Banking with Agents: Experimental Evidence from Senegal

This paper uses a randomized controlled trial to compare different banking delivery channels. Individuals living close to banking agents and branches of the same financial institution were encouraged to open a savings account and transact at either an agent or a branch. Compared with individuals sent to a branch, individuals sent to an agent increased their number of transactions at the agent and incurred lower transaction costs there. One year later, 42% of active users still used both channels but made transactions at the agent only half as large as those made at the branch.

and thus may deepen financial inclusion more cost-effectively than the traditional branch banking model. 1 The goal of this paper is to compare these two different delivery channels for financial services: agent banking and traditional branch banking.Understanding the differences in delivery channels is important because financial institutions are thinking about ways to effectively increase their outreach and ability to mobilize more savings.Building a large branch network is expensive and may not reach the same population segments as those that can be reached with an agent network.In contrast, expanding an agent network allows the bank to have many more access points across the country, but because agents operate on their own without regular supervision from bank staff, there may be concerns about the quality of service delivered.
As documented in the literature, banking with an agent may be different from banking at a branch along several additional dimensions, including transactions costs such as transport costs, account-opening fees, and fees per transaction; trust; privacy; restrictions on the use of the account; and social constraints (see Karlan, Ratan, and Zinman 2014 for reviews of interventions that use different channels to encourage savings mobilization).
Because the decision to bank with an agent or a branch is typically endogenous, we use a randomized controlled trial to identify the causal effects of agent banking in Senegal.In particular, a randomly selected group of individuals was given information about a savings account.Half of the individuals were encouraged to open the account at the nearest agent, while the other half were encouraged to do so at the closest branch. 2 Individuals thus fall into one of three groups: (i) those that were offered information about the account (Information)  and encouraged to open a savings account at the closest branch; (ii) those that were offered information about the account (Information) and encouraged to open a savings account with the closest agent (Agent); and (iii) a control group where individuals did not receive information to open the savings account.
Individuals in the study were thus offered the same account irrespective of the location where they were encouraged to open it, and deposit or withdrawal fees did not apply to nearly all transactions during the study period.In addition, unlike much of the literature that has focused on how physical proximity to agents can promote use of savings accounts (see, e.g., Ashraf, Karlan, and  Yin 2006; Brune et al. 2016; Bachas et al. 2021), the selected study areas are close to and equidistant from agents and branches alike.By holding distance to point of service fixed, which is arguably the most important difference between agent and branch banking, this paper tries to get at the "plumbing" of financial inclusion (Duflo 2017) by focusing on other aspects of the transaction costs such as waiting and face-to-face time and the social distance between the agent or branch teller and the client.
We find that visits to agents (including transport and waiting time with the agent) were around 10 minutes shorter than visits to branches, mostly due to a shorter waiting time.Transaction costs were thus lower when banking with agents.
In addition, providing account information increases the probability of opening the account by 12.9 percentage points, and this rate does not vary by the location where individuals were encouraged to open the account.It is however on the low side of the wide variation in account take-up found in the literature, which is tied to the availability of reliable savings alternatives.Take-up rates range from about 10% in Indonesia where multiple providers coexisted (Cole, Sampson, and Zia 2011) to about 80% among female household heads living in Nepalese slums (Prina 2015) where no other formal financial service providers were present.While many households in the areas we study are poor, they are better off than those studied in Nepal.Moreover, half of our subjects already had an account with another financial provider prior to the experiment, suggesting that the relatively low take-up rate in our study may be driven by the existence of other providers.
Because of the lower transaction costs of banking with an agent, individuals directed to agents should visit them more often compared with individuals directed to a branch and may have higher savings balances.The results broadly support these hypotheses.During the 12 months after the opening of the account, individuals directed to an agent made 0.4 more deposits and 0.4 more withdrawals than those directed to a branch (who had an average of 0.3 deposits and 0.4 withdrawals).In addition, average balances after 12 and 24 months of account opening across all individuals directed to an agent were 109% and 180% higher than those directed to a branch, who had an average balance of US$3.22 (US$2.20)12 (24) months after opening the account.Both increases in balances are statistically significant at conventional levels (p-values are .05and  .022,respectively).
If study participants chose locations on the basis of transaction costs, then those sent to an agent would bank exclusively with agents.Yet, individuals directed to the agent visited the branch as often as those encouraged to open the account at the branch.In fact, their overall increase in the number of transactions came only from visiting the agent more often.In addition, they made transactions of roughly US$219 (median is US$74) at the branch compared with transactions of about US$96.50 (median is US$31) when visiting the agent.Individuals sent to the agent, therefore, visited agents more often and made larger transactions at branches.In contrast, individuals sent to the branch visited the agent less often and made transactions of roughly the same size at agents and branches. 3o why did clients not bank exclusively with agents, thus minimizing transaction costs?Although we lack experimental variation to identify plausible causal mechanisms, we offer three potential explanations based on greater trust in branch staff, better security at branches, and less privacy regarding transactions conducted at agents.In section IV, we offer supporting evidence based on correlations and survey responses that supports some of these explanations better than others.But an overall message is that individuals sent to the agent seem to be better off because they use it more often and choose to visit both locations.
This work is related to two emerging strands in the literature.The first introduces ordinary savings accounts to poor households (e.g., Cole, Sampson,  and Zia 2011; Dupas and Robinson 2013; Prina 2015; Brune et al. 2016;  Dupas et al. 2016; Schaner 2017).Most of these studies subsidize the account-opening and maintenance fees (if they exist) and show significant effects on take-up and use of accounts.The second strand in the literature focuses on how poor households use agents to send and receive money within social networks to improve their financial management.Jack and Suri (2014)  show that mobile telephony reduces the costs of such within-network transfers in Kenya and find that proximity to "mobile money" agents has helped households to smooth consumption in the face of economic shocks. 4ecause we examine the effects of proximity to banking channels (branches and agents), our work is also related to the nonexperimental literature showing how expansion of bank branch networks is associated with increased account use among the poor, resulting in increases in their incomes and reduced poverty levels (Burgess and Pande 2005; Bruhn and Love 2014; Allen et al.  2021). 5What distinguishes our work from both the experimental and nonexperimental literature is that we provide a direct comparison of different banking channels, and in particular of how agents affect the take-up and use of ordinary savings accounts relative to bank branch staff.To our knowledge, this is the first such comparison in the literature.
The rest of the paper is organized as follows.Section II describes the experimental design and data sources.Section III explains the empirical framework and presents the results, section IV discusses possible explanations for the results, and section V concludes.

II. Experimental Design and Data
The experiment was a collaborative effort among Microcred Senegal (Microcred), MasterCard Foundation, the World Bank, and the International Finance Corporation.Microcred is a microfinance institution that offers microcredit as well as savings products.After entering the Senegalese market in 2007, Microcred quickly became one of the four largest microfinance institutions in Senegal.Having focused on lending to micro, small, and medium entrepreneurs in the past, Microcred has shifted focus over time to cater to low-income individuals, the illiterate, women, youth, and rural residents.The institution's agent network was launched in 2014 and has grown rapidly to support this vision: 439 agents complement a network of 55 branches serving clients all over Senegal in all large cities and several smaller ones as well as increasingly in rural areas. 6espite the arrival of Microcred, Senegal remained below the average for sub-Saharan Africa in terms of financial inclusion.According to the Global Findex Database (Demirguc-Kunt et al. 2015), prior to the intervention studied here only 11.9% of the adult Senegalese population had a bank account, to more easily draw on their social networks for support in trying circumstances.Relatedly, Blumenstock, Eagle, and Fafchamps (2016) show that Rwandan households affected by an earthquake received increased amounts of cellular "airtime" (a simple precursor to mobile money) from members of their social network, especially those with whom they had already established reciprocal relationships. 5 Burgess and Pande (2005) and Bruhn and Love (2014) find significant increases in income, output, and employment as a consequence of bank branch expansion in India and Mexico, respectively. 6Baobab Annual Report 2020: https://baobabgroup.com/rapports/2020/2020-EN/files/basic-html /page20.html.and only 6.6% of adults reported saving at a financial institution.7 Saving informally by using savings clubs or nonrelatives was more common and done by 29%.These numbers are lower than those reported in a census we conducted in the study of peri-urban areas with closer proximity to financial institutions.According to this census, 46% of households reported having a savings account (see panel A of table 1).
Our study sample consists of 2,200 individuals from nine different survey areas located in the suburbs of the capital, Dakar (six areas), in Thies, the third most populous city (two areas), and in a village outside of Thies (one peri-urban area).
These areas were chosen because Microcred had recently established an agent in each area.Once agents were selected by Microcred, we identified suitably populated areas that were equidistant to Microcred branches and agents.Figure 1 shows an example of the two survey areas in Thies.Each survey area is located between the Microcred branch and a Microcred agent and is typically 1-1.5 square kilometers in size.

A. Census and Baseline Data
Between September and October 2014, basic sociodemographic and financial data were collected from every household in the study areas.The enumerators revisited the household up to four times at different hours during the day in case the household was unavailable.From each household in this census, a member was selected at random using a Kish grid.Households with existing Microcred clients were excluded from the census, but those that were clients of other financial institutions were included.In total, 8,002 individuals were interviewed in the census.The timeline of the experiment is presented in figure 2. We used data from a pilot in which a small sample from the census was offered the possibility to open a savings account to identify respondents with a high propensity for take-up. 8The 2,500 individuals with the highest probability were selected and randomly allocated into a control and two different treatment groups of 1,000 each, as described above.
Household visits took place between January and March 2015.During the visit, tablets were used to collect baseline data about respondents' demographics, household characteristics, financial assets, credit and saving behavior, use of money transfer services and bill payment methods, and their awareness and use of mobile money services.A total of 2,200 respondents were successfully interviewed, 374 respondents in the control group and 1,827 respondents from the treatment groups.
Table 1 reports summary statistics from the census data for the whole sample (panel A) and the one selected for the study (panel B).The full census sample is comparable to the Global Findex Database because both use a Kish grid to select the sample among household members.Study participants, however, were selected because they had a high propensity to open a savings account, and thus, the baseline sample and the census differ.Panel B reports that 53% of respondents are female, less than in the full census sample in panel A. Fortyone percent of baseline respondents are heads of households, who are typically the managers of household finances.While that could have contributed to a high propensity to open an account, the proportion of household heads in panel B is actually similar to that of the full census in panel A.
The average and median proportion of self-employed respondents, respondents with credit, and respondents with a savings account are higher in the baseline sample than in the census sample.The same holds for respondents' average and median age and individual monthly income.
Baseline respondents in panel B of table 1 are on average about 38 years old, obtained 7 years of schooling, and have an average monthly income of about US$184.The average monthly income of all census respondents is comparatively lower at US$124.Literacy among baseline respondents in panel B is 64%, more than 10 percentage points higher than in the census.Further, 80% of baseline respondents report being self-employed compared with only 32% in the census.The use of financial services is again higher among baseline respondents with 52% already having a savings account and virtually everyone reporting to save informally.In contrast, only 46% of census respondents have a savings account.The baseline sample in panel B is thus richer on average, more educated, and more actively engaged with financial services than the full census sample in panel A.

B. Treatments
After the baseline data were collected, individuals in the treatment groups were offered the opportunity to open a Sukaliku account, which has a number of features designed to appeal to the poor: no opening or account maintenance fees, no minimum balance, and no limit to the number of deposits and withdrawals to and from the account (although there is a 500 CFA, or US$0.83, minimum amount for deposits and withdrawals).Account balances earn interest at 4% per year, and, importantly, there are no fees on deposits at either an agent or a branch or on withdrawals at the branch.During the study period, a holder of a Sukaliku account could make up to three free withdrawals per month at an agent, and for additional withdrawals, the account holder was charged a fee to withdraw money that depended on the amount of the withdrawal.Section OA1 of the appendix (available online) provides more details about the fee structure that clients faced during the study period. 9he treatment groups received information about the Sukaliku account features during the visit and were encouraged to open the Sukaliku account either at an agent or at a branch (depending on the treatment group) at an agreed day and time, although they had 2 weeks from the day of the visit to do it.They were handed a card with the address of the agent or branch, date, time, and a description of the documentation required to open the account. 10Section OA2 of the appendix provides examples of the cards handed out.At that time, however, agents did not have the authority to open Sukaliku accounts on behalf of Microcred, and as a result, a Microcred representative was present with the agent during the account-opening days.His or her role was to passively observe the account-opening process and to collect the paperwork needed to open the account after the client had left.Any information about the account was provided by the agent, who also answered all the clients' questions.The different treatment conditions and the number of observations in each group are reported in table 2. 11 Table 3 shows that experimental groups are balanced.Columns 2-4 report the mean for each group while column 5 reports the p-value of an F-test that all treatment dummies are jointly zero in a regression where each household or individual characteristic is the dependent variable.Panel A reports characteristics from the census data, and panel B reports characteristics from the baseline data.We do not observe systematic differences between the different treatment and control groups with respect to demographic characteristics, education level, monthly income, or the proportion of respondents with financial products.Similarly, there are no significant differences in the distance to the closest Microcred agent and branch.The number of observations drops to 2,089 because GPS data were incomplete for 112 households.We find significant differences between the different treatment and control groups at the 10% level for the proportion of self-employed individuals, individual monthly income, and an index of risk aversion.Following Bruhn and McKenzie (2009), we control for these variables in the analysis of section III.An F-test of the null hypothesis that all these variables are jointly insignificant in predicting whether an individual is in a particular experimental group cannot be rejected (all p-values are higher than .143).
C. Account Opening and the Neighborhood and End-Line Surveys Among individuals in the two treatment groups, a sample of individuals that opened the Sukaliku account was surveyed immediately afterward and asked about their experience opening the account.Note.Reported are means of treatment and control groups to check whether the randomization was successful.The means reported here are calculated for the individuals in the experimental sample.Panel A reports data from the filter survey.Panel B complements the analysis by using variables collected at baseline via the household survey.Column 1 reports the number of observations.Columns 2-4 report means for the control group and for the different treatment groups.Column 5 reports the p-value of an F-test that, in a regression of the variable against treatment indicators, all coefficients associated to treatment indicators are jointly zero.The last line of each panel reports the p-value of the F-test that in a regression of a treatment arm indicator against all filter survey or household survey variables, all coefficients associated to the filter survey or household survey variables are jointly zero.Has savings account (1 5 yes) is a dummy variable that takes the value of 1 if a household member reports having a savings account.Risk aversion (0-10) is an index of self-reported risk behavior that varies from 0 to 10, where 0 means "I always try to avoid taking risk" and 10 means "I am fully prepared to take risks."Raven's test score (0-3) varies from 0 to 3 and is equal to the number of correct answers given by a survey respondent in a Raven's test.Distance to branch (in kilometers) and distance to agent (in kilometers) are the distances from the household to the closest branch and to the closest agent, respectively.* p < .10.
In table 4, column 2 of panel A shows that the account-opening visit for individuals sent to the agent was 9.7 minutes shorter than that for individuals sent to the branch, which lasted about 1 hour.Most of this time reduction comes from shorter waiting time at the agent.The actual face-to-face meeting with the agent or teller to present the documentation and open the account did not differ by location and lasted 26.7 minutes.These numbers suggest that transaction costs when visiting an agent are lower.
In the account-opening survey, we also asked respondents about their level of trust in Microcred agents and branches after opening the accounts.Column 5 of panel A in table 4 suggests that 83.5% of respondents fully trust the branch tellers and that this trust is 1.2 percentage points higher for agents, although this increase is not statistically significant at conventional levels.This result suggests that Microcred agents are perceived to be as trustworthy as branch staff, at least initially.
At the time of the baseline survey and intervention, we also implemented a neighborhood survey to identify all bank and microfinance branches and banking and mobile money agents in the study areas.By design, a study area contained one Microcred branch and one Microcred agent.Combining all study areas, the neighborhood survey identified a total of 391 establishments, including 57 bank branches, 44 microfinance branches, 6 Microcred banking agents, and 251 mobile money agents, so the density of financial providers was high.All major banks, savings cooperatives, and microfinance institutions had branches, including CBAO, Banque Atlantique, Pamecas, SGBS, and our partner Microcred, among others.In contrast, only Microcred had banking agents at the time of the study, so all banking agents are from Microcred.It is important to note that while individuals in a study area had the choice of visiting the Microcred branch or the agent, in practice they had little choice over different Microcred agents or branches because those were located far apart from each other.Transaction data confirms that a given individual always visited the same agent and/or branch.
A Microcred banking agent can collect deposits into and make withdrawals from the Sukaliku savings account and other Microcred accounts.Both branches and banking agents often also provide various mobile money and money transfer services simultaneously, including Wari, Joni Joni, Orange Money, Western Union, MoneyExpress, and MoneyGram, among others.Microcred branches and agents typically work with two to three mobile money providers, while the median mobile money agent in our sample worked with four mobile money providers.Microcred branches are open from 8:30 a.m. to 7:00 p.m. from Monday to Friday, 8:30 a.m. to 5:30 p.m. on Saturdays, and are closed on Sundays.Microcred agents are typically open for longer hours (until 9:00 p.m. or 10:00 p.m.), and they also operate on Sundays.Most agents are retail shops (typically electronics), although some work exclusively as banking and mobile money agents.Banking agents receive from the bank a commission per transaction.Section OA1 of the appendix provides more details about the commission that agents received per transaction.
The neighborhood survey also included a module that collected basic individual characteristics for a sample of Microcred agents and Microcred branch tellers.Table OA2 compares the characteristics of respondents, Microcred agents, other agents, and branch tellers.Respondents and Microcred branch tellers are older than Microcred agents (38 and 32 years old, respectively, vs. 24.5 years old) and more likely to be married (66% and 69%, respectively, vs. 33%), but Microcred agents and bank tellers are far more educated than respondents (14 and 16 years of education, respectively, vs. 6.5 years), and, perhaps as a result, they enjoy higher monthly household income.It thus appears that both Microcred agents and branch tellers belong to a different socioeconomic group from that of respondents.Microcred agents also appear to be better off than mobile money agents in terms of education (14 vs. 11.7 years, respectively) and income (US$767.5 vs. US$487.4,respectively).
The end-line survey was conducted in March 2016, 1 year after the baseline, with the same respondents.The overall attrition rate between baseline and end line was 25% and, as reported in table OA3, it is almost identical and thus not statistically different across experimental arms.This suggests attrition bias is not a concern when examining the effect of banking with an agent or a branch on the outcomes measured in the end-line survey.
The end line included a module in which individuals were asked to recall their last transaction with either an agent or at a branch.This module recorded the total time involved in the visit (including time for transport, waiting to be served, and face to face with the agent or teller).Because the location of the last transaction is endogenous and individuals that chose to transact at the agent may be different from the people that chose to transact at the branch, we restrict the sample to transactions at the location where individuals were encouraged to open the account.The sample thus consists of 154 visits to a branch teller and 48 to an agent.The average overall time is about 33 minutes at the branch but only 19 minutes at the agent, while the face-to-face time with the agent was 7 minutes compared with 10 minutes at the branch. 13anel B of table 4 reports regressions of different outcomes against a dummy for whether the individual was encouraged to open the account with an agent and the distance from the respondent's house to the branch.
Column 1 of panel B in table 4 shows that visits to the Microcred agent are US$0.32cheaper on average than visits to the branch.This difference is not 13 The distribution of waiting times at the branch and the agent are similar.The 90th percentile waiting time at the branch is 40 minutes and at the agent is 35 minutes.In contrast, the 10th percentile waiting time is 10 minutes at the branch and 7 minutes at the agent.In addition, as part of the neighborhood survey, we asked that enumerators record the time they had to wait for the agent or a bank teller to answer the actual survey.The average waiting times reported by enumerators are lower than the 10th percentile waiting times reported by study participants that opened an account.Enumerators report a wait time of 5.6 minutes with tellers at the branch versus only 2.7 minutes with agents.Some enumerators told us that they announced themselves as they entered the establishment and so waiting times may have been shorter than for an average client waiting to make a transaction.economically large or statistically significant.In fact, the median visit by individuals to the agent or the branch does not involve a transport cost.Column 2 of panel B in table 4 confirms that the total visit time by an individual sent to an agent is shorter by 14.3 minutes.The bulk of the time reduction is waiting time by 11.3 minutes (col.3), followed by a reduction of 3 minutes in transaction time (col.4).None of these differences, however, are statistically significant at conventional levels.
Using individuals who report the waiting time for both the first visit during account opening and the last visit made a year later at the same location, we find that 70% of individuals experienced lower waiting times in their last visit.More interesting, this percentage increases to 77% for individuals who made more than two transactions by the end line at the location where they opened the account.In contrast, only 55% of individuals who made two or fewer transactions reported a decline in the waiting time.The actual waiting times were 2 minutes higher for individuals who made two or fewer transactions at the location where they opened the account and 23 minutes lower for individuals who made more than two transactions.Figure OA1 of the appendix (figs.OA1-OA9 are available online) compares the distribution of visits by day of the week for a client's first two visits and her or his last two visits-a point at which a client may have learned to avoid the potential congestion at the bank and at the agent.The p-value of the x 2 test that both distributions (of the first vs. last two visits) are equal are .143and .026for visits to the branch and to the agent, respectively.This suggests clients are more likely to have changed the day of the week that they visited agents over time than to have changed the day when they visited a branch.Perhaps this was because branches tended to always be more congested and so there was little gain to changing the day of the visit.This evidence supports the idea that individuals reduce the overall transaction costs as they learn to avoid the busy times when visiting the location multiple times.
As we will see in section III.B, the transaction costs of visiting the agent could be lower for individuals sent to the agent compared with the transaction costs for those sent to the branch simply because individuals sent to the agent were induced to visit the agent more often.Individuals sent to the branch would not have experienced different transaction costs at the branch because they made roughly the same (low) number of transactions at the branch as did individuals sent to the agent.
In the visit module of the end-line survey, we also asked respondents about their level of trust in Microcred agents and branches.Column 5 of panel B in table 4 suggests that 91% of respondents fully trust the branch tellers and that this trust is 3.2 percentage points higher for agents.This result is notable because it suggests that Microcred agents have been successful in gaining the trust of their clients, thus likely contributing to deeper financial inclusion.14

D. Administrative Data
Administrative data on account opening and the location and size of transactions were obtained from the start of the intervention until roughly 1 year later in March 2016. 15These administrative data are used to compare account opening across treatment groups (take-up) and to measure account activity, that is, whether a transaction was a deposit or a withdrawal and the location where it occurred (branch or agent).We also have data on money transfers, but virtually no client in our sample used the account to transfer money.Data on savings balances were also obtained from Microcred through April 2017, allowing us to compute balances 12 and 24 months after the intervention.

III. Empirical Framework and Results
By virtue of the design of the experiment, the effect of the different treatments on savings account opening, use, balances, and other outcomes can be estimated with the following equation: The variable Y i denotes the dependent variable for individual i, namely, whether the Sukaliku account was opened and variables related to account use such as the number of deposits and withdrawals as well as the transacted amounts and other measures of savings behavior.Unless otherwise noted, the variable Y i is assigned the value of 0 for individuals who did not open an account.The treatment dummy Agent i takes the value of 1 if respondent i was sent to an agent to open the account, and Information i takes the value of 1 if respondent i was provided information about the Sukaliku account during the household visit.Given the definition of the treatment dummies, the dummy Information i captures the effect of being sent to the branch to open the account.The main coefficient of interest is Agent i because it captures the difference between being sent to the agent and being sent to the branch.
Table 2 describes the different experimental arms and the value that the treatment dummies take.In addition, DistBranch i denotes the distance in kilometers from the respondent's house to the closest Microcred branch.We do not include the distance to both the branch and the agent because individuals are equidistant between the branch and the agent by design, and so the distance to either location is highly positively correlated (correlation coefficient is 0.96). 16inally, we include variables that were unbalanced across treatment arms at baseline, namely, a dummy for self-employment, individual monthly income, and the index of risk aversion in vector X i . 17Each observation corresponds to one individual or household, and the standard errors are robust to heteroscedasticity.Because we test several hypotheses at once, we correct the standard errors for multiple hypothesis testing and report the corrected p-values in brackets. 18 Account Take-Up Table 5 presents the intent-to-treat effects on account take-up and the number of transactions with the Sukaliku account.In column 1, the dependent variable is whether a Sukaliku account was opened.Only 11.7% of individuals who were not provided information about the account (control group) managed to open an account.Providing information about the account increases the probability of opening it by 12.9 percentage points, an increase of 110% relative to the take-up rate among individuals in the control group.Notably, take-up is not affected by the location where the individual was encouraged to open the Sukaliku account.
Column 2 of table 5 examines who used the account actively, that is, who made at least one transaction during the 12 months since the opening date.We find that encouraging individuals to open an account by providing information increases the probability of using the account by 3.3 percentage points (the coefficient of the dummy Information), an increase of 77% relative to the control group.More important, there are no differences in the probability of using the account actively by the location where individuals were encouraged to open the account (the coefficient of the dummy Agent is virtually zero and not statistically significant).
Column 3 of table 5 reports the take-up of other Microcred savings accounts opened outside of the study but during the study period among the 16 By virtue of the randomization, the point estimates on the treatment dummies do not vary if we instead include both distance variables, the difference in distances, or if we exclude the distance variable altogether. 17The vector X i does not include location fixed effects because the randomization was not stratified by location.However, the results are robust to the inclusion of location dummies. 18We use the Stata routine wyoung (for details, see Westfall and Young 1993).4. DistBranchi is the distance in kilometers between the household of the respondent and the closest branch.Xi is a vector of control variables evaluated at baseline, namely, a dummy for self-employment status, monthly income in US dollars, and an index for risk aversion.The dependent variable in col. 1 is a dummy variable that takes the value of 1 if the individual opened a Sukaliku savings account within a year after the intervention.In col.2, the dependent variable is a dummy that takes the value of 1 if the account is active within a year after the intervention, which means the individual made at least one transaction since he or she opened the account.In col.3, we use as dependent variable an indicator of opening any other Microcred account within the relevant period.For cols.1-3, yes is equal to 1. Columns 1-9 use the whole sample, whereas col. 10 uses the sample of individuals who have active accounts.In cols.4-6, the dependent variables are the total number of deposits and the number of deposits made at the branch and with an agent, respectively, within a year after the intervention.In cols.7-9, the dependent variables are the total number of withdrawals and the number of withdrawals made at the branch and with an agent, respectively, within a year after the intervention.In col. 10 we use as the dependent variable the share of all transactions within a year after the intervention that were made at the agent.individuals who opened a Sukaliku account.Among individuals being sent to the Microcred agent, the probability of opening another Microcred account during the study period increased by 1.8 percentage points relative to being sent to the branch, representing an increase of 128.6% relative to the 1.4% of individuals in the control group who opened another Microcred account (p-value is .012).

B. Number of Transactions
We predict that individuals encouraged to open an account with a Microcred agent will face lower transaction costs and thus make more transactions.Columns 4-6 (7-9) of table 5 report the total number of deposits (withdrawals) and those made at a branch or with a Microcred agent, respectively, from the time the account was opened until March 2016.In column 4, individuals in the control group made 0.30 deposits in total.The coefficient of the dummy Information is close to zero and not statistically significant, indicating that individuals directed to a branch did not make significantly more deposits relative to those in the control group.Individuals directed to a Microcred agent made 0.37 more deposits compared with those directed to a branch (corrected p-value is .064).Similarly, in column 7, individuals in the control group (and those directed to the branch) made 0.38 withdrawals on average while individuals directed to the agent made 0.39 more withdrawals (corrected p-value is .02).Columns 4 and 7 of table 5 therefore confirm that individuals sent to the agent made more transactions compared with those sent to the branch.Columns 5 and 6 of table 5 show that the increase in the overall number of deposits among individuals encouraged to open the Sukaliku account with a Microcred agent comes from an increase in deposits with the agent (col.6) rather than at the branch (col.5).We find a similar result when comparing withdrawals in columns 8 and 9.These results indicate that our treatment of encouraging individuals to open an account with a Microcred agent provided an impetus to make more deposits and withdrawals with an agent.
Column 10 of table 5 confirms the increased use of agents by active account users encouraged to open the account with an agent.Active individuals in the control group used a Microcred agent for 26.8% of their transactions.Individuals sent to the branch used the Microcred agent at the same rate.In contrast, individuals directed to the agent increased their share of transactions with the Microcred agent by 34.6 percentage points relative to individuals directed to the branch, an increase of 180%. 19We previously showed that transaction costs (waiting time) declined as the number of transactions increased.Because individuals encouraged to open the account with an agent ended up making more transactions, they experienced, on average, lower transaction costs.

C. Usage and Account Balances
Table 6 shows that for individuals sent to an agent, the increase in the number of transactions resulted in an increase in the sum of deposits (col. 1) and in the sum of withdrawals (col.2) by 202% and 162%, respectively, from a base of US$39.2 of accumulated deposits and US$24.33 of accumulated withdrawals (corrected p-values are .088and .066,respectively) for individuals encouraged to open an account at the branch. 20 reduction in transaction costs should lead to a reduction in optimal cash balances, and, as a result, savings balances in the account should increase because individuals carry less cash around on average.Columns 3-6 of table 6 report the balance in the Sukaliku account (cols.3 and 5) and in all Microcred accounts (including Sukaliku, in cols.4 and 6) 12 and 24 months after the intervention.Columns 3 and 4 suggest that the increase in balances 12 months after the intervention for individuals sent to the agent compared with balances for individuals sent to the branch was about US$0.95 in the Sukaliku account and US$2.4 across all Microcred accounts, although only the increase in the Microcred accounts is significant at conventional statistical levels (corrected pvalue is .06).After 24 months, the increase in balances in the Sukaliku account and across all Microcred accounts for individuals sent to the agent compared with balances for individuals sent to the branch was US$1.49 and US $2.91 (corrected p-values are .06and .026),representing an increase of 1,022% and 129%, respectively.
We conclude that balances among individuals with perceived lower transaction costs are larger, though that difference is more precisely estimated 24 months (rather than 12 months) after the intervention.

D. Savings and Other Outcomes
Table 7 and table OA4 go beyond account opening and use and report the effect of the different treatments on various savings and welfare outcomes.We estimate equation ( 1) and include as a regressor the value of the dependent among individuals directed to the branch (the mean of the control group of 26.8% minus the coefficient on the dummy Information of 7.6% in col. 10 of table 5).In short, 0:346=ð0:2682 0:076Þ 5 1:80. 20The average accumulated deposits and withdrawals for individuals sent to the branch is computed as the average for the control group plus the coefficient of the dummy for Information.Thus, 31:8 1 7:4 5 39:2 USD for deposits and 35:9 2 11:57 5 24:33 USD for withdrawals.variable at baseline.Column 1 of table 7 reports the willingness to pay elicited at end line for a Sukaliku account with a balance of 1,500 CFA (US$2.8 at the time).Individuals encouraged to open a Sukaliku account with an agent were willing to pay US$1.08 more for the account, an increase of 14% relative to those sent to the branch, although the increase is not statistically significant at conventional levels.
Column 2 of table 7 reports whether the individual saved formally, either with Microcred or any other financial institution.Thirty-seven percent of those in the control group saved formally.The provision of information to open a Microcred account increased the probability that the individual saved formally by 13 percentage points (corrected p-value is .002).In addition, the farther away from the branch (and the agent, as distances are correlated) the individual lived, the lower the probability that the individual saved formally.While treatments were assigned randomly and we thus interpret the coefficients associated with treatment dummies as causal effects, distance to the branch was not randomly assigned, Note.Presented are results on account balance by using administrative data at the individual level.
The specification we use is the following, estimated by ordinary least squares: yi 5 a 1 b1Agent i 1 b2Informationi 1 b3DistBranchi 1 Xi 1 εi.Each observation corresponds to an individual i.Informationi is defined in table 5, and Agenti is defined in table 4. DistBranchi is the distance in kilometers between the household of the respondent and the closest branch.Xi is a vector of control variables evaluated at baseline, namely, a dummy for self-employment status, monthly income in US dollars, and an index for risk aversion.In cols. 1 and 2, we use as dependent variables the sum of all deposits and withdrawals, respectively, made within a year after the intervention, evaluated in US dollars of January 2015.In cols. 3 and 5, the dependent variables are, respectively, the total balance in the Sukaliku account 12 and 24 months after the individual opened it, evaluated in US dollars of January 2015.In cols. 4 and 6, the dependent variables are, respectively, the total balance in all Microcred accounts 12 and 24 months after the individual opened the Sukaliku account, also evaluated in US dollars of January 2015.Family-wise corrected p-values are reported in brackets below the coefficients (for details, see Westfall and Young 1993).* p < .10. ** p < .05. *** p < .01.
and so there may have been other factors correlated with distance that directly contributed to the lack of formal savings.For example, individuals that are farther away may be poorer and have less disposable income than those that live near the branch.Because provision of information led to more account openings but there is no significant additional effect on savings of encouraging the opening of accounts with an agent, we conclude that similar to the take-up of the Sukaliku account, our intervention succeeded in increasing access to finance through formal savings (regardless of the channel).
In column 3 of table 7, we use the log of the self-reported total amount of savings, excluding savings at Microcred, as our dependent variable.The mean of the dependent variable for the control group was roughly US$28 and is mostly composed of savings with Susu collectors.21Savings outside Microcred neither Note.Presented are results on savings outcomes by using data at the household level collected at end line.
The specification we use is the following, estimated by ordinary least squares: yi 5 a 1 b1Agent i 1 b2Informationi 1 b3DistBranchi 1 Xi 1 Y0i 1 εi.Each observation corresponds to an individual i.Agenti and Informationi are the treatment dummies defined in tables 4 and 5, respectively.DistBranchi is the distance in kilometers between the household of the respondent and the closest branch.Xi is a vector of control variables evaluated at baseline, namely, a dummy for self-employment status, monthly income in US dollars, and an index for risk aversion.We also include the dependent variable collected at baseline, Y0i.
In col. 1, the dependent variable is the individual's willingness to pay (WTP) for a savings account with 1,500 CFA (US$2.8 at the time).In col.2, the dependent variable is a dummy variable that takes the value of 1 if the respondent reports saving money with a financial institution.For those individuals who reported not saving with a financial institution at end line but who opened a Sukaliku account according to administrative records, we input a value of 1.In col.3, the dependent variable is the sum of savings balances at end line in financial institutions (excluding Microcred) and other informal sources such as Susu collectors.In col.4, the dependent variable is a dummy that takes the value of 1 if the respondent reported spending his or her savings to manage the reduction of income caused by a shock within the past year.Family-wise corrected p-values are reported in brackets below the coefficients (for details, see Westfall and Young 1993).*** p < .01.
increased nor decreased among individuals encouraged to open the Sukaliku account.That is, individuals encouraged to open the account with an agent had higher savings outside Microcred than those sent to the branch, but the increase is not statistically significant at conventional levels.Finally, in column 4 we study whether access to savings enables individuals to respond more easily to unexpected shocks in income, but we find no statistically significant results.Table OA4 reports the effect of opening a Sukaliku account on outcomes such as whether the individual is self-employed, had a business that failed, opened a business, the individual's total monthly expenses, and an index of happiness based on questions about positive feelings and thoughts during the month prior to the end line.Once we correct for multiple hypothesis testing, none of the effects of the Sukaliku account are statistically significant.To conclude, encouraging individuals to open the account through the provision of information had some positive effects on savings behavior, but the precise channel used to open the account, while affecting usage, did not affect welfare (Dupas et al. 2018 also report limited evidence of welfare effects from opening bank accounts).

IV. Discussion
If individuals use only the channel that minimizes transaction costs, they should use only agents once the individuals learn about the lower visit times at agents.Table OA5, however, shows that 42% of active account users still used both channels by end line. 22lthough we lack experimental variation to causally identify the precise mechanism, we provide in this section several explanations that could account for such behavior. 2322 While 55% of individuals encouraged to open an account with an agent visited both locations regularly, only 27% of individuals encouraged to open an account with a branch teller visited both locations. 23We note that after the third monthly withdrawal, clients transacting with the agent faced a fee for additional withdrawals, and as a result, the decision of how much to withdraw and where to do it could have been influenced by the fee.This alternative explanation cannot explain the different amounts per channel observed.First, given the use of the accounts described, only 3% of Sukaliku holders ever paid withdrawal fees.When amounts higher than the third transaction a month are removed, the patterns in the data are similar.Second, both deposits and withdrawals were lower compared with transactions at the branch, and yet, there are no deposit fees.Finally, individuals do not seem to avoid paying the fee because the third withdrawal at an agent in a month is similar in size to the fourth withdrawal at the agent that same month, even though the fourth withdrawal requires a fee.If individuals were not aware of the fee and learned about it after the fourth withdrawal, one could compare the fourth with the fifth withdrawal that same month.Figures OA6 and OA7 show the average sizes of withdrawals at an agent and at a branch, respectively, while figs.OA8 and OA9 show the average sizes of deposits for comparison.Figure OA6 does show a smaller fifth withdrawal compared with a fourth withdrawal in a month, but because there are only 10 such fifth transactions (out of 369 transactions) in a given month, we cannot reject that the fourth and fifth withdrawals at an agent are equal.Finally, using the history of transactions and their location in a given month, we The first explanation is based on trust.While data described in section II suggest that by the end line, branch tellers and Microcred agents were similarly trusted, individuals may have been initially unfamiliar with agents and thus needed to gain trust before they fully engaged with them.In the beginning, individuals would have made few transactions with a Microcred agent, going to a branch for the rest of their transactions until they gained enough trust to use the agent exclusively.Figures 3 and 4 plot the average number of deposits and withdrawals, respectively, per customer and quarter with the Microcred agent (solid circles) and at the branch (solid triangles).Figure 3 shows the number of deposits declined slightly over time at the branch but remained roughly constant at Microcred agents.Figure 4 shows that the number of withdrawals at either location is very similar and roughly constant.Relatedly, transactions at the agent would have been initially smaller (and larger at the branch) until find that among individuals making more than three withdrawals in a given month (with at least three at an agent), the probability that the fourth withdrawal is also made at an agent is 95%, indicating that most individuals that make at least three transactions at the agent in a given month will also do the fourth transaction at the agent.For the 10 individuals making more than four withdrawals in a given month (at least four at an agent), all did the fifth withdrawal at an agent.Put differently, there is no significant increase in the probability that individuals will go to the branch when they face a fee at the agent.clients became more comfortable with the agent.Figures 5 and 6 plot the sizes of deposits and withdrawals, respectively, and neither shows a clear trend over time at either location.There is thus no clear trend over time in the number or size of transactions. 24ore important, after 1 year there were still transactions made at the branch, even when individuals reported trusting the agent completely.The data are therefore not consistent with an explanation based on an initial lack of trust in the agents.
We now explore whether individuals chose the channel on the basis of the size of the transaction they planned to make by using the sample of treated individuals.Given the lower transaction costs of agents, we should expect more frequent and smaller transactions.We have shown in table 5 that individuals encouraged to open an account with a Microcred agent made more transactions with Microcred agents.Table 8, however, shows that these individuals did not necessarily transact smaller amounts.The average transaction for individuals sent to the branch versus an agent were US$108 and US$146, respectively.Thus, transactions were, on average, larger among individuals sent to a Microcred agent, and that difference is statistically significant.While the average transaction by those sent to an agent was relatively large, there is a clear distinction between the size of their transactions made at agents versus those they made at branches.Specifically, columns 1 and 2 of table 8 show that individuals encouraged to open the Sukaliku account with a Microcred agent chose to make significantly smaller deposits and withdrawals with the agent compared with those made at the branch.Their average transaction at the branch was US$213 compared with US$97 for transactions with the agent (corrected p-value is .00).In contrast, individuals encouraged to open the account at the branch transacted on average US$102 at the branch compared with US$119 at the agent (corrected p-value is .530).
An issue with the averages in panel A of table 8 is that because an observation is a transaction, individuals with more transactions end up receiving a larger weight.We address this issue in panel B by first averaging transactions at the individual level and then averaging across individuals.The results in panel B are qualitatively similar to those in panel A but because there are fewer observations, the differences are no longer statistically significant at conventional levels.
Why would individuals encouraged to open an account with an agent make larger transactions at the branch?There could be at least three reasons.First, agents may be unable to accommodate requests for large withdrawals because of lack of liquidity and may refuse large deposits because of concerns about security and fears of theft.According to the end-line survey, however, none of the respondents who had visited a Microcred agent reported that the agent had ever refused a transaction. 25Similarly, none of the six Microcred agents reported problems completing transactions during the last week. 26Perhaps more important, this explanation is not very convincing because we only observe a difference by channel in the size of transactions among individuals sent to the agent but not among those sent to the branch.It is rather unlikely that agents would systematically refuse transactions to one group of clients and not another.Alternatively, clients rather than the agents may feel insecure depositing or withdrawing large amounts with the agents for fear of theft.While each branch has a security guard, some Microcred agents do not have one.However, the selection process to become a Microcred agent ensures that the establishment has brick walls and concrete ceilings.Again, because we only see differences in transaction sizes among those sent to the agent, this explanation suggests that these individuals would feel more insecure at agents than those sent to the branch, which is unlikely because both groups report trusting agents as much as branch tellers, if not more.And in fact, none of the clients in the study ever complained about lack of security when dealing with agents.
Finally, clients may be reluctant to make large transactions with the agent because of privacy concerns.Transactions with an agent happen in plain view of other customers who are present at the agent's shop.These other customers present at the time of the transaction may be part of their social network.At the branch, in contrast, only the teller and the client witness the transaction.The client may therefore be concerned that at the agent, information about the transaction may reach his or her family and friends, while at the branch, transactions are kept confidential.Focus groups conducted with long-time Microcred account holders living in the study areas, but who did not participate in this study, reveal a concern with making large transactions (both deposits and withdrawals) with a Microcred agent because of lack of privacy.Those individuals feared that neighbors at the shop during a transaction or the agent himself would tell relatives and friends about the large transaction and that this could prompt pressure from them to share cash holdings. 27ut why would individuals sent to the agent be concerned about privacy and therefore choose the location on the basis of the size of the transaction while those sent to the branch would not be concerned?We speculate that it may take time and several transactions to appreciate the fact that neighbors present at the shop may tell others about the transaction and that agents may violate the privacy of their clients.By end line, individuals sent to the branch transacted fewer times than individuals sent to the agent, and thus for them, the probability that a large transaction had been made public to their friends and family was lower.In short, individuals sent to the branch may not have yet realized that agents may violate their privacy simply because they had not yet made unusually large transactions with the agent enough times. 28lthough we lack the statistical power to test for learning effects directly, we compare the average size of transactions using each channel by individuals who transacted more than the median with an agent compared with those who transacted less.We find that regardless of the treatment status, only the individuals that transacted more than the median with the agent appeared to choose the location on the basis of the desired amount of the transaction.For these clients who transacted often, the difference in transaction amounts by channel is statistically significant (corrected p-values in col.8 of table 8 are .012 in panel A and .364 in panel B).For clients who did not transact often, there is no difference in transaction amounts by channel (corrected p-values in col.8 are .968 in panel A and .924 in panel B).
The concern for privacy expressed in the focus groups is consistent with a growing literature based on lab-in-the-field experiments from Africa that document the willingness to pay to keep income gains private.Jakiela and Ozier  (2016), for example, find that making returns to investment public lowers the willingness to take riskier but more profitable investments, especially among women with relatives who also participated in the experiment.In the same study areas as ours, Boltz, Marazyan, and Villar (2016) find that 65% of subjects prefer to receive the gains from the experiment in private rather than publicly and that they are willing to forgo 14% of the winnings to keep them secret.
In sum, the patterns are consistent with the idea that individuals chose the channel depending on the size of the transaction not because Microcred agents refused certain transactions because of liquidity constraints or lack of security; rather, anecdotal evidence suggests that they were less willing to make large transactions with an agent because friends and family could find out about their finances.The evidence is thus consistent with a trade-off when banking with agents.They entail lower transaction costs (time, travel) but also less privacy as transactions are made "in the open."We acknowledge, however, that our experimental design did not have treatments based on the amount of privacy clients received during visits to agents (e.g., by varying the type of training some agents received so that their behavior would mimic more closely that of bank tellers), and so these issues are left for future research.

V. Conclusion
During the past decade there has been a proliferation of alternative channels to the traditional banking model to deliver financial services.These alternative channels received even greater interest during the COVID-19 pandemic as academics and policy makers explored cost-effective mechanisms to transfer money to individuals otherwise excluded from the formal banking sector.Agents are often viewed as a promising alternative for deepening financial inclusion.And indeed, since the conclusion of this study, the network of Microcred agents increased from 99 in early 2015 to 522 in early 2017, while the number of branches increased from 11 to 32 during the same time period (see Buri, Cull, and Giné 2023 for a review of issues related to networks of banking agents).
This paper implements a randomized controlled trial to compare the effects of agent banking with those of traditional branch banking in a context where prospective clients are equidistant from a branch and an agent of the partner financial institution.Naturally, the findings come from one institution serving customers from one country.And as with all empirical research, questions persist as to whether findings might differ in other settings, such as more remote rural areas, where individuals would typically be far closer to agents than to branches, and in other countries, cultures, or financial institutions.However, because the main advantage of agents for most clients in other contexts is their geographic proximity, by equating proximity our results could be seen as a lower bound on the benefits of agent banking to the average client.

Figure 2 .
Figure 2. Timeline of the experiment.

Figure 3 .
Figure 3. Number of deposits over time.Plotted is the average number of deposits per quarter made by active customers at a Microcred branch (filled triangles) and with an agent (filled circles) after opening an account.Confidence intervals are also reported.

Figure 4 .
Figure 4. Number of withdrawals over time.Plotted is the average number of withdrawals per quarter made by active customers at a Microcred branch (filled triangles) and with an agent (filled circles) after opening an account.Confidence intervals are also reported.

Figure 5 .
Figure 5. Size of deposits over time.Plotted is the average size of deposits (in log USD) per quarter made by active customers at a Microcred branch (filled triangles) and with an agent (filled circles) after opening an account.Confidence intervals are also reported.

Figure 6 .
Figure 6.Size of withdrawals over time.Plotted is the average size of withdrawals (in log USD) per quarter made by active customers at a Microcred branch (filled triangles) and with an agent (filled circles) after opening an account.Confidence intervals are also reported.
Table OA1 of the appendix tables OA1-OA5 are available online) shows that the 274 individuals surveyed are comparable to the 227 individuals not surveyed for 12 of 14 characteristics.There are statistical differences in a dummy for whether the individual saves informally and the Raven's test, and these are included in the analysis of panel A of table 4 (though the respective coefficients are not shown).12 11Half of the individuals in each of the two treatment groups received a monetary incentive of 1,500 CFA (US$2.8)onlyif the savings account was opened before a certain date and was transferred as balance to the newly opened account.This monetary incentive was covered by Microcred.(

TABLE 5 ACCOUNT
TAKE-UP AND NUMBER OF TRANSACTIONS i 1 b2Informationi 1 b3DistBranchi 1 Xi 1 εi.Each observation corresponds to an individual i.Informationi is a dummy that takes the value of 1 if the individual received a visit with an encouragement to open the account, while Agenti is a treatment dummy defined in table Family-wise corrected p-values are reported in brackets below the coefficients (for details, seeWestfall and Young 1993).
This table reports average amounts of both deposits and withdrawals made at a branch or at an agent.In panel A, the mean amount is taken over all transactions.In panel B, the mean amounts are computed by first averaging over transactions of a given individual then by averaging across individuals.Under "corrected p-value" we report the family-wise corrected p-values (for details, seeWestfall and Young 1993).