Access to Finance FoRuM Reports by CGAP and Its Partners No. 2, December 2011 68101 Latest Findings from Randomized Evaluations of Microfinance Jonathan Bauchet, Cristobal Marshall, Laura Starita, Jeanette Thomas, and Anna Yalouris 1 © 2011 Consultative Group to Assist the Poor/The World Bank All rights reserved. Consultative Group to Assist the Poor 1818 H Street, N.W. Washington, DC 20433 USA Internet: www.cgap.org Email: cgap@worldbank.org Telephone: +1 202 473 9594 Acknowledgments Our thanks to the following people who reviewed and gave helpful input on this paper: Lasse Brune, Erica Field, Nathanael Goldberg, Dean Karlan, Asim Khwaja, Meng Lu, David McKenzie, Jonathan Morduch, Jonathan Robinson, and Dean Yang, and from CGAP’s publications committee Tilman Ehrbeck, Alexia Latortue, Kate McKee, and Richard Rosenberg. Latest Findings from Randomized Evaluations of Microfinance I n 2009, the results from two microcredit impact in yield high returns. Microcredit is not transform- studies in Hyderabad, India, and Manila, the ing informal markets and generating significantly Philippines were released to mixed responses higher incomes on average for enterprises. And yet (Banerjee, Duflo, Glennerster, and Kinnan 2010; the industry has focused almost exclusively on the Karlan and Zinman 2011). Some media declared mi- rhetoric of entrepreneurship and has overlooked crofinance a failure (Bennett 2009). Many in the the many important benefits to households that are microfinance community dismissed these random- using loans to accelerate consumption, absorb ized studies as too limited to be a true reflection of shocks, or make household investments, such as in- the entire sector.1 vestments in durable goods, home improvements, These first randomized studies caused a sensa- or education for their children. tion because they challenged the dominant impact Combined with other evidence, randomized narrative for microcredit—a narrative that rests on evaluations are contributing to an emerging body of loans to capital-constrained microentrepreneurs knowledge that is creating a new narrative around who earn a steep return on marginal capital and how financial services for the poor really work. As thus can repay a relatively high interest rate and re- the results from new studies have been released, invest to grow out of poverty—and the way in which the discussion has evolved, and randomized evalua- that narrative had been universalized in the popu- tions are being used to examine when particular lar imagination. In fact, the results were more nu- products and designs work, for what segments of anced. What the microcredit studies really showed people, and why. is that this model of microcredit works for some Today researchers are using randomized tech- populations—those who successfully grow busi- niques to better understand the underlying finan- nesses—but not for others. cial services needs of poor clients and what impacts Many now agree that the expectations for micro- are achieved when appropriate financial services credit in the popular discourse were overblown. are offered. Building on evidence from earlier non- For some, the pendulum had swung: far from a pan- randomized studies, researchers are increasingly acea against poverty, some argued that microcredit able to work with microfinance providers to apply was actually doing harm. The evidence supports these techniques to product innovation and to neither extreme view. In fact, the results of the tweak product design. In this way, randomized studies aligned with and confirmed some of the evi- techniques can make a significant contribution to dence from nonrandomized methods already in the the field by clarifying our understanding of precise- microfinance research literature that found modest ly how, and under what conditions, financial ser- but neither revolutionary nor deleterious impacts vices benefit poor people.2 (See Box 1.) from credit. While the concept of capital that will Poor households clearly have other financial allow poor people to unleash small business oppor- needs that go beyond working capital loans to mi- tunities remains valid for some poor clients, not ev- croentrepreneurs. They use a variety of informal ery borrower is a microentrepreneur—take-up rates and semi-formal mechanisms to cope with risk, for credit products are often surprisingly low, and not all economic activities that poor people engage 2. Naturally, not all settings are appropriate for randomized eval- uations. This paper does not discuss such methodological issues in detail, but it does share findings from settings where 1. See, e.g., Helms (2010). randomized evaluations were feasible and illuminating. 1 Box 1 Frequently Asked Questions about Randomized Evaluations Why are randomized trials considered the most it is not a specific number of studies that will allow researchers rigorous method of impact evaluation? to assert a theory to make predictions about what happens in In the case of evaluating a microfinance program, if we simply other places. The only way to generalize findings is to conduct compare clients to nonclients we are comparing two differ- a series of careful replications to evaluate similar approaches in ent types of people: those who choose to borrow or save, different contexts until a clear pattern emerges. and those who do not. The ones who choose to participate likely have different business acumen, tolerance for risk, and Why do researchers measure results after only one other characteristics, and studies have shown that they can or two years? Microentrepreneurs need much more be wealthier than nonclients—even before joining a microfi- time to establish their businesses or build up savings, nance program. By randomly assigning access to financial so we should perhaps not expect to see impact on services, randomized trials ensure that the only difference, on poverty within such a short timeframe. average, between clients and nonclients is access to the This is a limitation of randomized evaluations, as well as some program. Therefore any difference between the groups can other evaluation techniques. Most randomized evaluations of be confidently attributed as the impact of the program. microfinance programs measure results after two years or less, in part, because practitioners are typically reluctant to refrain Can results from randomized trials be generalized? from serving a control group for longer than a couple years. Randomized trials help to establish causality. But they do However, follow-up studies may be employed to estimate this only for the particular context of the evaluation (i.e., what long-run effects, and where practical, researchers may go back we learn from Kenya may or may not apply to Vietnam). and resurvey households after a longer period. This is a limitation of all types of evaluation, however. Ran- domized trials are no more or less vulnerable than other Why are randomized trials so expensive? methods. Randomized evaluations are not more expensive than other An increasing number of credible studies, using both types of evaluations. The high cost from many impact evalua- qualitative and quantitative techniques, are helping to build tions comes from collecting data, such as household surveys, a body of knowledge about how financial services work. But to measure household welfare, whether randomized or not. Note: For a short history of randomized evaluations, see http://www.povertyactionlab.org/methodology/when/when-did-randomized-evaluations-begin seize opportunities, manage the risks and incon- appropriately, the impact should nonetheless be veniences that come from having uneven cash welfare-enhancing. flows, and smooth household consumption. They Recent evaluations are helping us understand use credit or savings to pay school fees, they save when access to the appropriate product is welfare- to invest in businesses, and they use health and enhancing, and when it is not. And the emerging crop insurance, when available, to stave off risk. body of evidence is both promising and highly prac- While these uses of financial services are differ- tical, allowing practitioners to think more creative- ent from the uses initially anticipated, they are ly about options that will help their clients. While still valuable, and the ability to manage finances microcredit in the India study showed no discern- is a fundamental part of everyday life for all peo- ible impact on measures of health, education, and ple (Collins, Morduch, Rutherford, and Ruthven female empowerment, it led to more businesses be- 2009). The financial services needs of poor ing created and enabled poor households with busi- households may require different product fea- nesses to change their spending patterns. Further tures, and perhaps different payment and deliv- randomized evaluations of other products have ery structures, but when such needs are met been far more positive on welfare impacts. A study 2 conducted in Kenya shows that access to savings sponds to access to specific new financial services accounts for female market vendors allows them to against how a comparable group fares without keep higher levels of inventory and therefore have those services. (See Box 2.) This paper goes back a higher incomes. Consumer credit was shown to couple of years to the first studies that used this ap- have significant welfare benefits for wage earners proach, and summarizes a series of research studies in South Africa. A study conducted in Ghana pro- presented at the October 2010 Microfinance Im- vides evidence that rainfall insurance helps farmers pact and Innovation Conference in New York. use more fertilizers and increase their cultivation These studies evaluated product design for a range area, and results in fewer meals missed for the fam- of financial services, including credit, savings, and ily. (See Annex 1 for a summary of the research pa- insurance. The studies discussed here were under- pers discussed in this paper.) taken by research affiliates of Innovations for Pov- Perhaps one of the greatest contributions from erty Action (IPA), the Financial Access Initiative these first randomized evaluations of microcredit (FAI), and the Abdul Latif Jameel Poverty Action will be to help reset expectations. Far from offering Lab (J-PAL) at the Massachusetts Institute of Tech- the last word on the impact of microfinance, the ex- nology; they are all randomized evaluations unless isting evidence instead offers a foundation for otherwise specified. learning what works, for whom, and under what Part 1 of this paper reviews the main results from circumstances so that products and delivery ap- randomized evaluations that measure the impact of proaches can be better used and adjusted to meet microcredit and microsavings on business invest- the needs of poor people. ment and creation, consumption, and household This paper summarizes the latest research find- well-being. Part 2 presents evidence from evalua- ings from a new body of empirical evidence that tions of products and delivery design. Part 3 dis- uses randomized evaluations, similar to those used cusses the evidence on microinsurance products. in medical trials, to compare how one group re- 3 1 Pa r t Evaluating the Impacts of Microcredit and Microsavings 3 M any poor families in the developing world ularly poor women, they argue, microcredit has have limited access to formal financial the potential to increase households’ health and services, including credit, savings, and in- education, empower women, and reduce poverty. surance. They instead rely on a variety of informal credit relationships with moneylenders, relatives, What does the evidence say? friends, or merchants. Poor people also use a num- Recent experimental evidence from three random- ber of informal savings devices—for example, they ized impact evaluations suggests that while increas- may participate in rotating savings associations or ing access to credit does not produce the kind of dra- keep their savings at home. These options are not matic transformations conjured in the popular ideal. They tend to be unreliable, and it can be hard imagination, with millions of poor people springing to protect savers from the demands of relatives and out of poverty simply by taking out loans and apply- friends.4 Traditionally, banks and other formal fi- ing them to their microbusinesses, it does appear nancial service providers, such as insurance compa- to have some important—though more modest— nies, have not considered the poor a viable market, outcomes for some people. These include creating and penetration rates for formal financial services new businesses and tipping consumption away from in developing countries are extremely low.5 temptation goods, such as alcohol, tobacco, and Increasing access to financial services holds the snacks, so that households can invest in their busi- promise to help reduce poverty and improve de- nesses or buy more durable goods. This suggests velopment outcomes, by enabling the poor to that microloans help some households reprioritize smooth consumption, start or expand a business, their expenditures and smooth consumption—a cope with risk, and increase or diversify house- valuable function for poor households that suffer hold income. Microcredit stands to benefit poor from irregular and unpredictable income streams. individuals who lack collateral, steady employ- The results of these randomized evaluations find ment, verifiable credit history, or other require- little, if any, evidence of impact on use of healthcare,7 ments necessary to gain access to formal credit. In education, or female empowerment within the the past three decades, access to credit has ex- treatment period (Banerjee, Duflo, Glennerster, and panded dramatically. Now with nearly 200 million Kinnan 2010; Karlan and Zinman 2009; Crépon, De- borrowers, microcredit has been successful in voto, Duflo, and Parienté 2011). The groups that ben- bringing formal financial services to the poor.6 efited most from increased access to credit tended to Many believe it has done much more. By putting be men with relatively high incomes, not those typi- money into the hands of poor families, and partic- cally targeted by microfinance institutions (MFIs) (i.e., poor female entrepreneurs). Significant welfare 3. “Impacts� in this context refers to the effect that access to fi- benefits were found in a study that extended con- nance has on the well-being of poor people, as indicated by sumer credit to wage earners who were considered business income, household income, household consumption, health, children’s schooling, and other measures. marginally creditworthy because it enabled them to 4. Collins, Morduch, Rutherford, and Ruthven (2009) show that withstand shocks and keep their jobs. poor people use a variety of informal mechanisms to manage One evaluation of the impact of access to formal cash flow, cope with risk, and seize opportunities. They also find that at almost every turn poor households are frustrated savings for businesses in Kenya found increased by the poor quality—particularly the low reliability—of the in- struments they use to manage their meager incomes. 7. Microcredit did help families deal with health shocks, but it 5. See CGAP and the World Bank (2010, p. 4). did not lead to greater expenditure on healthcare or better 6. According to the Microcredit Summit, as of 31 December 2009, health outcomes for children. 3,589 microcredit institutions reported reaching 190,135,080 clients (Reed 2011, p. 5). 4 Box 2 Randomized Approaches to Measuring Impacta To evaluate the impacts of microfinance, researchers use ran- interested in opening a new branch, the MFI randomly selects domized techniques to assess how the lives of people in a some areas for opening new branches. Areas not selected for the program changed compared to how their lives would have opening of a new branch make up the comparison group. changed if the program had not existed. Simply comparing clients with nonclients cannot account for the fact that those Randomizing access at the margin who sign up are likely to have both observable and nonob- A second approach, used by Karlan and Zinman in South Africa servable characteristics that make them not comparable to (2010) and the Philippines (2011), is to randomize access to nonclients. Randomized assignment, whereby one group or credit among clients that the lending institution has identified individual gains access to a particular service while another as being marginally creditworthy. Applicants for loans are sort- group or individual does not, allows researchers to compare ed into groups based on a credit scoring mechanism that mea- two statistically equivalent groups. sures business capacity, personal financial resources, outside Existing evaluations follow one of two approaches: ran- financial resources, personal and business stability, and demo- domizing MFI branch placement in new areas, or randomiz- graphic characteristics. Those with high scores are automati- ing loan approval for marginally creditworthy applicants. cally approved, and those with low scores are automatically rejected. From the group that falls in the middle and is scored Randomizing MFI branch placement borderline creditworthy, applicants are randomly assigned a The approach used by Banerjee, Duflo, Glennerster, and Kinnan loan. This allows researchers to compare outcomes for groups (2010) in urban India and Crépon, Devoto, Duflo, and Pariente who received a loan with those who were denied credit. It also (2011) in rural Morocco is to partner with an MFI and randomize provides the lending institution with a way to judge what differ- the placement of new branches offering services. From a pool ence approving more risky loans might make to its business so of areas identified by the MFI as being places where it would be that it can fine-tune its approval threshold.b a. For more on randomized evaluation methodologies, see Bauchet and Morduch (2010) and Duflo, Glennerster, and Kremer (2008). b. For a discussion of the advantages and disadvantages of this approach, see Karlan and Zinman (2011). business investment and personal income growth domly selected for the opening of a new MFI branch among women, suggesting that savings could be an offering loans to self-formed groups of six to ten effective tool to help the poor accumulate funds for women. The typical loan averaged Rs. 10,000 investment or consumption. (US$200), for families where the average monthly expenditure was Rs. 5,000 (US$100) for a family of five (Banerjee, Duflo, Glennerster, and Kinnan 2010). Credit Twelve to 18 months after the introduction of an MFI branch, a comprehensive household survey was Impact of Grameen-style group lending conducted in a random sample of eligible households in an urban setting in both treatment and comparison areas. Demand for Starting in 2005, Banerjee, Duflo, Glennerster, and the credit product was not high: take-up was 18.6 per- Kinnan (2010) conducted the first randomized im- cent among households in the treatment group, 8.3 pact evaluation of expanding access to credit in a percentage points higher than in comparison areas. new urban market. Researchers partnered with People with access to microcredit were more Spandana, one of the largest and fastest growing likely to have started a business. The probability of MFIs in India, to identify 104 slums in Hyderabad as starting a business increased by 1.7 percentage places where Spandana would be interested in open- points relative to comparison areas, implying that ing new branches. Fifty-two communities were ran- approximately one in five of the additional MFI 5 loans in treatment areas was associated with the Impact of Grameen-style group lending opening of a new business. Beyond the impact on in a rural setting new business creation, there was no significant ef- In 2006 and 2007, Crépon, Devoto, Duflo, and Pari- fect on average business profits, monthly revenues, ente (2011) conducted the first randomized impact inputs spending, or number of employees. evaluation of microcredit in a rural setting. While Access to credit did not change the amount house- there are some differences, the results show some holds spent significantly, but researchers did find a notable parallels with the Spandana study. change in how households spent. Those with an ex- Al Amana, a Moroccan microcredit institution, isting business bought more durable goods for their opened 60 new branches serving 81 rural districts home and business. Households that did not start a that had no previous access to formal financial ser- business consumed more nondurable goods. But vices. Taking advantage of this expansion, research- those who started a new business cut back on tempta- ers selected two similar villages at the periphery of tion goods (tobacco, alcohol, tea, betel leaves, gam- each district and offered group loans of 1,000 to bling, and food consumed outside the home) and in- 15,000 DH (approximately US$124–1,855) to one vested more—tightening their belts to make the most randomly selected village, while the other village of the new opportunity.8 This switch from temptation would be served two years later, after the outcomes goods to investment and durable consumption in the of both groups had been compared. groups with businesses is an encouraging finding. After two years, loan take-up was fairly low. Only No evidence was found to suggest that micro- 16 percent of people borrowed from Al Amana, and credit was empowering women, at least along mea- many used loans to pay off existing debt. Similar to sured dimensions, such as exercising greater control the results found by Banerjee, Duflo, Glennerster, over how the household spent its money. Research- and Kinnan (2010) in urban India, there was no in- ers also found no evidence of improved indicators crease in consumption and no noticeable welfare for the use of healthcare services or education.9 improvements. Researchers did not find any evi- While media reports interpreted the lack of pos- dence that access to credit helped absorb income itive results along measurable dimensions of health, shocks. Fourteen percent of households experi- women’s empowerment, and education as signs enced health shocks, while 25 percent experienced that microcredit was a failure,10 Banerjee and Duflo shocks to business,11 yet there was no evidence that say this study presented clear evidence that micro- consumption decreased less for people with access credit was working along the dimension it was sup- to microcredit, as we would expect to see if access posed to. The new businesses created and the shift to credit helped families cope with financial shocks away from small “wasteful� expenditures implied (Crépon, Devoto, Duflo, and Parienté 2011, p. 16). that access to loans enabled households to make Contrary to what was found in India, the number clear choices to reprioritize, invest, and make the of new businesses did not increase in rural Morocco most of the new opportunity: “The main objective as a result of the loans, even though there was a lot of of microfinance seemed to have been achieved. It activity in the sample in terms of businesses starting was not miraculous, but it was working� (Banerjee and finishing. For individuals with existing farming and Duflo 2011, p. 171). activities, access to credit increased the volume of activity: more employees were hired from outside 8. Spending on temptation goods was reduced by Rs 9 per capi- the household, and sales, expenses, and profits in- ta per month (Banerjee, Duflo, Glennerster, and Kinnan 2010, p. 19). creased. In the case of livestock activities, most of 9. Households in treatment areas spent no more on medical and the expansion can be explained by higher savings sanitation items (e.g., medicines, soap) than comparison house- (livestock accumulation). There were some minor holds, and among households with children, households in effects on sales but no effect on profits. Animal hus- treatment areas were no less likely to report that a child had a major illness in the past year. There was no significant increase bandry also increased, and loans were used to diver- in levels of spending on school tuition, fees, and other educa- tion expenses or on school enrollment of teenage children (Ba- 11. Esther Duflo presentation at Microfinance Impact and Inno- nerjee, Duflo, Glennerster, and Kinnan 2010). vation Conference in New York, October 2010. 10. See, e.g., Bennett (2009). 6 sify the types of animals raised, increasing the asset access to credit did not increase investment in their value of the livestock. On the other hand, micro- business and they reduced their overall number of credit had no effect on nonagricultural businesses. business activities and employees. Subjective well- Those with an existing business at the start of the being slightly declined. study reduced consumption (presumably as they ex- But access to credit helped borrowers cope with panded their business) and considerably increased risk, strengthened community ties, and increased savings. But for those without prior business activi- their access to informal credit. Karlan and Zinman ties, consumption increased. These changes in con- (2011) conclude that microcredit may work, “but sumption patterns are similar to those of the Hyder- through channels different from those often hy- abad study (Banerjee, Duflo, Glennerster, and Kinnan pothesized by its proponents … and that start with 2010). The findings suggest that microcredit is an the household rather than with the business.� Ac- opportunity that different people will take advantage cess to credit lowered the demand for other kinds of of in different ways—whether because of disposition risk mitigation tools, a similar result to a study Kar- or circumstances. More evidence could help us un- lan and Zinman (2010) conducted in South Africa, derstand the factors that affect a person’s ability to where wage earners with access to consumer credit make good use of loans. were more able to absorb shocks, and therefore more likely to keep their jobs. Impact of individual microcredit loans in a peri-urban setting Impact of consumer credit Karlan and Zinman (2011) published the first ran- Karlan and Zinman (2010) worked with a consum- domized impact study to evaluate access to individ- er finance company in South Africa in designing an ual microcredit loans in the Philippines. Research- experiment to estimate the effects of expanding ers worked with First Macro Bank, a for-profit consumer credit to low-income workers in South lender offering small, short-term, uncollateralized Africa.13 The lender had operated for over 20 years credit with fixed repayment schedules to microen- as one of the largest, most profitable consumer trepreneurs on the outskirts of Manila. While the lenders in South Africa, offering small loans at high study focused on microentrepreneurs, the average interest rates, frequently to low-income workers income and education level of these customers is who have no collateral and must make payments on somewhat higher than that of traditional micro- a fixed schedule. Just over half of the sample of 787 credit borrowers.12 The bank used credit-scoring loan applicants who had narrowly failed to qualify software to rate applicants based on business capac- under the normal underwriting criteria was offered ity, personal financial resources, outside financial standard loans of US$127 (equivalent to 40 percent resources, and personal and business stability. Some of the median borrower’s gross monthly income) at applicants scored well above the bank’s base re- a 200 percent annual percentage rate (APR).14 quirements and some scored well below, but there The results were quite striking. Expanding ac- was a marginal group that just barely failed to meet cess to credit increased borrower well-being. Six to the bank’s criteria for lending. For the study, a num- 12 months after taking out the loan, incomes were ber of the 1,601 sample of marginally creditworthy higher for applicants in the treatment group, and applicants were randomly approved for a loan of applicants in the treatment group were more likely around 10,000 pesos (US$220), equivalent to 37 per- to have kept their jobs than those in the comparison cent of the average borrower’s net monthly income group. Twenty-six percent of treated households (Karlan and Zinman 2011). reported an improvement in food consumption. Eleven to 22 months later, the researchers found Subjective measures of decision-making within the that even though they borrowed more, those given household, community status, and overall optimism were also higher. In addition, the creation of a cred- 12. Households in the study had incomes averaging 5,301 pesos/ it history increased the probability of future loan month/household member (US$106/month/person, or about US$3.50/day). The Manila borrowers were far above the official Philippines poverty line of about 1,000 pesos/ 13. Income averaged about $300 a month. month/person, and well above the Hyderabad slum dwellers 14. APR describes the interest rate for a whole year (annualized). in the Banerjee, Duflo, Glennerster, and Kinnan (2010) study, In this case, effective APR is used (the fee plus compound in- who earned about $20/month/person. terest rate calculated across a year). 7 approval in the sample by 19 percent over a 15 to 27 Several categories of expenditures were also month horizon. All these outcomes were measured higher for women in the treatment group. Food ex- well after the loan had been taken out and repaid. penditures were 10–20 percent higher, suggesting that income had increased. Daily private expendi- tures were also 27–40 percent higher. This latter Savings 15 result also suggests higher income, though another possible explanation is that women were better To study the effects of savings constraints on the able to shield their income from others, thus spend- poor, Dupas and Robinson (2011) worked in collab- ing a higher share of their income for themselves oration with the Bumala village bank in Kenya to and their children. randomly provide small business owners with ac- Savings accounts also seemed to make women cess to savings accounts. somewhat less vulnerable to health shocks, which The accounts offered no interest on deposits and were particularly common in this sample. The log- included substantial withdrawal fees. There was books showed that women without savings ac- nonetheless high demand for these costly savings counts were forced to draw down their working strategies, which suggests that the available alter- capital in response to illness. In contrast, female natives were worse.16 savers did not have to reduce their business invest- The potential savers were market vendors, bicy- ment levels when dealing with a health shock, and cle taxi drivers, and self-employed artisans who did were better able to afford medical expenses for not already have a savings account, but were inter- more serious illness episodes. ested in opening one. The researchers had them The study suggests that the bank accounts of- keep daily logbooks with detailed information on fered were effective in increasing savings by over- business investments, expenditures, and health coming pressure on market women to share their shocks.17 From this information, Dupas and Robin- cash with others. Putting money into formal ac- son were able to examine the impact of the savings counts seemed to reduce the risk of appropriation accounts along a variety of dimensions. by relatives, friends, and neighbors.18 However, the Data from the bank showed that many women sample size of this study was too small to be defini- used the accounts quite intensively. For example, tive, so future work will be needed to explore how 25 percent of women saved more than 1,000 K Sh robust this finding is. (US$14.28) in the accounts, a substantial amount Despite the lack of evidence for positive effects given daily income of about $2 per day. Some on welfare from credit, the studies so far offer tanta- women saved much more. These savings translat- lizing evidence that there could be important poten- ed into other positive outcomes. Four to six tial benefits for some poor households to be gained months after account opening, women in the by helping the poor reprioritize their expenditures. treatment group had 45 percent higher daily in- Notably, the impact study for savings showed posi- vestment in their businesses than women in the tive outcomes for female savers. While it is still too comparison group. These findings suggest that early to reach any definitive conclusions, particu- women faced significant barriers to saving, and larly for savings where there is just one existing im- those constraints were important for the business- pact study with a small sample size, the findings give es they run. There was no measurable impact for researchers cause to explore further, and more stud- men in the study. ies are underway to see if these findings hold up in other contexts. The next generation of studies is ex- amining product design to see how small changes 15. In Part 2 we consider the impact of savings accounts com- bined with commitment devices that address the issue of self- can improve outcomes for poor clients. control among farmers. 16. From the whole sample only 8 percent of respondents re- 18. We use the term “formal� for consistency with the original fused to open an account; 39 percent opened an account but research paper, but the savings accounts offered by Bumala never made a deposit. village bank would often be called “semi-formal� since 17. Dupas, Karlan, and Robinson are currently replicating this Bumala village bank is not regulated by the Central Bank of study in four different settings: Chile, Malawi, Uganda, and Kenya. It is affiliated with K-Rep bank and has private de- the Philippines. posit insurance. 8 2 Pa r t Evaluations of Product Features— Design Matters A s the headlines and bloglines buzz with dis- screening and peer monitoring. Women, it is be- cussion of whether or not microfinance—or lieved, pay back their loans more reliably than men more precisely, microcredit—works, several (Armendariz and Morduch 2007). (Lending to wom- evaluations have started using randomized ap- en also supports the social mission of many MFIs, proaches to explore specific questions around since women are more likely than men to be poor, product design. What would be the impact of offer- and income in the hands of women is more often ing flexible repayment options, of allowing for grace spent to benefit the household and the children.)19 periods, or of replacing group liability with individ- And weekly repayment that begins right after the ual loans? This ongoing work reveals that small de- loan is given decreases credit risk by creating imme- tails matter, sometimes enormously. diate discipline and a pattern of repayment. This section reviews a series of recent studies Simply put, the model works for MFIs. But does that isolate specific features or attributes of prod- it work for borrowers? ucts to show how even small changes in their de- Increasing evidence suggests that some of these sign can yield significantly different results. These key design features may be far from optimal and studies offer insights for how financial service pro- may actually bring negative trade-offs. Some recent viders can tweak or improve their products to ben- studies look in turn at group liability, the effects of efit poor and low-income clients. We begin with lending to women, the importance of timing for en- some studies that have explored variations on tradi- suring repayment, and emerging tools for lenders to tional microcredit products and, indeed, challenge assess and monitor the credit worthiness of clients. some core tenets of the microcredit movement. We Collectively, these results provide insight into im- then discuss the latest evidence on the effect of portant product design options that may be used to commitment savings, reminders to save, and ac- improve financial outcomes for poor clients. count “labeling.� Questioning group liability Group liability has been at the center of the peer Microcredit Design— pressure model, which assumes that borrowers will Disrupting Tradition choose members they know to be reliable. Yet there are some very real disadvantages of group liability. Targeting women, group liability, and weekly re- If an emergency leads a group borrower to default, payments that start immediately have long been her social and community support system can un- considered defining attributes of a classic micro- ravel with it. credit model that is particularly strong throughout Beginning in 2004, Giné and Karlan (2011) ran a South Asia and in some other places. Some provid- study with Green Bank in rural areas of the Philip- ers see them as keys to success in keeping default pines to explore whether group liability was in fact rates close to zero. necessary for managing default risk. The study ex- Especially in the early days of the modern micro- amined what happened when the bank switched its credit movement, each of these features was seen as existing group liability model to an individual liabil- key to reducing the risk for the provider of uncollat- ity model, as well as when groups of new borrowers eralized lending, allowing many MFIs to operate as started out with individual liability loans. sustainable, even profitable, businesses. Group liabil- ity ensures repayment by enlisting the benefits of 19. See Thomas (1990), Engle (1993), and Schultz (1990). 9 The results showed that the shift to individual As mentioned, the vast majority of microcredit liability did not negatively affect loan repayment for programs nominally extend loans for the purpose of either group. The bank also saw an increase in out- starting or running a business. Business loans are reach, as more customers, attracted by the individ- seen as addressing a critical need, since formal- ual liability option, sought loans from the bank. The sector jobs are scarce in poor communities and study was extended to new areas, in which groups poor, unemployed women often do not have the either formed initially as individuals or as groups. necessary capital available all at once to invest in in- Here, too, no difference in repayment was observed, ventory or equipment to start a business or make although the credit officers were more reluctant to necessary investments for growth. Giving women open up lending groups without individual liability. credit, cash, or business inputs theoretically re- Given how such results likely rely heavily on cul- lieves capital constraints and helps them take busi- tural context and institutional incentives, these re- ness opportunities. sults should not be extrapolated without caution, In practice, however, access to capital does not but they do provide cause for challenging the pre- seem to be having as large an effect on increasing sumption that group liability is a key to successfully women’s incomes as development experts had lending to poor people. thought. The three microcredit impact studies con- Strengthening the case against group liability for ducted in India, the Philippines, and Morocco MFIs is the continued low demand for formal mi- showed that increasing the availability of credit had crocredit. As mentioned in the microcredit impact no impact on the profits of women-owned busi- studies described earlier, poor people are not nesses (Banerjee, Duflo, Glennerster, and Kinnan pounding down the doors of microlenders,20 de- 2010; Karlan and Zinman 2011; Crépon, Devoto, spite widespread, documented use among the poor Duflo, and Parienté 2011). of informal loans from friends, neighbors, or mon- A 2008 study by de Mel, McKenzie, and Woodruff eylenders.21 One possible reason why so few poor on returns to capital for businesses in Sri Lanka people take out formal loans is that the group liabil- found that the average real return to capital was 5.7 ity model repels risk-adverse individuals who are percent per month—substantially higher than the not willing to co-sign for their peers.22 market interest rate—and the returns varied with measures of ability, household liquidity, and the gen- Women, men, and returns to capital der of the owner. In a follow-up study (2009) the MFIs’ focus on lending to women is partly a conse- researchers show that women-owned businesses quence of commercial interest, given women’s earned no returns from either cash or in-kind grants, higher loan repayment rates. Development re- compared to men in the study who earned high re- search also suggests that women tend to put more turns from both. These results could have been en- of their earnings back into the home or into services tirely explained by environment, however, given for their children (health, education, etc.) than men that only 35 percent of women participate in the do.23 Serving women, therefore, is good for business work force in Sri Lanka and may choose low-return and good for fulfilling a social mission. sectors for their businesses (World Bank n.d.). To test whether the results held in an environ- ment with higher female participation, McKenzie 20. Loan take-up from MFIs was only 16 percent in rural Mo- and Woodruff partnered with Fafchamps and rocco (Crépon, Devoto, Duflo, and Parienté 2011) and 18.6 Quinn from the University of Oxford to study fe- percent in urban India (Banerjee, Duflo, Glennerster, and male and male entrepreneurs in Ghana, a country in Kinnan 2010). 21. See Collins, Morduch, Rutherford, and Ruthven (2009). which 74 percent of women participate in the 22. Context may make a big difference to the effects of the group workforce (World Bank n.d.). In Ghana, the re- lending approach, and not all of the literature points in the searchers gave either cash grants or grants of in- same direction. For example, a World Bank study (Carpela, Cole, Shapiro, and Zia 2010) exploits a natural experiment kind inventory or equipment to different male and and shows benefits of group lending. female entrepreneurs, to see whether cash had a 23. See, e.g., Engle (1991). different effect than in-kind capital, and whether 10 women responded differently than men.24 They ucts to acknowledge what many already know— did, on both counts. that loans are often used for nonbusiness purposes.25 Cash grants to female entrepreneurs in Ghana produced no return on capital, just as in Sri Lanka. The role of timing—delaying repayment Yet in Ghana, the in-kind gifts of inventory or Growing a business, no matter the size, often re- equipment showed a significant average return for quires entrepreneurs to make investments and then women. The researchers found that when given wait for those investments to mature. Yet the inflex- cash, women invested less of the gift in the busi- ible nature of the typical microcredit programs, in- ness, splitting pieces off for household purchases volving weekly, or monthly, repayments that begin or other expenses. They also found an important the first week or month after the loan is given, may nuance: the high returns from in-kind gifts came not provide the necessary time for investments to entirely from the women who had larger, higher show a yield. Many loan recipients, in fact, set aside profit businesses at the outset. Women with below- part of the loan from the beginning to ensure they average profits (around $1 a day) saw no benefit in can make the first two or more payments. So clients terms of profit from either form of grant. Male are not investing the full bulk of the funds, and they business owners, on the other hand, saw significant may be avoiding investments that require a longer returns to capital from both the in-kind grants and period to yield returns. the cash grants. When Field, Pande, Papp, and Rigol (2011) These findings from Ghana are certainly more looked at small business loan design in the United encouraging for female microentrepreneurs than States, they saw that business loans build in a grace the earlier findings from Sri Lanka. But even in period of a few months between when the funds are Ghana, it was only the larger female-owned busi- given and when the borrower has to begin paying nesses that benefited in terms of profit. Women the loan back. Between 13 percent and 15 percent of from the general population are not always, nor, in- U.S. business borrowers default, compared to be- deed, more likely to be, able to convert capital into tween 2 percent and 5 percent of microcredit bor- profits, and men tend to be more successful overall. rowers in developing countries, a significant in- These results suggest opportunities to adjust to crease in default risk for the lender. Yet the key whom MFIs lend and how they structure their question—does increased repayment flexibility cor- products. MFIs may have a greater impact on the relate with increased profit and still allow the lend- women they serve if they can filter their applicant er to manage default risks adequately—is important pool to identify and target high performers. It is rel- enough from the development perspective to war- evant not only for knowing which clients can excel, rant examination. but also for the MFIs’ ability to offer more flexible In West Bengal, India, the researchers (2011) products. Knowing what the client is likely to earn compared the outcomes of two groups of micro- can allow lenders to adjust the risk profile—and the credit borrowers with the Village Welfare Society. interest rate. It even may allow institutions to add One group received a traditional group microcredit microequity to their product portfolio, assuming product with semi-weekly payments that started they can find effective ways to accurately monitor immediately after receipt of the loan, and the sec- business performance. It may also be time for mi- ond group was awarded a two-month grace period crofinance providers to redesign their loan prod- before repayment began. 24. Fafchamps, McKenzie, Quinn, and Woodruff (2011) offer 25. As noted, in practice poor clients are not only using loans to grants instead of loans because many banks require that their invest in businesses, but also as a means to manage their clients already be business owners or have an idea for a start- household cash flow, for emergencies, and to smooth con- up that the banks deem worthy; likewise, entrepreneurs who sumption. While theory of impact is quite different, the use take loans may be more willing to take risks. Both factors po- of microcredit or savings for consumption smoothing may tentially create a study population that is more savvy or more nonetheless be important for the overall well-being of likely to be successful than the average population, so they clients. work with existing entrepreneurs and provide them with grants as a technique for preventing bias. 11 The grace period group members invested 6 per- derabad to identify some shared characteristics cent more of their loans in their businesses than among those individuals in their sample who were borrowers who received no grace period, and two more likely to start a business, but they have not years after the loans were given, those grace period tested whether that information predicts successful borrowers saw 30 percent higher average profits. use of a loan when used as a selection tool. Household income was also higher on average for Creating that selection tool is a high priority for the grace period borrowers. Khwaja, of Harvard’s Kennedy School. Khwaja’s However, the average result masks significant work focuses on the developing world’s small variation within the grace period group. The 25 firms—enterprises that have outgrown microcredit, percent average profit increase came about because but that still lack the collateral and the size to easily some of the women did extremely well with the de- secure financing from a mainstream bank. These layed payment loans. Unfortunately, big wins for businesses typically find it very difficult to grow some were matched by big losses for others—9 per- past the micro level for lack of investment capital. cent of the individuals in the grace period group ul- This absence of small, formal firms is known as timately defaulted on their loans, compared with a the “missing middle,� and it is a problem, not only for 2 percent default rate among individuals with the the high-potential poor who have the ability to grow standard weekly repayment structure. but lack the necessary capital, but also for poor en- In 2008, the Village Welfare Society participated trepreneurs who have trouble increasing their in- in a study measuring the effects of weekly versus come from self-employment. High-potential micro- monthly meetings on loan repayment (conducted entrepreneurs need financing, and the banks that by Feigenberg with Field and Pande). When the re- fund them need an inexpensive and reliable way to searchers found that the monthly meetings did sift through a pool of candidates and pull out those not affect repayment, the bank switched to month- with the highest potential for success. ly meetings, as the operational savings were sub- The challenge is not insignificant. The banks and stantial. venture capital firms that typically provide business These results again suggest significant opportu- financing screen ideas for their business value and nities for both high-functioning borrowers and the the entrepreneur’s ability to pay back by delving institutions that serve them. Banks could commer- into credit or business histories or conducting an cialize business loans with a two-month grace pe- in-depth evaluation of the business idea. These op- riod for all borrowers who want it by increasing the tions are not viable with microfirms, however, be- interest rate on those loans sufficiently to make up cause of their small size and smaller predicted re- the losses from default. It is not clear how much turns. For small firms, banks really need to know such high rates would affect demand. about the ideas, skills, and trustworthiness of the Another, more nuanced, approach is to identify individual borrower. who the high-potential borrowers are before set- Khwaja focused on the potential of automated ting product terms. Such individualized service of- psychographic evaluation tools for measuring an fers the possibility of creating a targeted product— entrepreneur’s ability and honesty. The psycho- whether loan or microequity—built around that graphic test is based on tools used by human re- person’s potential earnings, and tailoring loan source departments in developed countries. These amount, term, and price accordingly. tests are prevalent in other contexts, they are diffi- cult to game, and the results tend to correlate with The role of the borrower—client screening entrepreneurial success. as a product design tool To test their appropriateness for funding high- Financial service providers would be well-served potential microbusinesses, Khwaja has conducted a by any technique or tool that would allow them to number of tests around the world to see if the psy- predict in advance who the high performers might chographic tools work to identify high potential en- be. Banerjee, Duflo, Glennerster, and Kinnan (2010) trepreneurs with good ideas, strong business ability, delved into their microcredit impact data from Hy- and honest character. Khwaja and his colleagues 12 developed a 30–40 minute computerized psycho- metric identifiers. Applicants in the study answered graphic test to measure the test taker’s intelligence; questions about their business, their past borrow- implicit, practical business skills; and psychology or ing experiences, and their households, and they character (Is he honest? How does she view the were given a presentation on the importance of world, etc.?). To date, more than 2,000 entrepre- maintaining a clean credit history to ensure future neurs in seven countries have taken the test. They access to credit. Some of the borrowers then had have had different levels of experience, and they their thumbprint recorded and were given a further have sought loans of varying sizes (from $2,000 to demonstration of how the print would be used to $150,000). Khwaja’s pilot data show that the test identify them in the future. meets or exceeds the predictive ability of credit Data collected at the beginning of the study were scoring models used in developed countries, and it used to identify borrowers predicted to be high- effectively predicts financial success for micro or risk, based on their probability of business success small business entrepreneurs who do not have fi- and likelihood of repayment. As a result of the fin- nancial histories. gerprinting intervention, borrowers predicted to be The test is also uncovering some nonintuitive in- least likely to repay showed a significant change in dicators of business failure. For example, test takers behavior. Fingerprinted borrowers in this group who scored higher for intelligence actually achieve took smaller loans when they knew they could be lower profits; honesty also correlates with lower identified and were more likely to repay their loans than average profits—in both cases, these effects on time as well as eventually, compared to equiva- were stronger for women than men. The indicators lent borrowers in the comparison group. of success seem more obvious. Individuals with Fingerprinted borrowers in the high-risk group strong drive do much better, and those with busi- also allocated more of their land to the production of ness skills do moderately better than the average. paprika (the crop that the in-kind loan was intended to finance) and invested more inputs, such as fertil- Making the borrower do the work izer, in the paprika crop. In addition to improving the Khwaja’s approach puts the onus on the lender to repayment performance of high-risk borrowers, be- extract and evaluate that information and use it to ing in the fingerprinting system may have further make a lending decision. Giné, Goldberg, and Yang benefits for well-performing borrowers if their good take a different approach. credit histories can be stored and used to access bet- Giné, Goldberg, and Yang (2011) evaluated the ter borrowing conditions from other institutions. impact of improving the lending institution’s abil- ity to withhold credit from past defaulters and re- ward good borrowers with expanded credit on Savings Design borrower behavior. Their study focuses on paprika farmers in rural Malawi, where group liability and As early as 1999, Rutherford showed that poor peo- frequent repayments are impractical, since crop ple are active money managers: they look for ways failure usually affects everyone in a region, and to “save up� (to create a usefully large sum of money farm income from this cash crop arrives all at by storing it somewhere) or to “save down� (taking once. Likewise, there is no central identification a loan and repaying it later out of future savings). system in Malawi, so borrowers who default have Given that the poor do save, why don’t they use little problem accessing future loans, either by us- those savings to finance business investments? ing a different name or seeking financing some- Experts agree it might be hard for poor individ- where else. Together, these factors make it diffi- uals with variable income to get together enough cult for the lender to use loan access as an incentive money to start a business, but running a business to encourage repayment—the customer knows de- theoretically should not require further outside fault will likely have little consequence. funding, given that many microenterprises earn The researchers sought to improve the lender’s high returns. Ananth, Karlan, and Mullainathan ability to identify borrowers through the use of bio- (2007) conducted a survey that showed it would 13 not take much for vegetable market sellers in In- Commitment savings dia—who usually finance the purchase of daily in- Commitment savings accounts are one of the prime ventory with loans from the moneylender—to save innovations that have come out of recent efforts to a very small amount from the business every day, help poor people save. Commitment savings ac- equivalent to the amount needed to buy a cup of counts require the saver to deposit a certain amount tea. Within 28 days, those market sellers would of money in a bank account and relinquish access to have saved the same amount that they borrow ev- the cash for a period of time—usually until a certain ery day. At this point, they would no longer need date or until a certain dollar amount has accumulat- cash from the moneylender and could instead use ed. Such lack of access is valuable as a way of protect- savings for inventory purchases, thus saving even ing the cash both from the impulses of the savers more because they don’t have to pay the money- themselves, and from the hands of family and neigh- lender’s high interest rates (Ananth, Karlan, and bors. Ashraf, Karlan, and Yin (2006) conducted a Mullainathan 2007). Yet the vegetable vendors do study on commitment savings accounts in the Philip- not do it. Ananth, Karlan, and Mullainathan tried a pines that show they are effective at increasing sav- number of different techniques to nudge the mar- ings, especially for people with self-control issues. ket sellers to use savings for their businesses. For More recent studies have examined how a com- instance, they tried giving vendors a “top up� grant mitment savings product helps farmers to adopt the that restored savings after an emergency. They also use of fertilizer and invest more in their crops. As offered “financial literacy� training that taught the shown in randomized evaluations, farmers can earn vendors about the compounding effects of in- much stronger yields from their crops when they creased savings and decreased interest rate pay- take small steps, such as using fertilizer at specific ments, under the theory that the vendors did not points during the growing season. Duflo, Kremer, fully understand how much the moneylender was and Robinson (2008) show that, among maize farm- costing them. Yet nothing seemed to change the ers in western Kenya, the annualized return to ½ typical practice of frequent borrowing from the teaspoon of fertilizer at top dressing (when the maize moneylender. plant is knee high) was almost 70 percent per year. Psychology offers a number of theories for why Despite this evidence, few farmers consistently people do not save enough for productive invest- use fertilizer, largely because they earn all their in- ments, despite having the apparent means to do so. come for the year at harvest and do not have suffi- One theory suggests that some individuals simply cient funds left over to buy it at planting. In a fol- value the present more, and therefore prefer spend- low-up study, Duflo, Kremer, and Robinson (2010) ing available funds immediately rather than saving show how a simple commitment product can in- them. The future is unknown, so they don’t see crease fertilizer use. A field officer visited farmers much use in considering it. Another possibility is immediately after harvest and offered them an that people want to save, but self-control issues opportunity to buy a voucher for fertilizer, at the make it difficult for them to resist the temptation to regular price, but with free delivery. The results use extra cash today rather than save it for tomor- showed that free delivery early in the season in- row. Limited attention can also explain the lack of creases fertilizer use by 47–70 percent. savings as people fail to foresee the need for cash in To benchmark this effect, a second treatment the future. Last, there is the reality that not-quite-as- group was made the same offer of free delivery later poor individuals may receive a lot of pressure from in the season, while a third was offered a 50 percent friends and family members to share any (relative) subsidy later in the season. If farmers are complete- windfalls or help on a day-to-day basis pay for re- ly rational, then the effect of free delivery later in curring or emergency expenses, which eat away at the season should be the same as earlier, and the ef- savings. Innovations in savings product design fect of the subsidy should be greater. However, the therefore aim to help savers overcome one or more effect of the commitment device was greater than of these challenges. offering free delivery, even with a 50 percent subsi- dy on fertilizer, later in the season. 14 Based on that knowledge, Brune, Giné, Goldberg, commitment savings accounts had a higher impact and Yang (2011) estimated the impacts of facilitat- for wealthier households, a subgroup that may be ing access to a savings account coupled with a com- under more pressure to share. The existence of the mitment device as a mechanism to encourage sav- commitment device may have allowed farmers to ings among cash crop farmers in Malawi. The credibly claim that money was inaccessible. evaluation allowed farmers to put funds into a spe- cial account where withdrawals were restricted for Reminders to save—making tomorrow defined periods. The idea was to help farmers to real today buy inputs by better dealing with self-control prob- Beyond commitment savings, there are other inno- lems and cash demands from their social network. vations in savings product design that try to combat In the study, farmers in the treatment group were the tendency to spend, rather than save, limited re- randomly assigned to receive assistance in either sources, by making the future seem more real and opening an ordinary savings account or opening an relevant. These design innovations try to call the ordinary account with a commitment device. saver’s attention to her long-term goals, based on the The results of the evaluation showed the com- theory that people get distracted by the everyday mitment treatment had a large positive effect on the and need help remembering, and properly prioritiz- amounts of deposits and withdrawals made imme- ing, the future. For example, a number of efforts to diately before the planting season. On average, the promote savings accounts ask the savers what goals net effect on deposits (savings balance) was positive they have for their savings, and then find ways to although not statistically significant. Along with in- regularly remind them of that goal. One program creasing savings previous to the planting seasons, had savers bring a representative photograph of the commitment device also had positive effects in their goal—the new bicycle she wanted, for example, terms of a number of outcomes of interest. Farmers or the new cook stove. The bank then created puz- in this treatment group had a 26 percent increase in zles with the pictures and gave the saver a piece of agricultural input use, 22 percent increase in value the puzzle every time she made a deposit (Karlan, of crop output in subsequent harvest, and 17 per- McConnell, Mullainathan, and Zinman 2011). cent increase in household total expenditure re- Four recent studies have examined the effects of ported in the past 30 days. Farmers who had access two different approaches to making the present to only the ordinary account showed lower or non- more salient to savers: one approach is to use re- significant impacts in terms of those same out- minders, the other is to offer “labeled� accounts. comes, suggesting the commitment device played The studies on savings reminders took place in an important role for these results. Peru, Bolivia, and the Philippines, where savers Commitment savings accounts seem to help this were sent either letters (in Peru) or SMS text mes- community of farmers less by increasing their self- sages (in Bolivia and the Philippines) reminding control than by shielding funds from an individual’s them to save. Karlan, McConnell, Mullainathan, social network (for better or worse).26 The study and Zinman (2011) varied the messages to test the data show that actual amounts saved in the ac- effects of different wording. Some savers received counts were very low, ruling out that it helped indi- generic messages that said simply that they should viduals with self-control problems by restricting save; others received messages that referenced a their options to spend. Additionally, study partici- specific purchase that the saver said she wanted to pants who were identified as having self-control make with her savings. problems experienced no different effects from the The studies found that reminders increased av- commitment savings than their peers. Instead, the erage savings balances overall by 6 percent, but this impact increased substantially, to 16 percent, for the 26. Note that this important finding points to a tension between Peruvian savers when the reminder referred to a individual well-being and that of the community. It could be purchase goal. In the environments where SMS that the introduction of commitment savings devices work well for those who take them up but could also harm mem- text reminders were employed and automatically bers of their social network. executed, the cost of employing reminders to save 15 is very low, making text reminders a highly cost- A lack of financial literacy and basic accounting effective way to increase savings. skills offers one hypothesis for why the poor as a Account labeling offered an even greater return whole do not experience significant income gains for study participants in eastern Ghana.27 People when given access to formal credit—perhaps they have long used the technique of “labeling� to allo- do not have the skills to compare the likely returns cate funds for different purposes, and such labeling of different investments and account for them ac- can be highly effective at protecting the funds allo- cordingly. Yet there is little concrete evidence on cated for, say, the rent from being siphoned for oth- the effects of business or financial management er purposes. Some use mental tallying to allocate skills training on poor entrepreneurs. Karlan and the funds; others literally place different amounts Valdivia (2011) found positive impacts of a business in different jars or envelopes. With this approach in training program for microcredit borrowers in a mind, existing clients of the Mumuadu Rural Bank study they conducted in Peru. Yet a paper by Cole in eastern Ghana were asked about their savings and Shastry (2009) on U.S. participation in savings goals; some were given the opportunity to open and investment markets showed that the education separate, parallel savings accounts labeled “educa- level and cognitive ability of the participant corre- tion,� “business,� “housing,� or some other category. lated with positive gains, but that financial literacy The study found that savers eligible to open parallel training had no effect. accounts saved 31 percent more on average than With the jury still out on the value of providing those in the comparison group, with the greatest ef- financial and business training, Drexler, Fischer, fect seen for the education label. and Schoar (2011) tested whether financial literacy The strong effects of commitment savings, text training can improve business outcomes for small reminders, and account labeling show that small businesses in the Dominican Republic. The re- design changes can help poor people save more, and searchers tested the effect of two different sets of in some cases, to leverage those savings for positive content: one focused on traditional, principles- income-generating purposes. based accounting rules taught in the curriculum of- fered by organizations, such as Freedom from Hun- ger and BRAC; the other taught simple accounting Financial Product Plus: Improving rules of thumb, which essentially amounted to in- structing the business owners to keep personal and Results with Improved Skills business accounts separate. Many MFIs use the weekly or monthly repayment The researchers found that the business owners meetings they hold with their clients as an opportu- who received the rules-of-thumb training applied nity to teach some other relevant skill. No discus- sound accounting principles more often than their sion of financial product design would be complete peers. For example, they were more likely to keep without commenting on these add-ons as a func- their business and personal cash and accounts sep- tion of product design. arate, they were more likely to keep records, they Add-on services—sometimes referred to as “mi- were more likely to calculate their revenues, and crofinance plus�—vary significantly in their focus they were less likely to make mistakes when report- and goals. Some of these programs provide useful ing any of their results. Those who received the secondary skills, such as health education, with the rules-of-thumb curriculum also earned more reve- goal of helping customers avoid or lessen the im- nue than their peers, especially during “bad� weeks. pacts of disruptive events on income and savings. Though the researchers are careful not to extend Others aim to equip customers with business or their findings outside the specific group studied, financial management skills that they can use to the results seen from rules-of-thumb education in improve income generation. this one study suggest that less may well be more when it comes to training poor business owners in 27. Study by Karlan, Osei-Akoto, Osei, and Udry (forthcoming). sound financial practices. 16 3 Pa r t Microinsurance and Household Decision-Making P oor people face an enormous amount of risk (Karlan and Morduch 2009; Banerjee and Duflo in their lives. A major effort is currently un- 2011). Rainfall insurance pays a set amount when derway to expand access to insurance prod- rainfall, as measured by a local weather station, is ucts that improve upon traditional risk-sharing lower or higher than established thresholds. Be- arrangements and informal insurance networks to cause rainfall is not under the control of insurance help poor households deal with weather shocks clients, their behavior does not influence the possi- and irregular income from agriculture. In theory, bility of a payout (moral hazard is eliminated). microinsurance—insurance targeted to the poor Rainfall insurance is also simpler and cheaper to through low premiums and/or low coverage lim- administer than many other types of insurance; be- its—should be in strong demand to act as a safety cause rainfall is a public event, insured households net for poor families whose crops may fail, whose do not need to file claims, and insurance companies livestock may die, and who may suffer from the ef- do not need to spend time and resources verifying fects of bad weather and health shocks. This section the validity of claims.28 introduces some encouraging research results on in- Rainfall microinsurance holds promise to help novations in microinsurance. Recent findings sug- households reduce their exposure to risk, and may gest that microinsurance has positive impacts on modify farmers’ incentives to invest in riskier but poor households, but persistent low rates of take- more profitable crops or varieties. But take-up rates up, even for effective products, show that product remain puzzlingly low. design matters tremendously. Impact of rainfall microinsurance on Design matters household decision-making The difficulty in designing good insurance products Giné, Menand, Townsend, and Vickery (2010) mea- partly comes from problems of information asym- sured two different types of possible impacts of metry—when one party to a contract holds more rainfall microinsurance in Andhra Pradesh, India: information than the other. For one thing, people how well does having insurance help farmers cope who are insulated against risk may behave differ- with an agricultural shock (in their case, a drought), ently than they would have if they were fully ex- and how does having access to insurance affect posed to the risk (what insurers call moral haz- household decision-making, even in the absence of ard). Second, higher risk individuals may be more a claim? Preliminary results indicate that insurance likely to buy more insurance, which would not does not increase the use of inputs or change the matter as long as the insurer could charge that in- allocation of land, although having access to rainfall dividual higher premiums to cover the risk. But if insurance does cause farmers to shift toward more the insurance company is unable to identify the risky, rain-sensitive crops, which typically provide higher risk individuals, it responds by increasing higher profit. the premium for everyone. Both these information In an ongoing project in Ghana, Karlan, Osei- asymmetries can push up premiums and contribute Akoto, Osei, and Udry (forthcoming) focus on how to low take-up of products. No perfect design solution has been found to 28. One potential drawback in the design of rainfall insurance is eliminate issues of asymmetric information, but the possibility of a gap between the amount of rainfall mea- sured at the weather station and the actual losses suffered by some types of risk should be easier to insure than clients at their precise location, particularly if the two are far others. Rainfall insurance stands out among these from each other. 17 rainfall insurance helps rural households improve (2004), for example, showed that husbands and their farm decision-making, in particular whether wives fail to insure each other perfectly when a lack microinsurance can help lower the risks of agricul- of rainfall affects the yield of crops grown exclu- ture production and counter farmers’ risk aversion. sively by one or the other. In a survey in Andhra The study couples rainfall insurance with cash Pradesh (Cole et al. 2011), 89 percent of households grants in four randomly selected groups of farmers: reported that drought is the most significant risk in one group, farmers receive both insurance and they face. Asked why they do not purchase insur- subsidy, another group receives insurance only, an- ance, less than 25 percent of the surveyed house- other group receives capital only, and a final group holds (and as low as 3 percent in one sample) indi- receives neither capital nor insurance, serving as a cated that they did not need it. comparison group. Cole et al. presents the best evidence to date to Providing insurance and capital together (i.e., explain why take-up rates remain so low. Re- subsidizing the purchase) produced the most im- searchers measured take-up of a microinsurance pact. Farmers in the insurance and capital group product that protects farmers in Andhra Pradesh, increased their spending on farm chemical inputs India, against too little or too much rainfall. The by 47 percent, increased their cultivation area by 22 researchers assigned a sample of potential clients percent, and were less likely to have members of to receive various offers and information about the their household miss meals than the comparison product. Each offer or information set was de- group. Farmers in the insurance-only group signed to isolate one possible cause of low take-up: changed some of their farming decisions, but to a the price of the policy, the availability of cash in lower extent than farmers in the insurance and cap- the household to purchase the policy, the under- ital group. This suggests that reducing risk is bene- standing of rainfall measurements by the potential ficial by itself, but much greater impact may come client, the level of trust of the potential client to- by looking at poor households’ financial needs in a ward the insurance scheme or the insurance mar- more comprehensive way. keting agent, and the framing of the information describing the insurance. As expected, the price of the insurance policy is Challenges in promoting microinsurance a strong determinant of whether households buy products the product. Take-up increased by 10.4 percent on Why, if having insurance has such potentially large average (from an overall average of 24–29 percent) impacts, are take-up rates among poor farmers so when the premium decreased by 10 percent. Price, low?29 Cole et al. (2011) report take-up rates for a however, is not the only determinant of demand. In rainfall insurance product in India between 23 and Cole et al.’s survey, “lack of available funds� was the 29 percent, even though the households cited most commonly cited reason for not purchasing in- droughts as the most significant risk they face. surance. About 40 percent of households in the Karlan, Potential clients may also lack information about Osei-Akoto, Osei, and Udry Ghana project bought and understanding of how formal insurance works. a rainfall insurance product at the actuarially fair However, Cole et al. show that receiving additional price (the price that covers average payouts, but information about the product did not cause an in- not the costs of administering the product). crease in take-up. In a 1994 study in rural Thailand, Townsend Finally, lack of trust in the insurance provider showed that households use informal insurance may be another reason why poor households do not mechanisms to maintain a certain level of con- buy policies. To measure whether trust is indeed a sumption even when income fluctuates. Other re- significant driver of take-up, Cole et al. evaluated search, however, indicates that these informal takeup of the insurance product when the market- mechanisms do not cover all risks. Duflo and Udry ing team was accompanied on their visits to house- holds by an individual from Basix, a nongovern- 29. See also Karlan, McConnell, Mullainathan, and Zinman (2010). mental organization that the farmers know well, 18 versus when the team went out on its own. The re- The overall message from this body of work is that searchers found that households that were visited poor people face various limits, and their ability to by a marketer accompanied by a member of Basix capitalize on opportunities varies greatly. One of were 10 percentage points more likely to purchase a the next steps is to find simple ways to identify policy, suggesting that trust is a significant issue. those differences and cater to them with the right While far from providing a complete picture, products delivered with the right design. these studies together do provide a more nuanced Details matter. Purpose does as well—not all bor- and precise set of information on how to better de- rowers want to grow a business. The variable results sign, price, and market microinsurance products so seen can be as much a function of borrower intent as that the supply of products for poor clients can borrower ability. A one-size-fits-all product will not meet the real need in a cost-effective manner. bring benefit to the borrowers or profit to the provid- ers. Instead, the microfinance industry needs to con- tinue to mature in ways that allow it to view poor Conclusion customers as individuals. Some of those individuals will leverage financial services to smooth consump- While still based on a relatively small number of tion; some to manage risk; some to make investments studies, the work of researchers and participating they have the skill and resources to profit from; some microfinance providers is bringing new knowledge will do all of the above. With a view of serving all of about how clients use capital, what helps them to these needs, microfinance providers may evolve a save, and what constraints they face that prevent new generation of improved services and products them from benefiting more from financial access. that reliably and flexibly help poor people. 19 Annex 1 RESEARCHERS LoCATIon FInAnCIAL SERvICE InTERvEnTIon Abhijit Banerjee, Esther Duflo, India microcredit Researchers evaluate the impact of access to credit by randomizing Rachel Glennerster, and the placement of new Spandana MFI branches in Hyderabad, India. Cynthia Kinnan (2010) Lasse Brune, Xavier Giné, Malawi microsavings Researchers evaluate whether commitment devices can reduce self- Jessica Goldberg, and control problems and cash demands from social networks. Farmers Dean Yang (2011) were randomly assigned to receive either assistance to open an ordinary savings account, or to open an ordinary account with a commitment device. Shawn Cole, Xavier Giné, India microinsurance Researchers investigate the importance of price and nonprice Jeremy Tobacman, Petia determinants in the demand for rainfall insurance by randomly Topalova, Robert Townsend, varying the price of the insurance policy, randomly assigning certain and James Vickery (2011) households’ positive liquidity shocks, or randomly assigning endorse- ments by a trusted agent. Other experiments test the role of financial literacy, product framing, and other behavioral biases. Bruno Crépon, Florencia Devoto, Morocco microcredit Researchers evaluate the impact of access to credit in a rural setting Esther Duflo, and William by randomizing the placement of new Al Amana MFI branches in Pariente (2011) Morocco. Alejandro Drexler, Greg Fischer, Dominican financial literacy/ Researchers evaluate the impact of financial literacy training on and Antoinette Schoar (2011) Republic business training business outcomes for small enterprises in the Dominican Republic. Two methods of financial literacy training are tested: (1) classic accounting principles, and (2) simple accounting “rules of thumb.� Esther Duflo, Michael Kremer, Kenya return to capital/ Researchers measure the rates of return for different quantities of and Jonathan Robinson (2008) inputs fertilizer used on crops in Kenya. Esther Duflo, Michael Kremer, Kenya commitment device Researchers evaluate an intervention to test whether providing and Jonathan Robinson (2010) mechanisms to save harvest income for future fertilizer purchases could be effective in increasing fertilizer use among farmers. Pascaline Dupas and Kenya microsavings Researchers investigate the importance of savings constraints for Jonathan Robinson (2011) microenterprise development by randomly providing small business owners in Kenya with access to savings accounts. Erica Field, Rohini Pande, India microcredit Researchers investigate how the term structure of debt influences John Papp, and Natalie entrepreneurship among the poor. Borrowers were randomly assigned Rigol (2010) either the classic microfinance contract with repayment beginning immediately after loan disbursement or a contract that provides a two- month grace period prior to repayment. Xavier Giné, Jessica Goldberg, Malawi microcredit Researchers evaluate the impact of an improved personal identification and Dean Yang (2011) system on loan repayment. Randomly selected borrowers who applied for loans for agricultural inputs in rural Malawi had fingerprints collected as a part of the loan application process. 20 MAIn FInDIngS PAPER No discernible impact on measures of health, education, and female empowerment. The Miracle of Microfinance? Evidence from a More businesses were created. While some households increased nondurable Randomized Evaluation consumption, others reduced expenditure on temptation goods, such as alcohol, tobacco, tea, and snacks, and instead invested in their businesses, or bought more durable goods. The commitment treatment had a large positive effect on the amounts of deposits and Commitments to Save: A Field Experiment in withdrawals made immediately prior to the planting season and a positive effect on Rural Malawi agricultural input use, leading to a 22 percent increase in the value of the crop output, and a 17 percent increase in total household expenditure. Farmers who had access only to the ordinary account showed lower or nonsignificant impacts on the same outcomes. Insurance did not increase the use of inputs or change allocation of land, but having Barriers to Household Risk Management: Evidence access to rainfall insurance did cause farmers to shift toward more risky, rain-sensitive from India crops, which typically provide higher profit. No discernible impact on measures of health, education, and female empowerment. Impact of Microcredit in Rural Areas of Morocco: For individuals with existing farming activities, access to credit increased the volume Evidence from a Randomized Evaluation of activity. Microcredit had no impact on nonagricultural businesses. Those with an existing business at the start of the study reduced consumption and considerably increased savings. But for those without prior business activities, consumption increased. Business owners who received the rules-of-thumb training applied sound accounting Keeping It Simple: Financial Literacy and Rules of principles more than their peers. Those who received the rules-of-thumb training also Thumb earned more revenues than their peers, especially during “bad� weeks. Farmers earn much stronger yields by using fertilizer at specific points during the How High are Rates of Return to Fertilizer? growing season. The annualized return to half a teaspoon of fertilizer at top dressing Evidence from Field Experiments in Kenya (when the maize plant is knee high) was almost 70 percent per year. The commitment device offering an opportunity to buy a voucher for fertilizer but with Nudging Farmers to Use Fertilizer: Theory and free delivery early in the season increased fertilizer use by 47 to 70 percent. The effect Experimental Evidence from Kenya of the commitment device early in the season was greater than other offers, such as free delivery later in the season, and a 50 percent subsidy also offered later in the season. Access to formal savings accounts for market stallholders led to increased business Savings Constraints and Microenterprise investment and personal income growth. Four to six months after account opening, Development: Evidence from a Field Experiment in women in the treatment group had a 4.5 percent higher daily investment in their Kenya businesses than women in the comparison group. There was was no measurable impact for men in the study. Several categories of expenditures were higher for women in the treatment group. Savings accounts also seemed to make women somewhat less vulnerable to health shocks. The grace period group members invested 6 percent more of their loans in their Term Structure of Debt and Entrepreneurial Behavior: businesses than borrowers who received no grace period, and two years after the loans Experimental Evidence from Microfinance were given, they had 30 percent higher average profits. Household income was also higher. However, the average result masks significant variation within the grace period group: some of the women did really well, while others suffered losses. Nineteen percent of the individuals in the grace period group ultimately defaulted on their loans compared with 2 percent default for the individuals with the standard repayment structure. As a result of the fingerprinting intervention, borrowers predicted to be least likely to Credit Market Consequences of Improved Personal repay showed a significant change in behavior. Fingerprinted borrowers in this group Identification: Field Experimental Evidence from took smaller loans when they knew they could be identified, and were more likely to Malawi repay their loans on time as well as eventually, compared to equivalent borrowers in the comparison group. 21 Annex 1, continued RESEARCHERS LoCATIon FInAnCIAL SERvICE InTERvEnTIon Xavier Giné, Lev Menand, India microinsurance Researchers summarize results of previous research on rainfall Robert Townsend, and James insurance markets in India, which provides evidence that price, Vickery (2010) liquidity constraints, and trust all present significant barriers to increased take-up. Xavier Giné and Philippines microcredit Researchers investigate whether group liability is in fact necessary Dean Karlan (2011) for managing default risk. In one treatment, existing group-lending clients of the Green Bank of Caraga were randomly converted to an individual liability model. In a second treatment, new borrowers started out with individual liability loans. Dean Karlan, Edward Kutsoati, Ghana microsavings Researchers evaluated the impact of a new type of “labeled� savings Margaret McConnell, Margaret account that was intended to help clients save by focusing their attention McMillan, and Christopher on their savings goals. Existing clients of the Mumuada Rural Bank Udry (forthcoming) in Eastern Ghana were asked about their savings goals, and some were given the opportunity to open separate, parallel savings accounts labeled “education,� “business,� “housing,� or some other category. Dean Karlan, Margaret Peru, Bolivia, microsavings Researchers measure the effectiveness of sending savings reminders McConnell, Sendhil Philippines in the form of letters (in Peru) or SMS text messages (in Bolivia and Mullainathan, and Jonathan the Philippines) to clients holding programmed savings accounts. Zinman (2011) Dean Karlan, Isaac Osei-Akoto, Ghana microinsurance Researchers investigate the role of risk in constraining farmer Robert Osei, and Chris Udry investment and technology adoption choices, and to evaluate its (forthcoming) importance relative to constraints on credit, by coupling rainfall insurance with cash grants. Dean Karlan and Peru financial literacy/ Researchers evaluate the marginal impact of adding business Martin Valdivia (2011) business training training to a group lending program in Peru. Dean Karlan and South Africa microcredit Researchers estimate the effects of expanding access to expensive Jonathan Zinman (2010) consumer credit in South Africa by randomizing loan approval for clients identified by the cooperating lender as being marginally creditworthy. Dean Karlan and Philippines microcredit Researchers evaluate the impact of increasing access to credit in the Jonathan Zinman (2011) Philippines by randomizing loan approval for clients identified as marginally creditworthy. Suresh de Mel, Sri Lanka return to capital/inputs To evaluate whether there are high returns to capital for micro- David McKenzie, and enterprises, researchers randomize the provision of cash and equipment Christopher Woodruff (2008) grants to small firms in Sri Lanka, and measure the increase in profits arising from exogenous (positive) shock to capital stock. Marcel Fafchamps, Ghana return to capital/inputs Researchers evaluate the differential effects of providing either cash David McKenzie, grants or in-kind grants of inventory or equipment on both male and Christopher Woodruff, female entrepreneurs. and Simon Quinn (2011) 22 MAIn FInDIngS PAPER Preliminary results indicate that insurance does not increase the use of inputs or Microinsurance: A Case Study of the Indian change the allocation of land, although having access to rainfall insurance does Rainfall Index Insurance Market cause farmers to shift toward more risky, rain-sensitive crops, which typically provide higher profit. The shift to individual liability did not negatively affect loan repayment for either group. Group versus Individual Liability: Short and Long- The bank also saw an increase in outreach, as more customers, attracted by the Term Evidence from Philippine Microcredit Lending individual liability option, sought loans from the bank. Groups Savers eligible to open parallel accounts saved 31 percent more on average than those in the comparison group, with the greatest effect seen for the accounts labelled “Education.� Reminders increased average savings balances overall by 6 percent. This impact Getting to the Top of Mind: How Reminders Increase increased substantially, to 16 percent, for the Peruvian savers when the reminder Saving referred to a purchase goal. Farmers who receive both insurance and capital (i.e., subsidizing the purchase) Examining Underinvestment in Agriculture: increased their spending on farm chemical inputs by 47 percent, increased their Measuring Returns to Capital and Insurance cultivation area by 22 percent, and were less likely to have members of their household miss meals than the comparison group. Farmers who received insurance changed only some of their farming decisions, but to a lower extent than those who had the capital also. The results found positive impacts from business training. Teaching Entrepreneurship: Impact of Business Training on Microfinance Clients and Institutions Expanding access to credit increased borrower well-being: incomes increased, food Expanding Credit Access: Using Randomized Supply consumption went up, and measures of decision-making within the household Decisions to Estimate the Impacts went up, alongside community status and overall optimism. Net borrowing increased in the treatment group relative to the comparison. However, Microcredit in Theory and Practice: Using the number of business activities and employees in the treatment group decreased Randomized Credit Scoring for Impact Evaluation relative to the comparison, and subjective well-being declined slightly. However, microloans increased ability to cope with risk, strengthened community ties, and increased access to informal credit. The average real return to capital was 5.7 percent per month—substantially higher than Returns to Capital in Microenterprises: Evidence from the market interest rate. Returns varied with measures of ability, household liquidity, a Field Experiment and the gender of the owner (men fared better than women). Cash grants to women entrepreneurs produced no return on capital, whereas in-kind When Is Capital Enough to Get Female Enterprises gifts of inventory or equipment to women showed a significant average return. When Growing? Evidence from a Randomized Experiment given cash, women invested less of the gift in the business, splitting off pieces for in Ghana household purchases or other expenses. The high returns from in-kind gifts came entirely from the women who had larger, higher profit businesses at the outset. Women with below-average profits (around $1 a day) saw no benefit in terms of profit from either form of grant. Men-owned businesses, on the other hand, saw signficant returns from both the in-kind grants and the cash grants. 23 References Ananth, Bindu, Dean Karlan, and Sendhil Mullainathan. 2007. “Microentrepreneurs and Their Money: Three Anomalies.� Draft. http://karlan.yale.edu/p/AnomaliesDraft.v7.pdf Armendariz, Beatriz, and Jonathan Morduch. 2007. The Economics of Microfinance. Cambridge, Mass.: MIT Press. Ashraf, Nava, Dean Karlan, and Wesley Yin. 2006. “Tying Odysseus to the Mast: Evidence from a Commitment Savings Product in the Philippines.� Quarterly Journal of Economics, 121(2): 635–72. Banerjee, Abhijit V., and Esther Duflo. 2011. Poor Economics. New York, NY: Perseus Books, Public Affairs, USA. Banerjee, Abhijit V., Esther Duflo, Rachel Glennerster, and Cynthia Kinnan. 2010. “The Miracle of Microfinance? Evidence from a Randomized Evaluation.� Cambridge, Mass.: J-PAL and MIT, June. Bauchet, Jonathan, and Jonathan Morduch. 2010. “An Introduction to Impact Evaluations with Randomized Designs.� FAI Framing Note. New York: Financial Access Initiative. Bennett, Drake. 2009. “Small Change: Billions of Dollars and a Nobel Prize Later, It Looks Like ‘Microlending’ Doesn’t Actually Do Much to Fight Poverty.� The Boston Globe, 20 September. Brune, Lasse, Xavier Giné, Jessica Goldberg, and Dean Yang. 2011. “Commitments to Save: A Field Experiment in Rural Malawi.� World Bank Policy Research Working Paper, No. 5748. Washington, D.C.: World Bank. Carpena, Fenella, Shawn Allen Cole, Jeremy Shapiro, and Zia Bilal. 2010. “Liability Structure in Small-Scale Finance: Evidence from a Natural Experiment.� World Bank Policy Research Working Paper Series. Washington, D.C.: World Bank, September. http://www-wds. worldbank.org/external/default/WDSContentServer/IW3P/IB/2010/09/20/000158349_ 20100920115522/Rendered/PDF/WPS5427.pdf CGAP and the World Bank Group. 2010. Financial Access 2010: The State of Financial Inclusion through the Crisis. Washington, D.C.: CGAP and World Bank. http://www.cgap.org/gm/ document-1.9.46570/FA_2010_Financial_Access_2010_Rev.pdf Cole, Shawn, and Gauri Kartini Shastry. 2009. “Smart Money: The Effects of Education, Cognitive Ability and Financial Literacy on Financial Market Participation.� Harvard Business School Working Paper No. 09-071. Cambridge, Mass.: Harvard University. Cole, Shawn, Xavier Giné, Jeremy Tobacman, Petia Topalova, Robert Townsend, and James Vickery. 2011. “Barriers to Household Risk Management: Evidence from India.� Harvard Business School Finance Working Paper No. 09-116. Cambridge, Mass.: Harvard University. Collins, Daryl, Jonathan Morduch, Stuart Rutherford, and Orlanda Ruthven. 2009. Portfolios of the Poor: How the World’s Poor Live on $2 a Day. Princeton, N.J.: Princeton University Press. 24 Crépon, Bruno, Florencia Devoto, Esther Duflo, and William Parienté. 2011. “Impact of Micro- credit in Rural Areas of Morocco: Evidence from a Randomized Evaluation.� MIT Working Paper. Cambridge, Mass.: MIT, March. de Mel, Suresh, David McKenzie, and Christopher Woodruff. 2008. “Returns to Capital in Microenterprises: Evidence from a Field Experiment.� Quarterly Journal of Economics, Vol. 123(4): 1329–72. ———. 2009. “Are Women More Credit-Constrained? Experimental Evidence on Gender and Microenterprise Returns.� American Economic Journal: Applied Economics 1(3): 1–32. Drexler, Alejandro, Greg Fischer, and Antoinette Schoar. 2010. “Keeping It Simple: Financial Literacy and Rules of Thumb.� CEPR Working Paper. London, UK: Centre for Economic Policy Research. September. Duflo, Esther, and Chris Udry. 2004. “Intrahousehold Resource Allocation in Cote d’Ivoire: Social Norms, Separate Accounts and Consumption Choices.� NBER Working Paper No. 10498. Cambridge, Mass.: National Bureau of Economic Research. Duflo, Esther, Rachel Glennerster, and Michael Kremer. 2008. “Using Randomization in Development Economics Research: A Toolkit.� In T. Paul Schultz and John A. Strauss, ed., 2008, Handbook of Development Economics, Elsevier, edition 1, volume 4 (5): pp. 3895–962. http://ideas.repec.org/h/eee/devchp/5-61.html Duflo, Esther, Michael Kremer, and Jonathan Robinson. 2008. “How High Are Rates of Return to Fertilizer? Evidence from Field Experiments in Kenya.� American Economic Review, 98(2): 482–88. ———. 2010. “Nudging Farmers to Use Fertilizer: Theory and Experimental Evidence from Kenya.� NBER Working Papers 15131. Cambridge, Mass.: National Bureau of Economic Research. Dupas, Pascaline, and Jonathan Robinson. 2011. “Savings Constraints and Microenterprise Development: Evidence from a Field Experiment in Kenya.� NBER Working Paper No. 14693. Cambridge, Mass.: National Bureau of Economic Research. Economist, The. 2007. “In Praise of Usury.� The Economist, 2 August. ———. 2009. “A Partial Marvel: Microcredit May Not Work Wonders but It Does Help the Entrepreneurial Poor.� The Economist, 16 July. Engle, Patrice L. 1991. “Maternal Work and Child-Care Strategies in Peri-Urban Guatemala: Nutritional Effects.� Child Development. Vol. 62, 954–65. Fafchamps, Marcel, David McKenzie, Simon Quinn, and Christopher Woodruff. 2011. “When Is Capital Enough to Get Female Enterprises Growing? Evidence from a Randomized Experiment in Ghana.� World Bank Policy Research Working Paper No. 5706. Washington, D.C.: World Bank. Feigenberg, Benjamin, Erica Field, and Rohini Pande. 2010. “Building Social Capital through Microfinance.� HKS Working Paper. Cambridge, Mass.: Harvard University. Field, Erica, Rohini Pande, John Papp, and Natalia Rigol. 2011. “Term Structure of Debt and Entrepreneurial Behavior: Experimental Evidence from Microfinance.� Harvard University Working Paper, December. Cambridge, Mass.: Harvard University. http://www.economics. harvard.edu/faculty/field/files/repayment_default_Dec19.pdf 25 Freedman, David H. 2010. “Lies, Damned Lies, and Medical Science.� The Atlantic Monthly, November. http://www.theatlantic.com/magazine/archive/2010/11/lies-damned-lies- and-medical-science/8269/ Giné, Xavier, and Dean Karlan. 2011. “Group versus Individual Liability: Short and Long Term Evidence from Philippine Microcredit Lending Groups.� Working Paper. New Haven, Conn.: Yale University, June. Giné, Xavier, Jessica Goldberg, and Dean Yang. 2011. “Credit Market Consequences of Improved Personal Identification: Field Experimental Evidence from Malawi.� NBER Working Paper No. 17449. September. Cambridge, Mass.: National Bureau of Economic Research. Giné, Xavier, Lev Menand, Robert Townsend, and James Vickery. 2010. “Microinsurance: A Case Study of the Indian Rainfall Index Insurance Market.� World Bank Policy Research Working Paper Series, No. 5459. Washington, D.C.: World Bank. http://econ.worldbank.org/ external/default/main?pagePK=64165259&theSitePK=469372&piPK=64165421&menuPK= 64166093&entityID=000158349_20101025152029 Harford, Tim. 2008. “Conflicts of Interest.� Financial Times, 6 December. Helms, Brigit. 2010. “Microfinancing Changes Lives Around the World—Measurably.� Seattle Times, 7 April. Karlan, Dean, and Jacob Appel. 2011. More Than Good Intentions: How a New Economics Is Helping to Solve Global Poverty. New York: Dutton, Penguin Group, April. Karlan, Dean, Isaac Osei-Akoto, Robert Osei, and Chris Udry. Forthcoming. “Examining Underinvestment in Agriculture: Measuring Returns to Capital and Insurance.� Karlan, Dean, Margaret McConnell, Sendhil Mullainathan, and Jonathan Zinman. 2010. “Getting to the Top of Mind: How Reminders Increase Saving.� NBER Working Paper No. 16205. Cambridge, Mass.: National Bureau of Economic Research. ———. 2011. “Getting to the Top of Mind: How Reminders Increase Saving.� Yale University Working Paper, January. New Haven, Conn.: Yale University. http://karlan.yale.edu/p/ Top%20of%20Mind%202011jan.pdf Karlan, Dean, and Jonathan Morduch. 2009. “Access to Finance: Ideas and Evidence. Risk Management and Insurance.� Financial Access Initiative Note. New York: Financial Access Initiative. Karlan, Dean, and Jonathan Zinman. 2010. “Expanding Credit Access: Using Randomized Supply Decisions to Estimate the Impacts.� The Review of Financial Studies, Vol. 23 (1): 433–64. ———. 2011. “Microcredit in Theory and Practice: Using Randomized Credit Scoring for Impact Evaluation.� Science, 332(6035): 1278–84. Karlan, Dean, and Martin Valdivia. 2011. “Teaching Entrepreneurship: Impact of Business Training on Microfinance Clients and Institutions.� The Review of Economics and Statistics, 93(2): 510–27. Karlan, Dean, Jonathan Morduch, and Sendhil Mullainathan. 2010. “Take-Up: Why Microfinance Take-Up Rates Are Low & Why It Matters.� Financial Access Initiative Research Framing Note. New York: Financial Access Initiative. 26 Reed, Larry R. 2011. “The State of the Microcredit Summit Report 2011.� Washington, D.C.: Microcredit Summit Campaign. http://www.microcreditsummit.org/SOCR_2011_EN_ web.pdf Rutherford, Stuart. 1999. The Poor and Their Money. Oxford: Oxford University Press. Schultz, T. Paul. 1990. “Testing the Neoclassical Model of Family Labor Supply and Fertility.� Journal of Human Resources, Vol. 25(4): 599–634. Thomas, Duncan. 1990. “Intra-Household Resource Allocation: An Inferential Approach.� Journal of Human Resources, Vol. 25 (4): 635–64. Townsend, Robert. 1994. “Risk and Insurance in Village India.� Econometrica, 62 (3): 539–92. World Bank. n.d. “Labor participation rate, female (% of female population ages 15+).� Table. http://data.worldbank.org/indicator/SL.TLF.CACT.FE.ZS 27