tilburg university payment technology adoption and finance ... · innovations in nancial...
Post on 18-Jul-2020
1 Views
Preview:
TRANSCRIPT
Tilburg University
Payment Technology Adoption and Finance
Dalton, Patricio; Pamuk, Haki; Ramrattan, R.; van Soest, Daan; Uras, Burak
Document version:Early version, also known as pre-print
Publication date:2018
Link to publication
Citation for published version (APA):Dalton, P., Pamuk, H., Ramrattan, R., van Soest, D., & Uras, B. (2018). Payment Technology Adoption andFinance: A Randomized-Controlled-Trial with SMEs. (CentER Discussion Paper; Vol. 2018-042). CentER,Center for Economic Research.
General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
- Users may download and print one copy of any publication from the public portal for the purpose of private study or research - You may not further distribute the material or use it for any profit-making activity or commercial gain - You may freely distribute the URL identifying the publication in the public portal
Take down policyIf you believe that this document breaches copyright, please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.
Download date: 06. Sep. 2020
No. 2018-042
PAYMENT TECHNOLOGY ADOPTION AND FINANCE: A RANDOMIZED-CONTROLLED-TRIAL WITH SMES
By
Patricio S. Dalton, Haki Pamuk, Ravindra Ramrattan, Daan van Soest,
Burak Uras
16 October 2018
ISSN 0924-7815 ISSN 2213-9532
Payment Technology Adoption and Finance:
A Randomized-Controlled-Trial with SMEs∗
Patricio S. Dalton† Haki Pamuk‡ Ravindra Ramrattan§
Daan van Soest† Burak Uras†
October 16, 2018
Abstract
What determines the adoption of electronic-payment instruments? Do these instru-
ments impact business outcomes, in particular access to finance? To shed light on
these questions, we conducted a Randomized-Controlled-Trial with Kenyan SMEs.
Our experiment released barriers to adopt a novel payment instrument. We uncover
that the adoption barriers were binding for a large portion of the firms and that firms’
financial transparency interacted with the decision to adopt. After sixteen months,
treated businesses were more likely to feel safe and had more loans. The impact on
loans was especially pronounced for smaller size establishments, which also experi-
enced a reduction in sales-volatility.
Keywords: SME Finance; Transparency; Technology Adoption; Lipa Na M-Pesa.
JEL Classification: D22, G00, G21, O33.
∗We are grateful to Abhilash Maji, Wasike Nambuwani and Edoardo Totolo for early constructive discussions about thisproject, and to FSD, specially Wasike Nambuwani, for their assistance with the field work. We also thank Thorsten Beck,William Jack and Jeremy Tobacman for their recommendations on earlier drafts of this paper. This paper has benefitedfrom comments of participants at 2017 Wageningen Development Economics Workshop, 2018 IMF-DFID Conference onFinancial Inclusion and seminars at Bogazici University and Tilburg University. This paper is written in the framework ofthe research project “Enabling Innovation and Productivity Growth in Low Income Countries (EIP-LIC/PO5639)”, fundedby the Department for International Development (DFID) of the United Kingdom and implemented by Tilburg Universityand partners. Website: www.tilburguniversity.edu/dfid-innovation-and-growth.†Tilburg University, Department of Economics and CentER, Warandelaan 2, 5037 AB, Tilburg, The Netherlands. E-mail:
p.s.dalton@uvt.nl, D.P.vanSoest@uvt.nl and r.b.uras@uvt.nl .‡Wageningen University and Research, Hollandseweg 1 6706 KN Wageningen, The Netherlands. E-mail:
haki.pamuk@wur.nl.§This paper is dedicated to the memory of our dear friend Ravindra Ramrattan, who inspired us to begin research in
Mobile Money and lost his life at the tragic Westgate Mall terrorist attacks in Nairobi, Kenya. Ravi’s soul has guided us untilthe completion of this paper.
1 Introduction
Technological progress in financial intermediation plays a key role in economic development.
Innovations in financial technologies, such as electronic payment instruments, can foster
market exchange and expand financial connectedness by reducing transaction costs and
anonymity. Despite their economic advantages, the adoption of such technologies has been
slow across the globe, and this holds especially for small- and medium-sized enterprises
(SMEs).1 A better understanding of the reasons for the reluctance of SMEs to adopt these
technologies and the consequences of adoption for business outcomes is essential to inform
financial policy reform.
Most of the empirical evidence on the determinants of adoption of cashless payment
technologies comes from studies using observational data. These studies flag the importance
of such technologies for, among others, financial access, ATM usage and safety.2 However,
observational studies typically suffer from endogeneity issues. For this reason, scholars like
Camera, Casari, and Bortolotti (2016) and Arifovic, Duffy, and Jiang (2017) have recently
resorted to designed laboratory experiments to complement the insights obtained by the
observational analyses.3 While these lab-experiments are able to identify important barriers
to adopt payment instruments, it is an open question what impact the removal of these
barriers will have on actual business outcomes in the real world.
To the best of our knowledge, no field-experimental studies are available yet that identify
the determinants and consequences of the adoption of electronic payment technologies by
actual businesses. This study aims to fill this important gap by conducting a Randomized
Controlled Trial (RCT) in which we offer SMEs in the central business district of Nairobi,
Kenya, the opportunity to adopt a mobile payment technology, allowing us to analyze up-
take as well as the outcome effects of adoption. Our identification strategy takes advantage
of a particular moment in time in which the country’s major electronic money provider,
1Klee (2008), Bolt, Janker, and van Renselaar (2010), Arango, Huynh, and Sabetti (2011), Wakamoriand Welte (2017) provide evidence that in US, Netherlands and Canada cash is the dominant form ofpayment for small transactions (below $25).
2See Humphrey, Pulley and Vesela (1996) and Chakravorti (2007).3Using experimental settings that capture some essential features of prototypical retail markets, these
authors vary the available types of payment methods, and find that fixed-adoption costs and service feescould be relevant for the diffusion of a cashless instrument.
1
Safaricom, launched a new payment instrument, called Lipa Na M-Pesa. Lipa Na M-Pesa
is a product specially designed to facilitate retail transactions by SMEs.
For our study, we sampled 1222 restaurants and pharmacies and randomly assigned
half of each business-group to a treatment aimed at encouraging merchants to adopt the
Lipa Na M-Pesa technology. The technology was available for all interested businesses at
the time of our intervention, but we increased adoption in the treatment group compared
to the control group by mitigating (or even completely removing) three potential adoption
barriers: product information, registration costs and technology know-how. Specifically, we
visited all merchants in the treatment group, providing them with a) leaflets highlighting
the benefits and the costs of the technology, b) a short movie featuring the experiences
of successful similar merchants who use the technology, and c) the possibility to have a
Lipa Na M-Pesa account opened on their behalf, at zero cost. Regarding c, we informed
businesses in the treatment group that the registration paperwork required to sign up for
the technology would be done by our research team and also that the business owner would
be provided with a short training on technology know-how, if the owner is willing to open
an account.4
We focus on restaurants and pharmacies because they share a similar business structure:
they have high frequency daily transactions and the payment per transaction is relatively
large. This implies that adopting an electronic payment technology is expected to be
particularly relevant for these businesses. In addition, cash theft – both external and
internal – is an important concern for both restaurants and pharmacies, and the incidence
of theft may be reduced by using Lipa Na M-Pesa.5 By means of our experiment, we
analyze the determinants of take-up of Lipa Na M-Pesa, while we also measure the impact
of Lipa Na M-Pesa on business outcomes.
We obtain four sets of key results. First, we confirm earlier findings regarding the impor-
tance of seemingly small adoption barriers in preventing the diffusion of cashless payment
4This included filling out a registration form and handing in copies of additional required documents toSafaricom, collecting the technology from Safaricom (once the till-number is issued) and bringing it backto the shop premises with Lipa N M-Pesa advertisement flyers and posters.
5According to Global Retail Theft Barometer Survey 2014-2015 conducted in 24 countries from AsiaPacific, Europe, Latin America, and US, among retailers, pharmacies have highest rate of losses due tointernal employee theft and shoplifting (external theft) (an equivalent of about 1.99% of sales).
2
instruments. Our experimental intervention uncovered a significant unmet demand for the
technology: 62% of the restaurants and 21% of the pharmacies in the treatment group in-
dicated that they would like to sign up. This result is in line with Bertrand, Mullainathan
and Shafir (2004), who argue that small situational barriers could play a decisive role in
preventing people to take advantage of available technologies. Quoting from Bertrand et al.
(2004, p. 420) “[small situational] barriers might be a testy bus ride, challenging hours, or
the reluctance to face a contemptuous [agent].” Indeed, many businesses who expressed in-
terest in the technology reported that, prior to our intervention, these barriers were reasons
for not adopting the technology.
Second, we find that business owners who are more open about their businesses’ financial
situation are more likely to be willing to adopt the technology. This is consistent with a-
priori expectations, because every Lipa Na transaction is traceable by Safaricom as business
activity. We also find that businesses with past exposure to mobile money products and
those who are more future biased and more trusting are more likely to want to adopt the
technology.
Third, we find that sixteen months after the intervention, the financial connectedness of
SMEs in the treatment group had increased compared to the control group, especially in
the form of improved access to mobile loans. More specifically, we observe that the amount
of mobile loans provided by Safaricom had increased, without significantly affecting access
to loans by other financial institutions, formal and informal. Lipa Na M-Pesa’s electronic
payment mechanism allows Safaricom to track business transactions, observe creditworthi-
ness of SMEs and lend mobile loans to those with higher probability of repayment. In
connection to this, we also find that the access to mobile loans that our treatment stim-
ulated is significantly more pronounced for small scale establishments, which are likely to
suffer the most from opacity and the lack of hard information in proving creditworthiness.
We conclude that financial transparency induced by Lipa Na M-Pesa fosters fast access
to short-term loans and replaces the necessity to build-up long-term relationships with
providers of working capital finance. Our results also reveal that small scale establishments
in the treatment group end up having less volatile sales over the course of the past 12
months compared to the control group. This business performance outcome is consistent
3
with our financial access finding on mobile loans, because mobile loans are designed to cope
with short-term uncertainty.
Finally we find that, sixteen months after the intervention, safety concerns were less
of an issue in the treatment group. This outcome is significantly more pronounced for
businesses operating in relatively unsafe areas, providing us with an underlying mechanism
for the result at hand.
We interpret the changes induced by our treatment in financial connectedness, reduced
sales volatility and improved safety as welfare improving, because perceived safer and less
uncertain business environments typically induce growth in the long-run and also because
the change in financial structure is reflective of improved financial inclusion.
Overall, our results show the importance that seemingly small adoption barriers have in
preventing the diffusion of cashless payment instruments. In addition, we suggest that the
electronic traceability of business transactions can be an important factor slowing down
the adoption of high-tech payment instruments among business owners who prefer to keep
transactions anonymous. However, we also show that once the technology is adopted,
electronic visibility is likely to have a positive impact on financial integration, which we
interpret as an opportunity cost of anonymity. These key findings are important for design-
ers of electronic payment instruments and policy makers in both advanced and developing
societies, as small adoption barriers and anonymity of transactions could pose challenges
across the globe.
The rest of the paper is organized as follows. Section 2 relates this paper to the existing
literature. Section 3 introduces the context and the technology. Section 4 describes the
RCT design and data. Section 5 presents the results on the adoption, usage and impact of
the technology. Section 6 concludes.
2 Contribution to the literature
This paper primarily contributes to the literature of cashless payment instruments by con-
ducting what constitutes, to our knowledge, the first field experiment with a payment
technology. Existing studies in this literature utilize non-experimental data or data from
4
laboratory experiments. Using survey data, for instance, Humphrey, Pulley and Vesela
(1996) document a positive association between debit card usage and the availability of
ATMs and a negative association between aggregate crime rates and debit card adoption.
Chakravorti (2007) argues that market competition is important in determining firms’ elec-
tronic payment adoption behavior, while Schuh and Stavins (2010) show that technological
developments in debit-card payments drive out checks and suggest that there are strong
substitutions between comparable payment methods that differ in the degree of their effi-
ciency. Bolt, Jonker and v. Renselaar (2010) refer to the issue of surcharges and suggest
that allowing for surcharges at retailers could be an important determinant of electronic
payment adoption. Different from the previous research, Arifovic, Duffy, and Jiang (2017)
develop a game theoretic framework, which the authors bring to a laboratory environ-
ment, to show that fixed adoption fees could be important in inhibiting the adoption of
an electronic money instrument. Similarly, Camera, Casari, and Bortolotti (2016) design
a laboratory experiment and show that eliminating service fees or introducing rewards can
have significant implications on adoption of electronic money instruments. We contribute
to this literature in that we identify and measure the barriers preventing adoption of a
cashless payment technology by SMEs in their own environment, as well as we measure the
impact of adopting such technology.
We also contribute to the recent and rapidly growing literature on the economic effects
of mobile money technologies, such as M-Pesa, in developing countries. Most of the studies
in this line of research aim to understand the implications of M-Pesa on household finance.
For instance, Mbiti and Weil (2011) find that the increased use of M-Pesa lowers the use
of informal savings mechanisms (for instance ROSCAS), and raises the propensity to save
via formal bank accounts. Jack, Ray and Suri (2013) and Jack and Suri (2014) show that
M-PESA help households manage financial uncertainties caused by crop failures, or health
issues and smooth consumption. The overarching conclusion of these studies is that the
users of M-PESA can access a wider network of support whenever financial needs arise, and
receive funds more quickly. The conclusions of our paper complement those of the studies
on household finance effects of M-Pesa, which reveal that Lipa Na M-Pesa could improve
5
financial connectedness and reduce sales volatility especially for small-scale enterprises.6
Different from the papers on household finance, Beck, Pamuk, Ramrattan and Uras
(2018) develop a dynamic general equilibrium model - calibrated with firm-level survey data
from Kenya - in order to evaluate the interactions between mobile-money, entrepreneurial
finance and macroeconomic activity. The authors quantify a substantial impact of M-Pesa
on SME trade credit arrangements and aggregate outcomes. While we also study mobile
money in the context of SMEs, and also uncover a complementarity between payment-
technologies and access to finance, our focus and method is substantially different. Our
focus is on the identification of the determinants to adoption of a technology designed to
cater the needs of SMEs, and on the implications for the SMEs of adopting the technology.
We also relate to the vast literature on SME finance. In particular, the findings presented
in this paper and the key underlying mechanism are relevant to studies on relationship lend-
ing. Seminal papers in this line of research, such as Petersen and Rajan (1994), Berger and
Udell (1995) and Degryse and Cayseele (2000), suggest that because of apparent difficulty
associated with collateral-based lending, extending loans to small and opaque businesses
requires the build-up of soft information, necessitating the formation of long-term bank-
firm relations. In a recent paper, Beck, Degryse, De Haas and van Horen (2018) show, for
instance, that long-term lending relationships help especially during business downturns
and significantly more so the smallest firms with non-transparent operations. Our findings
in this paper are in line with this research, while we also propose a policy-relevant novel an-
gle: connecting payment and lending means of financial services spectrum - using electronic
financial instruments - can overcome transaction frictions associated with the formation of
long-term lending relations and help small businesses to have fast access to loan products.
Finally, our paper also contributes to the technology adoption literature in the context
of developing countries.7 The studies that are most closely related to our research are the
field experiments on adoption of efficient technologies. Most studies in this literature con-
6Also in this literature, Suri and Jack (2016) show evidence of notable long-term effects of mobile-moneyon poverty reduction in Kenya. The authors estimate that since 2007, access to mobile-money servicesincreased daily per capita consumption levels considerably, lifting thousands of Kenyan households out ofextreme poverty.
7A large literature argues that differences in technology adoption rates explain per-capita income dif-ferences across countries (Caselli and Coleman (2001) and Comin and Hobijn (2004)).
6
centrate on the agricultural sector and in particular on the adoption of farming techniques
by small and micro enterprises, such as the seminal papers by Duflo, Kremer and Robinson
(2004, 2008, 2011) and Foster and Rosenzweig (2010). We add to this important literature
by studying the financial technology adoption decisions of SMEs in the service sector and
understanding the barriers to adopt and business outcome effects to this end. Importantly,
several papers in this literature highlighted the role of behavioral factors on technology
adoption, such as complexity of information, present bias and loss aversion.8 Our exper-
imental design reduces the complexity of information required to evaluate the benefits of
Lipa Na M-Pesa, while our survey design allows us to measure important behavioral factors
such as present bias, future bias, trust and cognitive capacity.9
3 Institutional Context: Mobile Money in Kenya
Over the past decade, mobile money has created a profound transformation in cashless
money circulation in the developing world. In 2007, Kenya’s Safaricom introduced the
first mobile-phone based money instrument, called M-Pesa, to enhance P2P (Person to
Person) money transfers. In this section we provide an overview of the standard M-Pesa
and then focus on Lipa Na M-Pesa, an extension of M-Pesa designed specially to facilitate
P2B (Person to Business) money transfers.
3.1 Standard M-Pesa
M-Pesa is the brand name of the most commonly utilized P2P mobile money service in
Kenya, which allows users to transfer money via mobile-phone text messages (SMS) to other
mobile money users.10 The way standard M-Pesa works is as follows. Users sign up for an
8See Hanna et al. (2014), and Drexler et al. (2014).9Also relevant for our paper are the empirical studies on technology adoption, which are interested in
understanding the heterogenity in technology adoption decisions across firms, such as Jack and Suri (2011)and Foster and Rosenzweig (2013). This strand of literature finds a positive correlation between technologyadoption and firm characteristics. Our empirical findings also uncover a heterogeneity in the adoption ofthe Lipa Na M-Pesa payment instrument, based on which we argue that the heterogeneity in relative costsand benefits of the technology could be important to explain differences in payment instrument take-uprates.
10At the time of the study, there were other mobile money providers in Kenya like Airtel Money, OrangeMoney, Equitel, Mobikash, and Tangaza. However, according to the Communications Authority of Kenya(2015), 77% (about 21 million) of the people who hold a mobile money account in June 2015 were M-Pesa
7
M-Pesa account and top it up by converting cash in M-Pesa units at specialized agents,
called M-Pesa kiosks. The electronic units can then be transferred to any other mobile
money user, utilized to make payments, or stored in the mobile phone. The recipients of
M-Pesa transfers can use the units to make new transfers or can redeem them at M-Pesa
kiosks.11
Signing up for a standard M-Pesa account does not entail any monetary cost. All that is
required is to visit an M-Pesa kiosk with an official identity document and a mobile phone.
Money can be exchanged for M-Pesa units free of charge, and the account holder does not
incur any costs either when receiving M-Pesa units. But a transfer fee is charged when
spending units by transferring them to another account, and another fee also applies when
converting M-Pesa units into cash. Both fees are step-wise increasing in the size of the
transaction, as shown in Figures 1 and 2.
Figures 1 and 2 here
With this electronic money technology, Safaricom has revolutionized P2P money trans-
fers in Kenya. By 2013 and only six years after being launched, more than 70% of Kenyan
households had an M-Pesa account.12 In that year 733 million M-Pesa transactions took
place, with an aggregate value of about 1.9 trillion Kenyan shillings (about 22 billion US
dollars), an equivalent of 40% of Kenyan GDP at the time.13 However, despite the high
adoption rate of M-Pesa by households, businesses did not show the same enthusiasm to
use M-Pesa for P2B and B2B money transactions. According to a nation-wide survey
conducted by FSD-K, only 35 percent of SMEs accepted M-Pesa as a common method of
payment over the years of 2013-2014, especially because of technological challenges and
financial considerations.14 Safaricom was well-aware of these barriers and, recognizing the
niche in the P2B market, introduced Lipa Na M-Pesa in 2014, a mobile money facility
users.11In December 2014, there were about 124,000 M-Pesa kiosks scattered across all Kenya. Around 20
percent of them are located in Nairobi (FSP interactive maps, 2013) with approximately 25 million cus-tomers.
12See Jack and Suri (2014).13The annual transaction volumes of mobile money transactions are the sum of the
monthly transaction volumes reported by the Central Bank of Kenya National Paymenthttps://www.centralbank.go.ke/national-payments-system/mobile-payments/.
14See FinAccess Business Survey, 2014.
8
developed to stimulate P2B transactions. In what follows we briefly describe the main
characteristics of Lipa Na M-Pesa, which constitutes the core of this paper.
3.2 Lipa Na M-Pesa
Lipa Na M-Pesa is a product especially designed for merchants, allowing them to receive
payments in retail transactions. When a business owner signs up for a Lipa Na M-Pesa
account, unlike for the case of standard M-Pesa, the account gets registered under the
name of the business. For this reason, the transactions made through Lipa Na M-Pesa
become visible to Safaricom as business activity. In addition to reducing cash-based transfer
frictions in the same way as standard M-Pesa does, Lipa Na M-Pesa offers monetary and
technological benefits that make the product attractive for P2B purposes.
With respect to the monetary benefits, standard M-Pesa users can make Lipa Na M-Pesa
payments without incurring any monetary costs. This is a major difference compared to the
P2P transactions made between two standard M-Pesa accounts, where the account holder
making the transaction (in this case, the buyer) needs to pay a transaction fee (Figure 1).
Second, while the business has to incur a payment receipt fee of 1% of the transaction value,
the costs are lower than those incurred by the customer when using standard M-Pesa for
all transactions below 8500 KSh (more than US$ 80). Importantly, the typical transaction
values the merchants tend to have in our sample are well below that threshold, implying
that using Lipa Na M-Pesa raises the surplus generated by an economic transaction between
a customer and a business.
Lipa Na M-Pesa also has two key technological advantages compared to standard M-
Pesa. First, there are restrictions on the amount of money one can store in a standard
M-Pesa account, while such restrictions are virtually absent for Lipa Na M-Pesa. This
implies that a business owner with Lipa Na M-Pesa does not need to cash-out so often
as with standard M-Pesa, substantially reducing the costs of cash withdrawals.15 Second,
Safaricom records all the transactions made via Lipa Na M-Pesa, and allows the business
owner free-of-charge access to daily transaction-records over a six-month period; all that
15The fees charged for cash withdrawal from Lipa Na M-Pesa and standard M-Pesa accounts are thesame; see Figure 2.
9
is required is to send the request via a simple e-mail. In this way, Lipa Na M-Pesa offers
business owners an option for bookkeeping system at zero cost.
3.3 Potential Factors affecting Lipa Na M-Pesa Adoption
The positive features of Lipa Na M-Pesa do not imply that the adoption of this technology
would be immediate among merchants. The literature of technology adoption – in devel-
oped as well as in developing countries – has consistently shown that adoption of a new
technology, regardless of how efficient the technology might be, tends to be slow due to the
existence of pecuniary and non-pecuniary factors. In the context of Lipa Na M-Pesa we
identify a number of factors that might prevent adoption of this new payment instrument.
First, the merchants may lack information about the costs and benefits of the technology,
since the technology was new in the market. Second, even though there is no fee to opening
an account, there are transaction costs associated with the time needed to do the paperwork,
e.g. filling out a form and handing it in at a Safaricom office. Such costs, even if seemingly
low compared to the benefits of the technology, have proven to be consequential especially
in the context of developing countries (Bertrand, Mullainathan and Shafir, 2004). Third,
there could be technology implementation barriers: merchants might not know how to
(make optimal) use of the various Lipa Na M-Pesa features and services. We consider these
three as “soft barriers”, because they could be mitigated at relatively low cost.
The decision to adopt Lipa Na M-Pesa can also be explained by other potential factors.
Notably, Safaricom traces all transactions made through Lipa Na M-Pesa as business activ-
ity. And, in addition to its money transfer and payment services, Safaricom provides also
mobile loans (called M-Shwari). For financially constrained firms, which otherwise do not
have the chance to signal the flow of their business activity to external financiers, sharing
business transaction volumes with Safaricom could thus help with accessing external finance
in the form of M-Shwari loans. For other businesses, however, transparency could be more
of a concern if the owner is not willing to disclose business transactions to third parties.
Additional potential factors that can explain adoption relate to exposure to sophisticated
payment products in the past, exposure to cash theft and behavioral characteristics of the
business owner.
10
4 Research Design
The purpose of our research is twofold. The first is to test the relevance of the above-
mentioned soft adoption barriers by exogenously lifting them for randomly selected enter-
prises, and then to compare adoption rates between the treatment and control groups. The
second purpose is to estimate the impact of Lipa Na M-Pesa adoption on a number of busi-
ness outcomes. We do so by exploiting the exogenous difference in adoption rates between
the treatment and control groups our intervention gave rise to. Our research thus consists
of an RCT with detailed baseline and endline surveys. The RCT is an intention-to-treat
field experiment with two possible forms of non-compliance. Businesses in the treatment
group are of course free to not adopt the technology. Moreover, the technology is also freely
available on the market, which means that at least some businesses in the control group
may also end up adopting the technology. That means that our experimental design is
closest to a standard encouragement design (Bradlow 1998).
The detailed baseline survey allows us to identify the other potential factors that can
explain the decision to adopt the technology, and that are not released experimentally. We
complement the baseline survey with an endline survey sixteen months after the intervention
to evaluate the effect of releasing the barriers on business outcomes. In what follows, we
introduce the experimental design and the survey data.
4.1 Experimental Design
4.1.1 Sampling and Randomization
The study took place in the periphery of Nairobi’s central business district. We chose
this location because lower business densities mitigate possible spillover concerns between
businesses in the treatment and control groups. To construct a sample of comparable
businesses, we focused only on two sectors: restaurants and pharmacies. The choice of
these two activities was based on the fact that both share characteristics which potentiate
the benefits of a cashless payment technology. They both are sectors in which SMEs are
overrepresented, they handle relatively many transactions per day, have relatively many
customers, and are relatively vulnerable to internal and external theft.
11
To draw the list of eligible firms for the study, we randomly assigned enumerators to
visit specific areas in the city and requested them to list restaurants and pharmacies that
had one or more employees, that were located at a distance not less than 50 meters from
the closest other business in the same sector, did not have a Lipa Na M-Pesa account yet,
and whose owner or manager was willing to participate in a study on mobile money use by
businesses. Following this procedure, we listed in total 1222 merchants, 669 restaurants and
553 pharmacies. Out of the 1222 eligible merchants, we randomly assigned 607 merchants
(331 restaurants and 276 pharmacies) to the treatment group and 615 merchants to the
control group. Random assignment was stratified by geographic location and by firm size
- measured by the number of employees.16 Figures 3, 4 and 5 provide the geographic
distribution of the sample of businesses over districts of Nairobi, by treatment and control
groups.
Figures 3, 4 and 5 here
4.1.2 Experimental Intervention
The experimental intervention was conducted right after concluding the baseline survey
interview. Firms assigned to the treatment group were informed that the researchers in-
volved in this study had done research on Lipa Na M-Pesa’s potential as a new cashless
payment instrument, and they were interested in informing merchants about the costs and
benefits of the technology. After receiving this information, the merchant was asked if he
or she would like us to open an account on his or her behalf at no cost.17 The answer to
this question is what constitutes the measure of willingness to adopt the technology. The
timing of the events is decribed in the following timeline.
Timeline here
The intervention itself consisted of three components, which we describe next.
16We consider a restaurant (respectively pharmacy) with more than 5 (respectively 2) employees to bebig and we use this categorization to stratify the sample. The geographic division of the area was madead-hoc by the survey company for logistical reasons.
17We made it very clear to the merchant that we did not have any relationship or commercial ties withSafaricom. We explicitly stated that our aim was purely academic.
12
Provision of Information about the Technology
The objective of this component was to provide information on the advantages and
disadvantages of Lipa Na M-Pesa compared to other payment methods. The information
was provided with an information leaflet and a video. The leaflet consisted of concrete and
easy-to-understand information on Lipa Na M-Pesa’s costs and benefits. A recent literature
emphasizes that the complexity of information concerning the benefits of a new technology
can limit adoption. People only pay attention to the slice of the whole information set
which they think is the most relevant to them. Hanna, Mullainathan and Joshua (2014)
suggest that this issue could be handled with the provision of simplified information. We
design our experiment with this insight in mind.
The video complemented the leaflet by showing a short clip of an interview with a fel-
low business owner from the same sector who had already adopted Lipa Na M-Pesa and
benefited from its usage.18 The inclusion of the video as a component of the intervention is
motivated by an emerging literature highlighting the effectiveness of role models to induce
behavioral change. This literature shows that successful peers can act as role models and
are particularly effective in the context of low-income households in developing countries
(see for instance Bernard et al. (2014) and La Ferrara (2016)).
Support for the Registration Process
Besides providing information, we also offered to handle the paperwork needed to open a
Lipa Na M-Pesa account on behalf of business owners. The motivation for this component is
based on the literature pioneered by Bertrand, Mullainathan and Shafir (2004), who argue
that small situational barriers play a decisive role in preventing people to take advantage
of efficient investment opportunities. The paperwork associated with opening an account
can be perceived as a hassle for a small firm owner and can prevent him or her to adopt
the technology. Specifically, we offered to fill out the registration form on behalf of the
business owner, bring the required forms and copies of the documents to Safaricom, pick
18There were thus two videos – a 5.2 minute clip for the restaurant sector and a 3.2 minute clip forpharmacists. The videos as well as the information leaflet that we produced are available upon request.
13
up the SIM-card and advertisement material from Safaricom and deliver this material at
the business premises.
Technology Implementation Assistance
We offered the business owner to take the business to a “transaction ready” mode,
should he or she be willing to adopt the technology. Specifically, this component of the
intervention consisted of inserting the Lipa Na M-Pesa SIM-card in the mobile phone the
business owner would use for Lipa Na M-pesa transactions, testing whether the SIM-card
was functional, and performing a transaction test worth of 100 KShs to illustrate how to
use the account in practice.
4.2 Survey Measures
A key objective of our baseline survey is to measure the factors that could potentially be
associated with the adoption of Lipa Na M-Pesa but which were not exogenously varied by
our experimental intervention. This subsection describes these factors and the way they
were measured in the survey instrument.
4.2.1 Financial Transparency Aversion
Transactions made through Lipa Na M-Pesa get recorded as business activity by Safaricom.
Therefore, firms that are averse to disclose their financial activities with a third party
might be reluctant to adopt the technology. These transparency concerns are difficult
to measure directly, but they can be captured in an indirect way. For this purpose, we
proxied transparency concerns with a dummy variable indicating the willingness of the
business owner to disclose revenue and profit figures during the baseline survey interview.
This measure, of course, could also be correlated with trust of the survey respondent in
the interviewer. For this reason, we gather a measure for “trust in a person at a first-time
contact” and include it as a control variable in the empirical analysis.19
Financial transparency concerns may be correlated with key business characteristics,
19As complementary information, we also created a financial sophistication index, which covers formalloan use from banks and mobile money providers, as well as a dummy regarding keeping business records.
14
such as financial sophistication20, profitability, size, and holding an up-to-date business
license. To mitigate these concerns, we control for these variables too.
4.2.2 Behavioral Factors
Recent studies stress the importance of behavioral factors in the technology adoption de-
cisions of small enterprises. Our design is, in part, motivated by these behavioral insights,
as we include role models to foster social learning and account for limited attention and
seemingly small bureaucratic hassles as important potential determinants of technology
adoption. However, there are other potential behavioural aspects that can interfere the
decision to adopt the technology. We measure these aspects in the baseline survey as
follows.
First, we collect a proxy measure of each entrepreneur’s cognitive function, as a plau-
sible determinant of the ability to grasp the information on Lipa Na M-Pesa costs and
benefits. We measure cognitive function by means of the Digit Span Task, in which the
entrepreneur is read a list of numbers and then asked to repeat these numbers in the same
order.21 Outcomes for this task are the longest correctly remembered span, and also overall
accuracy.22
Second, we elicit time preferences, as well as present and future bias. The reason we are
interested in these is that the decision to adopt a new technology involves planning (costs)
in the present, while benefits are obtained in the future. We elicited these preferences and
biases in an incentive-compatible way adopting the method used by, among others, Dupas
and Robinson (2013).23 Merchants were asked to choose between receiving either 500 Ksh
(US$4.93) the next day, or receiving a larger amount in 31 days. The larger the amount
needed to induce the merchant to choose for the later payment, the more impatient he
or she is coded to be. To measure time inconsistency, we also asked merchants to choose
between Ksh 500 in 31 days and a larger amount in 61 days. A merchant is defined to be
20For example, by being connected to a formal credit network, offering the option to customers topurchase on credit and keeping business records.
21Daneman and Carpenter (1980 and 1983).22This task is widely used in field experiments in economics as a proxy for cognitive ability, particularly
in the context of developing countries. See Dean, Schilbach, and Schofield (2017) for a review.23By incentive-compatible measures we mean that the decisions were actually paid.
15
present biased if he or she is more impatient in the present than in the future, i.e. exhibits
a higher discount rate between tomorrow and 31 days than between 31 days and 61 days.
In contrast, a merchant that is more patient in the present than in the future is defined to
be future biased.
Finally, we measure trust, as we conjecture that trust is particularly relevant in explain-
ing adoption of Lipa Na M-Pesa. Access to finance is quite limited in developing economies,
also because trust in financial contract enforcement is weak.24 Therefore, specially in our
context, lack of trust in customers, in people met for the first time, in Safaricom, and in
institutions in general, are potential important explanatory factors. We measure all these
dimensions with direct survey questions.
4.2.3 Other Potential Factors
We measure additional factors that we conjecture can contribute to explain the decision to
adopt an electronic money instrument. First, based on results from a laboratory experiment
by Arifovic, Duffy, and Jiang (2017), we conjecture that merchants who have a positive
previous experience with electronic money are more likely to adopt Lipa Na M-Pesa. For
this purpose, we ask if the merchant uses the (standard) personal mobile money account
for business purposes.
Second, we conjecture that merchants who are exposed to high risk of theft may value
Lipa Na M-Pesa relatively more (see also Jack and Suri (2014) and Economides and
Jeziorski (2016)). To capture this, we ask the business owner about his or her own percep-
tion of safety in the business premises, and about the frequency and magnitude of theft he
or she experienced over the last six months.
Third, we also conjecture that Lipa Na M-Pesa is less appealing for businesses who
frequently transfer money into saving accounts. Exchanging Lipa Na M-Pesa credits for
cash is subject to a withdrawal fee, and so is sending money to a bank account. By the
time our experiment was conducted, there was only two banks that provided real time
cash transfer services from Lipa Na M-Pesa to business bank accounts. At other financial
institutions, when business owners wanted to transfer money from Lipa Na M-pesa accounts
24See e.g. La Porta et al. (1997 and 1998).
16
to their business bank accounts, they were required to use Pay-Bill or M-Pesa transfer
services, which are costly and take at least 6 hours to complete the transfer. To account
for this in our analysis, we ask the merchant about the use of bank accounts for business
purposes to deposit cash.
Finally, we conjecture that smaller firms are expected to benefit more from a release of
fixed adoption costs, but bigger firms are expected to benefit more from the technology
due to higher sales, large pool of customers and higher likelihood of theft by employees. In
our survey we proxy size with the number of employees and total sales.
4.3 Sample Characteristics
Table 1 describes the characteristics of the firms in our sample and compares the firms
assigned to control and treatment groups.
Table 1 here
The average business in our sample is rather small. On average, pharmacies employ
three workers and restaurants five. The average monthly sales and profits of pharmacies
is about 1470 and 600 U.S dollars PPP respectively, and restaurants have about 3225 US
dollars PPP in sales and 820 dollars in profits. Only 19% of the pharmacies and 36% of
restaurants have made investments in their businesses in the past six months and only few
businesses have received loans in the past twelve months. Moreover, 91% of pharmacies
and 54% of restaurants have a business license. Almost all business owners in our sample
posses a personal M-Pesa account at the time of the baseline. 43% of restaurant owners and
31% of pharmacy owners report mobile money to be the most frequent method of payment
to suppliers, while 40% of restaurants and 25% of pharmacies have received payments from
their customers through personal mobile money account of the business owner.
Moreover, more than 90% of pharmacies and restaurants in our sample knew about the
existence of Lipa Na M-pesa, although they had not adopted the technology yet at the time
of our baseline. The primary reason reported for not having a Lipa Na M-Pesa account was
perceived lack of net benefits. Business owners also perceived that it was too costly to open
the account, that Lipa Na M-pesa transaction fees were too high, and that they did not have
17
the time to do the registration paperwork. This pre-treatment information is enlightening
in two ways. First, it reveals that merchants had wrong beliefs about the cost and benefits
to adopt and use the technology, which our intervention aimed to correct. Second, it
becomes clear that the paperwork for the registration was perceived as an impediment to
open an account, an aspect that was directly targeted by our experimental intervention.
Finally, it is important to note that no business in the sample sells its main product for
more than 8500 Kenyan Shillings (about 80 US$ by the time of the baseline survey), which
is the threshold amount of payment above which transferring money to a Lipa Na M-Pesa
account becomes more expensive than transferring to a standard M-pesa account in a P2B
transaction (Figure 1). Since surcharging is allowed, this means that all businesses in our
sample could generate a transaction surplus by asking the customers to pay through the
business Lipa Na M-Pesa account instead of business owners’ personal M-Pesa.
Columns 4 and 8 of Table 1 report the differences in average baseline firm characteris-
tics between treatment and control groups and the statistical significance for the difference
between the two groups resulting from t-tests. Overall, the sample characteristics of restau-
rants are balanced between the treatment and control groups across the board. Treated
pharmacies are more likely to use standard mobile money vis-a-vis their customers and pay
salaries through the mobile money accounts of their employees more often. Likewise, more
pharmacies in the treatment group opened a Lipa Na M-pesa account on their own in the
period between the listing and the baseline survey.25 We implement only the information
component of our treatment in those businesses, which were assigned to the treatment
group and opened a Lipa Na M-Pesa account before our experimental intervention.
Finally, relative to the control group, the treated pharmacies have also higher sales
and profits. However, as we present in Table OA4 of the online appendix, statistically
significant differences in sales and profits between the two pharmacy groups cease to exist
when we concentrate on businesses, which participate in the endline survey and possess
a business license.26 In addition, we perform a test for joint orthogonality by estimating
2582 restaurants (12%) and 25 pharmacies (5%) adopted the Lipa Na M-pesa technology in the periodsbetween listing and baseline/intervention.
26Because of a Safaricom policy change that we will explain below our outcome analysis concentrates onbusinesses with a business license.
18
a linear regression where the dependent variable is being assigned to treatment and the
explanatory variables are all the variables in Table 1. We test the null that the coefficients
of explanatory variables are jointly zero and find that the joint orthogonality test cannot
reject the null hypothesis.27 The observed characteristics cannot jointly explain treatment
assignment, suggesting that the two groups are well-balanced. Notwithstanding this, to be
conservative, in our outcome regression analyses we control for the baseline values that are
unbalanced at baseline.
5 Analysis and Results
We present the empirical analysis and results in two parts. In Section 5.1 we investigate
the factors explaining the willingness to adopt the technology and in Section 5.2 we study
the impact of our intervention on business outcomes.
5.1 Willingness to Adopt the Technology
Interest of SMEs in the Lipa Na M-Pesa technology can be gauged from the response of the
firms in the treatment group to our offer to open an account on their behalf. Acceptance
rates to our offer were 62% among the restaurants and 21% among the pharmacies. These
rates are high, given that the technology had been available for more than a year before our
intervention. This means under normal market conditions there was a substantial “unmet
demand” for the technology. It also shows that the “soft” adoption barriers were quite
important in preventing actual adoption.
Figure 6 presents the main reasons why firms had not adopted Lipa Na M-Pesa prior to
the intervention. Perceived lack of benefits was an important reason for not having adopted,
and especially more so for the pharmacies. Additional reasons were the (time) costs of
opening an account, and perceived complexity of using the technology. Interestingly, our
intervention was targeted at (i) providing good information on the benefits of the technolgy,
(ii) reducing the (time) costs of opening an account, and (iii) explaining usage. We conclude
that our intervention indeed targeted the key factors preventing adoption. Similar insights
27An F-test for the joint significance of 23 coefficient estimates after a linear regression produces a p-valueof 0.18 for pharmacies and 0.70 for restaurants.
19
are obtained when comparing the main reasons for not having adopted yet between those
who ended up accepting our offer, and those who did not. Results are reported in Figures
7 and 8. As can be gauged from Figure 7, our intervention increased take-up among
restaurants especially because it solved the time constraint, but we did not manage to
convince all respondents of the benefits of using the technology. The latter was also an
issue with the pharmacies, as shown in Figure 8, and the same holds for the costs of opening
the account, and of using it.
Next, we analyze what types of firms are interested in receiving an account. Table 2
explores in which respects firms accepting our offer differ from those that rejected it. Prior
usage of electronic payment systems, and then especially standard M-Pesa, is an important
factor. In case of restaurants this holds for nearly all mobile money usages; for pharmacies
receiving payments and storing money are the most important ones. Pharmacies that
accepted our offer tend to be larger and more profitable, while restaurants that tend to
invest more are also more prone to be willing to receive an account.
Figures 6, 7 and 8 here
Table 2 here
Exploring in what respects adopters and non-adopters tend to differ is one approach
to gauge insight into the determinants of the adoption process; using formal regression
analysis is another. We specify the following regression model for the latter purpose:
Adopti = α+ β′Exposurei +ω′Transparencyi + γ′Behaviori +φ′Sizei +χd +µj + εi, (1)
where i denotes the business, Adopti is a dummy variable equal to one if the business accepts
the offer to open an account, Exposurei is a vector of measures of past experience with
mobile money, Transparencyi is a vector of financial transparency measures, Behaviori is
a vector of behavioral factors, Sizei is a vector of business size measures, µj controls for
sectoral differences (pharmacies vs restaurants) and χd is the enumerator-and-district fixed
effects.28
28Table OA1 in Online Appendix provides the full list of variables utilized in this analysis and TableOA2 provides summary statistics of the main variables.
20
We estimate three specifications of Equation (1). In the first specification, we use ag-
gregated exposure and transparency-concern proxies in addition to behavioral factors and
business size indicators. In the second and third specifications we replace the aggregated
indicators with the indices’ individual components. Table 3 provides results from the full
sample of firms, while tables 4 and 5 report the results for restaurants and pharmacies
separately. In all tables the variables in panel A measure ex-ante mobile money exposure,
variables in panel B measure transparency concerns, those in panel C capture behavioral
aspects and the variables in panel D are business size indicators.
Table 3 here
The results reported in Table 3 confirm that “pre-treatment exposure to mobile money”
is a significant determinant of a firm’s willingness to adopt the technology. This echoes
some recent findings from laboratory experiments.29 Column (1) shows that firms that
use mobile money for business purposes are 14.2 percentage points more likely to adopt
the technology compared to other businesses. The more detailed measures of prior mobile
money usage indicate for example that firms that use mobile money to buy inputs are
more prone to request to receive Lipa Na M-Pesa. Also importantly, we observe that those
businesses which do not surcharge when transacting with M-Pesa are also more eager to
accept our offer to adopt the Lipa Na M-Pesa technology.
As an interesting piece of evidence we obtain that businesses that are non-transparent
with respect to their profits and sales in the baseline interview are less likely to sign up.
While other transparency aversion proxies (like not having an official business license, or
measures of financial sophistication) turn out to be insignificant, all coefficient estimates
have expected signs. As far as we are aware, this is a very novel finding, which is critical
both for the design of financial products as well as for policies aiming to promote electronic
payment usage. We will return to this important adoption channel when analyzing the
outcome effects of Lipa Na M-Pesa.
The behavioral factors perform more unevenly across various specifications, though one
key behavioral aspect stands out: cognitive capacity of the business owner is a significant
29Camera, Casari, and Bortolotti (2016) and Arifovic, Duffy, and Jiang (2017).
21
determinant of adoption throughout regression specifications in Table 3, confirming a priori
expectations.
With respect to other determinants, we observe that the decision to adopt is by and
large independent of (indicators of) firm size. Finally, results in Table 3 also show that
businesses which save at a business bank account are less likely to adopt. This pattern
holds most likely because banked firms can access payment and other financial services
through their bank and also importantly because transferring funds from Lipa Na M-Pesa
to a bank account is costly, as highlighted before.
Tables 4 and 5 confirm that pre-treatment use of mobile money turns out to be impor-
tant in both sectors. Also transparency aversion has negative coefficient estimates for both
pharmacies and restaurants, while the coefficient is not statistically significant in restau-
rants sub-sample, it is significant at 5% level for pharmacies. Having a business saving
account at a bank is negatively associated with the adoption of Lipa Na M-Pesa for phar-
macies (Table 5), but not as much for restaurants (Table 4). Similar conclusions apply to
behavioral biases, which show up significantly for pharmacies but not for restaurants.
Tables 4 and 5 here
5.2 Adoption, Use and Impact of the Technology
As we will delineate below, our intervention was successful in increasing the take-up in the
treatment compared to the control group, and hence we can measure the impact of Lipa Na
M-Pesa on business outcomes. The sample that we use for the outcome analysis consists of
618 businesses – all those that possessed a business license at baseline and that participated
in the endline30, which took place more than 16 months after the baseline intervention. A
change in Safaricom’s operating procedure is the reason why we were forced to conduct
our outcome analysis using only licensed businesses: after our baseline research had been
completed and during the process of registering businesses for Lipa Na M-Pesa, Safaricom
changed one of the formal requisites to open a Lipa Na M-Pesa account – any applying
firm should have an up-to-date business license. Because pharmacies tend to be more
30Table OA3 in Online Appendix provides summary statistics for the endline survey.
22
likely to have a business license, the change in the rule had a larger impact on our pool
of restaurants than on our pool of pharmacies. Importantly, as Table 1 illustrates with
respect to possessing a business license treatment and control groups are balanced for both
restaurants and pharmacies.31
We first examine whether attrition at endline is non-random. We regress “not having
participated in the endline survey” on the treatment dummy as well as on other business
characteristics. The estimates reported in Table 6 indicate that assignment to treatment
is not significantly related to attrition, implying that attrition does not bias our impact
estimates for the treatment. However, we also find that there is more attrition among
restaurants than among pharmacies, and also that there is more attrition for establishments
that were larger at baseline, experienced more external theft, received mobile loans, or had
more trust in mobile money companies.
5.2.1 Lipa Na M-Pesa Adoption and Usage
We first check to what extent our intervention was successful in exogenously increasing
adoption in the treatment group compared to the control group. We do so for various
measures of take-up: “actual registration of the Lipa Na technology”, “usage of Lipa Na
M-Pesa over the last 30 days”, “having utilized Lipa Na M-Pesa to receive payments from
the customers over the last 30 days”, and “sales through Lipa Na M-Pesa over the last
30 days”. The results are presented in Table 7A. The coefficients can be interpreted as
percentage-point differences in actual take-up between the treatment and the control group
(in columns 1-3) or as the percentage increase in take-up itself (in the fourth column).32
All four of the adoption measures are significantly higher in the treatment group. About
7 percentage points more businesses in the treatment group ended up having a Lipa Na
M-Pesa account by the time of the endline, usage is about 8 percentage points higher
among treated businesses, and the same holds for the increase in the propensity to receive
31The change in the business license requirement made 292 restaurants - which participated in ourbaseline survey - ineligible to adopt the technology since they did not have a business license. 47 pharmaciesstated that they had a business license in the sampling visits; however, baseline data collection revealedthat they did not have one.
32For presentational purposes, the coefficients of control variables have been omitted from the Table.The control variables we use are baseline measures of sales, not reporting sales in baseline, number ofemployees, use of Lipa Na M-Pesa in the baseline, enumerator FE, gender and business sector.
23
payments via the new technology. According to these three measures the uptake was
increased by 30% to 35% through our treatment. As shown in column 4 of Table 7A, the
treatment increased monthly sales via Lipa Na M-pesa by 26%, or 32 US$ (or 3256 Ksh).
This amount corresponds to 15% of the average sales of an SME from our pooled sample
in the endline.
Table 7A here
Table 7B shows intention-to-treat (ITT) estimates for a split sample based on merchant’s
aversion towards financial transparency. Panel A shows the results for the transparent firms,
Panel B for the non-transparent ones. Consistent with the findings reported in Section
5.1, the intervention induced Lipa Na M-Pesa usage significantly more among transparent
firms that disclosed their sales figures to enumerators during the baseline interview. These
results indicate that financial transparency concerns might play an important role not only
to explain the adoption of the technology, but also to explain usage when it is adopted.
This reveals potentially persistent effects of transparency in understanding the diffusion of
electronic payment products.
Table 7B here
5.2.2 Business Outcomes
In Section 3.3 we hypothesized that Lipa Na M-Pesa usage may affect business outcomes,
such us perceived safety and access to finance. In what follows, we report results of the
ITT regressions to assess the average treatment effect on these business outcomes as well
as on sales and investment. We also report the treatment-on-the-treated estimates (ToT)
throughout our outcome analysis.
Business Safety
Table 8A presents ITT estimates on the impact of treatment assignment on perceived
safety, proxied by the endline measure of “feeling more safe when conducting the business
operations” - measured from a scale from 1 to 10.33 On average, the group of treated firms
33The control variables we use are baseline value of corresponding outcome variable, baseline measuresof sales, not reporting sales in baseline, number of employees, use of Lipa Na M-Pesa in the baseline,enumerator FE, gender and business sector.
24
experienced a 3.5% increase in perceived safety, an effect that is statistically significant.
Given that exposure to the intervention increased Lipa Na M-Pesa usage by about 40%,
we document a ToT effect of just below 9% increase in perceived safety. This is an impor-
tant result as it confirms the role of an electronic money instrument in raising the retail
transaction safety among SMEs.
Table 8A here
In order to understand the mechanism behind the increased perceived safety, we split the
sample by exposure to theft at baseline. Column 1 of Table 8B provides ITT estimates for
the sub-sample of firms which reported to have experienced external theft one year before
the intervention, while column 2 presents ITT estimates for those which did not experience
theft. We show that the effect documented in Table 8A is caused by a substantial increase
in perceived safety by firms that had experienced theft six months prior to the intervention.
This finding reinforces the relevance of Lipa Na M-Pesa on business safety, as it shows that
in areas where using cash is relatively more risky, safety outcome effect is stronger.
Table 8B here
Business Finance
Access to finance is an important driver of business performance. Table 3 already doc-
uments that non-banked firms are more likely to adopt Lipa Na M-Pesa. That possibly
means that firms considered that having access to mobile loans is one of the advantages of
having a Lipa Na M-Pesa account, because Safaricom is not just the provider of M-Pesa
products but also a provider of mobile-loans, M-Shwari. Lipa Na M-Pesa allows Safaricom
to have access to detailed information on electronic payments made to the firm, which is
likely to relax hard information constraints and stimulate access to external finance. We
test this mechanism in tables 9 and 10.
Table 9 reports ITT estimates of treatment impact on formal (Panel A) and informal
finance (Panel B). For each source of finance we report the impact of treatment assignment
on the extensive margin financial inclusion (columns 1 and 3) as well as on the intensive
25
margin financial deepening (measured in logarithm of the borrowed amount in columns 2
and 4).
A novel source of external finance that we study in Table 9 refers to mobile-loans, which
exhibit the following characteristics. The dominant mobile loan provider of the country is
Safaricom. Similar to the M-Pesa service, the provision of mobile loans are made to cellular
phones of borrowers. Safaricom charges 7.5% interest on mobile loans and this interest
charge does not go up with the loan amount. The loan limit is determined by the history
of transactions incurred by the borrower via M-Pesa and in particular through Lipa Na
M-Pesa payment history for the case of a business owner. In this respect, creditworthiness
signalled to Safaricom through Lipa Na M-Pesa use help borrowers to expand credit limits.
Finally, since mobile-loans are typically extended for 30 days, they are utilized to curb
short-term fluctuations in business activity - and not so much for long-term investment.
Results in Table 9 show that treatment assignment significantly influenced mobile-loan
usage at the extensive (column 1) margin, and also at the intensive margin (column 2). Both
effects are statistically significant at 5% level and economically large. Both the financial
inclusion and financial deepening effects correspond to an increase of 60% compared the
control group mean. Interestingly, we do not observe a contraction in any of the other
sources of external finance – neither formal nor informal. This implies that the exogenous
increase in Lipa Na M-Pesa usage unambiguously increased financial access for the treated
businesses.
Table 9 here
Having established that Lipa Na M-pesa increases financial access, we proceed with un-
derstanding the mechanism behind this result by studying heterogeneous treatment effects
with respect to business size. If Lipa Na M-Pesa use is resolving financial opacity and
implied hard information constraints, we should expect the mobile loan impact to be the
most significant among small-scale establishments. We classify a business as small if its
baseline number of employees is smaller than the median number of employees in the re-
spective sector and study heterogenous treatment effects with respect to business size.34
34In the baseline median business size is 3 employees for pharmacies and 5 employees for restaurants.
26
Table 10 shows that the business size is an important factor affecting the mechanism via
which the treatment affects financial connectedness. Specifically, as indicated by the posi-
tive and significant interaction terms associated with “Small x Treated” in Panel A, Lipa
Na M-Pesa treatment increases the probability of receiving a loan as well as the size of the
loan for small firms more than the rest of the sample: “Small x Treated” has a statistically
significant effect on having a mobile loan at 1% level, while the intensive margin effect on
mobile loan size is statistically significant at 5% level.35
This differential impact is consistent with the literature arguing that the smaller estab-
lishments are likely to suffer from credit market exclusion because of financial opacity. For
instance, the literature on relationship lending suggests that extending loans to small and
non-transparent businesses requires the build-up of soft information and trust and therefore
necessitates formation of long-term bank-firm relations.36 What we propose here is that an
electronic payment technology, such as Lipa Na M-Pesa, could induce transparency, reduce
the necessity to rely on a bank-firm relationship to prove creditworthiness and thereby
allow for fast and low-cost access to external finance - to compensate short-term liquidity
needs.
Another interesting piece of evidence we derive from Table 10 is that our treatment also
stimulated small businesses’ informal finance, such as trade credit and loans from informal
financial networks. Although coefficient estimates are not very significant, we see that
at both extensive and intensive margins there is an overall increase in access to informal
finance among treated small businesses. This result echoes the findings in the literature
that suggest that access to formal loans (such as mobile loans in our analysis) could work
as a signalling device for creditworthiness and enhance access to informal loans as well.37
Table 10 here
Business Sales
35In all heterogeneous treatment regressions we control for the dummy variable “Small” on the RHS.36See Petersen and Rajan (1994), Berger and Udell (1995), Degryse and Cayseele (2000), and Beck, De
Haas, Degryse and v. Horen (2018).37Such complementarities are highlighted, among others, in Demirguc-Kunt and Maksimovic (2001),
Burkart and Ellingsen (2004) and Burkart, Ellingsen and Giannetti (2011).
27
What is the impact of Lipa Na M-Pesa on business sales? As shown in the first column
of Table 11, our treatment did not have a significant impact on the overall level of sales.
We do not find this outcome surprising for two reasons. First, if the real performance effect
of Lipa Na M-Pesa works its way through stimulated financial integration (especially by
mobile loans), then this effect is expected to be relatively more present for small businesses
and not so much for medium-sized establishments. Indeed in column 2 of Table 11 we
observe a positive coefficient estimate for “Small x Treated” - though insignificant - in an
ITT regression with sales, a finding that is consistent with what we observed in Table 10.38
Table 11 here
Second, and more importantly, mobile loans are much more likely to help coping with
cyclical shocks, such as inventory shortages, and as a result with “production smoothing”:
as emphasized previously mobile loans are designed to compensate short-term liquidity
needs and are not likely to be associated with long-term capital investment in stimulating
business expansion. In columns 4 and 6 of Table 11 we find evidence for this argument.
Our treatment resulted with a reduction in the volatility of sales for small businesses,
where sales volatility is proxied by ln(Salesmax) − ln(Salesmin) with ln(Salesmax) and
ln(Salesmin) measuring the maximum and the minimum amount of the natural logarithm
of sales, respectively, that the business experienced during particular months over the
past 12 months. The volatility reduction that we capture is statistically significant at the
level of 10% and the coefficient estimate in column 4 corresponds to as high as 55% of
the sales volatility observed in the control group. This reduction in sales volatility is also
quantitatively in line with the expansion in financial access allowed through mobile loans.39
The sales smoothing induced by our Lipa Na M-Pesa treatment complements the findings
by Jack and Suri (2014), who show that the standard M-Pesa product smooths household
consumption by allowing easy access to liquidity. Our research shows that an analog
38We would like to note that if the dummy variable “Small” is dropped on the RHS of the specificationpresented in the second column of Table 11, “Small x Treated” turns out to have a significant coefficientestimate.
39Columns 5 and 6 in Table 11 perform a robustness check, where we fill missing information in thebaseline survey on salesmax or salesmin using average monthly sales data. When compared against column4, in column 6 we observe that the economic significance through “Small x Treated” goes down, but thestatistical significance of the coefficient estimate improves.
28
property exists for SMEs, where especially for small-scale establishments Lipa Na M-Pesa
causes financial access and a reduced sales volatility.
Finally, in Table 12 we present the impact of our treatment on total investment and
the inventory investment component of total investment. Total investment, a potential
determinant of overall business growth in the long-run, is not significantly influenced by
our treatment, though both “Treatment” and “Small x Treated” have positive coefficient
estimates. Complementing our findings on reduced sales volatility, we find that in inventory
investment regression the coefficient estimate associated with “Small x Treated” is positive,
and borderline statistically significant at 10% level. Inventory investment primarily serves
the purpose of absorbing shocks and short-run smoothing of sales.40 Therefore, this nearly
significant treatment effect of our experimental intervention on small firms’ inventory in-
vestment is in line with our highlighted mechanism. In this respect, this finding provides
an insight for how improved financial access and reduced sales volatility outcomes that our
experimental intervention caused might relate to each other.
Table 12 here
6 Conclusion
We used a Randomized-Controlled-Trial experiment to estimate the factors affecting the
decision to adopt a novel P2B mobile-money payment technology for small- and medium-
sized merchants in Nairobi, and measured the impact of using such technology on business
outcomes. We found causal evidence that seemingly small situational barriers to adoption
are decisive for the adoption of this instrument. Providing information about the product
and eliminating the paperwork to open an account was enough to increase the interest to
adopt the technology by 62% in restaurants and by 21% in pharmacies.
As a highlighted result we showed that financial transparency concerns of the business
owner persistently affect adoption and usage of the payment technology. Sixteen months
after our intervention, we observed that firms that were induced to adopt the technology
40Eichenbaum (1988) and Dynan, Elmendorf and Sichel (2006) provide theoretical and empirical argu-ments for the role of inventory investment in absorbing production shocks and smoothing sales over thebusiness cycle.
29
feel safer and have more mobile loans. The impact on mobile-loan usage is especially
pronounced for small-size establishments, which also experience a reduction in their sales
volatility.
Our key results are generalizable to a wide range of settings, because we studied the
adoption of a mainstream product provided by a large intermediary connecting “payment”
and “lending” means of financial services spectrum. Also, we conducted our research with
actual SMEs and (in particular) those which are licensed and hence are taxable. Therefore,
our findings are highly relevant for researchers and policy makers across the globe interested
in promoting both cashless transformations and financial access.
Finally, the key mechanism behind our results revealed that transparency aversion is
likely to be an important barrier for the diffusion of high-tech payment products. Fi-
nancial regulators and designers of financial instruments might need to pay attention to
this novel empirical finding. Our recommendation to regulators and advocates of financial
products is to emphasize the benefits of financial transparency when promoting electronic
payment instruments to SMEs - and especially to those which are expected to suffer the
most from opacity: as our impact analysis revealed, becoming transparent to a financial
service provider can be associated with better access to external finance and eventually with
real outcome effects, especially for small businesses which otherwise might be hindered by
barriers to visibility and hard information constraints.
30
References
[1] Arango, C. A., Huynh, K. P., and L. Sabetti, 2011. “How do you pay? The roleof incentives at the point-of-sale”, ECB Working Paper No. 1386.
[2] Arifovic, J., J. Duffy, and J.H. Jiang, 2017. “Adoption of a new payment method:theory and experimental evidence”, Bank of Canada Working Paper.
[3] Bertrand, M., Mullainathan, S. and E. Shafir, 2004. “A behavioral-economicsview of poverty”, American Economic Review, 94-2, 419-423.
[4] Beck, T., H. Pamuk, R. Ramrattan, and B.R. Uras, 2018. “Payment Instru-ments, Finance, and Development”, Journal of Development Economics, 133, 162-186.
[5] Beck, T., Degryse, H., De Haas, R., and N. Van Horen, N., 2018. “When arm’slength is too far: Relationship banking over the credit cycle”. Journal of FinancialEconomics, 127(1), 174-196.
[6] Berger, A. N., and G. F. Udell, 1995. “Relationship lending and lines of credit insmall firm finance.” Journal of Business, 351-381.
[7] Bolt, W., N. Jonker, and C. Van Renselaar, 2010. “Incentives at the counter:An empirical analysis of surcharging card payments and payment behaviour in theNetherlands”, Journal of Banking and Finance, 34(8), 1738-1744.
[8] Bradlow, E.T., 1998. Encouragement Designs: An Approach to Self-Selected Samplesin an Experimental Design. Marketing Letters 9(4), 383-391.
[9] Burkart, M., and T. Ellingsen, 2004. “In-Kind Finance: A Theory of Trade Credit”,American Economic Review, 94 (3): 569-590.
[10] Burkart, M., Ellingsen, T., and M. Giannetti, 2011. “What you sell is what youlend? Explaining trade credit contracts”, Review of Financial Studies, 24(4), 1261-1298.
[11] Camera, G., M. Casari, and S. Bortolotti, 2016. “An experiment on retail pay-ments systems”, Journal of Money, Credit and Banking, 48(2-3), 363-392.
[12] Caselli, F. and W.J. Coleman II, 2001. “The US structural transformation andregional convergence: A reinterpretation”, Journal of Political Economy, 109(3), 584-616.
[13] Chakravorti, S., and T. To, 2007. “A theory of credit cards”, International Journalof Industrial Organization, 25(3), 583-595.
[14] Comin, D. and B. Hobijn, 2004. “Cross-country technology adoption: making thetheories face the facts”, Journal of Monetary Economics, 51(1), 39-83.
[15] Communications Authority of Kenya, 2015. “Fourth quarter sector statisticsreport for the financial year 2014-2015”, retrieved from http://ca.go.ke/index.php/
statistics.
[16] Degryse, H., and P. V. Cayseele, 2000. “Relationship lending within a bank-based system: Evidence from European small business data.” Journal of FinancialIntermediation, 9, 1, 90-109.
31
[17] Demirguc-Kunt, A., and Maksimovic, V., 2001. “Firms as financial intermedi-aries: Evidence from trade credit data”. Policy Research Working Paper, No. 2696.World Bank.
[18] Drexler, A., Fisher, G. and A. Schoar, 2014. “Keeping It Simple: FinancialLiteracy and Rules of Thumb”, American Economic Journal: Applied Economics, vol.6, No. 2, pp. 1-31.
[19] Duflo, E., M. Kremer and J. Robinson, 2004. “Understanding technology adop-tion: Fertilizer in Western Kenya, preliminary results from field experiments”, Unpub-lished manuscript, Massachusetts Institute of Technology.
[20] Duflo, E., M. Kremer and J. Robinson, 2008. “How High are Rates of Return toFertilizer? Evidence from Field Experiments in Kenya”, American Economic Review,P&P, 98, 2, 482-488.
[21] Duflo, E., M. Kremer and J. Robinson, 2011. “Nudging Farmers to Use Fertilizer:Theory and Experimental Evidence from Kenya”, American Economic Review, 101, 6,2350-90.
[22] Dupas, P., and J. Robinson, 2013. “Savings constraints and microenterprise de-velopment: Evidence from a field experiment in Kenya”, American Economic Journal:Applied Economics, 5(1), 163-92.
[23] Dynan, K. E., Elmendorf, D. W., and D.E. Sichel, 2006. “Can financial innova-tion help to explain the reduced volatility of economic activity?”. Journal of MonetaryEconomics, 53(1), 123-150.
[24] Economides, N., and P. Jeziorski, 2017. “Mobile money in Tanzania”, MarketingScience, 36(6), 815-837.
[25] Eichenbaum, M. S., 1988. “Some empirical evidence on the production level andproduction cost smoothing models of inventory investment”.
[26] Finaccess, 2014. FinAccess Business Survey
[27] Foster, A. and M.R. Rosenzweig, 2010. “Microeconomics of Technology Adop-tion”, Annual Review of Economics, 2:394-424.
[28] Hanna, R., Mullainathan, S., S. Joshua, 2014. “Learning through noticing: The-ory and evidence from a field experiment”, The Quarterly Journal of Economics 129(3), 13111353.
[29] Humphrey, D. B., L.B. Pulley and J.M. Vesala, 1996. “Cash, paper, and elec-tronic payments: a cross-country analysis”, Journal of Money, Credit and Banking,28(4), 914-93
[30] Jack, W., A. Ray and T. Suri, 2013. “Transaction Networks: Evidence from MobileMoney in Kenya”, American Economic Review, 103, 3, 356-61.
[31] Jack, W. and T. Suri, 2011. “Mobile Money: The Economics of M-PESA”, NBERWorking Paper No. 16721.
32
[32] Jack, W. and T. Suri, 2014. “Mobile Risk sharing and transactions costs: Evidencefrom Kenya’s mobile money revolution”, The American Economic Review 104, 1, 183-223.
[33] Klee, E., 2008. “How people pay: Evidence from grocery store data”, Journal ofMonetary Economics 55(3), 526-541.
[34] Kremer, M. and E. Miguel, 2007. “The illusion of sustainability”, The QuarterlyJournal of Economics, 122(3), 1007-1065.
[35] La Ferrara, E., 2016. “Mass Media and Social Change: Can We Use Television toFight Poverty?” Journal of the European Economic Association, 14, 4(1), Pages 791827.
[36] La Porta, R., Lopez-de-Silanes, F., Shleifer, A., and R.W. Vishny, 1997.“Legal determinants of external finance”. The Journal of Finance, 52(3), 1131-1150.
[37] La Porta, R., Lopez-de-Silanes, F., Shleifer, A., and R.W. Vishny, 1998.“Law and finance”. Journal of Political Economy, 106(6), 1113-1155.
[38] Petersen, M.A., and R.G. Rajan, 1994. “The benefits of lending relationships:Evidence from small business data.” The Journal of Finance 49, no. 1: 3-37.
[39] Mbiti, I., and D.N. Weil, 2011. “Mobile banking: The impact of M-Pesa in Kenya(No. w17129)”, National Bureau of Economic Research.
[40] Schuh, S. and J. Stavins, 2010. “Why are (some) consumers (finally) writing fewerchecks? The role of payment characteristics”, Journal of Banking and Finance, 34(8),1745-1758.
[41] Suri, T. and W. Jack, 2016. “The long-run poverty and gender impacts of mobilemoney”, Science, 354, 6317, 1288-1292.
[42] Wakamori, N., and Welte, A., 2017. “Why do shoppers use cash? Evidence fromshopping diary data”, Journal of Money, Credit and Banking, 49(1), 115-169.
33
Figure 1: Lipa Na M-Pesa vs. standard personal M-Pesa transfer fees
050
100
150
200
mpe
sa tr
ansf
er fe
e, K
enya
n S
hilli
ngs
10 5000 8500 15000 20000amount of transfer, Kenyan Shillings
Personal Mpesa Lipa Na M-pesa
Notes: The figure depicts the transfer fees charged when transacting with Lipa Na M-Pesa andstandard M-Pesa. The red line shows the fee that Safaricom deducts from a merchant (y-axis)for the corresponding transfer amount made by a customer (x-axis) through Lipa Na M-Pesa(the marginal cost of transaction is constant at 1% of the payment). The blue line shows the feethat Safaricom deducts from a customer (y-axis) for the corresponding amount of transfer madefrom a customer to the merchant (x-axis) through a standard M-Pesa account (the marginalcost of transaction is a step function).
34
Figure 2: M-Pesa and Lipa Na M-Pesa withdrawal fees
050
100
150
200
with
draw
al fe
e, K
enya
n S
hilli
ngs
10 2500 5000 10000 15000 20000amount of withdrawal , Kenyan Shillings
Notes: The blue line depicts cash withdrawal fees (when converting M-Pesa or Lipa Na M-Pesaunits into cash). Same fees apply for M-Pesa and Lipa Na M-Pesa.
35
Figure 3: Geographic distribution of restaurants at baseline
Notes: The figure shows the geographic distribution of restaurants in treatment (blue) andcontrol (red) groups at baseline.
36
Figure 4: Geographic distribution of pharmacies at baseline
Notes: The figure shows the geographic distribution of pharmacies in treatment (blue) andcontrol (red) groups at baseline.
37
Figure 5: Geographic distribution of a random sub-sample of merchants at baseline
Notes: The figure shows the geographic distribution of a sub-sample of merchants in treatment(blue) and control (red) groups at baseline.
38
Figure 6: Stated reasons at baseline for not adopting Lipa Na: by sector
0.1
.2.3
Too co
mplex t
o use
Too co
stly t
o ope
n an a
ccou
nt
Not se
eing t
he be
nefits
of Li
pa N
a M-P
esa
Wou
ld no
t incre
ase m
y sale
s
Don't h
ave t
ime t
o ope
n an a
ccou
nt
No trus
t in m
obile
mon
ey pr
ovide
r
High tra
nsac
tion f
ees
Pharmacy Restaurant
95 percent confidence interval
Notes: The figure shows the fraction of businesses (restaurants and pharmacies), who provideda particular reason for not having a Lipa Na M-Pesa account before our study. We also report95% statistical confidence levels for each bar. Not seeing the benefits of Lipa Na M-Pesa, Toocostly to open an account, High transaction fees via Lipa Na M-pesa , Don’t have time to openan account, Would not increase my sales, No trust in mobile money provider, Too complex touse are reasons of not adopting the Lipa Na M-pesa before our experiment. These variablesequal 1 if the business stated the corresponding reason, 0 otherwise.
39
Figure 7: Stated reasons at baseline for not having adopting Lipa Na prior to the interven-tion, by Restaurants who accepted or rejected our offer to open an account
0.1
.2.3
.4
Too co
mplex t
o use
Too co
stly t
o ope
n an a
ccou
nt
Not se
eing t
he be
nefits
of Li
pa N
a M-P
esa
Wou
ld no
t incre
ase m
y sale
s
Don't h
ave t
ime t
o ope
n an a
ccou
nt
No trus
t in m
obile
mon
ey pr
ovide
r
High tra
nsac
tion f
ees
Wants to open account Does not want to open account
95 percent confidence interval
Notes: The figure compares the fraction of adopters against non-adopters after the treatment- with a particular reason for not having a Lipa Na M-Pesa account before our study. We alsoreport 95% statistical confidence levels for each bar.
40
Figure 8: Stated reasons at baseline for not having adopting Lipa Na prior to the interven-tion, by Pharmacies who accepted or rejected our offer to open an account
0.1
.2.3
.4
Too co
mplex t
o use
Too co
stly t
o ope
n an a
ccou
nt
Not se
eing t
he be
nefits
of Li
pa N
a M-P
esa
Wou
ld no
t incre
ase m
y sale
s
Don't h
ave t
ime t
o ope
n an a
ccou
nt
No trus
t in m
obile
mon
ey pr
ovide
r
High tra
nsac
tion f
ees
Wants to open account Does not want to open account
95 percent confidence interval
Notes: The figure compares the fraction adopters against non-adopters after the treatment -with a particular reason for not having a Lipa Na M-Pesa account before our study. We alsoreport 95% statistical confidence levels for each bar.
41
Restaurant
May 2015-June 2015 • ListingJuly 2015-August 2015 • Baseline survey and offers to open Lipa accountsSeptember 2015-February 2016 • Lipa account openingMarch 2017-May 2017 • Endline Survey
Pharmacy
August 2015 • ListingSeptember 2015-November 2015 • Baseline survey and offers to open Lipa accountsOctober 2015-February 2016 • Lipa account openingMarch 2017-May 2017 • Endline Survey
Timeline: Experiment and survey timeline.
42
Tab
le1:
Bas
elin
ebusi
nes
sch
arac
teri
stic
san
dbal
ance
test
Ph
arm
acy
Res
tau
rant
All
Con
tT
reat
Diff
All
Con
tT
reat
Diff
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Sta
ndard
mobi
lem
on
eyu
seU
sin
gm
ob
.m
oney
for
bu
sin
ess
pu
rpos
es(0
/1)
0.41
0.40
0.43
-0.0
30.
590.
570.
62-0
.05
Usi
ng
mob
.m
oney
tore
ceiv
ep
aym
ents
(0/1
)0.
250.
210.
28-0
.07*
0.40
0.38
0.42
-0.0
4U
sin
gm
ob
.m
oney
tost
ore
mon
ey(0
/1)
0.09
0.09
0.10
0.00
0.25
0.26
0.24
0.02
Usi
ng
mob
.m
oney
top
ayb
ills
(0/1)
0.26
0.26
0.27
-0.0
10.
360.
370.
350.
02U
sin
gm
ob
.m
oney
top
aysa
lari
es(0
/1)
0.04
0.02
0.06
-0.0
4**
0.07
0.07
0.07
0.00
Usi
ng
mob
.m
oney
top
ayin
pu
ts(0
/1)
0.31
0.32
0.29
0.03
0.43
0.41
0.45
-0.0
4A
ware
nes
sof
Lip
aan
dre
aso
ns
of
not
adopti
ng
Aw
are
of
Lip
aN
aM
-Pes
a(0
/1)
0.94
0.95
0.93
0.02
0.96
0.95
0.97
-0.0
2H
ave
Lip
aN
aM
-Pes
a(0
/1)
0.05
0.01
0.08
-0.0
6***
0.12
0.12
0.12
0.00
Not
seei
ng
the
ben
efits
ofL
ipa
Na
M-P
esa
(0/1
)0.
270.
270.
270.
000.
250.
270.
240.
03T
oo
cost
lyto
open
an
acco
unt
(0/1
)0.
170.
180.
160.
020.
050.
030.
06-0
.02
Hig
htr
an
sact
ion
fees
via
Lip
aN
aM
-pes
a(0
/1)
0.25
0.25
0.24
0.01
0.09
0.08
0.09
-0.0
1D
on
’th
ave
tim
eto
open
anacc
ount
(0/1
)0.
090.
110.
070.
040.
140.
140.
15-0
.01
Wou
ldn
otin
crea
sem
ysa
les
(0/1
)0.
090.
080.
11-0
.02
0.06
0.06
0.06
0.00
No
tru
stin
mob
ile
mon
eypro
vid
er(0
/1)
0.03
0.03
0.04
-0.0
10.
020.
020.
010.
00T
oo
com
ple
xto
use
(0/1
)0.
080.
080.
080.
000.
110.
090.
13-0
.03
Bu
sin
ess
size
loga
rith
m(S
ale
s,m
onth
ly(1
000
Ksh
.),
win
sori
zed
5%)
4.67
4.59
4.74
-0.1
6**
5.30
5.29
5.31
-0.0
1lo
gari
thm
(Pro
fits
,m
onth
ly,
(1000
Ksh
.),
win
sori
zed
5%)
3.74
3.65
3.83
-0.1
9**
3.93
3.93
3.94
-0.0
1lo
gari
thm
(Em
plo
yee
s)1.
181.
171.
20-0
.02
1.65
1.66
1.64
0.01
Inve
stm
ent
an
dacc
ess
tofi
nan
ceIn
vest
men
t(0
/1),
inth
ep
ast
6m
onth
s0.
190.
190.
20-0
.01
0.37
0.39
0.35
0.04
Ban
klo
an
(0/1
),in
the
pas
t12
month
s0.
070.
070.
060.
010.
120.
100.
14-0
.04
Info
rmal
loan
(0/1
),in
the
past
12
mon
ths
0.03
0.03
0.02
0.00
0.04
0.04
0.05
-0.0
1M
obil
elo
an
(0/1
)in
the
past
12
mon
ths
0.08
0.07
0.09
-0.0
20.
120.
120.
120.
00In
form
ali
tyB
usi
nes
sli
cen
se(0
/1)
0.92
0.93
0.91
0.02
0.56
0.55
0.57
-0.0
1
Notes:
Th
ista
ble
des
crib
esth
etr
eatm
ent
an
dco
ntr
ol
gro
up
mea
ns
for
ph
arm
aci
esan
dre
stau
rants
.V
ari
ab
led
escr
ipti
on
s:S
ale
sis
the
bu
sin
ess
tota
lre
ven
ues
inth
ep
ast
month
.P
rofi
tsis
the
tota
lin
com
eth
eb
usi
nes
sea
rned
du
rin
gth
ep
ast
month
aft
erp
ayin
gall
exp
ense
s.E
mp
loyee
sis
the
tota
lnu
mb
erof
emp
loyee
sp
lus
the
ow
ner
.In
ves
tmen
tis
the
tota
lca
pit
al
inves
tmen
tfo
rb
usi
nes
sp
urp
ose
s.b
ank
equ
als
1(0
oth
erw
ise)
ifth
eb
usi
nes
sev
erre
ceiv
eda
new
loan
from
aco
mm
erci
al
ban
k,
SA
CC
O,
or
oth
erfo
rmal
fin
an
cial
inst
itu
tion
over
the
last
12
month
s.In
form
al
loan
equ
als
1(0
oth
erw
ise)
ifth
eth
eb
usi
nes
sor
bu
sin
ess
ow
ner
sb
orr
ow
edm
on
ey(n
ewlo
an
)fo
rb
usi
nes
sp
urp
ose
sfr
om
any
bu
sin
ess
ass
oci
ati
on
,m
on
eyle
nd
er,
fam
ily
or
frie
nd
over
the
last
12
month
s.M
ob
ile
loan
equ
als
1(0
oth
erw
ise)
ifth
eb
usi
nes
sor
bu
sin
ess
ow
ner
sb
orr
ow
edm
on
ey(n
ewlo
an
)fo
rM
ob
ile
Mic
rofi
nan
ceso
urc
eslike
KC
B-M
pes
a,
M-K
esh
a,
M-S
hw
ari
over
the
last
12
month
s.A
ware
of
Lip
aN
aM
-Pes
aeq
uals
1(0
oth
erw
ise)
ifth
eb
usi
nes
sis
aw
are
of
Lip
aN
aM
-Pes
a.
Bu
sin
ess
lice
nse
equ
als
1if
the
bu
sin
ess
has
an
up
tod
ate
bu
sin
ess
lice
nse
from
an
au
thori
ty.
Usi
ng
mob
.m
on
eyfo
rb
usi
nes
sp
urp
ose
s(t
ore
ceiv
ep
aym
ent/
store
mon
ey/to
pay
bills
/to
pay
sala
ries
/to
pay
inp
uts
)eq
uals
1(0
oth
erw
ise)
.N
ot
seei
ng
the
ben
efits
of
Lip
aN
aM
-Pes
a,
Too
cost
lyto
op
enan
acc
ou
nt,
Hig
htr
an
sact
ion
fees
via
Lip
aN
aM
-pes
a,
Don
’th
ave
tim
eto
op
enan
acc
ou
nt,
Wou
ldn
ot
incr
ease
my
sale
s,N
otr
ust
inm
ob
ile
mon
eyp
rovid
er,
Too
com
ple
xto
use
are
reaso
ns
of
not
ad
op
tin
gth
eL
ipa
Na
M-p
esa.
Th
eyeq
ual
1(0
oth
erw
ise)
ifth
eb
usi
nes
sst
ate
dit
isa
reaso
nn
ot
toad
op
tL
ipa
Na
M-p
esa
bef
ore
the
trea
tmen
t.p<
0.1
.**
p<
0.0
5.
***
p<
0.0
1
43
Tab
le2:
Busi
nes
sch
arac
teri
stic
sin
the
trea
tmen
tgr
oup
by
willingn
ess
toad
opt
Lip
aN
aM
-pes
a
Ph
arm
acy
Res
tau
rant
Does
wan
tto
Wan
tsto
Diff
,D
oes
wan
tto
Wan
tsto
Diff
.op
enac
cou
nt
open
acco
unt
open
acco
unt
open
acco
unt
Per
son
al
mobi
lem
on
eyu
sefo
rbu
sin
ess
pu
rpose
sU
sin
gm
ob
.m
oney
for
bu
sin
ess
pu
rpos
es(0
/1)
0.41
0.48
-0.0
70.
520.
68-0
.17*
**U
sin
gm
ob
.m
oney
tore
ceiv
ep
aym
ents
(0/1
)0.
260.
35-0
.09
0.33
0.48
-0.1
5***
Sh
are
ofm
obil
em
on
eycu
stom
ers
0.02
0.04
-0.0
20.
010.
03-0
.02*
**U
sin
gm
ob
.m
oney
tost
ore
mon
ey(0
/1)
0.08
0.15
-0.0
70.
170.
29-0
.12*
**U
sin
gm
ob
.m
oney
top
ayb
ills
(0/1)
0.27
0.28
-0.0
10.
310.
38-0
.07
Usi
ng
mob
.m
oney
top
aysa
lari
es(0
/1)
0.06
0.07
-0.0
20.
060.
07-0
.01
Usi
ng
mob
.m
oney
top
ayin
pu
ts(0
/1)
0.28
0.35
-0.0
70.
350.
52-0
.17*
**A
ware
nes
sof
Lip
aN
aM
-Pes
aan
dre
aso
ns
of
not
adopti
ng
Aw
are
of
Lip
aN
aM
-Pes
a(0
/1)
0.93
0.96
-0.0
40.
980.
970.
01N
ot
seei
ng
the
ben
efits
ofL
ipa
Na
M-P
esa
(0/1
)0.
300.
130.
16**
0.31
0.18
0.13
***
Too
cost
lyto
open
an
acco
unt
(0/1
)0.
190.
080.
11*
0.05
0.05
0.00
Hig
htr
an
sact
ion
fees
via
Lip
aN
aM
-pes
a(0
/1)
0.27
0.13
0.14
**0.
100.
090.
00D
on
’th
ave
tim
eto
open
anacc
ount
(0/1
)0.
060.
13-0
.08*
*0.
070.
21-0
.14*
**W
ould
not
incr
ease
my
sale
s(0
/1)
0.13
0.06
0.07
0.11
0.03
0.08
***
No
tru
stin
mob
ile
mon
eypro
vid
er(0
/1)
0.05
0.00
0.05
*0.
030.
010.
02T
oo
com
ple
xto
use
(0/1
)0.
080.
12-0
.04
0.15
0.12
0.03
Bu
sin
ess
size
loga
rith
m(S
ale
s,m
onth
ly(1
000
Ksh
.),
win
sori
zed
5%)
4.75
4.75
0.00
5.30
5.34
-0.0
4lo
gari
thm
(Pro
fits
,m
onth
ly,
(1000
Ksh
.),
win
sori
zed
5%)
3.81
3.90
-0.0
83.
953.
940.
01lo
gari
thm
(Em
plo
yee
s)1.
181.
23-0
.04
1.67
1.62
0.06
Inve
stm
ent
an
dacc
ess
tofi
nan
ceIn
vest
men
t(0
/1),
inth
ep
ast
6m
onth
s0.
190.
24-0
.06
0.28
0.39
-0.1
1*B
ank
loan
(0/1
),in
the
pas
t12
month
s0.
060.
060.
010.
130.
16-0
.03
Info
rmal
loan
(0/1
),in
the
past
12
mon
ths
0.02
0.02
0.00
0.06
0.04
0.02
Mob
ile
loan
(0/1
)in
the
pas
t12
month
s0.
090.
080.
010.
110.
14-0
.02
Info
rmali
tyB
usi
nes
sli
cen
se(0
/1)
0.92
0.86
0.06
0.52
0.60
-0.0
7
Notes:
Th
ista
ble
com
pare
sth
ed
escr
ipti
ve
chara
cter
isti
csof
ou
rsa
mp
lefo
rp
harm
aci
esan
dre
stau
rants
by
willin
gn
ess
toad
op
tth
eL
ipa
Na
M-p
esa
tech
nolo
gy
inou
rex
per
imen
t.W
eon
lyu
setr
eatm
ent
gro
up
bu
sin
esse
san
db
usi
nes
ses
that
do
not
have
aL
ipa
Na
M-P
esa
acc
ou
nt.
We
rep
ort
mea
nvalu
esfo
rea
chch
ara
cter
isti
c.V
ari
ab
led
escr
ipti
on
s:S
ale
sis
the
bu
sin
ess
tota
lre
ven
ues
inth
ep
ast
month
.P
rofi
tsis
the
tota
lin
com
eth
eb
usi
nes
sea
rned
du
rin
gth
ep
ast
month
aft
erp
ayin
gall
exp
ense
s.E
mp
loyee
sis
the
tota
lnu
mb
erof
emp
loyee
sp
lus
the
ow
ner
.In
ves
tmen
tis
the
tota
lca
pit
al
inves
tmen
tfo
rb
usi
nes
sp
urp
ose
s.b
an
keq
uals
1(0
oth
erw
ise)
ifth
eb
usi
nes
sev
erre
ceiv
eda
new
loan
from
aco
mm
erci
al
ban
k,
SA
CC
O,
or
oth
erfo
rmal
fin
an
cial
inst
itu
tion
over
the
last
12
month
s.In
form
al
loan
equ
als
1(0
oth
erw
ise)
ifth
eth
eb
usi
nes
sor
bu
sin
ess
ow
ner
sb
orr
ow
edm
on
ey(n
ewlo
an
)fo
rb
usi
nes
sp
urp
ose
sfr
om
any
bu
sin
ess
ass
oci
ati
on,
mon
eyle
nd
er,
fam
ily
or
frie
nd
over
the
last
12
month
s.M
ob
ile
loan
equ
als
1(0
oth
erw
ise)
ifth
eb
usi
nes
sor
bu
sin
ess
ow
ner
sb
orr
ow
edm
on
ey(n
ewlo
an
)fo
rM
ob
ile
Mic
rofi
nan
ceso
urc
eslike
KC
B-M
pes
a,
M-K
esh
a,
M-S
hw
ari
over
the
last
12
month
s.A
ware
of
Lip
aN
aM
-Pes
aeq
uals
1(0
oth
erw
ise)
ifth
eb
usi
nes
sis
aw
are
of
Lip
aN
aM
-Pes
a.
Bu
sin
ess
lice
nse
equ
als
1if
the
bu
sin
ess
has
an
up
tod
ate
bu
sin
ess
lice
nse
from
an
au
thori
ty.
Usi
ng
mob
.m
on
eyfo
rb
usi
nes
spu
rpose
s(t
ore
ceiv
ep
aym
ent/
store
mon
ey/to
pay
bills
/to
pay
sala
ries
/to
pay
inp
uts
)eq
uals
1(0
oth
erw
ise)
ifth
eb
usi
nes
su
sep
erso
nal
mob
ile
acc
ou
nt
for
bu
sin
ess
pu
rpose
s(t
ore
ceiv
ep
aym
ent/
store
mon
ey/to
pay
bills
/to
pay
sala
ries
/to
pay
inp
uts
).N
ot
seei
ng
the
ben
efits
of
Lip
aN
aM
-Pes
a,
Too
cost
lyto
op
enan
acc
ou
nt,
Hig
htr
an
sact
ion
fees
via
Lip
aN
aM
-pes
a,
Don
’th
ave
tim
eto
op
enan
acc
ou
nt,
Wou
ldn
ot
incr
ease
my
sale
s,N
otr
ust
inm
obil
em
on
eyp
rovid
er,
Too
com
ple
xto
use
aer
reaso
ns
of
not
ad
op
tin
gth
eL
ipa
Na
M-p
esa.
They
equ
al
1(0
oth
erw
ise)
ifth
eb
usi
nes
sst
ate
dit
isa
reaso
nn
ot
toad
op
tL
ipa
Na
M-p
esa.
p<
0.1
.**
p<
0.0
5.
***
p<
0.0
1
44
Tab
le3:
Reg
ress
sion
esti
mat
esfo
rw
illi
ngn
ess
toad
opt
Lip
aN
aM
-Pes
a:F
ull
trea
tmen
tsa
mp
le
Var
iab
le(1
)(2
)(3
)P
an
elA
:M
-Mon
eyE
xposu
reM
eth
od
sof
Use
for
bu
sin
ess
0.1
41***
(0.0
44)
mob
ile
mon
eyu
seR
ecei
vep
aym
ents
0.0
94*
(0.0
53)
Sto
rem
on
ey0.0
93
(0.0
64)
Pay
bil
l-0
.073
(0.0
57)
Pay
inp
ut
0.1
23**
(0.0
56)
Pay
sala
ries
-0.0
25
(0.0
92)
Sav
ing
inm
ob
.m
on
.acc
ou
nt
-0.0
48
(0.0
74)
%of
uti
lity
exp
.via
per
s.m
ob
.-0
.067
(0.0
56)
%of
inp
ut
exp
.via
per
s.m
ob
.0.2
37***
(0.0
78)
Pay
ing
wages
via
mp
esa
-0.0
30
(0.1
18)
No
incr
ease
inp
rice
s0.2
01***
(0.0
61)
Th
eft
Th
eft
an
dsa
fety
0.0
28
(0.0
24)
Inte
rnal
thef
t0.0
12
(0.0
54)
-0.0
07
(0.0
62)
Exte
rnal
thef
t,fire
,et
c.0.0
40
(0.0
79)
0.0
82
(0.0
80)
Fee
lin
gn
ot
safe
0.0
08
(0.0
12)
0.0
09
(0.0
14)
Bu
sin
ess
savin
gS
avin
gat
ab
an
kor
mic
ro.
-0.0
32
(0.0
48)
acco
unts
Sav
ing
at
ap
ers.
ban
kacc
.0.0
29
(0.0
48)
0.0
34
(0.0
54)
Sav
ing
at
ab
us.
ban
kacc
.-0
.189***
(0.0
68)
-0.1
95***
(0.0
75)
Sav
ing
at
am
icro
f.in
st.
-0.0
31
(0.1
22)
0.0
36
(0.1
37)
Pan
elB
:T
ran
spare
ncy
Not
rep
orti
ng
Not
share
dsa
les
-0.1
34*
(0.0
68)
-0.1
50**
(0.0
69)
bu
sin
ess
acti
vit
yN
otsh
are
dp
rofi
ts-0
.214***
(0.0
74)
Bu
sin
ess
lice
nse
Bu
sin
ess
lice
nse
0.0
42
(0.0
54)
0.0
34
(0.0
58)
0.0
89
(0.0
67)
Fin
anci
alF
inan
cial
Soph
isti
cati
on
0.0
21
(0.0
20)
sop
his
tica
tion
Ban
klo
an
0.0
54
(0.0
61)
-0.0
26
(0.0
72)
Mob
ile
loan
0.0
22
(0.0
75)
0.0
60
(0.0
80)
Bu
sin
ess
reco
rds
0.0
83
(0.0
64)
0.0
70
(0.0
72)
Sal
escr
edit
Sel
lson
cred
itto
cust
.0.0
58
(0.0
42)
0.0
48
(0.0
41)
0.0
57
(0.0
46)
Pan
elC
:B
ehavi
ora
lP
rese
nt
and
Pre
sent
bia
s0.0
25
(0.0
72)
0.0
54
(0.0
69)
0.0
46
(0.0
78)
Fact
ors
futu
reb
ias
Fu
ture
bia
s-0
.021
(0.0
54)
-0.0
00
(0.0
55)
-0.0
23
(0.0
57)
Cog
nit
ive
cap
acit
y#
of
dig
its
rem
emb
ered
0.0
30*
(0.0
17)
0.0
28*
(0.0
17)
0.0
39**
(0.0
20)
Tru
stT
rust
infi
rst
tim
e-0
.024
(0.0
25)
-0.0
37
(0.0
25)
-0.0
16
(0.0
29)
Tru
stin
cust
om
ers
0.0
18
(0.0
35)
0.0
20
(0.0
34)
0.0
09
(0.0
37)
Tru
stin
cou
rts
-0.0
44*
(0.0
26)
-0.0
35
(0.0
28)
-0.0
49*
(0.0
29)
Tru
stin
mob
.m
on
.co
mp
.0.0
42
(0.0
35)
0.0
42
(0.0
35)
0.0
29
(0.0
38)
Pan
elD
:B
usi
nes
ssi
zelo
g(E
mp
loye
es)
0.0
48
(0.0
69)
0.0
97
(0.0
69)
log(
Sale
s,m
onth
ly,
win
.5%
)0.0
37
(0.0
30)
0.0
56*
(0.0
29)
log(
Pro
fits
,m
onth
ly,
win
.5%
)0.0
24
(0.0
31)
Enu
mer
ator
and
dis
tric
tF
EY
esY
esY
esO
bse
rvat
ion
s493
490
394
R-s
qu
ared
0.2
85
0.3
16
0.3
51
Notes:
This
table
show
sth
eest
imati
on
resu
ltfo
rth
ere
lati
onsh
ipb
etw
een
wil
lingness
toadopt
Lip
aN
aM
-pesa
,and
valu
ati
on,
vis
ibil
ity,
behavio
ral
facto
rs,
busi
ness
size
for
all
sam
ple
.W
eest
imateYi
=β0
+X
′ iβ1
+εi
thro
ugh
OL
Sfo
rall
specifi
cati
ons
wherei
denote
sth
ebusi
ness
andX
iis
the
vecto
rin
clu
din
gvari
able
sdesc
rib
ed
inT
able
sO
A1
and
OA
2.
We
rep
ort
coeffi
cie
nt
est
imate
sfo
rβ1
and
robust
standard
err
ors
inpare
nth
ese
s.T
ocontr
ol
for
unobse
rved
regio
nal,
enum
era
tor
facto
rs,
and
merc
hant
typ
e,
we
add
enum
era
tor
and
dis
tric
tand
merc
hant
fixed
eff
ects
toall
est
imati
ons.
*p<
0.1
.**
p<
0.0
5.
***
p<
0.0
1.
45
Tab
le4:
Reg
ress
sion
esti
mat
esfo
rw
illi
ngn
ess
toad
opt
Lip
aN
aM
-Pes
a:T
reat
edR
esta
ura
nts
Vari
ab
le(1
)(2
)(3
)P
an
elA
:M
-Mon
eyE
xposu
reM
eth
od
sof
Use
for
bu
sin
ess
0.1
63**
(0.0
65)
mob
ile
mon
eyu
seR
ecei
vep
aym
ents
0.0
50
(0.0
74)
Sto
rem
on
ey0.0
38
(0.0
82)
Pay
bil
l-0
.038
(0.0
74)
Pay
inp
ut
0.1
98**
(0.0
79)
Pay
sala
ries
-0.0
69
(0.1
22)
Sav
ing
inm
ob
.m
on
.acc
ou
nt
-0.0
53
(0.0
95)
%of
uti
lity
exp
.via
per
s.m
ob
.0.0
29
(0.0
77)
%of
inp
ut
exp
.via
per
s.m
ob
.0.2
41**
(0.0
99)
Pay
ing
wages
via
mp
esa
-0.0
54
(0.1
84)
No
incr
ease
inp
rice
s0.1
54**
(0.0
78)
Th
eft
Th
eft
an
dsa
fety
0.0
29
(0.0
31)
Inte
rnal
thef
t0.0
29
(0.0
62)
0.0
12
(0.0
77)
Exte
rnal
thef
t,fire
,et
c.-0
.017
(0.0
95)
0.0
42
(0.0
99)
Fee
lin
gn
ot
safe
0.0
04
(0.0
17)
-0.0
00
(0.0
21)
Bu
sin
ess
savin
gS
avin
gat
ab
an
kor
mic
ro.
0.0
32
(0.0
67)
acco
unts
Sav
ing
at
ap
ers.
ban
kacc
.0.0
77
(0.0
70)
0.1
22
(0.0
81)
Sav
ing
at
ab
us.
ban
kacc
.-0
.121
(0.1
05)
-0.0
94
(0.1
20)
Sav
ing
at
am
icro
f.in
st.
-0.0
33
(0.1
22)
0.0
41
(0.1
73)
Pan
elB
:T
ran
spare
ncy
Not
rep
orti
ng
Not
share
dsa
les
-0.1
06
(0.1
28)
-0.0
83
(0.1
31)
bu
sin
ess
acti
vit
yN
ot
share
dp
rofi
ts-0
.164
(0.1
77)
Bu
sin
ess
lice
nse
Bu
sin
ess
lice
nse
0.0
17
(0.0
63)
0.0
11
(0.0
71)
0.0
64
(0.0
83)
Fin
anci
alF
inan
cial
Soph
isti
cati
on
0.0
22
(0.0
25)
sop
his
tica
tion
Ban
klo
an
0.0
17
(0.0
81)
-0.0
56
(0.1
02)
Mob
ile
loan
0.0
64
(0.0
95)
0.1
65
(0.1
16)
Bu
sin
ess
reco
rds
0.0
74
(0.0
80)
0.0
45
(0.0
93)
Sal
escr
edit
Sel
lson
cred
itto
cust
.0.0
93
(0.0
61)
0.0
99
(0.0
62)
0.0
82
(0.0
71)
Pan
elC
:B
ehavi
ora
lP
rese
nt
and
Pre
sent
bia
s-0
.020
(0.1
06)
0.0
15
(0.1
00)
0.0
49
(0.1
15)
Fact
ors
futu
reb
ias
Fu
ture
bia
s0.0
95
(0.0
77)
0.1
13
(0.0
79)
0.0
98
(0.0
92)
Cog
nit
ive
cap
acit
y#
of
dig
its
rem
emb
ered
0.0
21
(0.0
24)
0.0
25
(0.0
24)
0.0
34
(0.0
28)
Tru
stT
rust
infi
rst
tim
e-0
.048
(0.0
39)
-0.0
69*
(0.0
39)
-0.0
42
(0.0
48)
Tru
stin
cust
om
ers
0.0
72
(0.0
47)
0.0
61
(0.0
48)
0.0
40
(0.0
57)
Tru
stin
cou
rts
-0.0
60
(0.0
37)
-0.0
52
(0.0
39)
-0.0
63
(0.0
45)
Tru
stin
mob
.m
on
.co
mp
.-0
.015
(0.0
51)
-0.0
10
(0.0
54)
-0.0
36
(0.0
59)
Pan
elD
:B
usi
nes
ssi
zelo
g(E
mp
loye
es)
0.0
44
(0.0
85)
0.0
87
(0.0
86)
log(S
ale
s,m
onth
ly,
win
.5%
)0.0
62
(0.0
40)
0.0
57
(0.0
40)
log(P
rofi
ts,
month
ly,
win
.5%
)0.0
18
(0.0
41)
Enu
mer
ator
and
dis
tric
tF
EY
esY
esY
esO
bse
rvat
ion
s277
276
216
R-s
qu
ared
0.2
13
0.2
49
0.2
82
Notes:
This
table
show
sth
eest
imati
on
resu
ltfo
rth
ere
lati
onsh
ipb
etw
een
willingness
toadopt
Lip
aN
aM
-pesa
,and
valu
ati
on,
vis
ibilit
y,
behavio
ral
facto
rs,
busi
ness
size
for
rest
aura
nt
sam
ple
.W
eest
imateYi
=
β0
+X
′ iβ1
+εi
thro
ugh
OL
Sfo
rall
specifi
cati
ons
wherei
denote
sth
ebusi
ness
andX
iis
the
vecto
rin
clu
din
gvari
able
sdesc
rib
ed
inT
able
sT
able
sO
A1
and
OA
2.
We
rep
ort
coeffi
cie
nt
est
imate
sfo
rβ1
and
robust
standard
err
ors
inpare
nth
ese
s.T
ocontr
ol
for
unobse
rved
regio
nal,
enum
era
tor
facto
rsw
eadd
enum
era
tor
and
dis
tric
tfi
xed
eff
ects
toall
est
imati
ons.
*p<
0.1
.**
p<
0.0
5.
***
p<
0.0
1.
46
Tab
le5:
Reg
ress
sion
esti
mat
esfo
rw
illi
ngn
ess
toad
opt
Lip
aN
aM
-Pes
a:T
reat
edP
har
mac
ies
Var
iab
le(1
)(2
)(3
)P
an
elA
:M
-Mo
ney
Exp
osu
reM
eth
od
sof
Use
for
bu
sin
ess
0.1
15*
(0.0
59)
mob
ile
mon
eyu
seR
ecei
vep
aym
ents
0.1
32
(0.0
84)
Sto
rem
on
ey0.1
67
(0.1
44)
Pay
bil
l-0
.057
(0.0
94)
Pay
inp
ut
0.0
53
(0.0
91)
Pay
sala
ries
0.0
95
(0.1
74)
Sav
ing
inm
ob
.m
on
.acc
ou
nt
-0.2
29**
(0.0
88)
%of
uti
lity
exp
.via
per
s.m
ob
.-0
.155*
(0.0
82)
%of
inp
ut
exp
.via
per
s.m
ob
.0.1
61
(0.1
38)
Pay
ing
wages
via
mp
esa
0.0
73
(0.2
01)
No
incr
ease
inp
rice
s0.2
25**
(0.1
13)
Th
eft
Th
eft
an
dsa
fety
0.0
21
(0.0
43)
Inte
rnal
thef
t-0
.022
(0.1
24)
-0.0
78
(0.1
29)
Exte
rnal
thef
t,fire
,et
c.0.1
79
(0.1
71)
0.0
39
(0.1
50)
Fee
lin
gn
ot
safe
0.0
05
(0.0
20)
0.0
35
(0.0
23)
Bu
sin
ess
savin
gS
avin
gat
ab
an
kor
mic
ro.
-0.1
05
(0.0
75)
acco
unts
Sav
ing
at
ap
ers.
ban
kacc
.-0
.038
(0.0
80)
-0.0
40
(0.0
85)
Sav
ing
at
ab
us.
bank
acc
.-0
.237**
(0.0
91)
-0.1
92*
(0.1
06)
Sav
ing
at
am
icro
f.in
st.
-0.0
97
(0.2
12)
-0.0
99
(0.1
87)
Pan
elB
:T
ran
spare
ncy
Not
rep
orti
ng
Not
share
dsa
les
-0.1
73**
(0.0
83)
-0.2
21**
(0.0
89)
bu
sin
ess
acti
vit
yN
ot
share
dp
rofi
ts-0
.261***
(0.0
86)
Bu
sin
ess
lice
nse
Bu
sin
ess
lice
nse
-0.0
22
(0.1
34)
-0.0
16
(0.1
40)
0.0
80
(0.1
68)
Fin
anci
alF
inan
cial
Sop
his
tica
tion
0.0
34
(0.0
38)
sop
his
tica
tion
Ban
klo
an
0.0
98
(0.1
27)
0.0
28
(0.1
21)
Mob
ile
loan
0.0
17
(0.1
50)
-0.1
26
(0.1
23)
Bu
sin
ess
reco
rds
0.1
72**
(0.0
83)
0.1
99
(0.1
30)
Sal
escr
edit
Sel
lson
cred
itto
cust
.0.0
08
(0.0
61)
-0.0
08
(0.0
61)
0.0
38
(0.0
68)
Pan
elC
:B
ehavi
ora
lP
rese
nt
and
Pre
sent
bia
s0.0
88
(0.1
24)
0.0
82
(0.1
19)
0.0
65
(0.1
35)
Fact
ors
futu
reb
ias
Fu
ture
bia
s-0
.199***
(0.0
75)
-0.1
82**
(0.0
79)
-0.1
64**
(0.0
76)
Cog
nit
ive
cap
acit
y#
of
dig
its
rem
emb
ered
0.0
45
(0.0
28)
0.0
40
(0.0
28)
0.0
58*
(0.0
31)
Tru
stT
rust
infi
rst
tim
e-0
.014
(0.0
36)
-0.0
11
(0.0
38)
-0.0
10
(0.0
39)
Tru
stin
cust
om
ers
-0.0
73
(0.0
63)
-0.0
57
(0.0
62)
-0.0
72
(0.0
60)
Tru
stin
cou
rts
-0.0
36
(0.0
42)
-0.0
29
(0.0
44)
-0.0
58
(0.0
45)
Tru
stin
mob
.m
on
.co
mp
.0.1
00*
(0.0
51)
0.0
94*
(0.0
52)
0.1
06**
(0.0
53)
Pan
elD
:B
usi
nes
ssi
zelo
g(E
mp
loye
es)
0.1
34
(0.1
31)
0.1
82
(0.1
33)
log(S
ale
s,m
onth
ly,
win
.5%
)0.0
26
(0.0
50)
0.0
63
(0.0
51)
log(P
rofi
ts,
month
ly,
win
.5%
)0.0
64
(0.0
61)
Enu
mer
ator
and
dis
tric
tF
EY
esY
esY
esO
bse
rvat
ion
s216
214
178
R-s
qu
ared
0.1
68
0.2
18
0.2
88
Notes:
This
table
show
sth
eest
imati
on
resu
ltfo
rth
ere
lati
onsh
ipb
etw
een
willingness
toadopt
Lip
aN
aM
-pesa
,and
valu
ati
on,
vis
ibilit
y,
behavio
ral
facto
rs,
busi
ness
size
for
pharm
acy
sam
ple
.W
eest
imateYi
=
β0
+X
′ iβ1
+εi
thro
ugh
OL
Sfo
rall
specifi
cati
ons
wherei
denote
sth
ebusi
ness
andX
iis
the
vecto
rin
clu
din
gvari
able
sdesc
rib
ed
inT
able
sT
able
sO
A1
and
OA
2.
We
rep
ort
coeffi
cie
nt
est
imate
sfo
rβ1
and
robust
standard
err
ors
inpare
nth
ese
s.T
ocontr
ol
for
unobse
rved
regio
nal,
enum
era
tor
facto
rsw
eadd
enum
era
tor
and
dis
tric
tfi
xed
eff
ects
toall
est
imati
ons.
*p<
0.1
.**
p<
0.0
5.
***
p<
0.0
1.
47
Table 6: Relationship between business attrition in the endline and baseline business character-istics for all businesses with business license.
(1) (2)Baseline characteristics coef se
Treatment =1 if assigned to treatment group -0.010 (0.031)Pharmacy =1 if business is a pharmacy -0.144*** (0.043)Methods of Have Lipa Na M-pesa account -0.130*** (0.049)
Saving in mob. mon. account -0.033 (0.061)% of utility exp. via pers. mob. -0.027 (0.038)% of input exp. via pers. mob. -0.002 (0.022)Paying wages via mpesa 0.036 (0.034)
Theft and safety Internal theft -0.007 (0.044)External theft -0.117** (0.053)Feeling safe 0.002 (0.009)
Business saving Saving at a pers. bank acc. -0.058 (0.036)accounts Saving at a bus. bank acc. 0.013 (0.040)
Saving at a microf. inst. 0.030 (0.124)Financial Bank loan 0.003 (0.058)sophistication Mobile loan -0.100** (0.052)
Business records -0.101 (0.074)Present and Present bias -0.050 (0.050)future bias Future bias -0.068* (0.044)Cognitive capacity # of digits remembered -0.007 (0.012)Trust Trust in first time 0.030 (0.020)
Trust in customers 0.028 (0.027)Trust in courts 0.027 (0.018)Trust in mob. mon. comp. -0.033 (0.024)
Business size Employees -0.135*** (0.035)Constant 0.669*** (0.172)
Observations 855R-squared 0.075
Notes: This table shows the estimation result for the relationship between business attrition in the endline and baseline business characteris-
tics. we use the sample of businesses with business license in the baseline. We estimate Yi = β0 +X′iβ1 +εi through OLS for all specifications
where i denotes the business and Xi is the vector including variables listed in the first column. Yi equals 1 if the business did not participatein the baseline survey. We report coefficient estimates for β1 and robust standard errors in parentheses. * p<0.1. ** p<0.05. *** p<0.01.
48
Table 7A: Intention to treat (ITT) estimates for LPN usage
(1) (2) (3) (4)Opened LPN (0/1) Used LPN (0/1) Received payment via LPN (0/1) LPN sales, log(1+x)
Treatment 0.07** 0.08** 0.08** 0.26**(0.03) (0.03) (0.03) (0.11)
Control Mean 0.23 0.21 0.20 0.63Control StDev 0.42 0.40 0.40 1.43
N 619 618 618 618Notes: This table shows the ITT estimates for Lipa Na use indicators. Dependent variables are having Lipa Na M-pesa account (0/1), usingLipa Na M-pesa account for business in the past 30 days, receiving payment via Lipa Na M-pesa in the past 30 days, and Lipa Na M-pesasales (log(1+x)). Control Vector in each ITT regression: Baseline value of ln(sales-winsorized), not reporting sales in baseline, ln(baselineemployee #), use of LPN in baseline, enumerator FE, dummy variables for gender and business type and small business type. * p<0.1. **p<0.05. *** p<0.01.
Table 7B: Intention to treat (ITT) estimates for LPN usage for visible and non-transparent busi-nesses
(1) (2) (3) (4)Panel A: Transparent firms: shared sales figures in the baselineTreatment 0.07* 0.09** 0.09** 0.32**
(0.04) (0.04) (0.04) (0.12)
Contol Mean 0.24 0.22 0.21 0.65Contol StDev 0.43 0.22 0.22 1.45
N 486 485 485 485Panel B: Non-transparent firms: did not share sales figures in the baselineTreatment 0.00 0.01 0.01 0.02
(0.08) (0.08) (0.08) (0.24)
Control Mean 0.19 0.18 0.18 0.58Contol StDev 0.40 0.39 0.39 1.35
N 131 131 131 131Notes: This table shows the ITT estimates for Lipa Na use indicators. Dependent variables are having Lipa Na M-pesa account (0/1), usingLipa Na M-pesa account for business in the past 30 days, receiving payment via Lipa Na M-pesa in the past 30 days, and Lipa Na M-pesasales (log(1+x)). Control Vector in each ITT regression: Baseline value of ln(sales-winsorized), not reporting sales in baseline, ln(baselineemployee #), use of LPN in baseline, enumerator FE, dummy variables for gender and business type and small business type. * p<0.1. **p<0.05. *** p<0.01.
49
Table 8A: ITT estimates for business safety
Feeling safe
Treatment 0.24*(0.14)
Control Mean 6.88Control StDev 1.84
N 619Notes: This table shows the ITT estimates for business safety feeling (1 not feel safe - 10 feel safe). Control Vector: Baseline values offeeling safe and ln(sales-winsorized), not reporting sales in baseline, ln(baseline employee #), use of LPN in baseline, enumerator FE, dummyvariables for gender and business type and small business type. * p<0.1. ** p<0.05. *** p<0.01.
Table 8B: ITT estimates for business safety and theft exposure in the baseline
(1) (2)Theft in Baseline No Theft in Baseline
Feeling safe Feeling Safe
Treatment 1.22*** 0.17(0.44) (0.15)
Control Mean 6.36 6.96Control StDev 1.95 1.82
N 75 543Notes: This table shows the ITT estimates for business safety feeling (1 not feel safe - 10 feel safe) for sub-samples of firms based on theftexposure in the baseline. The first column reports the coefficient estimate for the sub-sample of businesses which reported in the baselinethat they were exposed to external theft - over the last 12 months - and the second column for those which did not report any external theftin the baseline. Control Vector: Baseline values of feeling safe and ln(sales-winsorized), not reporting sales in baseline, ln(baseline employee#), use of LPN in baseline, enumerator FE, dummy variables for gender and business type and small business type. * p<0.1. ** p<0.05.*** p<0.01.
50
Table 9: ITT estimates for business finance
(1) (2) (3) (4)Panel A: External finance (Formal)
Mobile loans Mobile loans Bank loans Bank loans(Yes/No) (ln(Amount)) (Yes/No) (ln(Amount))
Treatment 0.06** 0.47** 0.01 0.07(0.03) (0.23) (0.02) (0.25)
Control Gr Mean 0.10 0.74 0.08 0.84Control Gr StDev 0.30 2.35 0.26 2.86
N 612 581 609 580
Panel B: External finance (Informal)
Trade credit Trade credit Informal Loan Informal Loan(Yes/No) (ln(Amount)) (Yes/No) (ln(Amount))
Treatment -0.02 -0.51 0.01 -0.03(0.04) (0.37) (0.02) (0.09)
Control Gr Mean 0.35 3.17 0.05 0.15Control Gr StDev 0.48 4.71 0.22 1.12
N 619 564 576 575Notes: This table shows the ITT estimates for financial access outcomes. Control Vector in each regression: Baseline values of each financialvariable and ln(sales-winsorized), not reporting sales in baseline, ln(baseline employee #), use of LPN in baseline, enumerator FE, dummyvariables for gender and business type and small business type. * p<0.1. ** p<0.05. *** p<0.01.
51
Table 10: Heterogeneous treatment effects for business finance
(1) (2) (3) (4)Panel A: External finance (Formal)
Mobile loans Mobile loans Bank loans Bank loans(Yes/No) (ln(Amount)) (Yes/No) (ln(Amount))
Treatment 0.04 0.38* 0.02 0.10(0.03) (0.23) (0.02) (0.26)
Small x Treated 0.28*** 1.88** -0.01 -0.27(0.11) (0.92) (0.11) (1.21)
Control Gr Mean 0.10 0.74 0.08 0.84Control Gr StDev 0.30 2.35 0.26 2.86
N 612 581 609 580
Panel B: External finance (Informal)
Trade credit Trade credit Informal Loan Informal Loan(Yes/No) (ln(Amount)) (Yes/No) (ln(Amount))
Treatment -0.03 -0.61 0.00 -0.08(0.04) (0.38) (0.02) (0.09)
Small x Treated 0.19 2.10 0.18* 1.01(0.15) (1.39) (0.10) (0.69)
Control Gr Mean 0.35 3.17 0.05 0.15Control Gr StDev 0.48 4.71 0.22 1.12
N 619 564 576 575Notes: This table shows the ITT estimates with heterogeneous treatment effects for financial access outcomes. Control Vector in eachregression: Baseline values of each financial variable and ln(sales-winsorized), not reporting sales in baseline, ln(baseline employee #), use ofLPN in baseline, enumerator FE, dummy variables for gender and business type and small business type. * p<0.1. ** p<0.05. *** p<0.01.A firm is classified as small in a respective sector if in the baseline it has # Employees<Median(# Employees). “Small x Treated” is theinteraction between being a small-firm in the baseline and being assigned to the treatment group. Regressions include the dummy variable“Small” on RHS.
52
Table 11: ITT estimates and heterogenous treatment effects for business sales
(1) (2) (3) (4) (5) (6)ln(Sales) ln(Sales) Sales Sales Sales Sales
Volatility Volatility Volatility Volatility[robust] [robust]
Treatment -0.04 -0.04 0.001 0.02 0.000 0.002(0.07) (0.07) (0.052) (0.05) (0.004) (0.004)
Small x Treated 0.13 -0.40* -0.037*(0.36) (0.21) (0.020)
Control Gr Mean 4.90 4.90 0.72 0.72 1.062 1.062Control Gr StDev 0.95 0.95 0.59 0.59 0.043 0.043
N 539 539 435 435 515 515Notes: This table shows the ITT estimates with heterogeneous treatment effects for sales outcomes. Control Vector in each regression:Baseline values of LHS variables, ln(sales-winsorized), not reporting sales in baseline, ln(baseline employee #), use of LPN in baseline,enumerator FE, dummy variables for gender and business type and small business type. * p<0.1. ** p<0.05. *** p<0.01. A firm is classifiedas small in a respective sector if in the baseline it has # Employees<Median(# Employees). “Small x Treated” is the interaction betweenbeing a small-firm in the baseline and being assigned to the treatment group. Sales Volatility is computed using the difference betweenln(Salesmax) and ln(Salesmin), where ln(Salesmax) and ln(Salesmin) stand for the maximum and the minimum sales, respectively,during a particular month over the last 12 months. The heterogenous treatment regressions include the dummy variable “Small” on RHSinstead of baseline employee #. Columns 5 and 6 perform a robustness check, where we fill missing information in the baseline survey onsalesmax or salesmin using average monthly sales data.
Table 12: ITT estimates and heterogenous treatment effects for business investment
(1) (2) (3) (4)Total Investment Total Investment Inventory Investment Inventory Investment
Treatment 5.30 4.78 1.51 0.96(6.71) (6.88) (2.62) (2.72)
Small x Treated 11.20 10.15(21.32) (7.38)
Control Gr Mean 32.44 32.44 14.14 14.14Control Gr StDev 75.32 75.32 32.79 32.79
N 525 525 584 584Notes: This table shows the ITT estimates with heterogeneous treatment effects for investment outcomes. Control Vector in each regression:Baseline values of each investment variable, ln(sales-winsorized), not reporting sales in baseline, ln(baseline employee #), use of LPN inbaseline, enumerator FE, dummy variables for gender and business type and small business type. * p<0.1. ** p<0.05. *** p<0.01. A firmis classified as small in a respective sector if in the baseline it has # Employees<Median(# Employees). “Small x Treated” is the interactionbetween being a small-firm in the baseline and being assigned to the treatment group. The heterogenous treatment regressions include thedummy variable “Small” on RHS instead of baseline employee #.
53
Online Appendix
54
Tab
leO
A1:
Var
iable
san
dth
eir
defi
nit
ions
use
din
the
regr
essi
onan
alysi
s
Var
iab
leD
efin
iton
Pan
elA
:V
alu
ati
on
Met
hod
sof
Use
for
bu
sin
ess
=1
ifth
eb
usi
nes
su
sep
erso
nal
mob
.m
on
eyin
the
last
30
day
s.m
obil
em
oney
use
Rec
eive
pay
men
ts=
1if
the
bu
sin
ess
rece
ives
pay
men
tsto
per
son
al
mob
ile
mon
ey.
Sto
rem
oney
=1
ifth
eb
usi
nes
sst
ore
mon
eyin
per
son
al
mob
ile
mon
ey.
Pay
bil
l=
1if
the
bu
sin
ess
pay
bil
lsvia
per
son
al
mob
ile
mon
ey.
Pay
inp
ut
=1
ifth
eb
usi
nes
sp
ayin
pu
tsvia
per
son
al
mob
ile
mon
ey.
Pay
sala
ries
=1
ifth
eb
usi
nes
sp
aysa
lari
esvia
per
son
al
mob
ile
mon
ey.
Sav
ing
inm
ob.
mon
.acc
ou
nt
=1
ifth
eb
usi
nes
ssa
ve
bu
sin
ess
reven
ues
inm
ob.
mon
eyacc
ou
nt.
%of
uti
lity
exp
.via
per
s.m
ob
.%
of
bu
sin
ess
uti
lity
exp
ense
sp
aid
via
per
son
al
mob
.m
on
inu
tili
tyex
pen
ses.
%of
inp
ut
exp
.via
per
s.m
ob
.%
of
bu
sin
ess
inp
ut
exp
ense
sp
aid
via
per
son
al
mob
.m
on
ein
inp
ut
exp
ense
s.P
ayin
gw
ages
via
mp
esa
=1
ifth
eb
usi
nes
spay
wages
toth
ep
erso
nal
M-p
esa
acc
ou
nts
of
work
ers.
No
incr
ease
inp
rice
s=
1if
the
bu
sin
ess
do
not
incr
ease
pri
cew
hen
cust
om
erp
ayvia
mob
ile
money
.T
hef
tT
hef
tan
dsa
fety
ind
exS
um
mati
on
of
inte
rnal,
exte
rnal
thef
t,an
dfe
elin
gn
ot
safe
vari
ab
les,
stan
dard
ized
.In
tern
alth
eft
=1
ifem
plo
yees
stole
som
eth
ing
from
the
bu
sin
ess
inth
ep
ast
6m
onth
s.E
xte
rnal
thef
t,fire
,et
c.=
1if
exp
erie
nce
dlo
sses
as
are
sult
of
fire
,th
eft,
rob
ber
y,et
c.in
the
past
6m
onth
s.F
eeli
ng
safe
Th
esa
fety
of
the
are
aw
her
eth
eb
usi
nes
sis
loca
ted
(1ve
ryu
nsa
fe-1
0sa
fe)
Bu
sin
ess
savin
gS
avin
gat
ab
an
kor
mic
ro.
=1
ifth
eb
usi
nes
ssa
ves
at
ban
kor
mic
rofi
nan
ceacc
ou
nt
acco
unts
Sav
ing
ata
per
s.b
an
kacc
.=
1if
the
bu
sin
ess
save
at
ap
erso
nal
ban
kacc
ou
nt.
Sav
ing
ata
bu
s.ban
kacc
.=
1if
the
busi
nes
ssa
veat
ab
usi
nes
sb
an
kacc
ou
nt.
Sav
ing
ata
mic
rof.
inst
.=
1if
the
bu
sin
ess
save
at
am
icro
fin
an
ceacc
ou
nt.
Pan
elB
:T
ran
spare
ncy
Not
rep
orti
ng
Not
shar
edsa
les
=1
ifth
eb
usi
nes
sh
as
not
share
dit
ssa
les.
bu
sin
ess
acti
vit
yN
otsh
ared
pro
fits
=1
ifth
eb
usi
nes
sh
as
not
share
dit
sp
rofi
ts.
Bu
sin
ess
lice
nse
Bu
sin
ess
lice
nse
=1
ifth
eb
usi
nes
sh
as
an
up
tod
ate
gov
ern
men
tb
usi
nes
sp
erm
it.
Fin
anci
alF
inan
cial
index
Su
mm
ati
on
of
ban
klo
an
,m
ob
ile
loan
an
dre
cord
sst
an
dard
ized
.so
ph
isti
cati
onB
ank
loan
=1
ifth
eb
usi
nes
sre
ceiv
eda
ban
cklo
an
inth
ep
ast
12
month
s.M
obil
elo
an=
1if
the
busi
nes
sre
ceiv
eda
mob
ile
mic
rom
on
eyp
rovid
erlo
an
inth
ep
ast
12
month
s.B
usi
nes
sre
cord
s=
1if
the
bu
sin
ess
keep
sb
usi
nes
sre
cord
s.S
ales
cred
itS
ells
oncr
edit
tocu
st.
=1
ifth
eb
usi
nes
ssa
veat
am
icro
fin
an
ceb
an
kacc
ount.
Pan
elC
:B
ehavi
ora
lP
rese
nt
and
Pre
sent
bia
s=
1if
bu
sin
ess
own
eror
man
ager
isp
rese
nt
bia
sed
.F
act
ors
futu
reb
ias
Fu
ture
bia
s=
1if
bu
sin
ess
own
eror
man
ager
isfu
ture
bia
sed
.C
ogn
itiv
eca
pac
ity
#of
dig
its
rem
emb
ered
maxim
um
nu
mb
erof
dig
its
rem
emb
ered
from
nu
mb
er.
Tru
stT
rust
infi
rst
tim
eT
rust
inp
eople
you
mee
tfi
rst
tim
e,1
(not
at
all
)-4
(com
ple
tely
).T
rust
incu
stom
ers
Tru
stin
cust
om
ers,
1(n
ot
at
all
)-4
(com
ple
tely
).T
rust
inco
urt
sT
rust
inco
urt
s,1
(not
at
all
)-4
(com
ple
tely
).T
rust
inm
ob.
mon
.co
mp
.T
rust
inm
ob
ile
mon
eyco
mp
an
ies,
1(n
ot
at
all
)-4
(com
ple
tely
).B
usi
nes
ssi
zeE
mp
loyee
sN
um
ber
of
tota
lem
plo
yes
s,in
logari
thm
s.S
ales
,in
log.
Sale
sin
the
past
month
,in
logari
htm
.P
rofi
t,in
log
Pro
fits
inth
ep
ast
month
,in
logari
htm
.
55
Tab
leO
A2:
Des
crip
tive
stat
isti
csof
the
vari
able
suse
din
the
regr
essi
ones
tim
atio
ns
All
Sam
ple
Ph
arm
acy
Res
tau
rant
Var
iab
leN
mea
nsd
Nm
ean
sdN
mea
nsd
Pan
elA
:V
alu
ati
on
Met
hod
sof
Use
for
bu
sin
ess
545
0.5
20.5
0251
0.4
20.4
9294
0.6
00.4
9m
obil
em
oney
use
Rec
eive
pay
men
ts543
0.3
40.4
7250
0.2
70.4
4293
0.4
00.4
9S
tore
mon
ey543
0.1
70.3
8249
0.0
90.2
8294
0.2
40.4
3P
ayb
ill
544
0.3
00.4
6250
0.2
60.4
4294
0.3
30.4
7P
ayin
pu
t544
0.3
70.4
8250
0.2
90.4
6294
0.4
30.5
0P
aysa
lari
es544
0.0
60.2
4250
0.0
60.2
4294
0.0
60.2
5S
avin
gin
mob
.m
on
.acc
ou
nt
542
0.1
00.3
0249
0.0
40.2
0293
0.1
50.3
6%
ofu
tili
tyex
p.
via
per
s.m
ob
.474
0.3
30.4
5225
0.3
40.4
6249
0.3
10.4
5%
ofin
pu
tex
p.
via
per
s.m
ob
.485
0.1
50.3
1224
0.1
40.2
8261
0.1
70.3
3P
ayin
gw
ages
via
mp
esa
538
0.0
50.2
3244
0.0
60.2
4294
0.0
50.2
1N
oin
crea
sein
pri
ces
546
0.1
30.3
3252
0.1
10.3
1294
0.1
50.3
5T
hef
tT
hef
tan
dsa
fety
index
537
0.0
01.0
0246
-0.1
10.8
5291
0.0
91.1
0In
tern
al
thef
t538
0.2
70.4
4247
0.0
70.2
6291
0.4
40.5
0E
xte
rnal
thef
t,fire
,et
c.545
0.1
00.3
0251
0.0
20.1
5294
0.1
70.3
7F
eeli
ng
safe
545
7.3
21.8
9251
7.0
61.8
6294
7.5
41.8
8B
usi
nes
ssa
vin
gS
avin
gat
ab
an
kor
mic
ro.
542
0.5
70.5
0249
0.5
30.5
0293
0.6
10.4
9ac
cou
nts
Sav
ing
at
ap
ers.
ban
kacc
.542
0.3
70.4
8249
0.2
90.4
5293
0.4
50.5
0S
avin
gat
ab
us.
ban
kacc
.542
0.2
00.4
0249
0.2
50.4
3293
0.1
50.3
6S
avin
gat
am
icro
f.in
st.
542
0.0
30.1
6249
0.0
20.1
5293
0.0
30.1
6P
an
elB
:T
ran
spare
ncy
Not
rep
orti
ng
Not
share
dsa
les
546
0.1
40.3
5252
0.2
10.4
1294
0.0
90.2
8b
usi
nes
sac
tivit
yN
otsh
are
dp
rofi
ts546
0.1
40.3
5252
0.2
10.4
1294
0.0
80.2
7B
usi
nes
sli
cen
seB
usi
nes
sli
cen
se538
0.7
10.4
6246
0.9
10.2
9292
0.5
40.5
0F
inan
cial
Fin
ance
ind
ex515
0.0
01.0
0229
0.0
40.7
7286
-0.0
31.1
5so
ph
isti
cati
onB
ank
loan
522
0.1
00.3
0234
0.0
60.2
5288
0.1
40.3
4M
obil
elo
an
517
0.1
00.3
1231
0.0
70.2
6286
0.1
30.3
4B
usi
nes
sre
cord
s546
0.8
70.3
4252
0.9
60.2
0294
0.7
90.4
1S
ales
cred
itS
ells
on
cred
itto
cust
.545
0.5
10.5
0252
0.4
50.5
0293
0.5
70.5
0P
an
elC
:B
eha
viora
lP
rese
nt
and
Pre
sent
bia
s543
0.1
20.3
3251
0.1
10.3
2292
0.1
30.3
4F
act
ors
futu
reb
ias
Fu
ture
bia
s543
0.1
50.3
5251
0.1
30.3
3292
0.1
60.3
7C
ogn
itiv
eca
pac
ity
#of
dig
its
rem
emb
ered
545
5.2
11.5
0252
5.4
41.4
8293
5.0
21.4
9T
rust
Tru
stin
firs
tti
me
543
2.1
10.8
5251
2.1
50.8
9292
2.0
70.8
2T
rust
incu
stom
ers
544
3.4
20.6
5251
3.4
20.6
1293
3.4
10.6
9T
rust
inco
urt
s544
2.4
00.9
4252
2.3
70.9
8292
2.4
20.9
1T
rust
inm
ob
.m
on
.co
mp
.544
3.5
80.6
4252
3.4
30.6
8292
3.7
10.5
7B
usi
nes
ssi
zeE
mp
loyee
s546
1.4
10.4
2252
1.1
90.2
7294
1.5
90.4
3S
ales
,in
log.
546
5.0
11.0
1252
4.7
40.8
9294
5.2
31.0
5P
rofi
t,in
log
546
3.8
40.9
6252
3.8
30.8
9294
3.8
51.0
2
Notes:
Th
ista
ble
sum
mari
zes
des
crip
tive
stati
stic
sfo
rth
evari
ab
les
for
regre
ssio
nan
aly
sis
pre
sente
din
Tab
les
3,
4,
an
d5.
Th
evari
ab
led
efin
itio
ns
are
giv
enin
Tab
leO
A1.
56
Tab
leO
A3:
Des
crip
tive
stat
isti
csfr
omen
dli
ne
surv
eyfo
ral
lb
usi
nes
ses
wit
hb
usi
nes
sli
cen
ses.
Var
iab
les
All
sam
ple
Contr
ol
Tre
atm
ent
Defi
nit
ion
NM
ean
Med
ian
NM
ean
Med
ian
NM
ean
Med
ian
Lip
aN
aM
-pes
au
seH
ave
Lip
an
aM
-pes
aac
cou
nt
(0/1
)=
1if
bu
sin
ess
hav
ea
regis
tere
dL
ipa
na
618
0.2
70
309
0.2
30
309
0.3
11
0M
-pes
aac
cou
nt
Use
dL
ipa
na
M-p
esa
acco
unt
for
bu
sin
ess
(0/1
)=
1if
bu
sin
ess
use
dL
ipa
na
618
0.2
49
0309
0.2
07
0309
0.2
91
0M
-pes
aac
cou
nt
over
the
last
30
day
sR
ecei
ved
pay
men
tvia
Lip
an
aM
-pes
a(0
/1)
=1
ifb
usi
nes
su
sed
Lip
an
aM
-pes
aacc
ou
nt
618
0.2
46
0309
0.2
04
0309
0.2
88
0re
ceiv
edp
aym
ents
over
the
last
30
day
sL
ipa
na
M-p
esa
sale
s,m
onth
ly(1
000
Ksh
.)T
otal
sale
sre
ceiv
edvia
Lip
aN
aM
-pes
a618
15.5
50
309
12.5
60
309
18.5
40
ata
typ
ical
month
div
ided
by
the
sale
sla
stm
onth
Man
age
men
tpra
ctic
esan
dsa
fety
Rec
ord
kee
pin
gvia
mob
ile
mon
ey(0
/1)
=1
ifth
eb
usi
nes
skee
ps
reco
rds
via
618
0.1
08
0309
0.1
0309
0.1
17
0p
erso
nal
mob
ile
mon
eyor
Lip
aN
aM
-pes
aN
oth
avin
gch
ange
(0/1
)=
1if
the
bu
sin
ess
exp
erie
nce
da
fore
gon
e618
0.2
54
0309
0.2
56
0309
0.2
52
0op
por
tun
ity
tose
llgood
sd
ue
ton
oth
avin
gch
an
ge
Saf
ety
(1n
otfe
elsa
fe-
10fe
elsa
fe)
Saf
ety
ofth
eare
aw
her
eth
eb
usi
nes
s618
7.0
26
7309
6.8
74
7309
7.1
78
8is
loca
ted
inte
rms
of
the
thre
ats
of
fire
,th
eft,
robb
ery,
etc.
Inve
stm
ent
an
dacc
ess
tofi
nan
ceC
apit
alin
ves
tmen
t,(1
000
Ksh
.)In
vest
men
tin
the
cap
ital
good
s599
10.1
40
301
9.2
33
0298
11.0
50
inth
ela
st6
month
sR
ecei
ved
loan
(0/1
)=
1if
the
bu
sin
ess
has
rece
ived
alo
an
from
605
0.2
23
0301
0.1
96
0304
0.2
50
info
rmal
ly,
or
thro
ugh
mob
ile
ab
ank,
orm
on
eyacc
ou
nts
inth
ep
ast
12
month
sB
ank
loan
(0/1
)=
1if
the
bu
sin
ess
rece
ived
alo
an
from
608
0.0
86
0303
0.0
76
0305
0.0
95
0a
ban
k.
inth
ep
ast
12
month
sIn
form
allo
an(0
/1)
=1
ifth
eb
usi
nes
sre
ceiv
eda
loan
from
605
0.0
55
0302
0.0
53
0303
0.0
56
0fr
ien
ds,
rela
tive
s,et
c.in
the
pas
t12
month
sM
obil
elo
an(0
/1)
=1
ifth
eb
usi
nes
sre
ceiv
eda
loan
thro
ugh
611
0.1
33
0305
0.1
02
0306
0.1
63
0m
obil
em
on
eym
on
eyco
mp
an
ies
inth
ep
ast
12
month
sB
usi
nes
ssi
zeS
ales
,m
onth
ly(1
000
Ksh
.)S
ales
over
the
past
month
538
213.3
120
266
218.8
135
272
208
120
Pro
fits
,m
onth
ly,
(100
0K
sh.)
Pro
fits
over
the
past
month
.530
66.9
946
265
68.8
345
265
65.1
448
Em
plo
yees
Nu
mb
erof
tota
lp
erm
an
ent
592
4.3
78
3299
4.4
05
3293
4.3
52
3an
dte
mp
ora
ryem
plo
yees
Notes:
This
table
sum
mari
zes
the
desc
ripti
ve
stati
stic
sfr
om
endline
surv
ey
for
all
busi
ness
es
wit
hbusi
ness
license
.
57
Tab
leO
A4:
Bas
elin
ech
arac
teri
stic
san
dbal
ance
test
for
the
busi
nes
ses
that
par
tici
pat
edin
the
endline
surv
eyan
dhav
ebusi
nes
slice
nse
Ph
arm
acy
Res
tau
rant
All
Con
tT
reat
Diff
All
Con
tT
reat
Diff
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Sta
ndard
mobi
lem
on
eyu
seU
sin
gm
ob
.m
oney
for
bu
sin
ess
pu
rpos
es(0
/1)
0.45
0.40
0.49
-0.0
90.
670.
650.
68-0
.03
Usi
ng
mob
.m
oney
tore
ceiv
ep
aym
ents
(0/1
)0.
270.
200.
34-0
.14*
**0.
490.
480.
50-0
.02
Usi
ng
mob
.m
oney
tost
ore
mon
ey(0
/1)
0.12
0.11
0.13
-0.0
20.
240.
290.
180.
11U
sin
gm
ob
.m
oney
top
ayb
ills
(0/1)
0.29
0.25
0.33
-0.0
70.
460.
440.
47-0
.03
Usi
ng
mob
.m
oney
top
aysa
lari
es(0
/1)
0.05
0.02
0.09
-0.0
7***
0.09
0.09
0.10
-0.0
1U
sin
gm
ob
.m
oney
top
ayin
pu
ts(0
/1)
0.34
0.33
0.36
-0.0
20.
500.
500.
500.
00A
ware
nes
sof
Lip
aan
dre
aso
ns
of
not
adopti
ng
Aw
are
of
Lip
aN
aM
-Pes
a(0
/1)
0.94
0.95
0.94
0.01
0.98
0.98
0.98
0.00
Hav
eL
ipa
Na
M-P
esa
(0/1)
0.05
0.02
0.09
-0.0
7***
0.23
0.25
0.20
0.05
Not
seei
ng
the
ben
efits
ofL
ipa
Na
M-P
esa
(0/1
)0.
260.
260.
250.
010.
190.
230.
150.
08T
oo
cost
lyto
open
an
acco
unt
(0/1
)0.
150.
170.
120.
040.
050.
030.
07-0
.04
Hig
htr
an
sact
ion
fees
via
Lip
aN
aM
-pes
a(0
/1)
0.24
0.23
0.24
-0.0
10.
090.
070.
11-0
.04
Don
’th
ave
tim
eto
open
anacc
ount
(0/1
)0.
070.
090.
050.
050.
190.
210.
170.
04W
ould
not
incr
ease
my
sale
s(0
/1)
0.10
0.09
0.11
-0.0
20.
080.
070.
08-0
.01
No
tru
stin
mob
ile
mon
eypro
vid
er(0
/1)
0.04
0.03
0.05
-0.0
20.
020.
030.
010.
02T
oo
com
ple
xto
use
(0/1
)0.
080.
080.
070.
010.
130.
090.
16-0
.06
Bu
sin
ess
size
loga
rith
m(S
ale
s,m
onth
ly(1
000
Ksh
.),
win
sori
zed
5%)
4.78
4.72
4.84
-0.1
25.
705.
755.
660.
09lo
gari
thm
(Pro
fits
,m
onth
ly,
(1000
Ksh
.),
win
sori
zed
5%)
3.88
3.82
3.94
-0.1
24.
324.
344.
290.
05lo
gari
thm
(Em
plo
yee
s)1.
211.
201.
22-0
.02
1.86
1.88
1.84
0.05
Inve
stm
ent
an
dacc
ess
tofi
nan
ceIn
vest
men
t(0
/1),
inth
ep
ast
6m
onth
s0.
210.
190.
22-0
.03
0.42
0.47
0.37
0.10
Ban
klo
an
(0/1
),in
the
pas
t12
month
s0.
080.
080.
080.
010.
090.
090.
10-0
.01
Info
rmal
loan
(0/1
),in
the
past
12
mon
ths
0.03
0.02
0.03
-0.0
10.
040.
040.
05-0
.01
Mob
ile
loan
(0/1
)in
the
past
12
mon
ths
0.11
0.11
0.11
0.00
0.09
0.09
0.09
0.00
Info
rmali
tyB
usi
nes
sli
cen
se(0
/1)
1.00
1.00
1.00
0.39
1.00
1.00
1.00
0.00
Notes:
This
table
sum
mari
zes
the
desc
ripti
ve
stati
stic
sfr
om
base
line
surv
ey
for
all
busi
ness
es
wit
hbusi
ness
license
and
who
rem
ain
ed
inth
eendline.
*p<
0.1
.**
p<
0.0
5.
***
p<
0.0
1.
58
top related