Can mobile money improve microfinance? Experimental. evidence from Uganda PRELIMINARY DRAFT - DO NOT CITE

Transcription

1 Can mobile money improve microfinance? Experimental evidence from Uganda PRELIMINARY DRAFT - DO NOT CITE Emma Riley Department of Economics, Manor Road Building, Oxford OX1 3UQ, UK ( emma.riley@economics.ox.ac.uk) July 31, 2018 Abstract I examine how changing the way microfinance loans are disbursed to utilise widespread mobile money services impacts the business and household financial outcomes of female microfinance borrowers. By keeping business funds separate from household funds both mentally and physically, mobile money may create behavioural impediments to acting in impulse and thereby facilitate saving and investment, while also serving to hide money from others. To test this, I randomly assigned 3,000 borrowers of BRAC Uganda in Kampala to receive either a mobile money account or a mobile money account and disbursement of their microfinance loan on this account. I compare these treatments to a control group who received no mobile money account. After 9 months, women who received their microfinance loan on the mobile money account had 16% high business profits and 19% higher levels of business assets. Women treated with the disbursement of their loan on a mobile money account hold balances of 14% of their savings on the mobile money account after 30 days and draw these down over 6 months. These results suggest that the manner in which borrowers receive their loan has an impact on how they save and invest it. I would like to thank my supervisors, Climent Quintana-Domeque and Stefan Dercon for their comments, advice and support. I would like the thank Mahreen Mahmud and Richard Sedlmayr for their comments and suggestions. I thank an anonymous donor for generous financial support of this research. The trial is registered at and the pre-analysis plan was uploaded there on 11th December 2017 before endline data collection had finished and analysis begun. An amendment to the pre-analysis plan documenting further intermediate outcomes based on admin data was lodged on the 31st July All analysis in this paper follows these pre-analysis plans unless clearly stated otherwise.

2 1 Introduction Microfinance loans are extremely popular in developing countries, with an estimated 200 million microfinance clients in Behavioural obstacles to saving are one potential reason for the large popularity of microfinance (Bauer et al., 2012). The poor find it hard to save incrementally and have shown a preference for lump-sum amounts either in the form of loans or from being a member of a ROSCA (Anderson and Baland, 2002; Afzal et al., 2014). In addition, the poor suffer relatively more from spending on temptation goods such as tea and sweets (Banerjee and Mullainathan, 2010), and demonstrate a preference for commitment devices which provide a small obstacle to access (Ashraf et al., 2006; Dupas and Robinson, 2013). Designing microfinance products to further take into account the behavioural obstacles the poor face could increase their benefits. Recent evidence showed that the benefits from microfinance loans have not been realised as hoped (Banerjee et al., 2015) and, in particular, investments did not lead to an increase in profits for female entrepreneurs or have other transformational impacts. However, women s businesses have been found to benefit from in-kind grants not cash, with the reason linked to self-control difficulties (Fafchamps et al., 2014). There is also evidence that women who are able to hide money from their spouse and family benefit from business grants and that pressure to share with family and give money towards a husband s business explains poor performance of women when given microfinance loans (Fiala, 2017; Bernhardt et al., 2017). This evidence suggests that providing loans in a way that provides a small obstacle to access, as opposed to as cash, would act as a commitment device and partially hide the money, reducing expenditures on temptation goods and pressure to share the money with family and spouses. Mobile money accounts, which are provided in an individual s name, can only be accessed or the balance checked by the individual and require the small barrier of going to an agent to withdraw money, can enable these behavioural and social obstacles to be overcome. In this paper I examine the impact of mobile money accounts on women s business and household financial situation using a Randomised Controlled Trial of 3000 female microfinance clients in Kampala, Uganda. I examine the impact of providing a mobile money account designated for their business to micro-entrepreneurs and the additional impact of also providing the microfinance loan on this account. This study individually randomised both existing and new clients of the NGO BRAC Uganda who applied for a new loan. All clients had an existing business. I find that providing a business-designated mobile money account, and also the microfinance loan on this account, leads to a 16% increase in business profits and an 19% increase in the value of business assets. I do not find any effects on savings or effects on any outcomes from just getting a mobile money account. This may be because over 95% of the sample had already used a mobile money account before at baseline and so simply designating an account for their businesses did 1

3 not have an impact. Looking at administrative data from the mobile money accounts provided as part of the two treatments, I find that those assigned to have their loan disbursed on the mobile money account make significantly more transactions on the accounts (2.4 on a mean in the mobile account treatment of 1.6) and hold significant balances on the account of 52,400 USH after 30 days, on average 14% of household savings. Examining this in more detail over time, those treated with disbursement of their loan on the mobile money account appear to hold some of the loan on the mobile money account and slowing draw this down over a 6 month period. By the end of 6 months there is no significant different between treatments in terms of the balances held on the accounts, and balances are on average a very low 190 USH ($0.05). This explains why by the 9 month point of the endline survey no differences are seen in terms of saving amounts between the two treatments and the control group. There are no differences between the mobile account only and loan disbursed on the mobile money account treatments in terms of deposits onto the accounts, again suggesting women are using the loan disbursement on the mobile money account as a way to safety and privately save some of the loan and draw on it as needed. This might also be explained by the fact that the majority of these women already having mobile money accounts. It is interesting that despite being prompted to use these accounts for their businesses, most women are primarily using the accounts as a way to save part of their loan if it was disbursed on the account, not for regular transactions. Only 11% of both treatment groups ever make a single deposit onto the account and there is no difference in this rate between the two treatment groups. Since the mobile account only treatment started with no balance on the account, and only a subset of them used the accounts, it is not surprising I find no effects from the mobile money accounts alone. This research builds upon the early work on mobile money services and the larger body of research examining how to make financial services meet the needs of the poor, particularly with regards to behavioural constraints. There is early evidence that mobile money accounts can serve as an alternative saving device to keeping money as cash or using a bank account (Blumenstock et al., 2015; Mas and Mayer, 2011; Mbiti and Weil, 2011). The evidence on mental accounting suggests that simply labelling something as a saving account can increase savings (Thaler, 1985, 1999). This research adds to this early evidence by showing that mobile money accounts can be used to facilitate business investment through labelling and the storage of funds on an account. Experimental evidence shows women are prepared to pay a high cost to hide money from family and friends (Jakiela et al., 2016). This has also been demonstrated through women having a preference for individual saving accounts (Schaner, 2015) and for improvements in female-decisionmaking power in the household when women are given access to their own saving accounts (Ashraf, 2

4 2009). This suggests that changes to the design of microfinance products to help women overcome social pressure and hide money (Fiala, 2017) could enable loans to have positive effects on borrower s businesses. The literature described above suggests that women in particular have difficulty using microfinance loans for their own businesses due to family pressure, and hence that providing loans on mobile money accounts, instead of as cash, could enable them to invest more of the loan in their business, possibly improving performance. Providing microcredit loans on a business designated mobile money account instead of as cash therefore acts as a nudge to using the loan for business purposes and helps to enable the borrower to use the loan for investment. The rest of this paper is organised as follows: Section 2 discusses the interventions and experiment design. Section 3 goes over the data used in this study. Section 4 contains the empirical specification and section 5 the results. Section 6 concludes. 3

5 2 Intervention and experiment Design 2.1 Setting The study location is Kampala, Uganda, which was chosen since it has both a high prevalence of microfinance borrowing and high mobile money penetration. The study took place in 6 microfinance branches of the NGO BRAC Uganda. BRAC Uganda is one of the largest providers of financial services to the poor in Uganda. This study took place at 6 of their microfinance branches in Kampala. BRAC offer microfinance loans to women of between 250,000 USH and 4mn USH ($70 - $1100) for expanding a small enterprise. Owning an existing enterprise is a pre-requisite for obtaining a microfinance loan, and a check of the business is carried out by credit officers before a loan is given. Loan durations varying between 20 and 40 weeks depending on the needs of the woman, with the interest rate set at 13% for the 20 week loan and 25% for the 40 week loan. Women apply for loans in groups of between 8 and 30 women, and they meet weekly with these women to repay their loans. While groups are not formally liable for repayment of their members loans, and women each have a guarantor from outside the group who is meant to repay the loan if a woman defaults, in practice a woman cannot get a new loans if other women from her group are in default. The subject population is composed of any microfinance client applying for a new loan (whether as a first time borrower or a repeat loan) who owns a mobile phone of her own. The mobile phone requirement was not often binding in this urban sample, and only 6 women were excluded from taking part in this study because they did not have their own mobile phone. 2.2 Interventions The study involved two interventions: Intervention One Women seeking a loan from BRAC were randomly offered a mobile money account designated for their business. Women were provided with a new sim card, helped in setting up their mobile money account and trained how to use it. The account was described as specifically for their business but no formal restrictions were placed on how they use the account nor money paid into the account. Women in this group continued to receive their microfinance loan as cash. Intervention Two Women seeking a loan from BRAC were offered the same business mobile money account as in Intervention One but, additionally, their microfinance loan will be paid directly into this account. These loans were paid through a mobile money provider and include an additional amount to cover 4

6 the fee of approximately 1% of the loan amount for withdrawing the money from an agent so as not to disadvantage women receiving the loan this way. A training fully explained this process so as to maximize take-up. 2.3 Experiment design The study involved 3,000 female micro-entrepreneurs, of which 1,000 act as controls receiving the microfinance loan in the usual way as cash and nothing else, 1,000 were signed up for a business designated mobile money account but still receive their loan as cash and 1,000 were signed up for the business designated mobile money account and receive their loan on that account. All other aspects of the BRAC microfinance loan product remained the same, including the requirement to be physically present at the branch for the disbursement of the loan and signing of final agreements and the repayment of the loans via weekly group collection meetings within the borrower s community. Randomisation took place weekly in blocks of women determined by the timing of requesting a new loan. All women who were both accepted for a loan with BRAC and who had a mobile phone were individually randomised into the treatment or control groups. This continued for approximately 5 months until the sample size of 3,000 was achieved. The randomisation was done in Stata. It was stratified by present bias and behaviour in a willingness-to-pay-to-hide-money game (see Section 3.4, first time borrower with BRAC, microfinance branch and also by business profits at baseline. The first two variables were chosen based on the idea that women who are present bias or show a desire to hide money from their spouse might benefit more from having their loan disbursed on a mobile money account instead of as cash. I stratified by first time borrower and branch in case there were systematic differences between these groups and to ensure an even amount of mobile disbursement by branch. I stratified by profit since Fafchamps et al. (2014) showed heterogeneous effects loans for women based on their profitability. For those assigned a treatment, the treatment was offered to the woman when she went to have her loan disbursed. At this point, if she was assigned to Treatment One she was offered a mobile money account and trained in how to use it. The account was framed as for her business but without any constraints. Women were free to refuse the account if they wanted. If she was assigned to Treatment Two, she was offered both the mobile money account and to have her loan disbursed on this account. She could refuse either the disbursement and/or the sim card, permitting partial compliance if she wanted the sim card but not the disbursement. The additional amount to cover fees was explained to the woman and the same training and framing as in Treatment One given. 5

7 3 Data A baseline survey was conducted on all women applying for a new loan at the 6 BRAC microfinance branches. Baseline surveys were conducted between January and June 2017 before randomisation and assignment to treatment group occurred. Approximately 1 week after the baseline survey, randomisation took place and a woman was disbursed her loan by BRAC in the assigned manner. Lists of treatment assignment were sent to the BRAC branches weekly, and only women who had been baselined and assigned a treatment could have a loan disbursed to them. This ensured that all women applying for loans during this 5 month period were part of the study. The endline survey began in October 2017 and ran until January This is approximately 8 months after the loan disbursement, and was chosen so that the women would still be repaying a 40 week loan when we went to conduct the endline survey with them, helping to reduce attrition. In addition to survey data, I have transaction records obtained from MTN Uganda of all the mobile money transactions between January 2017 and January 2018 made on the sim cards registered for the trial. All respondents gave their consent for the transaction records from these accounts to be used for the study. The mobile money account administrative data covers all transactions made using the mobile money accounts provided to clients as part of the study. This includes the type of transaction (including transfer, payment, cash-in, cash-out), account numbers for whom the transaction was from and to, date and time, amount, fee and balance on the account. The transaction records are available for both treatment groups but not the control group. 3.1 Balance test and summary statistics I confirm the validity of my randomisation by performing a balance test, results of which are shown in Table 1. None of the variables are significantly different across the 3 groups at the 10% level. Around 80% of the sample are taking out a second or further loan with BRAC and so are classed as repeat borrowers. The branches varied significantly in size, with 2 branches representing 50% of the sample. Looking at the game behaviour; 20% of the display hyperbolic preferences and 60% of them switch above the median in the hiding money game, meaning they are willing to give up at least $2 in order retain control over the money offered. Moving onto demographics; the sample was well educated with 80% of women completing primary school and 15% going onto complete secondary school. On average they are 35 with 3 other household members. Two-thirds of them were married and 20% had a job in addition to their business. The households earnt on average $274 a month and spent $245 a month. Nearly 90% had 6

8 Table 1: Summary statistics and balance test Mobile disbursement Mobile account Control mean sd obs mean sd obs mean sd obs p current client high profit hyperbolic hide money branch branch branch branch branch branch primary school secondary school household size age married job monthly income monthly consumption saves amount saved mobile account household asset value loan amount loan 40 weeks monthly profit calculated monthly profit reported All monetary amounts in 1000 USH and winsorised at the 99% level savings of on average $100. They owned nearly $1000 in household assets on average. 97% of women reported already having used mobile money before. The average loan was $380 and half the loans were for 40 weeks. Women reported making $120 a month in their businesses. 3.2 Take-up Since women were free to accept or reject the assigned treatment, take-up rates were a concern. However, the interventions had high take-up rates. 94% of the individuals assigned to the mobile money account (treatment one) received a mobile money account and 71% of those assigned to 7

9 receive a mobile money account and their loan on the mobile money account (treatment two) received this. Additionally, 14% of those assigned to receive a mobile money account and their loan on the mobile money account received only a mobile money account and their loan as cash (they were assigned to receive treatment two and got treatment one). The reasons for those assigned to treatment two getting treatment one were both refusal of treatment two but also problems completing mobile disbursement, such as power cuts or networks outages. Lastly 15% of women assigned to mobile disbursement refused the entire treatment (sim card and mobile disbursement). This is summarized in Table 2 below. Table 2: Treatment compliance mobile account mobile disburse Received mobile money account and loan as mobile money (71%) Received mobile money account and loan as cash 931 (94%) Refused mobile disbursement 51 (5%) Technical problem for mobile disbursement 88 (9%) Received no mobile money account (refused) (6%) (15%) Total (100%) (100%) 3.3 Attrition The survey team made a great effort to follow up with this highly mobile population of women. Even though the endline survey was only on average 8 months after the baseline, half the sample had taken loans of a shorter duration than this and so were not necessarily still attending their microfinance groups. Despite this 89% of the sample were found and re-surveyed for endline. Of these 25 refused to be surveyed and 292 couldn t be found. Attrition rates of approximately 10% are common in mobile populations such as this urban sample. However, of concern is whether treatment was correlated with attrition. I test for this in Table 3 and find no significant differences in attrition rates across treatment arms. 8

10 Table 3: Attrition (1) attrition Mobile account (0.014) Mobile disbursement (0.014) Constant 0.101*** (0.010) Observations 2,959 R-squared Standard errors in parentheses *** p<0.01, ** p<0.05, * p< Behavioural games In order to test the hypothesis that the women who benefit most from receiving the loan on a mobile money account are those who are most likely to give in to temptation goods or most subject to pressure to transfer money to others, incentivised games were played at baseline to elicit time preferences and willingness to pay to hide money from the spouse. The time preference game used were standard multiple price lists (Andersen et al., 2008), which have been used frequently in a developing country context (Ashraf et al., 2006). Individuals were asked to choose between a fixed monetary reward in one period and various larger rewards in a later period. The periods were either today and 2 weeks or 2 weeks and 4 weeks time. The near payment was fixed at $2 and the far payment varies between $1.8 and $8. One in five respondents was randomly chosen to be paid one of her choices from this game at the specified time period. The propensity to pay to hide money from others has been used as a measure of women s empowerment in the literature (Almas et al., 2015; Fiala, 2017; Mani, 2011). Here I expand upon the version used in Fiala (2017) by conducting a variant of the (Almas et al., 2015) game with multiple choices between whether the woman or her spouse receives set amounts of money the next day. Women had to make a series of 8 choices between receive a fixed amount of money themselves ($2) or having their spouse receive varying amount of money between $1.8 and $8. This set up allowed an approximate willingness to pay to be calculated for those who did switch to their spouse receiving as the mid-point of the difference between the payments on the choice she switched on and the previous choice. 9

11 One in five respondents was randomly chosen to be paid one of her choices from this game to either herself or her spouse tomorrow. Tomorrow was chosen to always be the payment date to remove effects of strong present bias and to allow the enumeration team time to contact and find the spouse if necessary. 10

12 4 Empirical strategy McKenzie (2012) showed that in the case of a single baseline and follow-up with an autocorrelation less than 0.5 (as is the case for business profits, saving and spending), power is highest when regressing an outcome measure at endline on baseline covariates, the treatment measure and the baseline value of the outcome measure. There are large power gains from using ANCOVA rather than a difference-in-difference specification. The study will therefore be analysed using an OLS regression of the form: Y i1 = α 0 + α 1 T 1i + α 2 T 2i + α X X i 0 + Y i0 + ɛ i1 (1) Where Y 1 is the outcome of interest, T 1 the mobile money account only treatment dummy, T 2 the mobile money account and loan on the mobile money account dummy, X a set of randomization strata dummies (Bruhn and McKenzie, 2009), Y 0 is the baseline value of the outcome (if measured at baseline, otherwise excluded) and ɛ random error for individual i. OLS estimation of the above regression will return the unbiased estimate of the Intent to Treat (ITT) effects, α 1 and α 2. To estimate the local average treatment effect, the above equation will be estimated where assignment to treatment is replaced with actual take-up, which is instrument by assignment, giving the two-stage least squares estimator. Because I am considering three primary outcome summary measures (profit, saving and business assets), I adjust the p-values of the coefficients of interest for multiple statistical inference by calculating sharpened q-values that control for the false discovery rate (FDR). These q-values correct for the fact that I conduct 3 tests across the 3 primary outcomes. Rather than prespecifying a single q, I report the minimum q-value at which each hypothesis is rejected, following Anderson (2008) and Benjamini et al. (2006). For some summary measures of outcome families, I will group several related variables into index variables following Anderson (2008). I will construct the indices in three steps. First, I will re-code all contributing outcomes so that higher values correspond to treatment effects in the same direction ( better outcomes). Second, I will standardize the individual outcomes using the baseline mean and standard deviation for that outcome. Third, I will calculate the average of the standardized constituent outcomes, weighted by the inverse covariance matrix. Where an outcome value is missing for a respondent, I will omit this outcome from the index construction. When looking at secondary and intermediate outcomes I do not correct for multiple testing as this analysis is informative for exploratory analysis of additional impacts and mechanisms analysis, not the main impact. When looking at additional outcomes I do not correct for multiple testing as this analysis is informative for robustness checks and understanding which components are aggregate and indexes might be driving any impacts. 11

13 4.1 Administrative data I also have administrative data collected from the mobile telecoms operator detailing the transaction records for all the mobile money accounts used in this study. This data is only available for the two treatment groups, not the control group, and hence will involve a comparison of being given the loan on a mobile money account compared to being given a mobile money account and the loan as cash. The administrative data will be analysed using an OLS regression of the form: Y i = α 0 + α 2 T 2i + +α X X i + ɛ i1 (2) Where Y is the outcome of interest, T 2 the mobile disbursement treatment dummy, X a set of randomization strata dummies and ɛ random error, for individual i. OLS estimation of the above regression will return the unbiased estimate of the Intent to Treat (ITT) effect, α 2. To estimate the local average treatment effect, the above equation will be estimated where assignment to treatment is replaced with actual active use of the account, which is instrument by assignment, giving the two-stage least squares estimator. defined as any transaction during the first 180 days of account opening 12

14 5 Results 5.1 Primary outcomes As outlined in my pre-analysis plan, the primary outcomes of this study are profits, savings and the value of enterprise assets (capital). The results for treatment effects using intent-to-treat Table 4: Treatment effects on primary outcomes (1) (2) (3) profit savings capital Mobile account (17.55) (42.83) (65.39) Mobile disbursement 64.75*** ** (17.77) (31.36) (67.07) Observations 2,601 2,630 2,601 Control mean Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Intent-to-treat estimates. 000 Ugandan Shillings Profits refers to the self-reported monthly business profit. Savings is individual savings held by the woman. Enterprise asset is the value of all assets the woman uses in her business. All outcomes are winsorized at the 99% level. All regressions include strata dummies and include the baseline value of the outcome. Differences in observations are due to women who shut down their businesses between baseline and endline estimates on those 3 outcomes are shown in Table 4. I find no effects from treatment with just the mobile money account but strong and significant effect when the loan is also disbursed on the mobile money account. Those in the mobile disbursement treatment experience a 15% increase in their profits and an 19% increase in the value of their business assets. There are no effects on saving amounts. These results are consistent with the hypothesis that disbursing the loan on a mobile money account increased the amount of the loan used to buy business assets and that these increased businesses assets lead to gains in profit. 13

15 5.2 Intermediate outcomes As outlined in the amendment to the pre-analysis plan, I look at intermediate outcomes based on administrative data collected from the mobile telecoms operator, MTN. Intermediate outcomes are shown in Table 5. I look first at the average balance over the first 30 days of account ownership. This takes the daily balance at the end of the day for every day of the first 30 after the account was given to the client. Secondly I look at the number of transactions on the account during the first 180 days. I cap transactions at 180 since the last mobile money accounts were given out in June 2017 and the administrative data ends in January A transaction includes depositing money, withdrawing money, transferring money to other mobile money accounts and paying bills. I do not include the loan disbursement as a transaction for the mobile disbursement group. Table 5: Treatment effects on intermediate outcomes (1) (2) Average balance 30 days Number of transactions Mobile disbursement *** 2.387*** (5.448) (0.423) Observations 1,925 1,925 Control mean Intent-to-treat estimates. Average balance in 000 Ugandan Shillings. Monetary outcomes are winsorized at the 99% level. All regressions include strata dummies. Control mean refers to the mean in the mobile account group. Average balance is the average end of day balance on the account for the first 30 days after the account was provided to the client. Number of transactions is the count of transactions over the first 180 days of account ownership. It excludes the disbursement of the loan onto the account for the mobile disbursement group. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 I find that treatment with the loan disbursed on the mobile money account compared to just providing a mobile money account results in average balances during the first 30 days being 52,400 USH higher compared to balances in the mobile account group of 1,599 USH. This is highly significant and tells me both that the mobile account group deposited little money onto the accounts and that the mobile disbursement group did not withdraw all the loan deposited on the account immediately, but kept some of the loan on the account for a period of time. The value of the money held on the account is large, comprising 5% of the average loan value and 14% of household savings. 14

16 Secondly I find treatment with the mobile disbursement results in the client performing 2.4 more transactions on the account compared to the mobile account group who performed 1.6 transactions on average. Again, this shows that the mobile disbursement group used the accounts more intensively than the mobile account group did. I examine both of these effects in more detail in Table 6 and Table 7. In Table 6, I show the average balance on the mobile money account for various time periods. These are the first 15 days, 15-30days, 30-45days, 45-60days, 60-90days and days. I also show the final balance from the last transaction before the 180 day cut-off. I show the summary outcome in the first column for comparison. Table 6 clearly shows that the average balance on the mobile money account is large and statistically significant for the mobile disbursement treatment. The average balance also declines over time, though remains significantly different than than mobile account treatment until the final balance. This indicates that microfinance clients treated with mobile disbursement are choosing to hold balance on their accounts, which they are slowly dipping into and running down over time. The mobile account group appears to initially deposit a small amount on the account, 2360 USH (approx $0.6) which then declines to around 1000 USH where it remains. Table 6: Treatment effects on intermediate balances outcomes (1) (2) (3) (4) (5) (6) (7) (8) Average Average Average Average Average Average Average Final balance balance balance balance balance balance balance bal ance Mobile disbursement Observations Control mean 52.40*** 80.92*** 21.91*** 10.82*** 9.40*** 7.34*** 6.04*** (5.45) (7.61) (4.00) (2.56) (2.44) (2.32) (2.26) (0.13) 1,925 1,925 1,925 1,925 1,925 1,925 1,925 1, Intent-to-treat estimates. Average balance in 000 Ugandan Shillings. Monetary outcomes are winsorized at the 99% level. All regressions include strata dummies. Control mean refers to the mean in the mobile account group. Average balance is the average end of day balance on the account the specified number of days after the account was given to the client. Final balance is the balance at the last transaction made within 180 days of account opening. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 15

17 Table 7: Treatment effects on intermediate usage outcomes (1) (2) (3) (4) (5) (6) (7) Number Ever Ever Number Number Total Total transac- de- with- deposit with- deposit with- tions posit draw drawals drawals Mobile disbursement Observations Control mean 2.387*** *** *** *** (0.423) (0.015) (0.018) (0.147) (0.302) (7.678) (30.38) 1,925 1,925 1,925 1,925 1,925 1,925 1, Intent-to-treat estimates. All regressions include strata dummies. Monetary outcomes are winsorized at the 99% level and in 000 Ugandan Shillings. Control mean refers to the mean in the mobile account group. All variables are defined over the first 180days after the account was provided. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Turning to Table 7, I show a series of variables capturing whether the mobile money account was used and how intensely. First I examine activity variables, whether the account ever had a deposit or withdrawal made into it during the first 180 days of use. Then I look at the number a value of deposits and withdrawals. Again I show the summary measure, number of transactions, for comparison. Table 7 shows that it is only for withdrawals that there are significant differences between the mobile account and mobile disbursement treatments. Mobile disbursement treated women are 59 percentage points more likely to make a withdrawal than the mobile account treatment. This seems reasonable considering they needed to withdraw the loan and 70% of them took-up the treatment according to the survey data. In the mobile account group, 12% of them ever make a deposit and 11% of them ever make a withdrawal. This shows the accounts had generally low levels of use, consistent with the lack of effects found on business outcomes. Similarly for the number of deposits and withdrawals, there is no difference between the treatments on deposits, which are low at half a deposit on average, but a large difference for withdrawals. On average, mobile disbursement treated clients make 2.3 withdrawals, significantly different from the mobile account mean of 1 withdrawal. The fact that withdrawals for the mobile disbursement 16

18 group are greater than 1 corroborate the finding that clients are leaving a balance on the accounts which they are slowly drawing down over time. Finally, the amount of deposits and withdrawals are the total value of all deposits and withdrawals. For the mobile disbursement group, the value of deposits excludes the loan being disbursed, whereas the withdrawals value captures the withdrawal of the loan. Again, deposits is not significantly different between the mobile account and mobile disbursement groups, at 24,000 USH (approximately $10). Total withdrawals for the mobile disbursement group are 900,000 USH, 450,000 USH less than the average loan size, again showing that money is being left on the accounts. This will not add up to the balance on the accounts due to fees paid on transactions 17

19 6 Conclusion This paper shows that the manner in which loans are disbursed to microfinance clients leads to significant differences in how those loans are used: clients who receive their loan on a mobile money account invest in 19% more business assets and as a result have 16% higher profits. They also hold significant savings equal to 14% of household savings on their account, which they draw down over a 6 month period. Neither treatment group used the accounts for regular deposits, with only 11% of either group ever making a deposit. This suggests the benefits to women s business from the mobile disbursement treatment come from a safe and private way to store the loan, saving it to invest when needed. With the increasing spread of mobile money services, this intervention is a low cost way to raise the benefits of microenterprise loans to women. 18

20 References Afzal, U., Adda, G., Fafchamps, M., Quinn, S., and Said, F. (2014). Two Sides of the Same Rupee? CSAE Working Paper Series. Almas, I., Carneiro, P., Attanasio, O., Armand, A., Attanasio, O., Carneiro, P., and Attanasio, O. (2015). Measuring and Changing Control : Women s Empowerment and Targeted Transfers. Working paper. Andersen, S., Harrison, G. W., Lau, M. I., and Rutström, E. E. (2008). Eliciting risk and time preferences. Econometrica, 76(3): Anderson, M. L. (2008). Multiple Inference and Gender Differences in the Effects of Early Intervention: A Reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects. Journal of the American Statistical Association. Anderson, S. and Baland, J.-m. (2002). The Economics of ROSCAs and Intrahousehold resource allocation. The Quarterly Journal of Economics, 117(3): Ashraf, N. (2009). Spousal Control and Intra-Household Decision Making : An Experimental Study in the Philippines Spousal An Experimental Study in the Philippines Control and Intra- Household Decision Making :. American Economic Review, 99(4): Ashraf, N., Karlan, D., and Yin, W. (2006). Tying odysseus to the mast: Evidence from a commitment saving product in the Philippines. Quarterly Journal of Economics, (May). Banerjee, A., Karlan, D., and Zinman, J. (2015). Six randomized evaluations of microcredit: Introduction and further steps. American Economic Journal: Applied Economics, 7(1):1 21. Banerjee, A. and Mullainathan, S. (2010). The shape of temptation: implications for the economics lives of the poor. NBER Working paper. Bauer, M., Chytilová, J., and Morduch, J. (2012). Behavioral Foundations of Microcredit : Experimental and Survey Evidence from Rural India. American Economic Review, 102(2): Benjamini, Y., Krieger, A. M., and Yekutieli, D. (2006). Adaptive linear step-up procedures that control the false discovery rate. Biometrika, 93(3): Bernhardt, A., Field, E., Pande, R., and Rigol, N. (2017). Household matters: revisiting the return to capital among female micro-entrepreneurs. Blumenstock, J., Callen, M., and Ghani, T. (2015). Mobile-izing Savings with Automatic Contributions : Experimental Evidence on Dynamic Inconsistency and the Default Effect in Afghanistan. pages

21 Bruhn, M. and McKenzie, D. (2009). In Pursuit of Balance: Randomization in Praction in Development Field Experiments. American Economic Journal: Applied Economics, 1(4): Dupas, P. and Robinson, J. (2013). Why Don t the Poor Save More? Evidence from Heath Savings Experiments. American Economic Review, 103(February): Fafchamps, M., Mckenzie, D., Quinn, S., and Woodruff, C. (2014). Microenterprise growth and the flypaper effect: Evidence from a randomized experiment in Ghana. Journal of Development Economics, 106: Fiala, N. (2017). Business is tough, but family is worse: Household bargaining and investment in microenterprises in Uganda. Jakiela, P., Ozler, O., and Ozier, O. (2016). Does Africa need a rotten kin theorem? Experimental evidence from village economies. Review of Economic Studies, (June). Mani, A. (2011). Mine yours or ours? The efficiency of household investment decisions: An experimental approach. Mas, I. and Mayer, C. (2011). Savings as Forward Payments:Innovations on Mobile Money Platforms. Forthcoming as Chapter 10 in Financial inclusion for poverty alleviation: Banking on the unbanked. Essam Yassin Mohammed and Zenebe Bashaw Uraguchi (eds)., (September):1 14. Mbiti, I. M. and Weil, D. (2011). Mobile banking: The impact of M-Pesa in Kenya. National Bureau of Economic Research. McKenzie, D. (2012). Beyond baseline and follow-up: the case for more T in experiments. Journal of Development Economics, 99(2): Schaner, S. (2015). Do Opposites Detract? Intrahousehold Preference Heterogeneity and Inefficient Strategic Savings. American Economic Journal: Applied Economics, 7(2): Thaler, R. (1985). Mental Accounting and Consumer Choice. Marketing Science, 4(3): Thaler, R. H. (1999). Menta accounting matters. Journal of Behavioral Decision Making, 12(3):