: Post-adoption Behavior Changes
In this subsection, we discuss our three main hypotheses on post-adoption behavior. We
consider three outcomes, i.e., transaction activities, consumer profitability, and channel
substitution and complementarity.
Transaction Activities. One key advantage of mobile channel is the convenience of car-
rying. Compared to mobile phones, devices with larger screens have a negative affect on
the convenience of carrying and on usage rates (Kim et al. (2011)). Similarly, although
PCs enable consumers to hold a global perspective when shopping via the PC (Overby
and Lee (2006)), the restriction of portability offsets this benefit. Therefore, facing the key
advantage of convenience, mobile channel adopters are likely to increase the number of
transactions they perform. Furthermore, we expect transaction activities to be different
across different product values and consumers. For example, low-value products, such as
groceries, are generally more likely to be hedonic shopping targets and e-channel appears
more attractive to small buyers (Langer et al. (2012)); therefore mobile channel adopters
might be more likely to purchase low value products. We next specify our hypothesis on
transaction activity as follows.
Hypothesis 3 (H3). Mobile channel adoption is associated with increased credit card
transaction activity, especially for low-value items.
Consumer Profitability. On one hand, the introduction of mobile payment channel makes
consumption much easier, and increased transaction activities are likely to lead to greater
consumer profitability. Prior research has shown that the online banking adoption is associ-
ated with increasing consumer profitability (Xue et al. (2011)), and multichannel customers
are more profitable than they would be if they were not multichannel (Montaguti et al.
(2016)). However, on the other hand, if most increased activities are associated with lower
Electronic copy available at: https://ssrn.com/abstract=3270523
Author:Article submitted to Management Science; manuscript no. 11
value items, then the overall profitability of consumer is not necessarily increasing. Fol-
lowing Xue et al. (2011), we use the monthly transaction amount to measure consumer
profitability, and we thus test the following hypothesis.
Hypothesis 4 (H4). Mobile channel adoption is associated with increased credit card
consumer profitability.
Channel Substitution and Complementarity. Prior research has shown that mobile and
PC channels can substitute for and complement each other simultaneously, depending on
the product category (Bang et al. (2013)), and the tablet channel acts as a substitute for
the PC channel and a complement for the smartphone channel (Xu et al. (2016)). Recent
work also finds that the mobile banking channel serves as a complement to the PC channel
(Liu et al. (2017)). In the next hypothesis, we want to explore the impact of the mobile
payment channel on consumer consumption, and find the intrinsic interdependencies that
are related to mobile, PC, and offline payment channels. We test the following hypothesis.
Hypothesis 5 (H5). Mobile channel serves as substitute of physical card channel; but
as complement of PC channel.
4. DATA
Our credit card data set contains three parts. The first part records 2,016,132 credit card
issuances between 1989 and 2013 from one of the largest banks in China7 to consumers
in Jiangsu Province. This issuance data discloses detailed information on the card-issuing
date, type of card, annual fee, credit limit, and so on. Here, by knowing the credit card
issuance date, we can easily compute the length of the usage of each credit card, which
later we use as an important control variable in our model.
7 Note that the first credit card in China was issued on February 1, 1987.
Electronic copy available at: https://ssrn.com/abstract=3270523
Author:12 Article submitted to Management Science; manuscript no.
The second part of our data set contains 1,541,769 consumers’ demographic informa-
tion, for example, age, gender, job position (manager or not), education (graduate degree,
bachelor’s degree, high school degree, middle high school degree, elementary school degree,
and non-educated), marriage status (married or not), living status (rent or own), and so
on.
The third part discloses the consumer transaction-level information from September
2010 to February 2013 (29 months), including transaction amount, account balances, trans-
action date, type of transaction (education, service, healthcare, entertainment, groceries,
and travel), interest rate, reward points, and so on. In this part of our data set, we have
159,954,104 observations. By knowing the transaction dates, we can later easily compute
the number of transactions (frequency) per month. Moreover, for each transaction obser-
vation, we can observe an associated label indicating the payment channel, i.e., online PC,
Alipay (mobile channel), offline, etc.8 Note that the payment here for each transaction is
still made through credit card, and each credit card is used through one of the three (online
PC, Alipay, and offline) channels. If the payment channel of one transaction is labeled as
Alipay, then it means that the credit card used for this transaction is linked to Alipay, and
when a consumer makes a payment, he can scan mobile phone codes in restaurants or stores
instead of using a physical card. Note that Alipay also allows consumers to make payments
or transfers to their friends and relatives (as in Paypal), however our study only focuses on
commercial transactions and does not include transactions made to any individual person.
Table 1 shows the summary statistics of key variables in our data set.
8 Note that during our observation period between September 2010 and February 2013, Alipay is the only dominating
mobile payment app in China. The second largest mobile payment app, WeChat Pay, was launched in September
2014.
Electronic copy available at: https://ssrn.com/abstract=3270523
Author:Article submitted to Management Science; manuscript no. 13
Table 1 Summary Statistics
Variable Definition Mean Sd. Min Max N
Numerical Variables
CDT LMT Credit limit in RMB 12,878 52,766 0 5,000,000 4,541,583CDT LEN Card usage length in months 25 14 0 192 4,541,583CDT TYP Main (1) or attached (0) card 0.9700 0.1700 0 1.000000 4,500,933GENDER Male (1) or female (0) 0.4600 0.5000 0 1.000000 3,181,418AGE Age of card holder 29 5.7 12 93 3,181,418TXN AMT Transaction amount per month in RMB 2,239.4 2,133.8 0 4,986,592 159,954,104TXN AMT 5 Average transaction amount last 5 months in RMB 2,478.3 2,259.6 0 4,295,375 159,954,104TXN CNT Transaction frequency per month 72.1 69.8 0 1,500.00 28,619,173ANU FEE Annual fee in RMB 3.2100 56.300 0 3,600.000 28,634,369CDT BAL Account credit balances 4,520.2 23,742 0 2,000,000 335,395
Binary Variables
JOB POS Manager (1) or not (0) 0.35 0.13 0 1 3,179,303EDU0 Non-educated (1) or not (0) 0.11 0.05 0 1 2,381,090EDU1 Elementary school (1) or not (0) 0.15 0.06 0 1 2,381,090EDU2 Middle high school (1) or not (0) 0.16 0.10 0 1 2,381,090EDU3 High school (1) or not (0) 0.15 0.06 0 1 2,381,090EDU4 Bachelor (1) or not (0) 0.27 0.17 0 1 2,381,090EDU5 Graduate (1) or not (0) 0.10 0.05 0 1 2,381,090MAR STS Married (1) or not (0) 0.48 0.27 0 1 2,788,549LIV STS Own (1) or rent (0) 0.44 0.23 0 1 3,154,055
Next, we describe some initial analysis of our data set. First, we find that around 35.14% of
the consumers in the data set have used online payment methods, including both PC and mobile
channels. Among the online channels, Alipay has 204,547 users, and the PC channel has 244,981
users during our entire observation period. Therefore, we can see that by the end of February 2013,
Alipay had almost as many users as the traditional PC payment channel.
Figure 1 shows the number of new consumers who adopted Alipay each month during our
observation period. We can see that, starting from December 2010, when our focal bank started
to collaborate with Alipay, the number of accumulated new adoptions has been growing steadily.
In February 2013, the number of new adoptions had grown to 204,547 (13.3%), which shows that
mobile payment has become more and more popular. Figure 2 compares the percentage of the online
transaction (including both mobile and PC channels) amount with the percentage of the offline
transaction amount made in each month. We can see that the fraction of the online transaction
amount has grown steadily over time, especially since October 2010, whereas the offline transaction
amount has decreased gradually. Figure 2 shows the initial analysis on the substitution effect of
Electronic copy available at: https://ssrn.com/abstract=3270523
Author:14 Article submitted to Management Science; manuscript no.
the online channel on the offline channel, and later we will show this effect more rigorously from
our DID estimation.
0
50,000
100,000
150,000
200,000
250,000
Number of New Adoptions
Figure 1 Number of New Alipay Adoptions
0%
20%
40%
60%
80%
100%
1009
1010
1011
1012
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1301
1302
Ratio between Online and Offline Transactions
线上交易 线下交易Online Offline
Figure 2 Ratio between Online (PC and Alipay) and Offline Transaction Amount
5. ECONOMETRIC MODEL
In this section, we show the econometric models used in both pre- and post-adoption analyses.
Section 5.1 describes the survival analysis that we used in the pre-adoption study. Section 5.2
presents the details of our DID method in post-adoption analysis. Section 4.3 discusses our two
matching strategies.
Electronic copy available at: https://ssrn.com/abstract=3270523
Author:Article submitted to Management Science; manuscript no. 15
5.1. Mobile Channel Adoption Analysis
To understand the determinants of mobile payment adoption, we conduct survival analysis, where
a subject leaves the panel when the adoption event happens. Therefore, this approach links the
explanatory variables with the time when consumers adopt the mobile payment channel. Note that
a number of approaches can be used for survival analysis, for instance, the accelerated failure time
(AFT) and proportional hazard (PH) models. Following Tellis et al. (2003) and Xue et al. (2011),
we use a log-logistic distribution AFT model, and consider both the regular regression coefficients
in the log-time format and the time-ratio coefficient, which is computed as the ratio of fail time to
normal time. Note that parametric models have been shown to be more statistically efficient than
nonparametric or semi-parametric models in settings where the focus is on the effects of covariates
that evolve over time.
We now specify our log-logistic distribution AFT model. Denote ti as the event time of adoption
for consumer i, xi as a set of covariates, and εi as the error term. We then have the following
regression function:
ln(ti) = xiβx + εi, (1)
where βx is the set of parameter weights to be estimated. Here, if a regular coefficient is negative
or a time ratio coefficient is less than 1, the result implies faster adoption.
To test our two hypotheses on service demand and local penetration, we now define the key
measures. We first define two measures to compute service demand. The first measure (denoted as
M1) is the same as Xue et al. (2011), which computes each consumer’s total transaction frequency
in the month prior to adoption. As a robustness check, we compute a second measure (denoted
as M2) that is the monthly average transaction frequency before adoption. Next, to define the
frequent and infrequent travelers, we compute the percentage of transactions made in cities other
than the focal consumer’s residential city. We then compute the sample average of this percentage
(0.16 on average). Consumers with values above the sample average are defined as frequent traveler,
Electronic copy available at: https://ssrn.com/abstract=3270523
Author:16 Article submitted to Management Science; manuscript no.
and the ones below the sample average are defined as infrequent traveler. Finally, to discuss the
impact of local penetration on mobile channel adoption, we use the number of existing Alipay
adopters within the same zip code area as the adopter under study to measure the level of local
Standard errors in parentheses∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Note: Trend here refers to the control of time dummies.
Electronic copy available at: https://ssrn.com/abstract=3270523
Author:Article submitted to Management Science; manuscript no. 29
6.3. Post-adoption: Subset Analysis
In this subsection, we try to explore the driven forces of the increasing transaction amount and fre-
quency. We conduct subset analysis based on purchasing product category and value, and consumer
credit limit level and observation periods.
Product Category. Based on the practice of our focal bank, the credit card department classifies
consumer purchasing products into six categories as shown in the following Table 9. We can see that
grocery shopping has the highest consumption frequency in our data set, followed by education and
entertainment. Intuitively, service, entertainment, groceries, and travel are more related to hedonic
shopping. To examine the impact of mobile channel adoption on the consumption for each category,
we compute the total consumption amount, frequency, and shares of different channels for each of
the six product categories, and then conduct our DID analysis for each category separately and show
our results in Table 10. From the results in Table 10, we can see that both the overall consumption
amount and frequency increase for groceries, entertainment, travel, and service categories. Second,
for these four categories, the complementarity and substitution effects still hold. Third, mobile
channel adoption affects the groceries category the most, followed by the entertainment, travel,
and service categories. Finally, mobile channel adoption seems unlikely to affect education and
healthcare categories, because the estimated coefficients are not statistically significant with our
data set.
Table 9 Consumption Categories
Category Definition Number PercentageEducation School education, professional training, etc. 16,193,466 14.84%Service Cleaning, repairing, etc. 1,462,219 1.34%Healthcare Hospital service, medicine, etc. 1,002,069 0.92%Entertainment Restaurant, gym, etc. 4,326,357 3.96%Groceries Supermarket, wholesale, etc. 43,122,235 39.51%Travel Flight ticket, hotel rooms, subway, etc. 3,832,095 3.51%
Electronic copy available at: https://ssrn.com/abstract=3270523
Author:30 Article submitted to Management Science; manuscript no.
Table 10 Impacts under Different Consumption Categories