Microfinance promise

Journal of Economic Literature Vol. XXXVII (Decmber 1999), pp. 1569–1614

Morduch: The Microfinance PromiseJournal of Economic Literature, Vol. XXXVII (December 1999)

The Microfinance Promise

Jonathan Morduch1

1. Introduction

ABOUT ONE billion people globallylive in households with per capita in-

comes of under one dollar per day. Thepolicymakers and practitioners who havebeen trying to improve the lives of thatbillion face an uphill battle. Reports ofbureaucratic sprawl and unchecked cor-ruption abound. And many now believethat government assistance to the pooroften creates dependency and disincen-tives that make matters worse, not bet-ter. Moreover, despite decades of aid,communities and families appear to be

increasingly fractured, offering a fragilefoundation on which to build.

Amid the dispiriting news, excite-ment is building about a set of unusualfinancial institutions prospering in dis-tant corners of the world—especiallyBolivia, Bangladesh, and Indonesia. Thehope is that much poverty can be allevi-ated—and that economic and socialstructures can be transformed funda-mentally—by providing financial ser-vices to low-income households. Theseinstitutions, united under the banner ofmicrofinance, share a commitment toserving clients that have been excludedfrom the formal banking sector. Almostall of the borrowers do so to financeself-employment activities, and manystart by taking loans as small as $75, re-paid over several months or a year. Onlya few programs require borrowers toput up collateral, enabling would-be en-trepreneurs with few assets to escapepositions as poorly paid wage laborersor farmers.

Some of the programs serve just ahandful of borrowers while others servemillions. In the past two decades, a di-verse assortment of new programs hasbeen set up in Africa, Asia, Latin Amer-ica, Canada, and roughly 300 U.S. sitesfrom New York to San Diego (The Econo-mist 1997). Globally, there are nowabout 8 to 10 million households servedby microfinance programs, and somepractitioners are pushing to expand to

1569

1 Princeton University. [email protected]. I have benefited from comments fromHarold Alderman, Anne Case, Jonathan Conning,Peter Fidler, Karla Hoff, Margaret Madajewicz,John Pencavel, Mark Schreiner, Jay Rosengard,J.D. von Pischke, and three anonymous referees. Ihave also benefited from discussions with AbhijitBanerjee, David Cutler, Don Johnston, AlbertPark, Mark Pitt, Marguerite Robinson, ScottRozelle, Michael Woolcock, and seminar partici-pants at Brown University, HIID, and the OhioState University. Aimee Chin and Milissa Day pro-vided excellent research assistance. Part of the re-search was funded by the Harvard Institute forInternational Development, and I appreciate thesupport of Jeffrey Sachs and David Bloom. I alsoappreciate the hospitality of the Bank Rakyat In-donesia in Jakarta in August 1996 and of Grameen,BRAC, and ASA staff in Bangladesh in the sum-mer of 1997. The paper was largely completedduring a year as a National Fellow at the HooverInstitution, Stanford University. The revisionwas completed with support from the Mac-Arthur Foundation. An earlier version of the pa-per was circulated under the title “The Microfi-nance Revolution.” The paper reflects my viewsonly.

100 million poor households by 2005.As James Wolfensohn, the president ofthe World Bank, has been quick topoint out, helping 100 million house-holds means that as many as 500–600million poor people could benefit. In-creasing activity in the United Statescan be expected as banks turn to mi-crofinance encouraged by new teethadded to the Community ReinvestmentAct of 1977 (Timothy O’Brien 1998).

The programs point to innovationslike “group-lending” contracts and newattitudes about subsidies as the keys totheir successes. Group-lending con-tracts effectively make a borrower’sneighbors co-signers to loans, mitigat-ing problems created by informationalasymmetries between lender and bor-rower. Neighbors now have incentivesto monitor each other and to excluderisky borrowers from participation, pro-moting repayments even in the absenceof collateral requirements. The con-tracts have caught the attention of eco-nomic theorists, and they have broughtglobal recognition to the group-lendingmodel of Bangladesh’s Grameen Bank.2

The lack of public discord is striking.Microfinance appears to offer a “win-win” solution, where both financial in-stitutions and poor clients profit. Thefirst installment of a recent five-part se-ries in the San Francisco Examiner, forexample, begins with stories about fourwomen helped by microfinance: a tex-tile distributor in Ahmedabad, India; astreet vendor in Cairo, Egypt; an artistin Albuquerque, New Mexico; and a

furniture maker in Northern California.The story continues:

From ancient slums and impoverished vil-lages in the developing world to the tired in-ner cities and frayed suburbs of America’seconomic fringes, these and millions of otherwomen are all part of a revolution. Somemight call it a capitalist revolution . . . Aslittle as $25 or $50 in the developing world,perhaps $500 or $5000 in the United States,these microloans make huge differences inpeople’s lives . . . Many Third World bank-ers are finding that lending to the poor is notjust a good thing to do but is also profitable.(Brill 1999)

Advocates who lean left highlight the“bottom-up” aspects, attention to com-munity, focus on women, and, most im-portantly, the aim to help the under-served. It is no coincidence that the riseof microfinance parallels the rise of non-governmental organizations (NGOs) inpolicy circles and the newfound attentionto “social capital” by academics (e.g.,Robert Putnam 1993). Those who leanright highlight the prospect of alleviat-ing poverty while providing incentivesto work, the nongovernmental leadership,the use of mechanisms disciplined bymarket forces, and the general suspicionof ongoing subsidization.

There are good reasons for excite-ment about the promise of microfi-nance, especially given the politicalcontext, but there are also good reasonsfor caution. Alleviating poverty throughbanking is an old idea with a checkeredpast. Poverty alleviation through theprovision of subsidized credit was a cen-terpiece of many countries’ develop-ment strategies from the early 1950sthrough the 1980s, but these experi-ences were nearly all disasters. Loan re-payment rates often dropped well below50 percent; costs of subsidies ballooned;and much credit was diverted to the po-litically powerful, away from the in-tended recipients (Dale Adams, DouglasGraham, and J. D. von Pischke 1984).

2 Recent theoretical studies of microfinance in-clude Joseph Stiglitz 1990; Hal Varian 1990; Timo-thy Besley and Stephen Coate 1995; AbhijitBanerjee, Besley, and Timothy Guinnane 1992;Maitreesh Ghatak 1998; Mansoora Rashid andRobert Townsend 1993; Beatriz Armendariz deAghion and Morduch 1998; Armendariz and Chris-tian Gollier 1997; Margaret Madajewicz 1998;Aliou Diagne 1998; Bruce Wydick 1999; JonathanConning 1997; Edward S. Prescott 1997; and LoïcSadoulet 1997.

1570 Journal of Economic Literature, Vol. XXXVII (December 1999)

What is new? Although very few pro-grams require collateral, the major newprograms report loan repayment ratesthat are in almost all cases above 95percent. The programs have also provenable to reach poor individuals, particu-larly women, that have been difficult toreach through alternative approaches.Nowhere is this more striking than inBangladesh, a predominantly Muslimcountry traditionally viewed as cultur-ally conservative and male-dominated.The programs there together serveclose to five million borrowers, the vastmajority of whom are women, and, inaddition to providing loans, some of theprograms also offer education on healthissues, gender roles, and legal rights.The new programs also break from thepast by eschewing heavy government in-volvement and by paying close attentionto the incentives that drive efficientperformance.

But things are happening fast—andgetting much faster. In 1997, a highprofile consortium of policymakers,charitable foundations, and practitionersstarted a drive to raise over $20 billionfor microfinance start-ups in the next tenyears (Microcredit Summit Report 1997).Most of those funds are being mobi-lized and channeled to new, untestedinstitutions, and existing resources arebeing reallocated from traditional pov-erty alleviation programs to microfi-nance. With donor funding pouring in,practitioners have limited incentives tostep back and question exactly how andwhere monies will be best spent.

The evidence described below, how-ever, suggests that the greatest promiseof microfinance is so far unmet, and theboldest claims do not withstand closescrutiny. High repayment rates haveseldom translated into profits as adver-tised. As Section 4 shows, most pro-grams continue to be subsidized di-rectly through grants and indirectly

through soft terms on loans from do-nors. Moreover, the programs that arebreaking even financially are not thosecelebrated for serving the poorest cli-ents. A recent survey shows that evenpoverty-focused programs with a “com-mitment” to achieving financial sustain-ability cover only about 70 percent oftheir full costs (MicroBanking Bulletin1998). While many hope that weak fi-nancial performances will improve overtime, even established poverty-focusedprograms like the Grameen Bank wouldhave trouble making ends meet withoutongoing subsidies.

The continuing dependence on subsi-dies has given donors a strong voice,but, ironically, they have used it topreach against ongoing subsidization.The fear of repeating past mistakes haspushed donors to argue that subsidiza-tion should be used only to cover start-up costs. But if money spent to supportmicrofinance helps to meet social objec-tives in ways not possible through alter-native programs like workfare or directfood aid, why not continue subsidizingmicrofinance? Would the world be bet-ter off if programs like the GrameenBank were forced to shut their doors?

Answering the questions requiresstudies of social impacts and informa-tion on client profiles by income andoccupation. Those arguing from theanti-subsidy (“win-win”) position haveshown little interest in collecting thesedata, however. One defense is that, as-suming that the “win-win” position iscorrect (i.e., that raising real interestrates to levels approaching 40 percentper year will not seriously underminethe depth of outreach), financial viabil-ity should be sufficient to show socialimpact. But the assertion is strong, andthe broader argument packs little punchwithout evidence to back it up.

Poverty-focused programs counterthat shifting all costs onto clients would

Morduch: The Microfinance Promise 1571

likely undermine social objectives, butby the same token there is not yet di-rect evidence on this either. Anecdotesabound about dramatic social and eco-nomic impacts, but there have been fewimpact evaluations with carefully cho-sen treatment and control groups (orwith control groups of any sort), andthose that exist yield a mixed picture ofimpacts. Nor has there been much solidempirical work on the sensitivity ofcredit demand to the interest rate, noron the extent to which subsidized pro-grams generate externalities for non-borrowers. Part of the problem is thatthe programs themselves also have littleincentive to complete impact studies.Data collection efforts can be costly anddistracting, and results threaten to un-dermine the rhetorical strength of theanecdotal evidence.

The indirect evidence at least lendssupport to those wary of the anti-sub-sidy argument. Without better data, av-erage loan size is typically used to proxyfor poverty levels (under the assump-tion that only poorer households will bewilling to take the smallest loans). Thetypical borrower from financially self-sufficient programs has a loan balanceof around $430—with loan sizes oftenmuch higher (MicroBanking Bulletin1998). In low-income countries, bor-rowers at that level tend to be amongthe “better off” poor or are even slightlyabove the poverty line. Expanding fi-nancial services in this way can fostereconomic efficiency—and, perhaps,economic growth along the lines ofValerie Bencivenga and Bruce D. Smith(1991)—but it will do little directly toaffect the vast majority of poor house-holds. In contrast, Section 4.1 showsthat the typical client from (subsidized)programs focused sharply on poverty al-leviation has a loan balance close to just$100.

Important next steps are being taken

by practitioners and researchers whoare moving beyond the terms of earlyconversations (e.g., Gary Woller, Chris-topher Dunford, and Warner Wood-worth 1999). The promise of microfi-nance was founded on innovation: newmanagement structures, new contracts,and new attitudes. The leading pro-grams came about by trial and error.Once the mechanisms worked reason-ably well, standardization and replica-tion became top priorities, with contin-ued innovation only around the edges.As a result, most programs are not opti-mally designed nor necessarily offeringthe most desirable financial products.While the group-lending contract is themost celebrated innovation in microfi-nance, all programs use a variety ofother innovations that may well be asimportant, especially various forms ofdynamic incentives and repaymentschedules. In this sense, economic the-ory on microfinance (which focusesnearly exclusively on group contracts) isalso ahead of the evidence. A portion ofdonor money would be well spent quan-tifying the roles of these overlappingmechanisms and supporting efforts todetermine less expensive combinationsof mechanisms to serve poor clients invarying contexts. New managementstructures, like the stripped-down struc-ture of Bangladesh’s Association for So-cial Advancement, may allow sharp cost-cutting. New products, like the flexiblesavings plan of Bangladesh’s SafeSave,may provide an alternative route to fi-nancial sustainability while helping poorhouseholds. The enduring lesson of mi-crofinance is that mechanisms matter:the full promise of microfinance canonly be realized by returning to theearly commitments to experimentation,innovation, and evaluation.

The next section describes leadingprograms. Section 3 considers theoret-ical perspectives. Section 4 turns to


financial sustainability, and Section 5takes up issues surrounding the costs andbenefits of subsidization. Section 6 de-scribes econometric evaluations of im-pacts, and Section 7 turns from creditto saving. The final section concludeswith consideration of microfinancein the broader context of economicdevelopment.

2. New Approaches

Received wisdom has long been thatlending to poor households is doomedto failure: costs are too high, risks aretoo great, savings propensities are toolow, and few households have much toput up as collateral. Not long ago, thenorm was heavily subsidized credit pro-vided by government banks with repay-ment rates of 70–80 percent at best. InBangladesh, for example, loans targetedto poor households by traditional bankshad repayment rates of 51.6 percent in1980. By 1988–89, a year of bad flood-ing, the repayment rate had fallen to18.8 percent (M. A. Khalily and RichardMeyer 1993). Similarly, by 1986 repay-ment rates sank to 41 percent for subsi-dized credit delivered as part of India’shigh-profile Integrated Rural Develop-ment Program (Robert Pulley 1989).These programs offered heavily subsi-dized credit on the premise that poorhouseholds cannot afford to borrow athigh interest rates.

But the costs quickly mounted andthe programs soon bogged down gov-ernment budgets, giving little incentivefor banks to expand. Moreover, manybank managers were forced to reduceinterest rates on deposits in order tocompensate for the low rates on loans.In equilibrium, little in the way of sav-ings was collected, little credit was de-livered, and default rates accelerated asborrowers began to perceive that thebanks would not last long. The repeated

failures appeared to confirm suspicionsthat poor households are neither credit-worthy nor able to save much. More-over, subsidized credit was often di-verted to politically-favored non-poorhouseholds (Adams and von Pischke1992). Despite good intentions, manyprograms proved costly and did little tohelp the intended beneficiaries.

The experience of Bangladesh’s Gra-meen Bank turned this around, and nowa broad range of financial institutionsoffer alternative microfinance modelswith varying philosophies and targetgroups. Other pioneers described belowinclude BancoSol of Bolivia, the BankRakyat Indonesia, the Bank Kredit Deasof Indonesia, and the village banksstarted by the Foundation for Interna-tional Community Assistance (FINCA).The programs below were chosen withan eye to illustrating the diversity ofmechanisms in use, and Table 1 high-lights particular mechanisms. The func-tioning of the mechanisms is describedfurther in Section 3.3

2.1 The Grameen Bank, Bangladesh

The idea for the Grameen Bank didnot come down from the academy, norfrom ideas that started in high-incomecountries and then spread broadly.4

3 Sections 4.1 and 5.1 describe summary statis-tics on a broad variety of programs. See also MariaOtero and Elisabeth Rhyne (1994); MicroBankingBulletin (1998); Ernst Brugger and Sarath Rajapa-tirana (1995); David Hulme and Paul Mosley(1996); and Elaine Edgcomb, Joyce Klein, andPeggy Clark (1996).

4 Part of the inspiration came from observingcredit cooperatives in Bangladesh, and, interest-ingly, these had European roots. The late nine-teenth century in Europe saw the blossoming ofcredit cooperatives designed to help low-incomehouseholds save and get credit. The cooperativesstarted by Frederick Raiffeisen grew to serve 1.4million in Germany by 1910, with replications inIreland and northern Italy (Guinnane 1994 and1997; Aidan Hollis and Arthur Sweetman 1997). Inthe 1880s the government of Madras in South In-dia, then under British rule, looked to the Germanexperiences for solutions in addressing poverty in


Programs that have been set up inNorth Carolina, New York City, Chi-cago, Boston, and Washington, D.C.cite Grameen as an inspiration. In addi-

tion, Grameen’s group lending modelhas been replicated in Bolivia, Chile,China, Ethiopia, Honduras, India, Ma-laysia, Mali, the Philippines, Sri Lanka,

India. By 1912, over four hundred thousand poorIndians belonged to the new credit cooperatives,and by 1946 membership exceeded 9 million (R.Bedi 1992, cited in Michael Woolcock 1998). Thecooperatives took hold in the State of Bengal, theeastern part of which became East Pakistan at in-dependence in 1947 and is now Bangladesh. Inthe early 1900s, the credit cooperatives of Bengalwere so well-known that Edward Filene, the Bos-

ton merchant whose department stores still bearhis name, spent time in India, learning about thecooperatives in order to later set up similar pro-grams in Boston, New York, and Providence(Shelly Tenenbaum 1993). The credit cooperativeseventually lost steam in Bangladesh, but the no-tion of group-lending had established itself and,after experimentation and modification, becameone basis for the Grameen model.

TABLE 1CHARACTERISTICS OF SELECTED LEADING MICROFINANCE PROGRAMS

GrameenBank,

Bangladesh

Banco-Sol,

Bolivia

Bank Rakyat

Indonesia Unit Desa

Badan Kredit Desa,

Indonesia

FINCA Village banks

2 millionborrowers;

Membership 2.4 million 81,503 16 million 765,586 89,986depositors

Average loan balance $134 $909 $1007 $71 $191Typical loan term 1 year 4–12 3–24 3 months 4 months

months monthsPercent female members 95% 61% 23% — 95%Mostly rural? Urban? rural urban mostly rural mostly

rural ruralGroup-lending contracts? yes yes no no noCollateral required? no no yes no noVoluntary savings emphasized? no yes yes no yesProgressive lending? yes yes yes yes yesRegular repayment schedules weekly flexible flexible flexible weeklyTarget clients for lending poor largely non-poor poor poor

non-poorCurrently financially sustainable? no yes yes yes noNominal interest rate on 20% 47.5– 32–43% 55% 36–48% loans (per year) 50.5%Annual consumer price inflation, 1996 2.7% 12.4% 8.0% 8.0% —

Sources: Grameen Bank: through August 1998, www.grameen.com; loan size is from December 1996, calculatedby author. BancoSol: through December 1998, from Jean Steege, ACCION International, personal communica-tion. Interest rates include commission and are for loans denominated in bolivianos; base rates on dollar loansare 25–31%. BRI and BKD: through December 1994 (BKD) and December 1996 (BRI), from BRI annual dataand Don Johnston, personal communication. BRI interest rates are effective rates. FINCA: through July 1998,www.villagebanking.org. Inflation rate: World Bank World Development Indicators 1998.


Tanzania, Thailand, the U.S., and Viet-nam. When Bill Clinton was still gover-nor, it was Muhammad Yunus, founderof the Grameen Bank (and a Vander-bilt-trained economist), who was calledon to help set up the Good Faith Fundin Arkansas, one of the early microfi-nance organizations in the U.S. AsYunus (1995) describes the beginning:

Bangladesh had a terrible famine in 1974. Iwas teaching economics in a Bangladesh uni-versity at that time. You can guess how diffi-cult it is to teach the elegant theories of eco-nomics when people are dying of hunger allaround you. Those theories appeared likecruel jokes. I became a drop-out from formaleconomics. I wanted to learn economicsfrom the poor in the village next door to theuniversity campus.

Yunus found that most villagers wereunable to obtain credit at reasonablerates, so he began by lending themmoney from his own pocket, allowingthe villagers to buy materials for proj-ects like weaving bamboo stools andmaking pots (New York Times 1997).Ten years later, Yunus had set up thebank, drawing on lessons from informalfinancial institutions to lend exclusivelyto groups of poor households. Commonloan uses include rice processing,livestock raising, and traditional crafts.

The groups form voluntarily, and,while loans are made to individuals, allin the group are held responsible forloan repayment. The groups consist offive borrowers each, with lending firstto two, then to the next two, and thento the fifth. These groups of five meettogether weekly with seven othergroups, so that bank staff meet withforty clients at a time. According to therules, if one member ever defaults, allin the group are denied subsequentloans. The contracts take advantage oflocal information and the “social assets”that are at the heart of local enforce-ment mechanisms. Those mechanisms

rely on informal insurance relationshipsand threats, ranging from social isola-tion to physical retribution, that facili-tate borrowing for households lackingcollateral (Besley and Coate 1995). Theprograms thus combine the scale advan-tages of a standard bank with mecha-nisms long used in traditional, group-based modes of informal finance, suchas rotating savings and credit associa-tions (Besley, Coate, and Glenn Loury1993).5

The Grameen Bank now has over twomillion borrowers, 95 percent of whomare women, receiving loans that total$30–40 million per month. Reported re-cent repayment rates average 97–98percent, but as Section 4.2 describes,relevant rates average about 92 percentand have been substantially lower inrecent years.

Most loans are for one year with anominal interest rate of 20 percent(roughly a 15–16 percent real rate).Calculations described in Section 4.2suggest, however, that Grameen wouldhave had to charge a nominal rate ofaround 32 percent in order to becomefully financially sustainable (holding thecurrent cost structure constant). Themanagement argues that such an in-crease would undermine the bank’s so-cial mission (Shahidur Khandker 1998),

5 In a rotating savings and credit association, agroup of participants puts contributions into a potthat is given to a single member. This is repeatedover time until each member has had a turn, withorder determined by list, lottery, or auction. Mostmicrofinance contracts build on the use of groupsbut mobilize capital from outside the area.ROSCA participants are often women, and in theU.S. involvement is active in new immigrant com-munities, including among Koreans, Vietnamese,Mexicans, Salvadorans, Guatemalans, Trinidadi-ans, Jamaicans, Barbadans, and Ethiopians. In-volvement had been active earlier in the centuryamong Japanese and Chinese Americans, but itis not common now (Light and Pham 1998).Rutherford (1998) and Armendariz and Morduch(1998) describe links of ROSCAs and microfinancemechanisms.


but there is little solid evidence thatspeaks to the issue.

Grameen figures prominently as anearly innovator in microfinance and hasbeen particularly well studied. Assess-ments of its financial performance aredescribed below in Section 4.2, of itscosts and benefits in Section 5.1, andof its social and economic impacts inSection 6.3.

2.2 BancoSol, Bolivia

Banco Solidario (BancoSol) of urbanBolivia also lends to groups but differsin many ways from Grameen.6 First, itsfocus is sharply on banking, not on so-cial service. Second, loans are made toall group members simultaneously, andthe “solidarity groups” can be formed ofthree to seven members. The bank,though, is constantly evolving, and ithas started lending to individuals aswell. By the end of 1998, 92 percent ofthe portfolio was in loans made to soli-darity groups and 98 percent of clientswere in solidarity groups, but it is likelythat those ratios will fall over time. Bythe end of 1998, 28 percent of the port-folio had some kind of guarantee beyondjust a solidarity group.

Third, interest rates are relativelyhigh. While 1998 inflation was below 5percent, loans denominated in bolivi-anos were made at an annual base rateof 48 percent, plus a 2.5 percent com-mission charged up front. Clients withsolid performance records are offeredloans at 45 percent per year, but this isstill steep relative to Grameen (but notrelative to the typical moneylender,who may charge as much as 10 percentper month). About 70–80 percent of

loans are denominated in dollars, how-ever, and these loans cost clients 24–30percent per year, with a 1 percent feeup front.

Fourth, as a result of these rates, thebank does not rely on subsidies, mak-ing a respectable return on lending.BancoSol reports returns on equity ofnearly 30 percent at the end of 1998and returns on assets of about 4.5 per-cent, figures that are impressive relativeto Wall Street investments—althoughadjustments for risk will alter the pic-ture. Fifth, repayment schedules areflexible, allowing some borrowers tomake weekly repayments and others todo so only monthly. Sixth, loan dura-tions are also flexible. At the end of1998, about 10 percent had durationsbetween one and four months, 24 per-cent had durations of four to sevenmonths, 23 percent had durations ofseven to ten months, 19 percent haddurations of ten to thirteen months,and the balance stretched toward twoyears.

Seventh, borrowers are better offthan in Bangladesh and loans are larger,with average loan balances exceeding$900, roughly nine times larger than forGrameen (although first loans may startas low as $100). Thus while BancoSolserves poor clients, a recent study findsthat typical clients are among the “rich-est of the poor” and are clustered justabove the poverty line (where povertyis based on access to a set of basicneeds like shelter and education; SergioNavajas et al. 1998). Partly this may bedue to the “maturation” of clients frompoor borrowers into less poor borrow-ers, but the profile of clients also looksvery different from that of the ma-ture clients of typical South Asianprograms.

The stress on the financial side hasmade BancoSol one of the key forcesin the Bolivian banking system. The

6 The financial information is from Jean Steege,ACCION International, personal communication,January 1999. Claudio Gonzalez-Vega et al. (1997)provide more detail on BancoSol. Further infor-mation can also be found at http://www.accion.org.


institution started as an NGO(PRODEM) in 1987, became a bank in1992, and, by the end of 1998, served81,503 low-income clients. That scalegives it about 40 percent of borrowersin the entire Bolivian banking system.

Part of the success is due to impres-sive repayment performance, althoughdifficulties are beginning to emerge.Unlike most other microfinance institu-tions, BancoSol reports overdues usingconservative standards: if a loan repay-ment is overdue for one day, the entireunpaid balance is considered at risk(even when the planned payment wasonly scheduled to be a partial repay-ment). By these standards, 2.03 percentof the portfolio was at risk at the end of1997. But by the end of 1998, the frac-tion increased to 4.89 percent, a trendthat parallels a general weakeningthroughout the Bolivian banking systemand which may signal the negativeeffects of increasing competition.BancoSol’s successes have spawnedcompetition from NGOs, new nonbankfinancial institutions, and even formalbanks with new loan windows for low-income clients. The effect has been arapid increase in credit supply, and aweakening of repayment incentives thatmay foreshadow problems to comeelsewhere (see Section 3.3).

Still, BancoSol stands as a financialsuccess, and the model has been repli-cated—profitably—by nine of the eigh-teen other Latin American affiliates ofACCION International, an NGO basedin Somerville, Massachusetts. ACCIONalso serves over one thousand clients inthe U.S., spread over the six programs.Average loan sizes range from $1366 inNew Mexico to $3883 in Chicago, andoverall nearly 40 percent of the clientsare female. As of December 1996, pay-ments past due by at least thirty daysaveraged 15.5 percent but ranged ashigh as 21.2 percent in New York and

32.3 percent in New Mexico.7 ACCION’sother affiliates, including six in the UnitedStates, have not, however, achieved fi-nancial sustainability. The largest im-pediments for U.S. programs appear tobe a mixed record of repayment, andusury laws that prevent microfinance in-stitutions from charging interest ratesthat cover costs (Pham 1996).

2.3 Rakyat Indonesia

Like BancoSol, the Bank Rakyat In-donesia unit desa system is financiallyself-sufficient and also lends to “betteroff” poor and nonpoor households, withaverage loan sizes of $1007 during1996. Unlike BancoSol and Grameen,however, BRI does not use a grouplending mechanism. And, unlike nearlyall other programs, the bank requiresindividual borrowers to put up collat-eral, so the very poorest borrowers areexcluded, but operations remain small-scale and “collateral” is often definedloosely, allowing staff some discretion toincrease loan size for reliable borrowerswho may not be able to fully back loanswith assets. Even in the wake of the re-cent financial crisis in Indonesia, repay-ment rates for BRI were 97.8 percent inMarch 1998 (Paul McGuire 1998).

The bank has centered on achievingcost reductions by setting up a network

7 Data are from ACCION (1997) and hold as ofDecember 1996. Five of the six U.S. affiliates haveonly been operating since 1994, and the group as awhole serves only 1,695 clients (but with capitalsecured for expansion). A range of microfinanceinstitutions operate in the U.S. Among the oldestand best-established are Chicago’s South ShoreBank and Boston’s Working Capital. The Cal-Meadow Foundation has recently provided fund-ing for several microfinance programs in Canada.Microfinance participation in the U.S. is heavilyminority-based, with a high ethnic concentration.For example, 90 percent of the urban clients ofBoston’s Working Capital are minorities (and 66percent are female). Loans start at $500. Clientstend to be better educated and have more job ex-perience than average welfare recipients, and just29 percent of Working Capital’s borrowers werebelow the poverty line (Working Capital 1997).


of branches and posts (with an averageof five staff members each) and nowserves about 2 million borrowers and 16million depositors. (The importance ofsavings to BRI is highlighted below inSection 7.) Loan officers get to knowclients over time, starting borrowers offwith small loans and increasing loansize conditional on repayment perfor-mance. Annualized interest rates are 34percent in general and 24 percent ifloans are paid with no delay (roughly 25percent and 15 percent in real terms—before the recent financial crisis).

Like BancoSol, BRI also does not seeitself as a social service organization,and it does not provide clients withtraining or guidance—it aims to earn aprofit and sees microfinance as goodbusiness (Marguerite Robinson 1992).Indeed, in 1995, the unit desa programof the Bank Rakyat Indonesia earned$175 million in profits on their loans tolow-income households. More striking,the program’s repayment rates—andprofits—on loans to poor householdshave exceeded the performance of loansmade to corporate clients by other partsof the bank. A recent calculation sug-gests that if the BRI unit desa programdid not have to cross-subsidize the restof the bank, they could have brokeneven in 1995 while charging a nominalinterest rate of just 17.5 percent peryear on loans (around a 7 percent realrate; Jacob Yaron, McDonald Benjamin,and Stephanie Charitonenko 1998).

2.4 Kredit Desa, Indonesia

The Bank Kredit Desa system(BKDs) in rural Indonesia, a sister insti-tution to BRI, is much less well-known.The program dates back to 1929, al-though much of the capital was wipedout by the hyper-inflation of the middle1960s (Don Johnston 1996). Like BRI,loans are made to individuals and theoperation is financially viable. At the end

of 1994, the BKDs generated profits of$4.73 million on $30 million of net loansoutstanding to 765,586 borrowers.8

Like Grameen-style programs, theBKDs lend to the poorest households,and scale is small, with an emphasis onpetty traders and an average loan size of$71 in 1994. The term of loans is gener-ally 10–12 weeks with weekly repay-ment and interest of 10 percent on theprincipal. Christen et al. (1995) calcu-late that this translates to a 55 percentnominal annual rate and a 46 percentreal rate in 1993. Loan losses in 1994were just under 4 percent of loansoutstanding (Johnston 1996).

Also as in most microfinance programs,loans do not require collateral. The in-novation of the BKDs is to allocatefunds through village-level managementcommissions led by village heads. Thisworks in Indonesia since there is a clearsystem of authority that stretches fromJakarta down to the villages. The BKDspiggy-back on this structure, and themanagement commissions thus build inmany of the advantages of group lend-ing (most importantly, exploiting localinformation and enforcement mecha-nisms) while retaining an individual-lending approach. The commissions areable to exclude the worst credit risksbut appear to be relatively democraticin their allocations. Through the late1990s, most BKDs have had excesscapital for lending and hold balances inBRI accounts. The BKDs are now su-pervised by BRI, and successful BKDborrowers can graduate naturally tolarger-scale lending from BRI units.

2.5 Village Banks

Prospects for replicating the BKDsoutside of Indonesia are limited, how-ever. A more promising, exportable

8 Figures are calculated from Johnston (1996)and data provided by BRI in August 1996.


village-based structure is provided bythe network of village banks started inthe mid-1980s in Latin America byJohn Hatch and his associates at theFoundation for International Commu-nity Assistance (FINCA). The villagebanking model has now been replicatedin over 3000 sites in 25 countries byNGOs like CARE, Catholic Relief Ser-vices, Freedom from Hunger, and Savethe Children. FINCA programs aloneserve nearly 90,000 clients in countriesas diverse as Peru, Haiti, Malawi,Uganda, and Kyrgyzstan, as well as inMaryland, Virginia, and Washington,D.C.

The NGOs help set up village finan-cial institutions in partnership with lo-cal groups, allowing substantial localautonomy over loan decisions and man-agement. Freedom from Hunger, forexample, then facilitates a relationshipbetween the village banks and local com-mercial banks with the aim to createsustainable institutional structures.

The village banks tend to serve apoor, predominantly female clientelesimilar to that served by the GrameenBank. In the standard model, the spon-soring agency makes an initial loan tothe village bank and its 30–50 members.Loans are then made to members, start-ing at around $50 with a four monthterm, with subsequent loan sizes tied tothe amount that members have on de-posit with the bank (they must typicallyhave saved at least 20 percent of theloan value). The initial loan from thesponsoring agency is kept in an “exter-nal account,” and interest income isused to cover costs. The deposits ofmembers are held in an “internal ac-count” that can be drawn down as de-positors need. The original aim was tobuild up internal accounts so that exter-nal funding could be withdrawn withinthree years, but in practice growingcredit demands and slow savings accu-

mulation have limited those aspirations(Candace Nelson et al. 1995).

Like the Indonesian BKDs, the vil-lage banks successfully harness local in-formation and peer pressure without us-ing small groups along BancoSol orGrameen lines. And, as with the BKDs,sustainability is an aim, with nominal in-terest rates as high as 4 percent permonth. Most village banks, however,still require substantial subsidies tocover capital costs. Section 4.1 showsevidence that village banks as a groupcover just 70 percent of total costs onaverage. Partly, this is because many vil-lage banks have been set up in areasthat are particularly difficult to serve(e.g., rural Mali and Burkina Faso), andthe focus has been on outreach ratherthan scale. Worldwide, the number ofclients is measured in the tens of thou-sands, rather than the millions servedby the Grameen Bank and BRI.

3. Microfinance Mechanisms

The five programs above highlightthe diversity of approaches spawned bythe common idea of lending to low-income households. Group lending hastaken most of the spotlight, and theidea has had immediate appeal for eco-nomic theorists and for policymakerswith a vision of building programsaround households’ “social” assets, evenwhen physical assets are few. But itsrole has been exaggerated: group lend-ing is not the only mechanism that dif-ferentiates microfinance contracts fromstandard loan contracts.9 The programsdescribed above also use dynamic in-centives, regular repayment schedules,and collateral substitutes to help main-tain high repayment rates. Lending to

9 Ghatak and Guinnane (1999) provide an excel-lent review of group-lending contracts. MonicaHuppi and Gershon Feder (1990) provide an earlyperspective. Armendariz and Morduch (1998) de-scribe the functioning of alternative mechanisms.


women can also be a benefit from afinancial perspective.

As shown in Table 1, just two of thefive use explicit group-lending con-tracts, but all lend in increasingamounts over time (“progressive” lend-ing), offer terms that are substantiallybetter than alternative credit sources,and cut off borrowers in default. Mostalso require weekly or semi-weekly re-payments, beginning soon after loan re-ceipt. While we lack good evidence onthe relative importance of these mecha-nisms, there is increasing anecdotal evi-dence on limits to group lending per se(e.g., the village studies from Bangla-desh in Aminur Rahman 1998; ImranMatin 1997; Woolcock 1999; Sanae Ito1998; and Pankaj Jain 1996). This sec-tion highlights what is known (or oughtto be known) about the diversity oftechnologies that underlie repaymentrates and screening mechanisms.

3.1 Peer Selection

Group lending has many advantages,beginning with mitigation of problemscreated by adverse selection. The key isthat group-lending schemes provide in-centives for similar types to group to-gether. Ghatak (1999) shows how thissorting process can be instrumental inimproving repayment rates, allowing forlower interest rates, and raising socialwelfare. His insight is that a group-lending contract provides a way to pricediscriminate that is impossible with anindividual-lending contract.10

To see this, imagine two types of po-tential investors. Both types are riskneutral, but one type is “risky” and theother is “safe”; the risky type fails moreoften than the safe type, but the riskytypes have higher returns when success-ful. The bank knows the fraction of

each type in the population, but it isunable to determine which specific in-vestors are of which type. Investors,though, have perfect information abouteach other.

Both types want to invest in a projectwith an uncertain outcome that requiresone unit of capital. If they choose not toundertake the project, they can earnwage income m. The risky investors havea probability of success pr and net re-turn Rr. The safe investors have a prob-ability of success ps and net return Rs.When either type fails, the return is zero.Returns are statistically independent.

Risky types are less likely to be suc-cessful (pr < ps), but they have higher re-turns when they succeed. For simplic-ity, assume that the expected netreturns are equal for both safe and riskytypes: prRr = psRs ≡ R

__. The projects of

both types are socially profitable in thatexpected returns net of the cost of capi-tal, ρ, exceed earnings from wage labor:R__

− ρ > m.Neither type has assets to put up as

collateral, so the investors pay the banknothing if the projects fail. To breakeven, the bank must set the interestrate high enough to cover its per-loancapital cost, ρ. If both types borrow, theequilibrium interest rate under compe-tition will then be set so that rp– = ρ,where p– is the average probability ofsuccess in the population. Since thebank can’t distinguish between borrow-ers, all investors will face interest rate,r. As a result, safe types have lower ex-pected returns than risky types—sinceR__

− rps < R__

− rpr —and the safe types willenter the market only if their expectednet return exceeds their fallback posi-tion: R

__ − rps > m. If the safe types enter,

the risky types will too.But the safe types will stay out of the

market if R__

− rps < m, and only riskytypes might be left in the market. Inthat case, the equilibrium interest rate

10 Armendariz and Gollier (1997) also describethis mechanism in parallel work.


will rise so that rpr = ρ. Risky types driveout the safe. The risky types lose theimplicit cross-subsidization by the safetypes, while the safe types lose access tocapital. This second-best scenario is in-efficient since only the risky types bor-row, even though the safe types alsohave socially valuable projects.

Can a group-lending scheme improveon this outcome? If it does, it mustbring the safe types back into the mar-ket. For simplicity, consider groups oftwo people, with each group formedvoluntarily. Individuals invest indepen-dently, but the contract is written tocreate joint liability. Imagine a contractsuch that each borrower pays nothing ifher project fails, and an amount r∗ ifher project is successful. In addition,the successful borrower pays a joint-liability payment c∗ if the other mem-ber of the group fails.11 The expectednet return of a safe type teamed with arisky type is then R

__ − ps(r∗ + (1 − pr)c∗),

with similar calculations for exclusivelysafe and exclusively risky groups.

Will the groups be homogeneous ormixed? Since safe types are always pre-ferred as partners (since their prob-ability of failure is lower), the questionbecomes: will the risky types be willingto make a large enough transfer to thesafe types such that both risky and safetypes do better together? By comparingexpected returns under alternative sce-narios, we can calculate that a safe typewill require a transfer of at leastps(ps − pr)c∗ to agree to form a partner-ship with a risky type. Will risky typesbe willing to pay that much? Their ex-

pected net gain from joining with a safetype is as much as pr(ps − pr)c∗. But sincepr < ps, the expected gains to risky typesare always smaller than the expectedlosses to safe types. Thus, there is nomutually beneficial way for risky andsafe types to group together. Grouplending thus leads to assortative match-ing: all types group with like types(Gary Becker 1991).12

How does this affect the functioningof the credit market? Ghatak (1999)demonstrates that the group-lendingcontract provides a way to charge dif-ferent effective fees to risky and safetypes—even though all groups face ex-actly the same contract with exactly thesame nominal charges, r∗ and c∗. Theresult arises because risky types will beteamed with other risky types, whilesafe types team with safe types. Riskytypes then receive expected net returnsof R

__ − pr(r∗ + (1 − pr)c∗), while safe types

receive expected net returns ofR__

− ps(r∗ + (1 − ps)c∗). Thus, a successfulrisky type is more likely to have to paythe joint-liability payment c∗ than asuccessful safe type. If r∗ and c∗ are setappropriately, the group-lending con-tract can provide an effective way toprice discriminate that is impossibleunder the standard second-best indi-vidual-lending contract. If ps = 0.9 andpr = 0.8, for example, the safer typescan expect to pay less than the riskiertypes as long as the joint liabilitypayment is set so that c∗ > 1.4r∗.

Efficiency gains result if the differenceis large enough to induce the safe typesback into the market. When this hap-pens, average repayment rates rise, andthe bank can afford to maintain a lowerinterest rate r∗ while not losing money.

11 In typical contracts, group members are re-sponsible for helping to pay off the loan in diffi-culty, rather than having to pay a fixed penalty fora group member’s default. While clients lack col-lateral, they are assumed to have a large enoughincome flow to cover these costs if needed. Inpractice this may impose a constraint on loan sizesince individuals may have increasing difficultypaying c∗ + r∗ when loan sizes grow large.

12 Ghatak (1998) extends the results to groupslarger than 2, a continuum of types, and prefer-ences against risk. See also Varian (1990) and Ar-mendariz and Gollier (1997) on related issues ofefficiency and sorting.


3.2 Peer Monitoring

Group lending may also providebenefits by inducing borrowers not totake risks that undermine the bank’sprofitability (Stiglitz 1990; Besley andCoate 1995). This can be seen byslightly modifying the framework inSection 3.1 to consider moral hazard.Instead, consider identical risk averseborrowers with utility functions u(x).

Each borrower may do either risky orsafe activities, and each activity againrequires the same capital cost. Thebank, as above, has imperfect informa-tion about borrowers—in particular,here it cannot tell whether the borrow-ers have done the safe or risky activity.Moral hazard is thus a prime concern.When projects fail, borrowers have a re-turn of zero, and a borrower’s utilitylevel when projects fail is normalized tozero as well.

We start with the standard individual-lending contract. Borrowers either haveexpected utility psu(Rs − r) or pru(Rr − r),depending on whether they do the safeor risky activity. If everyone did thesafe activity, the bank could charge aninterest rate of r = ρ/ps and break even.But, since the bank cannot see whichactivity is chosen (and thus cannot con-tract on it), borrowers may fare betterdoing the risky activity and getting ex-pected utility E[Usr] = pru(Rr − ρ/ps). Thebank then loses money. Thus, the bankraises interest rates to r = ρ/pr. Now theborrower gets expected utility ofE[Urr] = pru(Rr − ρ/pr), and she is clearlyworse off than with a lower interestrate. In fact, if the borrower couldsomehow commit to doing the safe ac-tivity, she could be better off—with ex-pected utility E[Uss] = psu(Rs − ρ/ps). Thusthe borrower prefers E[Usr] to E[Uss] toE[Urr], but the information problemand inability to commit means that shealways gets the worst outcome, E[Urr].

How can a group-lending contractimprove matters? The key is that it cancreate a mechanism that gives borrow-ers an incentive to choose the safe ac-tivity. Again consider groups of two bor-rowers and group-lending contracts likethose in Section 3.1 above. The borrow-ers in each group have the ability toenforce contracts between each other,and they jointly decide which typesof activities to undertake. Now theirproblem is to choose between both do-ing the safe activity, yielding each bor-rower expected utility of ps2u(Rs − r∗) +ps(1 − ps)u(Rs − r∗ − c∗), or doing therisky activity with expected utilitypr2u(Rr − r∗) + pr(1 − pr)u(Rr − r∗ − c∗). Ifthe joint-liability payment c∗ is set highenough, borrowers will always choose todo the safe activity (Stiglitz 1990).

This is good for the bank, but it sad-dles borrowers with extra risk. Thebank, though, knows borrowers will nowdo the safe activity, and it earns extraincome from the joint-liability pay-ments. The bank can thus afford tolower the interest rate to offset theburden.

Thus, through exploiting the abilityof neighbors to enforce contracts andmonitor each other—even when thebank can do neither—the group-lendingcontract again offers a way to lowerequilibrium interest rates, raise expectedutility, and raise expected repaymentrates.

3.3 Dynamic Incentives

A third mechanism for securing highrepayment rates with high monitoringcosts involves exploiting dynamic incen-tives (Besley 1995, p. 2187). Programstypically begin by lending just smallamounts and then increasing loan sizeupon satisfactory repayment. The re-peated nature of the interactions—andthe credible threat to cut off any futurelending when loans are not repaid—can


be exploited to overcome informationproblems and improve efficiency,whether lending is group-based orindividual-based.13

Incentives are enhanced further ifborrowers can anticipate a stream of in-creasingly larger loans. (Hulme andMosley 1996 term this “progressivelending,” and the ACCION networkcalls it “step lending.”) As above, keep-ing interest rates relatively low is criti-cal, since the advantage of microfinanceprograms lies in their offering servicesat rates that are more attractive thancompetitors’ rates. Thus, the Bank Rak-yat Indonesia (BRI) and BancoSolcharge high rates, but they keep levelswell below rates that moneylenderstraditionally charge.

However, competition will diminishthe power of the dynamic incentivesagainst moral hazard—a problem thatboth the Bank Rakyat Indonesia andBancoSol are starting to feel as othercommercial banks see the potentialprofitability of their model. In practice,though, real competition has yet to befelt by most microfinance institutions(perhaps because so few are actuallyturning a profit). As competition grows,the need for a centralized credit ratingagency will also grow.

Dynamic incentives will also workbetter in areas with relatively low mo-bility. In urban areas, for example,where households come and go, it maynot be easy to catch defaulters whomove across town and start borrowingagain with a clean slate at a differentbranch or program. BRI has facedgreater trouble securing repayments intheir urban programs than in their ruralones, which may be due to greaterurban mobility.

Relying on dynamic incentives alsoruns into problems common to all finiterepeated games. If the lending relation-ship has a clear end, borrowers have in-centives to default in the final period.Anticipating that, the lender will notlend in the final period, giving borrow-ers incentives to default in the penulti-mate period—and so forth until the en-tire mechanism unravels. Thus, unlessthere is substantial uncertainty aboutthe end date—or if “graduation” from oneprogram to the next is well-established(ad infinitum), dynamic incentives havelimited scope on their own.

One quite different advantage of pro-gressive lending is the ability to testborrowers with small loans at the start.This feature allows lenders to developrelationships with clients over time andto screen out the worst prospects beforeexpanding loan scale (Parikshit Ghoshand Debraj Ray 1997).

Dynamic incentives can also help toexplain advantages found in lending towomen. Credit programs like those ofthe Grameen Bank and the BangladeshRural Advancement Committee (BRAC)did not begin with a focus on women.In 1980–83, women made up 39 percentand 34 percent of their respective mem-berships, but by 1991–92, BRAC’smembership was 74 percent female andGrameen’s was 94 percent female (AnneMarie Goetz and Rina Sen Gupta 1995).As Table 2 shows, many other programsalso focus on lending to women, and itappears to confer financial advantageson the programs. At Grameen, for ex-ample, 15.3 percent of male borrowerswere “struggling” in 1991 (i.e., missingsome payments before the final duedate) while this was true for just 1.3percent of women (Khandker, BaquiKhalily, and Zahed Kahn 1995).

The decision to focus on women hassome obvious advantages. The lowermobility of women may be a plus where

13 See the general theoretical treatment in Bol-ton and Scharfstein (1990) and the application tomicrofinance contracts in Armendariz and Mor-duch (1998).


ex post moral hazard is a problem (i.e.,where there is a fear that clients will“take the money and run”). Also, wherewomen have fewer alternative borrow-ing possibilities than men, dynamicincentives will be heightened.14

Thus, ironically, the financial successof many programs with a focus onwomen may spring partly from the lackof economic access of women, while, atthe same time, promotion of economicaccess is a principal social objective(Syed Hashemi, Sidney Ruth Schuler,and Ann P. Riley 1996).

3.4 Regular Repayment Schedules

One of the least remarked upon—butmost unusual—features of most microfi-nance credit contracts is that repay-ments must start nearly immediately af-ter disbursement. In a traditional loancontract, the borrower gets the money,invests it, and then repays in full withinterest at the end of the term. But atGrameen-style banks, terms for a year-long loan are likely to be determined byadding up the principal and interest duein total, dividing by 50, and startingweekly collections a couple of weeks af-ter the disbursement. Programs likeBancoSol and BRI tend to be more flex-ible in the formula, but even they donot stray far from the idea of collectingregular repayments in small amounts.

The advantages are several. Regularrepayment schedules screen out undis-ciplined borrowers. They give earlywarning to loan officers and peer groupmembers about emerging problems.

TABLE 2PERFORMANCE INDICATORS OF MICROFINANCE PROGRAMS

ObservationsAverage loanbalance ($)

Avg. loan as% of GNPper capita

Averageoperational

sustainability

Averagefinancial

sustainability

Sustainability All microfinance institutions 72 415 34 105 83 Fully sustainable 34 428 39 139 113Lending method Individual lending 30 842 76 120 92 Solidarity groups 20 451 35 103 89 Village bank 22 94 11 91 69Target Group Low end 37 133 13 88 72 Broad 28 564 48 122 100 High end 7 2971 359 121 76Age 3 to 6 years 15 301 44 98 84 7 or more years 40 374 27 123 98

Source: Statistical appendix to MicroBanking Bulletin (1998). Village banks have a “B” data quality; all others aregraded “A”. Portfolio at risk is the amount in arrears for 90 days or more as a percentage of the loan portfolio.Averages exclude data for the top and bottom deciles.

14 Rahman (1998) describes complementary cul-tural forces based on women’s “culturally pat-terned behavior.” Female Grameen Bank borrow-ers in Rahman’s study area, for example, are foundto be much more sensitive to verbal hostilityheaped on by fellow members and bank workerswhen repayment difficulties arise. The stigma isexacerbated by the public collection of paymentsat weekly group meetings. According to Rahman(1998), women are especially sensitive since theirmisfortune reflects poorly on the entire household(and lineage), while men have an easier time shak-ing it off.


And they allow the bank to get hold ofcash flows before they are consumed orotherwise diverted, a point developedby Stuart Rutherford (1998).

More striking, because the repaymentprocess begins before investments bearfruit, weekly repayments necessitatethat the household has an additional in-come source on which to rely. Thus, in-sisting on weekly repayments meansthat the bank is effectively lendingpartly against the household’s steady,diversified income stream, not just therisky project. This confers advantagesfor the bank and for diversified house-holds. But it means that microfinancehas yet to make real inroads in areas fo-cused sharply on highly seasonal occu-pations like agricultural cultivation.Seasonality thus poses one of the largestchallenges to the spread of microfi-nance in areas centered on rainfedagriculture, areas that include some ofthe poorest regions of South Asia andAfrica.

3.5 Collateral Substitutes

While few programs require collat-eral, many have substitutes. For exam-ple, programs following the Grameenmodel require that borrowers contrib-ute to an “emergency fund” in theamount of 0.5 percent of every unit bor-rowed (beyond a given scale). Theemergency fund provides insurance incases of default, death, disability, etc.,in amounts proportional to the length ofmembership. An additional 5 percent ofthe loan is taken out as a “group tax”that goes into a group fund account. Upto half of the fund can be used by groupmembers (with unanimous consent).Typically, it is disbursed among thegroup as zero-interest loans with fixedterms. Until October 1995, GrameenBank members could not withdrawthese funds from the bank, even uponleaving. These “forced savings” can nowbe withdrawn upon leaving, but only af-ter the banks have taken out what they

TABLE 2 (Cont.)

Avg. return on equity

Avg. percent ofportfolio at risk

Avg. percentfemale clients

Avg. number ofactive borrowers

Sustainability All microfinance institutions –8.5 3.3 65 9,035 Fully sustainable 9.3 2.6 61 12,926Lending method Individual lending –5.0 3.1 53 15,226 Solidarity groups –3.0 4.1 49 7,252 Village bank –17.4 2.8 92 7,833Target Group Low end –16.2 3.8 74 7,953 Broad 1.2 3.0 60 12,282 High end –6.2 1.9 34 1,891Age 3 to 6 years –6.8 2.2 71 9,921 7 or more years –2.4 4.1 63 16,557


are owed. Thus, in effect, the fundsserve as a form of partial collateral.

The Bank Rakyat Indonesia’s unitdesa program is one of the few pro-grams to require collateral explicitly. Itsadvocates, however, emphasize insteadthe role of dynamic incentives in gener-ating repayments (Richard Patten andJay Rosengard 1991; Robinson 1992). Itis impossible, though, to determine eas-ily which incentive mechanism is mostimportant in driving repayment rates.While bank officials point out that col-lateral is almost never collected, thisdoes not signal its lack of importance asan incentive device. If the threat of col-lection is believable, there should befew instances when collateral is actuallycollected.

BancoSol also stresses the role ofsolidarity groups in assuring repay-ments, but as its clients have prosperedat varying rates, lending approacheshave diversified as well. As noted inSection 2.2, by the end of 1998, 28 per-cent of its portfolio had some kind ofguarantee beyond the solidarity group.

3.6 Empirical Research Agenda

Do the mechanisms above function asadvertised? Is there evidence of assorta-tive matching through group lending aspostulated by Ghatak (1999)? Are fu-ture loan terms predicted by laggedperformance, as suggested by the the-ory of dynamic incentives? Extendingthe theory further, does the group-lend-ing contract heighten default prob-abilities for the entire group when somemembers run into difficulties, as pre-dicted by Besley and Coate (1995)?Does group lending lead to excessivemonitoring and excessive pressure toundertake “safe” projects rather thanriskier and more lucrative projects(Banerjee, Besley, and Guinnane1992)? Is the group-lending structureless flexible than individual lending for

borrowers in growing businesses andthose that outstrip the pace of theirpeers (Madajewicz 1997; Woolcock1998)? Are weekly meetings particularlycostly (for both borrowers and bankstaff) in areas of low population densityand at busy agricultural seasons? Do so-cial programs enhance economic perfor-mance? When default occurs, do bankstaff follow the letter of the law and cutoff good clients with the misfortune tobe in groups with unlucky neighbors?Or is renegotiation common (Hashemiand Sidney Schuler 1997; Matin 1997;Armendariz and Morduch 1998)?

Most of the theoretical propositionsare supported with anecdotes from par-ticular programs, but they have notbeen established as empirical regulari-ties. Better research is needed to sharpenboth the growing body of microfinancetheory and ongoing policy dialogues.

Empirical understandings of microfi-nance will also be aided by studies thatquantify the roles of the various mecha-nisms in driving microfinance perfor-mance. The difficulty in these inquiries isthat most programs use the same lend-ing model in all branches. Thus, there isno variation off of which to estimate theefficacy of particular mechanisms. Well-designed experiments would help (e.g.,individual-lending contracts to some ofthe sample, group-lending contracts toothers; weekly repayments for some,monthly or quarterly schedules for others).

Lacking well-designed experiments, acollection of studies instead presentsregressions in which repayment ratesare explained by proxies for forces be-hind particular mechanisms. The vari-ation thus arises from features of theeconomic environment that affect theefficacy of particular program features:How good are information flows? Howcompetitive are credit markets? Howstrong are informal enforcement mech-anisms? The variation in answers to


these questions allows econometric esti-mation, but the evidence is indirect andsubject to multiple interpretations sincethe strength of information flows, mar-kets, and enforcement mechanisms isunlikely to matter only through theform of credit contract. In addition, se-lection biases of the sort raised in Sec-tion 6.1 are likely to apply. Still, someresults are provocative.

For example, Wydick (1999) reportson a survey of an ACCION Interna-tional affiliate in western Guatemalatailored to elicit information aboutgroups. He finds that improvements inrepayment rates are associated withvariables that proxy for the ability tomonitor and enforce group relation-ships, such as knowledge of the weeklysales of fellow group members. Hefinds little impact, though, of social tiesper se: friends do not make more reli-able group members than others. In fact,members are sometimes softer on theirfriends, worsening average repaymentrates.

Mark Wenner (1995) investigates re-payment rates in 25 village banks inCosta Rica affiliated with FINCA. Hefinds active screening that successfullyexcludes the worst credit risks, workingin a more straightforward way than inthe simple model of peer selection inSection 3.1 above. He also finds thatdelinquency rates are higher in betteroff towns. This lends support to the the-ory of dynamic incentives: where bor-rowers have better alternatives, they arelikely to value the programs less, andthis drives up default rates.

The result is echoed by ManoharSharma and Manfred Zeller (1996) in theirstudy of three programs in Bangladesh(but not Grameen). They find that re-payment rates are higher in remotecommunities—i.e., those with fewer al-ternative credit programs. Khandker etal. (1995, Table 7.2), however, find the

opposite in considering other Bangla-desh banks (including Grameen). Bothdrop-out rates and repayment rates in-crease in better-developed villages.This may be a product of improved li-quidity and better business opportuni-ties in better-off villages, but it mightalso reflect selection bias.

These bits of evidence show thatgroup lending is a varied enterprise andthat there is much to microfinance be-yond group lending. Narrowing the gapbetween theory and evidence will be animportant step toward improving andevaluating programs.

4. Profitability and FinancialSustainability

Microfinance discussions pay surpris-ingly little attention to particular mech-anisms relative to how much attentionis paid to purely financial matters. Ac-cordingly, this section considers fi-nances, and social issues are taken upagain in Section 5.

How well in the end have microfi-nance programs met their financialpromise? A recent survey finds 34 prof-itable programs among a group of 72with a “commitment” to financial sus-tainability (MicroBanking Bulletin1998). This does not imply, however,that half of all programs worldwide areself-sufficient. The hundreds of pro-grams outside the base 72 continue todepend on the generosity of donors(e.g., Grameen Bank and most of itsreplicators do not make the list of 72,although BancoSol and BRI do). Someexperts estimate that no more than 1percent of NGO programs worldwideare currently financially sustainable—and perhaps another 5 percent of NGOprograms will ever cross the hurdle.15

15 The figures are based on an informal polltaken by Richard Rosenberg at a microfinanceconference (personal communication, Nov. 1998).


The other 95 percent of programs inoperation will either fold or continuerequiring subsidies, either because theircosts are high or because they choose tocap interest rates rather than to passcosts on to their clients. Although subsi-dies remain integral, donors and practi-tioners have been reluctant to discussoptimal subsidies to alleviate poverty,perhaps for fear of appearing retro-grade in light of the disastrous experi-ences with subsidized government-runprograms. Instead, rhetoric privilegesfinancial sustainability.

4.1 International Evidence

Table 2 gives financial indicators forthe 72 programs in the MicroBankingBulletin survey.16 The 72 programs havebeen divided into non-exclusive catego-ries by age, lending method, targetgroup, and level of sustainability.17

(There is considerable overlap, for ex-ample, between the village bank cate-gory and the group targeting “low end”borrowers.)

The groups, divided by lendingmethod and target group, demonstratethe diversity of programs marching be-hind the microfinance banner. Averageloan balances range from $94 to $842when comparing village banks to thosethat lend exclusively to individuals. Thefocus on women varies from 92 percentto 53 percent. The target group cate-gory makes the comparison starker,

with average loan balances varying from$133 to $2971. Averages for the 34 fullysustainable institutions are not, how-ever, substantially different from theoverall sample in terms of average loanbalance or the percentage of femaleclients.

Sustainability is generally consideredat two levels. The first is operationalsustainability. This refers to the abilityof institutions to generate enough reve-nue to cover operating costs—but notnecessarily the full cost of capital. Ifunable to do this, capital holdings aredepleted over time. The second level ofconcern is financial sustainability. Thisis defined by whether or not the in-stitution requires subsidized inputs inorder to operate. If the institution isnot financially sustainable, it cannotsurvive if it has to obtain all inputs (es-pecially capital) at market, rather thanconcessional, rates.

Most of the programs in the surveyhave crossed the operational sustain-ability hurdle. The only exceptions arethe village banks and those with lowend targets, both of which generateabout 90 percent of the requiredincome.18

Many fewer, however, can cover fullcapital costs as well. Overall, programsgenerate 83 percent of the required in-come and the village bank/low end tar-get groups generate about 70 percent.Strikingly, the handful of programs thatfocus on “high end” clients are just asheavily subsidized as those on the lowend. Similarly, the financial perfor-mance of programs with individual

16 The project started as a collaboration with theAmerican Economic Association’s Economics In-stitute in Boulder, Colorado.

17 Those with low end target groups have aver-age loan balances under $150 or loans as a per-centage of GNP per capita under 20 percent (theyinclude, for example, FINCA programs). Thosewith broad targets have average balances that are20–85 percent of GNP per capita (and includeBancoSol and the BRI unit desa system). The highend programs make average loans greater than 120percent of GNP per capita. The solidarity groupmethodology is based on groups with 3–5 borrow-ers (like BancoSol). The village banks have groupswith over five borrowers.

18 See Mark Schreiner (1997) and Khandker(1998) for discussions of alternative views of sus-tainability. Unlike other reported figures, thosehere make adjustments to account for subsidies oncapital costs, the erosion of the value of equity dueto inflation, and adequate provisioning for non-re-coverable loans. To the extent possible, the figuresare comparable to data for standard commercialenterprises.


loans is roughly equivalent to that ofprograms using solidarity groups, eventhough the former serve a clientele thatis more than twice as rich.

The greatest financial progress hasbeen made by broad-based programslike BancoSol and BRI that serve cli-ents across the range. Financial pro-gress also improves with age (althoughcomparisons of young and old groupscan only be suggestive as their orienta-tions tend to differ).19

The returns to equity echo the dataon financial sustainability. The numbersgive profits relative to the equity putinto the programs. The table shows thatthis is not a place to make big bucks.While average returns to equity of 9.3percent for the financially-sustainableprograms are respectable, they do notcompete well with alternative invest-ments and often carry considerable risk.At the same time, social returns maywell be high even if financial returnsare modest (or negative). On average,the broad-based programs, for example,cover all costs and serve a large pool ofclients with modest incomes, most ofwhom are women. Wall Street wouldsurely pass by the investment opportu-nity, but socially-minded investorsmight find the trade-off favorable.

If returns to equity could be in-creased through more effective leverag-ing of equity, however, Wall Street mighteventually be willing to take a look. In-creasing leverage is thus the cuttingedge for financially-minded microfinanceadvocates, and it has taken microfi-nance discussions to places far fromtheir original focus on how to make$100 loans to Bolivian street vendors.

If donors tire of footing the bill formicrofinance, achieving financial sus-tainability and increasing returns to eq-uity is the only game to play. The issue is:will donors tire if social returns can beproven to justify the costs? Answeringthe question puts impact studies and cost–benefit analyses high on the researchagenda. It also requires paying close at-tention to the basis of self-reportedclaims about financial performance.

4.2 The Grameen Bank Example

The data above have been adjusted tobring them into rough conformity withstandard accounting practices. This isnot typical: microfinance statistics areoften calculated in idiosyncratic waysand are vulnerable to misinterpretation.The Grameen Bank has been relativelyopen with its data, and it provides a fullset of accounts in its annual reports.

Table 3 provides evidence on theGrameen Bank’s performance between1985 and 1996.20 The table shows Gra-meen’s rapid increase in scale, with thesize of the average annual loan portfolioincreasing from $10 million in 1985 to$271 million by 1996. Membership hasexpanded 12 times over the sameperiod, reaching 2.06 million by 1996.

The bank reports repayment ratesabove 98 percent and steady profits—and this is widely reported (e.g., NewYork Times 1997). All accounting defi-nitions are not standard, however. Thereported overdue rates are calculatedby Grameen as the value of loans over-due greater than one year, divided by

19 None of the U.S. programs that I know of areprofitable, and some are very far from financialsustainability, held back by legal caps on interestrates (Michael Chu 1996). None of the U.S. pro-grams are included in the MicroBanking Bulletinsurvey.

20 The base data are drawn from Grameen Bankannual reports. This section draws on Morduch(1999). Summaries of Grameen’s financial perfor-mance through 1994 can be found in Hashemi andSchuler (1997) and Khandker, Khalily, and Kahn(1995). Schreiner (1997) provides alternative cal-culations of subsidy dependence with illustrationsfrom Grameen. The adjustments here capture themost critical issues, but they are not comprehen-sive—for example, no adjustment is made for theerosion of equity due to inflation.


the current portfolio. A problem is thatthe current portfolio tends to be muchlarger than the portfolio that existedwhen the overdue loans were firstmade. With the portfolio expanding 27times between 1985 and 1996, reporteddefault rates are considerably lowerthan standard calculation of arrears(which instead immediately capturesthe share of the portfolio “at risk”). Theadjusted rates replace the denominatorwith the size of the portfolio at the timethat the loans were made.

Doing so can make a big difference:overall, overdues averaged 7.8 percentbetween 1985 and 1996, rather than thereported 1.6 percent. The rate is stillimpressive relative to the performanceof government development banks, but

it is high enough to start creating finan-cial difficulties. More dramatically, thebank reported an overdue rate of 0.8percent in 1994, while at the same timeI estimate that 15 percent of the loansmade that year were unrecovered.

Similarly, reported profits differ con-siderably from adjusted profits in Table3. The main adjustment is to make ade-quate provision for loan losses. Until re-cently, the bank had been slow to writeoff losses, and the adjusted rates ensurethat in each year the bank writes off amodest 3.5 percent of its portfolio (still,considerably less than the 7.8 percentaverage overdue rate). The result islosses of nearly $18 million between1985 and 1996, rather than the bank’sreported $1.5 million in profits.

TABLE 3GRAMEEN BANK: SELECTED FINANCIAL INDICATORS

(Millions of 1996 U.S. dollars)

1985 1990 1992 1994 1996

1985–1996

average

Size Average annual loans outstanding 10.0 58.3 83.8 211.5 271.3 108 Members (thousands) 172 870 1,424 2,013 2,060 1,101Overdues rates (%) Reported overdues rate 2.8 3.3 2.5 0.8 13.9 1.6A

Adjusted overdues rate 3.8 6.2 1.9 15.0 — 7.8A

Profits Reported profits 0.02 0.09 –0.15 0.56 0.46 1.5B

Adjusted profits –0.33 –1.51 –3.06 –0.93 –2.28 –17.8B

Subsidies Direct grants 0.0 2.3 1.7 2.0 2.1 16.4B

Value of access to soft loans 1.1 7.0 5.8 9.0 12.7 80.5B

Value of access to equity 0.0 0.4 2.7 8.0 8.8 47.3B

Subsidy per 100 units outstanding 11 21 16 7 9 11Interest rates (%) Average nominal on-lending rate 16.8 11.1 15.8 16.7 15.9 15.9 Average real on-lending rate 5.9 3.0 11.6 13.1 10.1 10.1 Benchmark cost of capital Average nominal cost of capital

15.07.9

15.02.2

13.52.1

9.45.5

10.33.4

11.33.7

Subsidy dependence index 80 263 106 45 65 74 Avg. nominal “break-even” rate 30.2 40.2 32.6 24.2 26.2 25.7

Source: Morduch (1999) based on data from various years of the Grameen Bank Annual Report. Notes: A: average for 1985–94, weighted by portfolio size. B: Sum for 1985–96.


Grants from donors are consideredpart of income in the profit calcula-tions. If the bank had to rely only onincome from lending and investments,it would have instead suffered losses of$34 million between 1985 and 1996.

The bulk of the bank’s subsidies en-ter through soft loans, however. Gra-meen paid an average of 3.7 percent onborrowed capital (a –1.7 percent realrate). Had it not had access to conces-sional rates, it would have had to payconsiderably more. Here, an alternativebenchmark capital cost measure is ap-proximated as the Bangladesh depositrate from IMF International FinancialStatistics (1996) plus a 3 percent adjust-ment for transactions costs. The differ-ence in rates yields a total value of ac-cess to soft loans of $80.5 millionbetween 1985 and 1996. An additionalimplicit subsidy of $47.3 million was re-ceived by Grameen through access toequity which was used to generatereturns below opportunity costs.

Although subsidies have increasedover time in absolute quantities, thebank’s scale has grown even morequickly. As a result, the annual subsidyper dollar outstanding has fallen sub-stantially, leveling off at about ten centson the dollar.

The subsidy dependence index sum-marizes the subsidy data by yielding anestimate of the percentage increase inthe interest rate required in order forthe bank to operate without subsidies ofany kind (Yaron 1992). The result for1985–96 indicates that in the early1990s Grameen would have had to in-crease nominal interest rates on its gen-eral loan product from 20 percent toabove 50 percent. Overall, the averagebreak-even rate is 32 percent (the aver-age on-lending rate is lower than 20percent since about one quarter of theportfolio is comprised of housing loansoffered at 8 percent interest per year).

While borrowers would not be happy, itis not obvious that they would defect.Clients of the Bangladesh Rural Ad-vancement Committee, a Grameencompetitor with a similar client base,are already paying 30 percent nominalbase interest rates, for example.

Alternatively, radically strippingdown administrative costs would pro-vide breathing room. In the early 1990ssalary and personnel costs accountedfor half of Grameen’s total costs, whileinterest costs were held below 25 per-cent. Decreasing wages has been impos-sible since they are linked to govern-ment wage scales, so the emphasis hashad to be on increasing efficiency. By1996, salary and personnel costs wereroughly equal to interest costs (Mor-duch 1999). Training costs have alsobeen high. One study found that in1991, 54 percent of female trainees and30 percent of male trainees dropped outbefore taking up first positions withGrameen—and much of Grameen’s di-rect grants are funneled to supportingtraining efforts (Khandker, Khalily, andKahn 1995).

The Association for Social Advance-ment (ASA), another large microfinancepresence in Bangladesh, demonstrates amore radical approach to cost control.They have streamlined record keepingand simplified operations so that, forexample, only one loan type is offered—versus Grameen’s choice of generalloans, housing loans, collective loans,seasonal loans, and, more recently,lease/loan arrangements. ASA thus feelscomfortable hiring staff with fewer for-mal qualifications than Grameen, andstaff retention is aided. ASA has alsoeliminated mid-level branch offices andhas centered nearly exclusively on thelarger groups of forty village members,rather than the five-member subgroups.

The Grameen Bank’s current path, pur-suing cross-subsidization and alternative


income generation projects (includingan internet provision service and otherfor-profit spin-offs) is appealing in themedium term, but it has its own perils:the bank’s mission risks getting diluted,and profitable sectors are vulnerable tocompetition over time.

Grameen’s self-reported successeshave been exaggerated, but even if thebank is not the economic miracle thatmany have claimed, it is not obviousthat its failure to reach financial self-sufficiency is in itself a problem. Aslong as benefits sufficiently exceedcosts and donors remain committed tothe cause, Grameen could hold up as awise social investment.

5. Costs and Benefits of Credit Subsidies

Nearly all programs espouse financialsustainability as a key principle. At thesame time, nearly all programs rely onsubsidies of one sort or another. Thesesubsidies are typically viewed as tem-porary aids that help programs over-come start-up costs, not as ongoing pro-gram features. It is the familiar “infantindustry” argument for protection.

The anti-subsidy stance springs froma series of worries. First, donors can befickle, and programs that aim to existinto the future feel the need for inde-pendence. Second, donor budgets arelimited, restricting the scale of opera-tions to the size of the dole. Self-suffi-cient programs, on the other hand, canexpand to meet demand. Third, subsi-dized programs run the risk of becom-ing inefficient without hard bottomlines. Fourth, in the past subsidies haveended up in the wrong hands, ratherthan helping poor households.

The view that subsidies should just betemporary has meant that calculatingthe costs and benefits of subsidies hasnot been an important part of microfi-nance practice, and there have been no

careful cost-benefit studies to date. Butthe fact is that subsidies are an ongoingreality: some “infants” are getting old.Moreover, many of the worries aboutproblems associated with subsidies canlikely be overcome.21

It is true that donors can be fickle,but governments will remain committedto poverty alleviation well after interna-tional agencies have moved on to thenext Big Idea. If subsidized microfi-nance proves to deliver more bang forthe buck than other social investments,should subsidies be turned down?

Scale certainly matters, but often asmall well-targeted program may domore to alleviate measured poverty thana large, poorly-targeted program. Con-sider this example from Morduch(2000). Assume that the typical client ina subsidized program has an income of,say, 50 percent of the poverty line,while the typical client of a sustainable(high interest rate) program has an in-come of 90 percent of the poverty line.To clarify the comparison, assume thatthe net impacts on income per borrowerare identical for the programs (afterrepaying loans with interest).

Minimizing poverty as measured bythe commonly-used “squared povertygap” of James Foster, Joel Greer, andErik Thorbecke (1984), for example,suggests that raising the poorer bor-rower’s income by one dollar has fivetimes greater impact than doing thesame for the less poor borrower. If thesustainable program has 63,000 clients(roughly the size of Bolivia’s BancoSolin the early 1990s), the subsidized pro-gram would need to reach just 12,600clients to have an equivalent impact. The

21 This section draws heavily on Morduch(2000). Adams, Graham, and von Pischke (1982)present a well-argued alternative perspective.Schreiner (1997) presents a framework for consid-ering cost-effectiveness applied to BancoSol andthe Grameen Bank.


comparison is too simple, but it amplyillustrates how social weights and depthof outreach can outweigh concerns withscale.

The third issue, the danger of slip-ping into inefficiency, has been demon-strated many times over by large publicbanks in low-income countries. But thekey to efficiency is the maintenance ofhard budget constraints, not necessarilyprofits. Several donors already use strictand explicit performance targets whenlending to microfinance institutions,conditioning future tranches on perfor-mances to date. The lessons can be ap-plied more widely and used to promoteefficiency and improve targeting in abroader range of subsidized programs.

5.1 Simple Cost–Benefit Ratios

How should costs and benefits becompared? A simple gauge can beformed by dividing the value of subsi-dies by a measure of benefits accruingto borrowers. For example, Khandker(1998) reports a cost–benefit ratio of0.91 with respect to improvements inhousehold consumption via borrowingby women from the Grameen Bank.This means that it costs society 91 centsfor every dollar of benefit to clients. Asimilar calculation leads to a cost–bene-fit ratio of 1.48 for borrowing by men.The ratio is higher, since lending tomen appears to have a smaller impacton household consumption (Mark Pittand Khandker 1998), but Khandkerstresses that even the ratio for maleborrowers compares favorably to alter-native poverty alleviation programs inBangladesh, like the World Food Pro-gramme’s Food-for-Work scheme (cost–benefit ratio = 1.71) and CARE’s simi-lar program (cost–benefit ratio = 2.62).The microfinance programs of theBangladesh Rural Advancement Com-mittee (BRAC) compare less favorably,however. Khandker (1998) reports ra-

tios of 3.53 and 2.59 for borrowing fromBRAC by women and men, respectively.

These calculations provide an impor-tant first-cut at taking costs and bene-fits seriously. They suggest that invest-ing in microfinance is not a universalwinner, but some programs beat alter-natives. Like all quick calculations,though, they rest on a series of simplifi-cations. Most immediately, only mea-surable benefits can be considered, thusexcluding much-discussed social im-pacts like “gender empowerment.”Other limits hinge on how the measur-able impacts are quantified. For exam-ple, the 0.91 ratio for lending to womenby Grameen draws on an estimated 18cent increase in household consumptionfor every additional dollar borrowed bywomen from Grameen (Pitt and Khand-ker 1998). The estimate is a marginalimpact of an additional dollar lent, butthe average impact is more appropriatehere since the entire program is beingevaluated, not just the expansion ofscale.22 If average benefits were usedinstead and if marginal returns diminishwith amounts borrowed, the cost–bene-fit ratio will be overstated. Supportingthe Grameen Bank will then yield agreater impact than $1 benefit for each$0.91 spent. On the other hand, if thereare large fixed costs in production tech-nologies, marginal returns may well behigher than average returns, weakeningsupport for Grameen. There is evidenceto suggest that this may be the case: asdiscussed further in Section 6.3, aver-age impacts estimated with the samedata are close to zero (Morduch 1998b).

Putting aside the average-marginal

22 The econometric structure required for iden-tification in fact rests on the assumption that mar-ginal and average impacts are equated (see Sec-tion 6.3 below), although Pitt and Khandkerinterpret the impacts as marginal. Average impactsestimated with more limited econometric struc-ture turn out to look very different (Morduch1998).


distinction, simple cost–benefit ratiosfail to capture dynamics. Imagine thatborrowing allows a client to purchase asewing machine. Owning the machine(and being able to set up a small-scaletailoring business) creates benefits intothe future, and using impacts on cur-rent household consumption does notcapture the full value of borrowingsince cost is a stock variable, whilebenefit is a flow. In principle, costsshould be compared to the presentvalue of the flow of future impacts, notthe current impact, and doing so willlower cost–benefit ratios.

Perhaps the most difficult problem—and the one most relevant from the van-tage of the current debate into microfi-nance—is that simple cost–benefitcalculations fail to provide insight aboutthe relevant counterfactual scenario.Cost–benefit ratios might be improved(or worsened) by reducing subsidiesslightly, and the simple cost–benefit ra-tios provide no sense of the optimalityof such a move. Thus, while it might bethat a dollar used to subsidize an exist-ing microfinance program helps poorhouseholds more than other uses, itmight also be that the microfinanceprogram would ultimately help morepoor people if it was not subsidized (orif it was subsidized at a much lowerlevel). This kind of argument has beenput forward often by observers skepticalof subsidies (e.g., CGAP 1996).

Again consider the hypothetical com-parison above of a sustainable programwith 63,000 clients versus a subsidizedprogram with just 12,600. If societyjudged that one dollar in the hands of apoor borrower is worth just ten timesthe value of a dollar to a less poor bor-rower (rather than 5 times the value asassumed above), the larger (nonsubsi-dized) program will now do more to im-prove social welfare than the one withsubsidies. Resolving the issues requires

making explicit social valuations andevaluating the sensitivity of impacts andcredit demand to the rate of subsidy.

The assertion that borrowers desireaccess to credit, not subsidized credit,implies that this hypothetical examplemisses another aspect that militatesagainst subsidies. The example assumesthat the subsidized program reachesmuch poorer clients, but if it is truethat credit demand by poor borrowers isnot very sensitive to the interest rate,the profile of borrowers should be simi-lar at subsidized and non-subsidizedprograms. Pushing for financial sustain-ability should not limit the depth ofoutreach by much, and the case forsubsidization weakens considerably.

Anecdotal evidence for this claim,however, tends to rely on either partialanalytics (e.g., application of the princi-ple of declining marginal returns tocapital when all else is not held con-stant) or incomplete views of demandconditions (e.g., seeing demand at highinterest rates but overlooking the poolof potential borrowers that are discour-aged by high costs). Some argue that in-terest costs are a small fraction of over-all production costs, so households canabsorb high interest rates. But, all thesame, net profits could remain sensitiveto interest rates.

Practitioners in Bangladesh tend tobelieve that the elasticity of credit de-mand with respect to the interest rate ishigh, and accordingly they keep interestrates relatively low (below 25 percentreal). Practitioners in Latin Americatend to believe that the elasticity is low,and they set interest rates as high asneeded (approaching 60 percent real).Both could be correct in their contexts,but serious empirical work is lacking.

The issue relates to a broader con-cern with impacts on non-borrowers.Specifically, why is it assumed that achoice must be made between program


types? Why can’t different types of pro-grams coexist? More generally, how willthe existence of a subsidized programaffect the profitability of both formaland informal institutions operatingnearby?

Theorists have made progress here,although solid empirical evidence re-mains scant. Karla Hoff and Stiglitz(1998) and Pinaki Bose (1998), for ex-ample, illustrate cases in which the en-try of a subsidized program worsens theterms and availability of loans offeredby moneylenders in the informal sector.The negative impacts occur because thesubsidized programs reduce optimalscale and siphon off the best borrowers,leaving the non-subsidized lenders witha riskier pool of clients and higher en-forcement costs than before. Sanjay Jain(1999) similarly considers the case inwhich clients might borrow simultane-ously from both formal and informalsources but where the scale advantagesof the formal sector outweigh the infor-mational advantages of local money-lenders. The three papers give examplesof cases in which the clients of subsi-dized programs do well but at theexpense of borrowers (and lenders)elsewhere. However, the papers alsodescribe the possibility of favorablecounter-examples in which everyonebenefits. In this line, Maria Floro, andRay (1997) and Gabriel Fuentes (1996)provide cases in which increasing for-mal sector credit may eventually per-colate down to the informal sector,increasing credit availability there aswell. Unfortunately, for now policy-makers have little to go on beyonda handful of small-scale case studiesand these theoretical examples andcounterexamples.

5.2 Empirical Research Agenda

The issues above can be put togetherformally to show the kinds of informa-

tion that are needed to put numberson the ideas under debate. The start-ing point is a social welfare functionW = W(w1,w2, …, wN), which is assumedto be additively separable and indexedover the entire population i = 1,2, ..., N;social weights are given by αi, and wi isa measure of the lifetime welfare ofhousehold i:23

W = ∑αi

i = 1

N

wi.

The total amount borrowed from allsources is Li, and the borrower’s aver-age return per unit is δi, where returnsare both pecuniary and non-pecuniary.Borrowers pay an average interest rateri, depending on the sources of loans.Those who borrow only from the subsi-dized source pay an interest rate ri = r.The net return from borrowing is thusyi = Li(δi – ri), and I assume that house-hold welfare is derived from base in-come Y plus income from borrowing:wi = w(Y + yi). The change in socialwelfare for a small decrease in subsidi-zation (i.e., a small increase in r) isthus:

dWdr

= ∑αi

i = 1

N

dwi

dyi dri

dr

dLi

dri (δi − ri) +

Li

dδi

dri − 1

.

23 To see the key issues most easily, I ignore theheterogeneity in capital and non-credit serviceslike savings. In thinking about the place of subsi-dized microfinance institutions more generally, wewould also want to consider impacts on the funda-mental economic and social structures in rural vil-lages: the role microfinance can play in empower-ing women, in encouraging better health practices,in promoting education, in reducing vulnerability,and in encouraging community cohesion. It is herethat many microfinance programs may make thegreatest impacts, but it is also the set of impactsthat are hardest to measure. The focus is on im-pacts on poverty rather than purely efficiency—al-though there may be pure efficiency-based justifi-cations for intervention (Besley 1994).


The equation illustrates points of con-tention and priorities for empirical re-search. The first issue, moving from theleft, is the need to make explicit socialjudgements about the distribution of so-cial weights αi, and this will hinge onknowledge of the baseline welfare levelsof all households—a critical determinantof how income affects welfare, dwi/dyi. Astarting point is documentation of thebaseline income levels and demographiccharacteristics of both participants and non-participants, a task possibly made easierby linking surveys of participants withexisting randomized household surveys.

The term dri/dr reflects externalitiesassociated with the impact of subsidizedinterest rates on interest rates in othersectors, as well as the degree to whichclients of subsidized programs get abreak on average capital costs. Bothsupply and demand factors are reflectedin the sensitivity of equilibrium creditto interest rates, dLi/dri. The sort ofsupply-side spillovers that drive dri/drare at play here, as well as the sensitiv-ity of credit demand to the interestrate—will reducing subsidies makecredit too costly for borrowers? Thesupply-side issues could be evaluatedwith household surveys that have infor-mation on the availability and terms ofcredit; for example, it would be naturalto gauge spillover effects by matchingthose surveys to information on the tim-ing and scope of new microfinance pro-grams. Those same household surveyscould also be used to measure the sensi-tivity of credit demand to interest rates.Selection problems are notorious inthese kinds of studies, but instrumentalvariables like inherited assets have beenshown to have potential.

Finally, the term dδi/dri reflects theinteraction of average returns, produc-tion technologies, risk, and capitalcosts. Will increased interest rates pushborrowers toward riskier but more prof-

itable technologies? Will it reduce equi-librium credit demand and thus limitscale economies (and thus reduce aver-age returns)? Do better-off householdshave projects with higher returns thanpoorer households? Household surveyswith disaggregated production data canbe used to address these questionsthrough estimates of profit functions,again with an eye to the responsivenessto capital availability and capital costs.

Debates about microfinance subsidi-zation have often been stymied by dif-ferences of opinion about the levels ofthese parameters. Those who opposesubsidization tend to assume a relativelyflat distribution of social weights αi, lowsensitivity of credit demand to interestrates dLi/dri, positive impacts of inter-est rates on returns dδi/dri, very low re-turns to investments by poorer house-holds, and negative externalities ofsubsidized credit programs on otherlenders: dri/dr < 0. Those who supportsubsidization, on the other hand, tendto put much greater social weight onconsumption by the poor, assume highlysensitive credit demand to interestrates, low impacts or perhaps negativeimpacts of interest rates on returns,moderately high (but not extremelyhigh) returns to investments by poorhouseholds, and small or beneficialspillovers onto other lenders.

Despite the lack of evidence, experi-enced practitioners on both sides of thedebate hold their views strongly. Dis-cussion about the role of microfinancein development thus remains stale-mated early in the game, with assertionschecked by counter-assertions and noimmediate route to resolution. Fortu-nately, apart from the social judge-ments, these are all issues that can beresolved by fairly straightforward em-pirical studies. It is the peculiar circum-stance of the microfinance policy con-text—with donors eager to spend on


new programs and ample funds avail-able for subsidization—that has pre-vented further progress in getting tothe roots of these most basic issues.

6. Social and Economic Impacts

In principle, self-employment activi-ties started due to microfinance partici-pation can affect households in manyways (if, indeed, that is what house-holds actually do with loans). First,there should be an income effect, push-ing up consumption levels and, holdingall else the same, increasing the de-mand for children, children’s education,and leisure. But there will also be ef-fects on the value of time, yielding a va-riety of counterbalancing effects. Withincreased female employment, havingmore children becomes costlier, push-ing fertility rates downward. The needto have children help at home (to com-pensate for extra work taken on by par-ents) could decrease schooling levels,and, most obviously, leisure may fall ifopportunity costs are sufficiently in-creased. On top of these forces, manyprograms directly advocate family plan-ning and stress the importance ofschooling, so participation may alsobring shifts in attitudes, as well as shiftsin the relative bargaining positions ofhusbands and wives. Thus, while con-sumption and income levels ought to in-crease, it is not clear a priori what willhappen to fertility, children’s education,and leisure.

Moreover, the extent of net impactsdepends on the opportunities open tohouseholds in the absence of microfi-nance. Households that do not partici-pate in microfinance programs mayhave access to a wide range of informalfinancial mechanisms and other servicesprovided by NGOs and governmentsocial programs.

Not long ago, similar claims to those

made for microfinance were made forpublicly-funded job training programsand for Head Start in the U.S.—thatthey could ultimately pay for them-selves while generating fundamentalchanges in the lives of poor households.And they too received enthusiastic bi-partisan support from the outset. HeadStart, which aims to help 3–5 year-oldchildren with disadvantaged back-grounds get an extra leg up on earlyeducation, has proved to be a success ingeneral. For African-American chil-dren, however, it has been largely inef-fective, with average impacts rapidly di-minishing over time (Janet Currie andDuncan Thomas 1995). Publicly-fundedjob training has also had real successes,especially when programs center onteaching basic job skills. However, moreintensive programs have been expensiveand seldom justify their costs (RobertLalonde 1995). These mixed reports donot overshadow the argument that bothprograms have played important rolesfor many beneficiaries, but they suggestthat marginal dollars would have beenmore effectively used by alternativeprograms.

As noted above, microfinance pro-grams have yet to receive that kind ofscrutiny. Visits to areas served by mi-crofinance programs show what cannotbe seen in books of accounts—earningsfrom microfinance participation arefunding new houses, further educationfor children, new savings accounts, andnew businesses. But are these changesmore remarkable than those occurringelsewhere?

Simple measures can be deceiving.For example, a recent survey shows that57 percent of the school-age sons ofGrameen Bank borrowers are enrolledin school—versus 30 percent of the sonsof eligible households that do not bor-row. The difference is sharp, but doesGrameen attract households with greater


propensities for education, or is this dif-ference a result of the program? A dif-ferent view of the data is obtained bypooling information on all children invillages served by the Grameen Bank.Taken together, the average enrollmentrate for sons from a random sample ofall eligible households is 46 percent(combining those that borrow and thosethat do not). But the fraction is 48percent in a random sample of compa-rable households in control villageswithout program access. Assuming thatcontrol and treatment groups arecomparable, the Grameen educationadvantage disappears.24

Unfortunately, there are few reliableestimates of the net impacts of pro-grams. The failures which dot the mi-crofinance landscape are also frequentlyoverlooked, overshadowed by the impres-sive claims that arise from successfulprograms.

Why the lack of sound statistical evalu-ations? First, many donors and practitio-ners argue that as long as programs covercosts and appear to serve poor house-holds, serious evaluations are a waste oftime and money—a diversion from run-ning the programs themselves. But asthe simple education example abovedemonstrates, quick looks can mislead.Moreover, almost no programs are cov-ering costs. Second, sound evaluationspose difficult statistical issues.

Many evaluations, not surprisingly,stress the banking side. As above, theevaluations generally measure perfor-mance by on-time repayment rates andthe ability to generate revenues which

cover costs.25 More recently, evalu-ations have included simple measures ofoutreach—the number of borrowers be-low official poverty lines, the gender ofborrowers, and the average size of loansand savings accounts (e.g., MicroBank-ing Bulletin 1998; Robert Peck Christenet al. 1995).

But nothing is ever truly simple.When money is fungible within thehousehold and fungible between differ-ent activities and assets, the net impacton women and saving cannot be gaugedwithout taking into account realloca-tions between men and women and be-tween multiple forms of saving and in-vestment. For example, although 95percent of Grameen borrowers are fe-male, Goetz and Sen Gupta (1995) findthat in just 37 percent of cases do fe-male borrowers from Grameen Bank re-tain significant control over loan use(Hashemi, Schuler, and Riley 1996,however, find control is retained in 63percent of cases). Addressing these is-sues—as well as selection bias—requiresevaluations with carefully constructedcontrol and treatment groups.

6.1 Selection

Microfinance programs can boast thattheir mechanisms ensure that borrowersare more entrepreneurial, better con-nected, more dedicated, and less riskythan non-participants. This success inscreening applicants makes addressingselection biases due to non-random par-ticipation that much more important.Would borrowers have done just as wellwithout the programs?

The biases can be large. In evaluatingthe Grameen Bank, Signe-Mary McKer-nan (1996, p. 31) finds that not control-ling for selection bias can lead to over-estimation of the effect of participation

24 Comparisons are from Morduch (1998) andare restricted to households with less than half anacre. The Grameen advantage remains elusiveeven after controlling for child-specific, house-hold-specific, and village-specific variables. Pittand Khandker (1998a), however, find some posi-tive effects on male schooling using a structuraleconometric model to estimate parameters withthe same data set.

25 See, e.g., Richard Patten and DonaldSnodgrass (1987), Yaron (1992), Bruce Bolnick(1988), and Mahabub Hossain (1988).


on profits by as much as 100 percent. Thisis a testament to the great success that theGrameen Bank has had in identifyingand targeting good clients. It also meansthat every dollar lent by Grameen maybe responsible for as little as half of theprofits reported by their clients.

Selection bias may also go in the op-posite direction. Many microfinance in-stitutions target women and poor house-holds. Pitt and Khandker (1998a), forexample, find that poorer householdsare more likely to be Grameen borrow-ers than their neighbors, conditional onvillage of residence and other observ-able characteristics. In cross-sectionalstudies, this outreach can lead to adownward bias on the estimated effect ofcredit on earnings. At the extreme, theeffective targeting of poor householdscan yield the impression that participa-tion in the program makes clients poorer.Addressing the selection bias revealshow participation increases earnings.

The second important source of biasis non-random program placement.Many programs are set up specificallyto serve the under-served. Thus, theyare located where there has long beenweak financial service. This may lead toapparent negative impacts relative tocontrol areas. Alternatively, the pro-grams may set up where there is goodcomplementary infrastructure (von Pis-chke 1991, pp. 305–306), biasing esti-mates upward. The size and signs of thebiases are likely to change as programsexpand over time into new areas.

A natural response has been to ex-ploit variation over time by collectinginformation on borrowers before and af-ter program participation.26 The ap-

proach has been popular because datacollection is simple, with recall dataoften used in the absence of a baselinesurvey, and because it promises to con-trol for both non-random participationand non-random program placement.Even so, it is subject to potential biasesdue to time-varying unobservables(James Heckman and Jeffrey Smith1995).27

The best known examples are thestudies collected in the Hulme andMosley (1996) volumes. The studies of-fer before-after comparisons, as well ascomparisons between participants andcontrol groups (where the controlgroups are often households that havebeen selected for program participationbut that have yet to begin borrowing).

Two results are striking. Comparisonof the second and final columns of Ta-ble 4 shows that programs that haveachieved higher levels of financial sus-tainability make larger net impacts onchanges in their borrowers’ incomes. (Itis not incidental that those programstend to cater to wealthier households.)Table 4 orders the programs in thestudy by their degree of subsidy depen-dence, ranging from –9 percent (fullprofitability) to 1884 percent (dire fi-nancial straits). The ranking is nearlyidentical to that based on the ratio ofparticipant-control comparisons of in-come changes, ranging from 544 per-cent to 117 percent (a negligible netimpact). Their second result is that,even within given programs, wealthierhouseholds benefit more than poorerhouseholds.

These results combine to suggest thatmicrofinance programs targeted to poor

26 Microfinance evaluations based on before-after comparisons include Eric Nelson and Bol-nick (1986), Barbara MkNelly and ChatreeWatetip (1993), Craig Churchill (1995), RichardVengroff and Lucy Greevey (1994), and J. R.Macinko et al. (1997)

27 The reliability of methods based on differ-ences is reduced as the time periods get closertogether, reducing temporal variation. Differenc-ing noisy data can also exacerbate measurementerror; in the “classical” case this leads to attenu-ation bias. Noisy recall may thus bias downwardcoefficients which show program impacts.


households may offer only limited bene-fits. The results have been used to but-tress arguments that pursuing full finan-cial sustainability is the surest way todeliver the most bang for the buck—and that poorer households should beserved by other interventions thancredit.

But observers have too quicklypointed to the apparent dichotomy. Theunresolved empirical issue is whetherthere is often an important group in themiddle—neither the destitute nor pettyentrepreneurs able to pay high interestrates. Is the typical middle-rung bor-rower at the Grameen Bank the normor the exception?

Given the sharpness of the results,the Hulme-Mosley studies deserve tobe read carefully. Unfortunately, doingso yields as many questions as answers.Corners were cut in the rush to get thevolumes out, and substantial inconsis-tencies slipped by. Key results vary by

as much as 40 percent even where the(ostensibly) identical series is presentedin more than one table (e.g., increasesin family income in their tables 4.1 and4.2 or a similar series in tables 4.1, 4.3,and 8.1). Even if the calculations wereconsistent, sample sizes are small forsome of the most important studies.The distribution of impacts in Mosley’s(1996) BancoSol study, for example,rests on evidence on just 24 borrowers.In addition, the quality of controlgroups is inconsistent. For example, theIndonesian studies draw on a controlgroup with fewer women and less accessto formal financial services than theoverall borrower group (p. 55). Table 4shows that the average control group in-come for BRI is 40 percent lower thanfor the borrower group—and the Banco-Sol control group has income one thirdthe level of the borrower group. Even ifthe income levels started closer to-gether, one is left to wonder why some

TABLE 4IMPACT AND FINANCIAL PERFORMANCE INDICATORS FOR SELECTED MICROFINANCE PROGRAMS, 1992

Program

Number ofborrowers

(1992)

1988–92 Subsidydependence

indexPercent female

clients

Average loan size for casestudies ($)

BRI unit desa, Indonesia 2,400,000 –9 24 600BKK, Indonesia 499,000 32 55 38Two RRBs, India 25,000 106 9 99BancoSol, Bolivia 45,000 135 74 322TRDEP, Bangladesh 25,000 199 38 —KREP Juhudi, Kenya 2,400 217 51 72Nine PTCCSs, Sri Lanka 700,000 226 50 50SACA, Malawi 400,000 398 28 70BRAC, Bangladesh 650,000 408 75 —Mudzi Foundation, Malawi 223 1884 82 57KIE-ISP, Kenya 1,700 — 23 —

Source: Data are from Hulme and Mosely (1996), volume 1, tables 3.3, 3.7, 4.1, and 5.1. The final four columnspertain to case studies. Abbreviations—BancoSol: Banco Solidario. BRI: Bank Rakyat Indonesia. BRAC:Bangladesh Rural Advancement Committee. BKK: Badan Kredit Kecamatan. PTCCS: Primary Thrift and CreditCooperative Society. RRB: Regional rural banks. KREP: Kenya Rural Enterprise Programme. KIE-ISP: KenyaIndustrial Estates-Informal Sector Programme. SACA: Smallholder Agricultural Credit Administration. The subsidydependence index gives the percentage increase in the interest rate required if the program is to exist withoutsubsidies; negative numbers indicate profitability without subsidies (Yaron 1992).


households were already borrowing butthe control groups had yet to receiveloans. Selection bias associated withfixed characteristics will be eliminatedthrough the differencing procedure, butselection bias associated with growthprospects remains. The Hulme-Mosleyhypotheses are provocative, but policydecisions should wait for more carefulstudies.

6.2 The Search for Instruments

More careful work would be helpedby the availability of instrumental vari-ables. The search for convincing instru-mental variables for credit has yieldedlittle, however. The problem is com-pounded since variables that may be un-related in more developed economies—such as the structure of production andconsumption—may be integrally linkeddue to non-separabilities driven by im-perfect and incomplete markets (Mor-duch 1995). It then becomes less likely

that a production-side variable thatexplains credit use does not also helpexplain expenditure-related outcomesindependently.

The interest rate is a potential identi-fying variable, but since achieving uni-formity across branches is a commongoal of microfinance programs, interestrates are unlikely to vary within a givenprogram area—and estimation is impos-sible without some variation. Even if in-terest rates vary, it is likely that thevariation will at least partly reflect un-observed attributes of the borrower, un-dermining their use as instruments (Pittand Khandker 1998a).

Other likely identifying variables arethose which affect the supply of creditbut not demand. Zeller et al. (1996, p.60), suggest community-level variables,proxies for “social capital,” lender char-acteristics, and program eligibility re-quirements. The first two suggestionswork as long as the community-level

TABLE 4 (Cont.)

Program

1992 familyincome—

borrowers ($)

1992 familyincome—

controls ($)

Average annual% change in

borrower income

As ratio ofcontrol group

percent change

BRI unit desa, Indonesia 1722 1074 20.7 544BKK, Indonesia 702 570 5.2 216Two RRBs, India 505 496 46.0 191BancoSol, Bolivia 3028 1121 28.1 193TRDEP, Bangladesh 1138 816 38.7 126KREP Juhudi, Kenya 1756 1307 1.5 133Nine PTCCSs, Sri Lanka 1301 981 15.6 157SACA, Malawi 830 276 2.8 175BRAC, Bangladesh 517 552 19.8 143Mudzi Foundation, Malawi 665 669 1.4 117KIE-ISP, Kenya 2807 1759 0.5 125


variables and social capital do not di-rectly affect profitability, investment,etc. This is a high hurdle for socialcapital to pass.

Lender characteristics have appeal.Like community-level variables, though,they will be wiped out when using vil-lage-level fixed-effects methods if thereis no variation in program access withina village. When there is within-villagevariation in program access, however,rules determining eligibility can be thebasis of an identification strategy, a tacktaken by Pitt and Khandker (1998a,b).The approach is considered in greaterdetail below.

6.3 Exploiting a Quasi-Experiment: An Example

The Hulme-Mosley (1996) studiesand similar small-scale evaluations buildon simple methods with small samples.In contrast, a series of recent papers onprograms in Bangladesh exploit a sam-ple of 1800 households and carefullyconsidered econometric approaches(Pitt and Khandker 1998a,b; Pitt et al.1999; McKernan 1996; Morduch 1998).The programs include the GrameenBank, Bangladesh Rural AdvancementCommittee (BRAC), and the Bangla-desh Rural Development Board (BRDB).All use a Grameen-style lending modeland nominally restrict access to house-holds holding under half an acre ofland.

The Bangladesh survey includessamples from villages with no access toprograms, and the approach exploitsprogram rules that bar wealthier house-holds from participating. These two fea-tures form the core of the quasi-experi-ment, offering two types of controlgroups. The main constraint is restric-tion to cross-sectional information onoutcomes.

The range of questions asked in thesestudies is ambitious, and answering the

questions requires technical sophistica-tion and a series of identifying assump-tions tied to the structure of the econo-metric models. The basic insight issimple, however. The fact that programrules restrict participation to house-holds owning a half acre of land or lesssuggests a source of variation to exploitfor identification. A natural first cut atevaluation would be a straightforwardcomparison of outcomes of householdsclustered just below the cut-off line tothose just above, a standard application ofregression discontinuity design (DonaldCampbell 1969).

Pitt, Khandker, and their associatesskip that step, however, and take twosteps ahead. First, rather than usingjust the simple eligibility rule for identi-fication, their instruments are effec-tively a series of household charac-teristics interacted with an indicatorvariable for whether each householdboth lives in a program village and isdeemed eligible to borrow. Identifica-tion thus comes from differences in theway that age, education, etc. affect out-comes for the sample as a whole versustheir effects for the eligible subsamplewith program access. Any differencesare assigned to program participation.Identification thus rests with the as-sumption that there are not importantnon-linearities in the ways that age,education, and the other variablesinfluence outcomes of interest.28

Pitt and Khandker (1998a) take oneadditional step, exploring the impactby gender and by each of the three

28 Pitt and Khandker (1998a) explain outcomesusing a linear functional form for the right handside variables, with the exception of land holdingsand program credit which are in logs. The lefthand side consumption and labor supply variablesare in logs. Pitt and Khandker demonstrate thattheir results are robust to allowing flexibility in thespecification for the land holdings variable but donot show results with flexible treatments of othervariables.


programs. Concern with gender is moti-vated by the observation that womentend to be more reliable borrowers thanmen (section 3.3 above) and thatwomen may allocate resources differ-ently from their spouses (GeoffreyWood and Iffath Sharif 1997; Goetz andSen Gupta 1995). The question is im-portant both for improving program im-pact and for helping better understandhousehold behavior more generally.

A more complicated selection prob-lem emerges now since participation inthe program entails not just a choiceabout whether a member of the house-hold should participate but also specifi-cally who in the household should par-ticipate. Pitt and Khandker exploit thefact that credit groups are never mixedby gender (by regulation), and not allvillages have groups of both genders.Thus, men in villages with no malegroups will not be eligible to borrow;likewise for women. In the 87 villagessurveyed, 10 have no female groups and22 have no male groups (and 40 haveboth, leaving 15 villages with nogroups). Identification now comes fromcomparing how the roles of age, educa-tion, etc. for men with access to malegroups compare to their roles for menwithout access; likewise for the charac-teristics of women with and withoutaccess.

Of course, the fact that a man is in avillage with no male groups may saysomething about the unobserved quali-ties of the men and the strength of theirpeer networks in that village. If, for ex-ample, the men are poor credit risks,the evaluation will overstate the pureimpact on men who do participate.Similarly, if having a strong peer groupincreases impacts directly, the estimateswill reflect the role of peer groups inaddition to the role of the program.

Pitt and Khandker partly address theproblem by estimating with village-level

fixed effects, thus sweeping out anyunobserved village-level heterogeneity(estimating with fixed effects withoutsoaking up most of the program impactis made possible by the fact that a frac-tion of residents in each village is ineli-gible to borrow from the programs—and by the assumption that spilloversare small). However, the fixed effectsdo not control for features of peernetworks—or other relevant charac-teristics—that are specific just to targethouseholds in program villages. The vil-lage-level fixed effects will only controlfor those unobservables that affect allhouseholds in a village identically (andlinearly). Non-random program place-ment thus remains an issue if, as isplausible, the functionally landless arenoticeably different from their wealth-ier neighbors (noticeable to bank staffbut not the econometrician), and if theprograms take this into account whendeciding where to locate.

The final structural detail stems fromthe use of a first stage Tobit to explaincredit demand. The Tobit requires thatsecond stage impacts must be assumedto be homogeneous across borrowers, acommon assumption in the evaluationliterature, but one that researchers arekeen to relax (Joshua Angrist, GuidoImbens, and Donald Rubin 1996). Italso implies, for example, that marginaland average impacts are equated. Theassumption poses difficulties if the dis-tribution of returns is anything like thatfor BancoSol borrowers, where staffpredict that in any given cohort roughly25 percent show spectacular gains toborrowing, 60–65 percent stay about thesame, and 10–15 percent go bankrupt(Mosley 1996).

Entertaining these assumptions, how-ever, offers the chance to estimate im-portant quantities like gender-specificmarginal impacts that would otherwisebe impossible. The structure cleverly


exploits the eligibility rules that barlending to households owning over halfan acre of land. Coupled with the often-cited observation that land markets arevery thin in South Asia, there is the ba-sis for an instrument that is plausiblyexogenous and that is associated witha sharp discontinuity in treatments,providing a substantial advantage overprevious studies.

These identifying assumptions do nothold up in the data, however. Mostcritically, leakages are evident whenlooking more closely at what the pro-grams do, rather than just at what theysay. Hassan Zaman (1997), for example,finds that 28 percent of borrowers fromBRAC are above the half acre cut-off,and I find a similar number for Gra-meen (30 percent) using the data col-lected by Pitt and Khandker (Morduch1998). The discrepancies are not minor:average land holdings are 1.5 acres forthose households that borrow fromGrameen but are over the half acre lineand some borrowers hold over fiveacres. Consequently, nonparametric re-gression yields no obvious discontinuityin the probability of borrowing forhouseholds across the relevant range oflandholdings. Contrary to the evidencefor India, the data also show consider-able activity in the land market, withnearly one eighth of borrowers makingpurchases during their tenure with theprograms.29

The results, putting aside the ques-tions about identification, are striking.Pitt and Khandker (1998a) estimate that

household consumption increases by 18taka for every 100 taka lent to a woman.The increase is just 11 taka for every100 taka lent to a man. Lending towomen has little effect on labor supply,but men take more leisure—explainingpart of the shortfall in consumption in-creases. Conversely, non-land assets in-crease substantially when borrowing isby women, but not by men. Results onschooling are mixed. Schooling of boysis increased whether men or womenborrow. When women borrow fromGrameen, schooling of girls also in-creases, but it does not do so whenwomen borrow from the other pro-grams. This may suggest that girls arecalled upon to help take care of workthat their mothers had done prior toborrowing.

Pitt and Khandker (1998a) interprettheir finding that loans to women havehigher marginal impacts than loans tomen as an indication of a lack of fungi-bility of capital and income within thehousehold. But since loans to males arelarger on average, in principle the pat-terns of impacts on consumption canalso be explained by the standard theoryof declining marginal returns to capital.30

Using a similar methodology, Pitt etal. (1999) find that Grameen participa-tion by women had no effect on contra-ceptive use and a slight positive effecton fertility. Participation by men, how-ever, reduced fertility and moderatelyincreased contraceptive use. The mixedfindings should perhaps not be surpris-ing, given the treatment-control set-up.Bangladesh underwent a broad fertilitydecline in the 1970s and 1980s, so con-trol villages were also in the process of

29 In choosing control groups, the survey strictlyfollows the half-acre cut-off rule. But the Gra-meen Bank’s eligibility requirement is in fact halfan acre of land or total net assets under the valueof one acre of single cropped, non-irrigated land(Hatch and Frederick 1998). Some householdswith over half an acre may still qualify under thesecond criterion, but mistargeting is so extensivethat considerable leakage remains even under theexpanded definition.

30 When calculated conditional on borrowingpositive amounts, males borrow slightly more onaverage from Grameen (15,797 taka versus 14,128taka). For BRAC, males cumulatively borrowed5,842 taka versus 4,711 taka for women, and forBRDB, males borrowed 6020 taka versus 4118taka for women (Morduch 1998).


reducing family size (Monica Das Guptaand D. Narayana 1996). Hashemi,Schuler, and Riley (1997) find positiveeffects of program participation on con-traceptive use in a sample of 1300women, but they do not control fornonrandom program placement.

Pitt and Khandker (1998b) extend theframework to consider impacts on sea-sonality, taking advantage of data on la-bor supply and consumption followingthe three main rice seasons. Microfi-nance borrowing is shown to improvethe ability to smooth consumptionacross seasons, and entry into the pro-grams is partly driven by insuranceconcerns.

McKernan (1996) builds on the Pittand Khandker research to investigatenon-credit impacts of the microfinanceprograms in Bangladesh. The questionis important since the programs putconsiderable energy into vocationaltraining and education about health andsocial issues. Beyond these direct “so-cial development programs,” participa-tion can also provide borrowers withdiscipline, a sense of empowerment,and shared information. Focusing juston credit misses these potentiallyimportant program aspects.

McKernan investigates these aspectsby estimating the determinants of self-employment profits while controllingfor capital, other inputs, and a dummyvariable for microfinance participation.The dummy variable indicates that par-ticipation in Grameen Bank is associ-ated with a 126 percent increase in self-employment profits beyond the directimpact of the capital. Thus, on averagehouseholds more than double their self-employment earnings (bearing in mind,though, that self-employment activitiesstart at a low base and are for mosthouseholds a minor share of total in-come). McKernan also finds that non-credit impacts alone raise profits by 50–

80 percent. Taking the two results to-gether, she argues that the provision ofcredit alone explains roughly half of theaverage increases in self-employmentprofits brought by Grameen.

Identification here is complicated bythe fact that both microfinance partici-pation and capital use are endogenous.The quasi-experiment is used to iden-tify program participation, and the in-struments for capital are the numbersof land-owning relatives of potentialborrowers. The latter instruments aremotivated by the suggestion that havingmore land-owning relatives is likely as-sociated with having greater access tointerhousehold transfers, a commoncredit substitute. However, the instru-ments prove invalid if, as is likely, prof-its are affected directly by the businessconnections, implicit insurance, andfamily responsibilities associated withthe size and characteristics of one’sextended family.31

As a result of questions raised aboutthe identifying assumptions, I take an-other look at the data, focusing insteadon simple comparisons across treatmentand control villages, controlling as wellfor household- and village-level charac-teristics and not making distinctions bygender (Morduch 1998). The resultsdiffer considerably from the Pitt–Khandker studies. After limiting sam-ples just to comparable households, I findno increase in consumption or educa-tion (and a slight increase in labor sup-ply) when using the data to measure theimpact of program access. For example,households with access to Grameenhave per capita consumption levels that

31 Under the maintained assumptions, Mada-jewicz (1997) finds that when disaggregating capi-tal types, the “non-credit impact” loses statisticalsignificance, suggesting that the impact could re-flect roles of specific types of capital use (in thiscase, greater use of working capital by programparticipants), rather than factors like education or“empowerment.”


are 7 percent below those of compara-ble control groups, a finding that is ro-bust to controlling for village-level un-observables (although the latter resultis not significantly different from zero).The weak findings are consistent withthe presence of a rich variety of alterna-tive institutions available to non-partici-pants: the programs may make impor-tant absolute differences in the lives ofparticipants, even if they have madenegligible relative differences.

But like Pitt and Khandker (1998b), Ifind some signs of consumption smooth-ing across seasons, a result that can betraced to increased smoothing of laboracross seasons. Taken together, the evi-dence is decidedly mixed. Pitt andKhandker have set out an important re-search agenda and have demonstratedthe sensitivity of results to methodologi-cal assumptions. But my results showthat the quasi-experiment turns out tobe much less clean than it appeared atfirst, and that using village-level fixedeffects is not a panacea for addressingbias due to non-random program place-ment. Substantively, the results suggestthat benefits from risk reduction maybe as important (or more important)than direct impacts on average levels ofconsumption. More generally, themixed results show that much morework is required to establish the casefor strong microfinance benefits in thiscontext.

7. Savings

One additional means for promotinghousehold welfare is the developmentof facilities for safe but liquid savingsdeposits. Early microfinance programswere not effective in mobilizing savingsand showed little interest in doing so.Partly, it was thought that poor house-holds were too poor to save. But recentmicrofinance experience shows that

even poor households are eager to saveif given appealing interest rates, a con-veniently located facility, and flexibleaccounts—with bankers in Indonesiaand South Asia finding that conveniencegenerally trumps interest rates.

A recent study of the expansion of ru-ral banking in Mexico shows this possi-bility clearly. Fernando Aportela (1998)measures the impact on savings rates ofthe expansion of Pahnal, a Mexican sav-ings institute targeted to low-incomeclients. Pahnal expanded rapidly in theend of 1993, setting up branches in postoffices, a model that follows the postalsavings programs of Japan and Ger-many. Pahnal also introduced simplersavings instruments with much lowerminimum balances and lower fees thanwere offered by earlier programs. Ex-ploiting the fact that Pahnal’s expansionwas not uniform across regions,Aportela uses a differences-in-differ-ences framework to estimate impacts,finding that expansion of program avail-ability pushed up savings rates by al-most five percentage points—and by al-most seven percentage points for someof the poorest households.

But how much is new savings andhow much is reallocations from otherassets (and from under mattresses)? Ifportfolio reallocations are substantial,net benefits to depositors may besmaller than it appears at first. Aportela(1998) finds little evidence for crowdingout of other savings instruments, how-ever, suggesting that much of theincrease is due to new saving.

From an institutional viewpoint, in-corporating savings mobilization in mi-crofinance programs makes sense for avariety of reasons (Robinson 1995).First, it can provide a relatively inex-pensive source of capital for re-lending.Second, today’s depositors may be to-morrow’s borrowers, so a savings pro-gram creates a natural client pool.


Third, building up savings may offer im-portant advantages to low-income house-holds directly: households can build upassets to use as collateral, they canbuild up a reserve to reduce consump-tion volatility over time, and they maybe able to self-finance investments ratherthan always turning to creditors. Onthe other hand, handling lots of smalldeposit accounts can be prohibitivelyexpensive.32

Grameen and BancoSol are just start-ing to mobilize savings more aggres-sively, but BRI has made it a major partof their program in Indonesia. A turn-ing point in Indonesia was introductionof the SIMPEDES saving program in1986. Before SIMPEDES, householdshad to save in accounts run by Bank In-donesia that limited withdrawals totwice a month but which offered rea-sonable interest rates—households re-ceived 15 percent on deposits and paid12 percent on loans! SIMPEDES offersunlimited withdrawals, and this hasturned out to be a boon to risk-aversedepositors. The bank has also success-fully implemented a lottery, such thatchances for prizes increase with theamount on deposit (Mexico’s Pahnal hasalso had success with a savings-basedlottery, a system that echoes Britain’slong-running lottery-based premiumbonds). These two features have madethe SIMPEDES program very popular,even if interest rates are zero for smalldeposits, 0.75 percent monthly for me-dium deposits, and 1.125 percentmonthly for larger deposits (over about$100) with inflation knocking out muchof the interest for poorer households(Patten and Rosengard 1991, p. 71).

By 1988, over four million poorhouseholds were saving through theprogram, and by December 1996, oversixteen million had deposits. Depositsizes are small, with average balances in1996 of $184, suggesting that the aver-age depositor is considerably less welloff than the average borrower (with anaverage loan balance over $1000). Thisrepresents over $3 billion in savings andgives BRI a relatively cheap source offunds for relending while providinghouseholds with means to build up as-sets and better smooth consumption. Asabove, the question is open, though, asto how much is new savings. (Of course,even if the increased savings rates weredue only to simple portfolio realloca-tions, there could still be substantial ef-ficiency benefits from promoting thescale of intermediation and enhancingflexibility for depositors.)

Like many programs, Grameen didnot focus on mobilizing voluntary sav-ings until recently. The bank now pro-vides opportunities for voluntary sav-ings, but total deposits remain small. Incontrast, SafeSave, an innovative newprogram in Bangladesh, has made vol-untary saving the core of its program.Staff solicit savings from members on adaily basis with the aim to help house-holds convert their ability to save inregular but small amounts into a usefullump of money, much as is achieved(less flexibly) through participation ininformal rotating savings and credit as-sociations (Rutherford 1998). In fact,the program was founded in part by Ra-beya Islam, a housewife in Dhaka whohad long experience running ROSCAs.By the end of 1998, SafeSave had over2000 clients, and it appears to havegood prospects for becoming financiallysustainable, although it remains smalland subsidized for now (SafeSave 1998).

Part of the reason that subsidized pro-grams have not been more aggressive in

32 In the U.S., however, banks find that servic-ing a $500 deposit balance can cost as much as $7per month. Costs, though, may be lowered sub-stantially through encouraging emerging technolo-gies like electronic “smart cards” that are used inautomatic teller machines (that might possibly betrucked from site to site).


mobilizing savings rests with interestrate spreads. Part of the trap that manyearly programs fell into involved bankscharging interest rates r on loans andpaying depositors a rate d, which wasless than r to avoid further losses. Sincer was kept artificially low in the name ofwelfare maximization, d was often kepteven lower, and incentives for saving werediminished. The spread (r – d) has thusbeen the focus of those interested insavings mobilization. Increasing lendingrates is clearly helpful here.

But this is not the appropriate spreadto maximize if capital is subsidized andthe objective is to enhance welfare in acost-effective manner. A more appro-priate spread to watch is m − (d + a),where m is the rate at which donors ob-tain funds and a reflects the per unitadministrative costs of managing andmobilizing savings deposits. Thus, mgives the donor’s opportunity cost ofraising funds and (d + a) gives the pro-gram’s opportunity costs. For example,in the mid-1990’s the Grameen Bankobtained funds from the BangladeshBank at just 5–6 percent while alterna-tive sources of funds would have cost12–15 percent. If Grameen could havemobilized savings at a cost below theBangladesh Bank’s opportunity cost offunds, the social cost of subsidizationcould have been reduced.

Savings mobilization at deposit ratesabove lending rates can reduce thecosts of programs, rather than add tothem—if donors reward microfinanceprograms for generating funds at costslower than they face. One way to do thisis to split the difference between do-nors and programs of (m − c) − (d + a)per dollar of savings mobilized and re-lent (where c is the concessional inter-est rate that subsidized microfinanceprograms pay for capital)—and to re-duce concessional lending by donors byone dollar for each dollar of lending

thus generated. Under earlier subsi-dized credit schemes, everyone lost outthrough savings mobilization. By imple-menting the proposed scheme, how-ever, clients, microfinance programs,and donors can share benefits fromsavings mobilization.

Promoting saving will not alwaysbenefit clients, however. Most impor-tant, rapid inflation can quickly erodethe holdings of poor households (whilebenefiting those holding debt). How-ever, even if individual households findit impossible to adequately protectthemselves, the bank can invest in ap-propriate foreign currencies and assetsto create a hedge. While I know of nomicrofinance institution that has yetdone so, there is no reason not to inprinciple.

There may also be practical con-straints. Only tightly-regulated institu-tions should be entrusted to hold sav-ings, but this would exclude mostmicrofinance programs (except, for ex-ample, for BRI, BancoSol, and Gra-meen, which are chartered banks).Large, traditional commercial banksmay also have cost advantages in han-dling deposits. One answer is that fully-chartered savings banks could operateindependently but alongside NGOs en-gaged in lending. The savings banksshould have fully independent accountsand funds, and the savings that are col-lected should in no way be tied to lend-ing operations. However, a contractuallink to exploit the rebate opportunityabove could still be used to reduce costsof subsidization on the lending side.

Both the rebate proposal and the sav-ings bank/microcredit partnership ideaneed further thought before implemen-tation. But both ideas appear sound inprinciple and suggest that there may becreative ways around roadblocks.

The evidence on savings raises an im-portant question for economists. Poor


households often appear to be con-strained in their ability to borrow (Mor-duch 1995). This is puzzling, though—as long as households are not tooimpatient, they should be able to savetheir way out of borrowing constraints(Angus Deaton 1992, section 6.2). Thenew institutions can provide a way to doso.

8. Conclusions

The microfinance movement hasmade inroads around the world. In theprocess, poor households are beinggiven hope and the possibility to im-prove their lives through their own la-bor. But the “win-win” rhetoric promis-ing poverty alleviation with profits hasmoved far ahead of the evidence, andeven the most fundamental claimsremain unsubstantiated.

Even if the current enthusiasms ebb,the movement has demonstrated theimportance of thinking creatively aboutmechanism design, and it is forcingeconomists to rethink much receivedwisdom about the nature of poverty,markets, and institutional innovation. Inthe end, this may prove to be the mostimportant legacy of the movement.

In particular, the movement hasshown that, despite high transactionscosts and no collateral, in some cases itis possible to lend profitably to low-in-come households. The experiences haveshown as well that many relatively poorhouseholds can save in quantity whengiven attractive saving vehicles; thissuggests that one way to address theborrowing constraints faced by poorhouseholds may be to address savingconstraints instead of addressing justthe credit side. But the experienceshave also confirmed how difficult it isto create new institutions, even thosethat are ultimately profitable. In Bo-livia, Bangladesh, and Indonesia it took

strong leadership and special legal ac-commodations. Elsewhere, it has takenpersistent prodding by donors and mi-crofinance advocates. Demonstrationeffects and subsidized experimentationhave also been integral.

The microfinance movement has alsolifted the profile of NGOs. While gov-ernment failures become increasinglyevident, NGOs have had the energy,dedication, and financial resources topursue required legislative changes andinstitutional experimentation. Increas-ingly, NGOs can be expected to takeover social tasks once the exclusive do-main of state ministries, and interna-tional organizations like the WorldBank are adapting accordingly.

This is all new, but some receivedwisdom holds. Most important, all elsethe same it remains far more costly tolend small amounts of money to manypeople than to lend large amounts to afew. As a result, the programs arehighly cost-sensitive, and most rely onsubsidies. Initiating a serious discussionabout next steps necessitates first facingup to the exaggerated claims for finan-cial performance that have characterizedsome leaders in the movement.

If the movement plans not to aban-don the promise of substantial povertyalleviation through finance, it mustmake hard choices. One avenue is totake another hard look at managementstructures and mechanism design in or-der to lower costs while maintainingoutreach. Doing so will be far from sim-ple, and it is hard to imagine substantialprogress without a second major waveof innovation. Donors can contribute byencouraging further experimentationand evaluation, rather than just replica-tion and adherence to a narrow setof “best practices” based on existingprograms.

The other path is to reopen theconversation on ways that ongoing


subsidies can benefit both clients andinstitutions. The movement has shownsome successes in coupling efficient op-erations with subsidized resources, andthese lessons can be expanded. Someobservers speculate that if subsidies arepulled and costs cannot be reduced, asmany as 95 percent of current programswill eventually have to close shop. Theremaining 5 percent will be drawn fromamong the larger programs, and theywill help fill gaps in financial markets.But, extrapolating from current experi-ence, the typical clients of these finan-cially sustainable programs will be lesspoor than those in the typical programfocused sharply on poverty alleviation.

No one argues seriously that finance-based programs will be the answer fortruly destitute households, but thepromise remains that microfinance maybe an important aid for households thatare not destitute but still remain consid-erably below poverty lines. The tensionis that the scale of lending to this groupis not likely to permit the scale econo-mies available to programs focused onhouseholds just above poverty lines.Subsidizing may yield greater socialbenefits than costs here, and Section 5outlines a framework for integratingcompeting arguments.

This prospect is exciting, especiallygiven the dearth of appealing alterna-tives, but the promise of microfinanceshould be kept in context. Even in thebest of circumstances, credit from mi-crofinance programs helps fund self-employment activities that most oftensupplement income for borrowersrather than drive fundamental shifts inemployment patterns. It rarely gener-ates new jobs for others, and successhas been especially limited in regionswith highly seasonal income patternsand low population densities. The bestevidence to date suggests that making areal dent in poverty rates will require

increasing overall levels of economicgrowth and employment generation.Microfinance may be able to help somehouseholds take advantage of those pro-cesses, but nothing so far suggests thatit will ever drive them.

Still, by forging ahead in the face ofskepticism, microfinance programs nowprovide promise for millions of house-holds. Even critics have been inspiredby this success. The time is right for as-sessing next steps with candor—andbetter evidence.

REFERENCES

ACCION International. 1997. Credit Lines 3:1(An Update of the ACCION U.S. Network),March.

Adams, Dale and J.D. von Pischke. 1992. “Mi-croenterprise Credit Programs: Déja Vu,”World Devel., 20:10, pp. 1463–70.

Adams, Dale; Douglas Graham, and J.D. von Pi-schke. 1984. Undermining Rural Developmentwith Cheap Credit. Boulder, CO: WestviewPress.

Angrist, Joshua; Guido Imbens, and Donald Ru-bin. 1996. “Identification of Causal Effects Us-ing Instrumental Variables,” J. Amer. Stat. As-soc., 91:34, pp. 444–56.

Aportela, Fernando. 1998. “Effects of FinancialAccess on Savings by Low-Income People,”MIT Econ. Dept., draft.

Armendariz de Aghion, Beatriz and Christian Gol-lier. 1997. “Peer Group Formation in an Ad-verse Selection Model,” Econ. Dept., U. Col-lege, London, draft.

Armendariz de Aghion, Beatriz and Jonathan Mor-duch. 1998. “Microfinance Beyond GroupLending,” Harvard and Princeton, draft.

Banerjee, Abhijit; Timothy Besley, and TimothyGuinnane. 1994. “Thy Neighbor’s Keeper: TheDesign of a Credit Cooperative with Theoryand a Test,” Quart. J. Econ., 109:2, pp. 491–515.

Becker, Gary. 1981. A Treatise on the Family.Cambridge: Harvard U. Press.

Bedi, R. 1992. Theory, History, and Practice ofCooperation. Meerut (U.P., India): R. LallBook Depot.

Bencivenga, Valerie and Bruce D. Smith. 1991.“Financial Intermediation and EndogenousGrowth,” Rev. Econ. Stud., 58:2, pp. 195–209.

Besley, Timothy. 1994. “How Do Market FailuresJustify Interventions in Rural Credit Markets?”World Bank Res. Observer, 9:1, pp. 22–47.

———. 1995. “Savings, Credit, and Insurance,” inHandbook of Development Economics, vol. 3A.T.N. Srinivasan and Jere Behrman, eds. Am-sterdam: Elsevier. pp. 2125–207.


Besley, Timothy and Stephen Coate. 1995. “GroupLending, Repayment Incentives, and SocialCollateral,” J. Devel. Econ., 46.

Besley, Timothy; Stephen Coate, and GlennLoury. 1993. “The Economics of Rotating Sav-ings and Credit Associations,” Amer. Econ.Rev., 83:4, pp. 792–810.

Bolnick, Bruce R.. 1988. “Evaluating Loan Collec-tion Performance: An Indonesian Example,”World Devel., 16, pp. 501–10.

Bolton, Patrick and David Sharfstein. 1990. “ATheory of Predation Based on Agency Problemsin Financial Contracting,” Amer. Econ. Rev., 80,pp. 93–106.

Bose, Pinaki. 1998. “Formal-Informal Sector In-teraction in Rural Credit Markets,” J. Devel.Econ., 56, pp. 256–80.

Brill, Betsy. 1999. “A Little Credit, A Huge Suc-cess,” San Francisco Examiner, May 2.

Brugger, Ernst and Sarath Rajapatirana, eds. 1995.New Perspectives on Financing Small Business inDeveloping Countries. San Francisco: ICS Press.

Burtless, Gary. 1995. “The Case for RandomizedField Trials in Economic and Policy Research,”J. Econ. Perspect., 9:2, pp. 63–84.

Campbell, Donald T. 1969. “Reforms as Experi-ments,” Amer. Psychologist, 24, pp. 409–29.

Chu, Michael. 1996. “The Numbers Begin to AddUp for Microcredit; Poor Entrepreneurs FindDoors Opening,” WorldPaper, April.

Churchill, Craig F. 1995. “The Impact of Crediton Informal Sector Enterprises in South Africa:A Study of Get Ahead Foundation’s StokvelLending Programme.” Clark U. MA Thesis,Worcester MA.

Christen, Robert Peck; Elizabeth Rhyne, RobertVogel, and C. McKean. 1995. “Maximizing theOutreach of Microenterprise Finance: TheEmerging Lessons of Successful Programs,”USAID Program and Operations AssessmentReport No. 10, Washington, DC

Conning, Jonathan. 1997. “Joint Liability, PeerMonitoring, and the Creation of Social Collat-eral,” Econ Dept., Williams College, draft.

Consultative Group to Assist the Poorest [CGAP].1996. “Microcredit Interest Rates,” OccasionalPaper No. 1, Washington, DC, August.

Currie, Janet and Duncan Thomas. 1995. “DoesHead Start Make a Difference?” Amer. Econ.Rev., 85:3, pp. 341–64.

Das Gupta, Monica and D. Narayana. 1996.“Bangladesh’s Fertility Decline from a RegionalPerspective,” Harvard Center for Populationand Development Studies WP 96.03.

Deaton, Angus. 1992. Understanding Consump-tion. Oxford: Clarendon Press.

Diagne, Aliou. 1998. “Default Incentives, PeerPressure, and Equilibrium Outcomes in Group-Based Lending Programs,” IFPRI FCNDWorking Paper.

Durham, Brad. 1996. “A Movement May Face aBattle for its Soul; Microcredit Plays for HighStakes with the World Bank,” WorldPaper,April.

Edgcomb, Elaine; Joyce Klein, and Peggy Clark.1996. The Practice of Microenterprise in theU.S.: Strategies, Costs, and Effectiveness. Wash-ington, DC: The Aspen Institute.

Edgcomb, Elaine and Laura Barton. 1998. “SocialIntermediation and Microfinance: A LiteratureReview,” Microenterprise Best Practices report,USAID/Development Alternatives, Inc., Aug.[www.mip.org].

The Economist. 1997. “Microlending: From San-dals to Suits,” Feb. 1, p. 75.

Floro, Maria and Debraj Ray. 1997. “VerticalLinks Between Formal and Informal FinancialInstitutions,” Rev. Devel. Econ. 1:1.

Foster, James; Joel Greer, and Erik Thorbecke.1984. “A Class of Decomposable Poverty Mea-sures,” Econometrica, 52:3, pp. 761–66.

Fuentes, Gabriel. 1996. “The Use of VillageAgents in Rural Credit Delivery,” J. Devel.Stud., 33:2, pp. 188–209.

Fugelsang, Andreas and Dale Chandler, with DayaAkuretiyagama. 1993. Participation as Pro-cess—Process as Growth: What We Can Learnfrom Grameen Bank, Bangladesh. Dhaka: Gra-meen Trust.

Ghatak, Maitreesh. 1999. “Group Lending, LocalInformation and Peer Selection,” J. Devel.Econ., forthcoming.

Ghatak, Maitreesh and Timothy Guinnane. 1999.“The Economics of Lending with Joint Liability:Theory and Practice,” J. Devel. Econ., forth-coming.

Ghosh, Parikshit and Debraj Ray. 1996. “Coopera-tion in Community Interaction without Infor-mation Flows,” Rev. Econ. Stud., 63, pp. 491–519.

———. 1997. “Information and Repeated Interac-tion: Application to Informal Credit Markets,”Texas A&M and Boston U., draft.

Goetz, Anne Marie and Rina Sen Gupta. 1996.“Who Takes the Credit? Gender, Power, andControl over Loan Use in Rural Credit Pro-grams in Bangladesh,” World Devel. 24:1, pp.45–63.

Gonzalez-Vega, Claudio; Mark Schreiner, RichardMeyer, Jorge Rodriguez-Meza, and SergioNavajas. 1997. “BancoSol: The Challenge ofGrowth for Microfinance,” in Microfinance forthe Poor? Hartmut Schneider, ed. Paris:OECD.

Grameen Bank (various years), Annual Report.Dhaka: Grameen Bank.

Guinnane, Timothy. 1994. “A Failed InstitutionalTransplant: Raiffeisen’s Credit Cooperatives inIreland, 1894–1914,” Explor. Econ. Hist. 31, pp.38–61.

———. 1997. “Cooperatives as Information Ma-chines: German Rural Credit Cooperatives,1883–1914,” Econ. Dept., Yale U., draft.

Gurgand, Marc; Glenn Pederson, and JacobYaron. 1994. “Outreach and Sustainability of SixRural Finance Institutions in Sub-Saharan Af-rica.” Washington, DC, World Bank Discuss.Paper 248.


Hashemi, Syed M. and Sidney Schuler. 1997a.“Sustainable Banking with the Poor: A CaseStudy of the Grameen Bank,” WP #10, JSI Re-search and Training Institute, Washington, DC

———. 1997b. “The Influence of Women’sChanging Roles and Status in Bangladesh’s Fer-tility Transition: Evidence from a Study ofCredit Programs and Contraceptive Use,”World Devel., 25:4, pp. 563–75.

Hatch, John and Laura Frederick. 1998. “PovertyAssessment by Microfinance Institutions: A Re-view of Current Practice,” Microenterprise BestPractices report, USAID/Development Alterna-tives, Inc., Aug. [www.mip.org].

Heckman, James J. and Jeffrey A. Smith. 1995.“Assessing the Case for Social Experiments,” J.Econ. Perspect. 9:2, pp. 85–110.

Hoff, Karla and Joseph Stiglitz. 1998. “Money-lenders and bankers: price-increasing subsidiesin a monopolistically competitive market,” J.Devel. Econ. 55, pp. 485–518.

Hollis, Aidan and Arthur Sweetman. 1997. “Micro-credit: What Can We Learn from the Past?”Econ. Depts., U. Calgary and U. Victoria, draft.

Hossain, Mahabub. 1988. “Credit for Alleviationof Rural Poverty: The Grameen Bank in Bangla-desh,” Int. Food Policy Research Institute Res.Report 65, Washington, DC.

Hulme, David and Paul Mosley, eds. 1996. Fi-nance Against Poverty. London: Routledge.

Huppi, Monica and Gershon Feder. 1990. “TheRole of Groups and Credit Cooperatives in Ru-ral Lending,” World Bank Research Obs. 5:2,pp. 187–204.

International Monetary Fund. 1996. Intl. Finan-cial Stat. Yearbook XLIX. Washington, DC:IMF.

Ito, Sanae. 1998. “The Grameen Bank and PeerMonitoring: A Sociological Perspective,” paperpresented at the Workshop on Recent Researchon Microfinance, U. Sussex, 13 February, 1998.

Jain, Pankaj. 1996. “Managing Credit for RuralPoor: Lessons from the Grameen Bank,” WorldDevel. 24:1, pp. 78–89.

Jain, Sanjay. 1999. “The Interaction of Formal andInformal Credit Markets in Developing Coun-tries: Symbiosis versus Crowding Out,” J. Devel.Econ., 59:2, pp. 419–44.

Johnston, Don. 1996. “Notes on Bank KreditDesa,” Bank Rakyat Indonesia/Harvard Insti-tute for Int. Development (Jakarta).

Khalily, M.A. and R. Meyer. 1993. “The PoliticalEconomy of Loan Recovery,” Saving and Devel.17:1, pp. 23–38.

Khandker, Shahidur. 1998. Fighting Poverty withMicrocredit: Experience in Bangladesh. NewYork: Oxford U. Press for the World Bank.

Khandker, Shahidur R., Baqui Khalily, and ZahedKahn. 1995. Grameen Bank: Performance andSustainability, World Bank Discuss. Paper 306,Washington DC.

Lalonde, Robert. 1995. “The Promise of PublicSector-Sponsored Training Programs,” J. Econ.Perspect. 9:2, pp. 149–68.

Light, Ivan and Michelle Pham. 1998. “Micro-credit Credit and Informal Credit in the UnitedStates: New Strategies for Economic Develop-ment,” J. Devel. Entrepreneurship, 3:1.

Macinko, J.; R. Rodriguez-Garcia, B. Schwethelm,and Stephen Smith. 1997. “Village Banking, In-come Generation and Maternal and ChildHealth: Design and Evaluation of a Social Sci-ence Experiment in Ecuador and Honduras,”George Washington U., draft.

Madajewicz, Margaret. 1997. “Capital for thePoor: The Role of Monitoring,” Harvard U.Econ. Dept, draft.

Matin, Imran. 1997. “Repayment Performance ofGrameen Bank Borrowers: the ‘Unzipped’State,” Savings and Devel. 4.

McGuire, Paul. 1998. “The Asian Financial Cri-sis—Some Implications for Microfinance,” Mi-crobanking Bulletin, Boulder, CO: EconomicsInstitute.

McKernan, Signe-Mary. 1996. “The Impact of Mi-cro-Credit Programs on Self-Employment Prof-its: Do Non-Credit Program Aspects Matter,”Brown U. (draft).

MicroBanking Bulletin. 1998. Boulder, CO: Eco-nomics Institute.

Microcredit Summit Report. 1997. Washington,DC: RESULTS Educational Fund.

MkNelly, Barbara and Chatree Watetip. 1993.“Impact Evaluation of Freedom from Hunger’sCredit with Education Program in Thailand,”Freedom From Hunger Foundation. Davis,CA.

Moffit, Robert. 1991. “Program Evaluation withNonexperimental Data,” Evaluation Rev. 15:3,June, pp. 291–314.

Morduch, Jonathan. 1995. “Income Smoothingand Consumption Smoothing,” J. Econ. Per-spect. 9:3, pp. 103–14.

———. 1998. “Does Microfinance Really Help thePoor? Evidence from Flagship Programs inBangladesh,” Hoover Institution, Stanford U.,July, draft.

———. 1999. “The Role of Subsidies in Microfi-nance: Evidence from the Grameen Bank,” J.Devel. Econ., Oct.

———. 2000. “The Microfinance Schism,” WorldDevel., forthcoming.

Mosley, Paul. 1996. “Metamorphosis from NGO toCommercial Bank: the Case of BancoSol in Bo-livia,” chapter 10 in Finance Against Poverty.David Hulme and Paul Mosley, eds. London:Routledge.

Navajas, Sergio; Mark Schreiner, Richard Meyer,Carlos Gonzalez-Vega, and Jorge Rodriguez-Meza. 1998. “Microcredit and the Poorest ofthe Poor: Theory and Evidence from Bolivia,”Dept. Agricultural, Environmental, and Devel-opment Econ., Ohio State U., draft.

Nelson Eric R. and Bolnick Bruce R. 1986. “Sur-vey Methods for Assessing Small Credit Pro-grams Evaluating the Economic Impact ofKIK/KMKP Credits.” J. Devel. Stud.

Nelson, Candace; Barbara MkNelly, Kathleen


Stack, and Lawrence Yanovitch. 1995. VillageBanking: The State of the Practice. New York:Small Enterprise Education and Promotion(SEEP) Network.

The New York Times. 1997. “Micro-loans for theVery Poor,” (editorial), February 16.

O’Brien, Timothy. 1998. “For Banks, A Big Nudgeto do More: Communities Fear Lending Drop-Off,” New York Times, July 5, pp. D1, D9.

Otero, Maria and Elisabeth Rhyne. 1994. The NewWorld of Microenterprise Finance. West Hart-ford, CT: Kumarian Press.

Patten, Richard and Jay Rosengard. 1991. Progresswith Profits: The Development of Rural Bankingin Indonesia. San Francisco: ICS Press.

Patten, Richard H. and Donald R. Snodgrass.1987. “Monitoring and Evaluating Kupedes(General Rural Credit) in Indonesia,” HIIDDevelopment Discuss. Paper 249, Harvard U.

Pham, Alex. 1996. “Profile: Michael Chu; A Manof Principles—and Principals,” Boston Globe,July 21, pp. D1, D5.

Pitt, Mark and Shahidur Kandker. 1998a. “TheImpact of Group-Based Credit Programs onPoor Households in Bangladesh: Does the Gen-der of Participants Matter?” J. Polit. Econ.106:5, pp. 958–996.

———. 1998b. “Credit Programs for the Poor andSeasonality in Rural Bangladesh,” Brown U. andWorld Bank, draft.

Pitt, Mark; Shahidur Khandker, Signe-MaryMcKernan, and M. Abdul Latif. 1999. “CreditPrograms for the Poor and Reproductive Behav-ior in Low Income Countries: Are the ReportedCausal Relationships the Result of Heterogene-ity Bias?” Demography, Feb.

Prescott, Edward S. 1997.”Group Lending and Fi-nancial Intermediation: An Example,” Fed. Re-serve Bank of Richmond Econ. Quart. 83:4,Fall, pp. 23–48.

Pulley, Robert. 1989. “Making the Poor Credit-worthy: A Case Study of the Integrated RuralDevelopment Program in India,” World BankDiscuss. Paper 58, Washington, DC

Putnam, Robert. 1993. Making Democracy Work.Princeton, NJ: Princeton U. Press.

Rahman, Aminur. 1998. “Microcredit Initiativesfor Equitable and Sustainable Development:Who Pays?” World Devel. 26:12.

Rashid, Mansoora and Robert M. Townsend. 1993.“Targeting Credit and Insurance: Efficiency,Mechanism Design, and Program Evaluation,”The World Bank and U. Chicago, draft.

Robinson, Marguerite. 1992. “Rural Financial In-termediation: Lessons from Indonesia, PartOne. The Bank Rakyat Indonesia: Rural Bank-ing, 1970–91,” Harvard Institute for Int. Devel-opment Discuss. Paper 434.

———. 1995. “Indonesia: The Role of Savings inDeveloping Sustainable Commercial Financingof Small and Micro Enterprises,” in New Per-spectives on Financing Small Business in Devel-oping Countries. Ernst A. Brugger and SarathRajapatirana, eds. San Francisco: ICS Press.

Rutherford, Stuart. 1995. ASA: The Biography ofan NGO. Dhaka: Association for Social Ad-vancement.

———. 1998. The Poor and Their Money, draftmanuscript.

Sadoulet, Loïc. 1997. “The Role of Mutual Insur-ance in Group Lending,” Department of Eco-nomics, Princeton U., manuscript.

SafeSave Cooperative, Ltd. 1998. 31st October1998 Update. Dhaka: SafeSave.

Schreiner, Mark. 1997. A Framework For theAnalysis of the Performance and Sustainabilityof Subsidized Microfinance Organizations withApplication to BancoSol of Bolivia and Gra-meen Bank of Bangladesh. Unpublished Ph.D.Dissertation, The Ohio State U..

Schuler, Sidney Ruth; Syed M. Hashemi, and AnnP. Riley. 1996. “Rural Credit Programs andWomen’s Empowerment in Bangladesh,” WorldDevel., 24:4, pp. 635–53.

Sharma, Manohar and Manfred Zeller. 1996. “Re-payment Performance in Group-Based CreditPrograms in Bangladesh: An Empirical Analy-sis,” FCND Discuss. Paper 15, Int. Food PolicyResearch Institute, Washington, D.C.

Stiglitz, Joseph. 1990. “Peer Monitoring andCredit Markets,” World Bank Econ. Rev. 4:3,pp. 351–66.

Stiglitz, Joseph and Andrew Weiss. 1981. “CreditRationing in Markets with Imperfect Informa-tion,” Amer. Econ. Rev. 71:3, pp. 393–419.

Tenenbaum, Shelly. 1993. A Credit to Their Com-munity: Jewish Loan Societies in the UnitedStates, 1880–1945. Detroit: Wayne State U.Press.

Varian, Hal. 1990. “Monitoring Agents with OtherAgents,” J. Instit. Theoretical Econ., 146, pp.153–74.

Vengroff, Richard and Lucy Greevey. 1994.“Evaluation of Project Impact: ACEP Compo-nent of the Community Enterprise Develop-ment Project,” Report to USAID/Dakar. U.Connecticut: Storrs.

von Pishke, J.D. 1991. Finance At The Frontier:Debt Capacity and The Role of Credit in thePrivate Economy, Washington, DC: The WorldBank, EDI Development Studies.

Wenner, Mark D. 1995. “Group Credit: A Meansto Improve Information Transger and Loan Re-payment Performance,” J. Devel. Stud. 32:2, pp.264–81.

Woller, Gary; Christopher Dunford, and WarnerWoodworth. 1999. “Where to Microfinance?”Int. J. Econ. Devel. Forthcoming.

Wood, Geoffrey and Iffath Sharif, eds. 1997. WhoNeeds Credit? Poverty and Finance in Bangla-desh. Dhaka: U. Press Ltd.

Woolcock, Michael. 1998. “Social Theory, Devel-opment Policy, and Poverty Alleviation: A Com-parative-Historical Analysis of Group-BasedBanking in Developing Economies,” unpub-lished Ph.D. Dissertation, Dept. Sociology,Brown U.

Woolcock, Michael. 1999. “Learning from Failures


in Microfinance: What Unsuccessful Cases TellUs About How Group-Based Programs Work,”Amer. J. Econ. Sociology, forthcoming.

Working Capital. 1997. “Information Packet,”Winter/Spring Materials, Cambridge, MA.

Wydick, W. Bruce. 1999. “Can Social Cohesion BeHarnessed to Repair Market Failures? Evi-dence from Group Lending in Guatemala,”Econ. J., 109:457, pp. 463–75.

Yaron, Jacob. 1992. “Successful Rural Finance In-stitutions,” World Bank Discuss. Paper 150,Washington, DC.

Yaron, Jacob; McDonald Benjamin, and StephanieCharitonenko. 1998. “Promoting Efficient RuralFinancial Intermediation,” World Bank Res.Observer, Aug.

Yunus, Muhammad. 1995. Testimony before theHouse [of Representatives of the U.S.] Com-mittee on Int. Relations, Fed. News Service,June 27.

Zaman, Hassan. 1997. “Micro-Credit Programmes:Who Participates and What Does it Matter?”Who Needs Credit? Poverty and Finance inBangladesh. Geoffrey Wood and Iffath Sharif,eds. Dhaka: U. Press Ltd.

Zeller, Manfred; Akhter Ahmed, Suresh Babu, Su-miter Broca, Aliou Diagne, and ManoharSharma. 1996. “Rural Financial Policies forFood Security for the Poor: Methodologies fora Multicountry Research Project,” Int. FoodPolicy Research Inst., FCND Discuss. Paper11.


Documents

Microfinance promise