www.elsevier.

com/locate/worlddev

World Development Vol. 32, No. 11, pp. 19091926, 2004 2004 Elsevier Ltd. All rights reserved Printed in Great Britain 0305-750X/$ - see front matter

doi:10.1016/j.worlddev.2004.05.011

Micronance Repayment Performance in Bangladesh: How to Improve the Allocation of Loans by MFIs
MARIE GODQUIN * Paris IPanthe on Sorbonne, and CNRS, France TEAM, Universite
Summary. The aim of this article is to produce a comprehensive analysis of the performance of micronance institutions (MFIs) in terms of repayment. We focus the analysis on the impact of group lending, nonnancial services and dynamic incentives on repayment performance. We test for endogeneity of loan size and use instrumental variables to correct for it. In the second section of the paper, we use a comparative analysis of the determinants of the repayment performance and of loan size in order to make policy recommendations on the allocation of loans by MFIs. 2004 Elsevier Ltd. All rights reserved. Key words micronance, social ties, group homogeneity, nonnancial services, Asia, Bangladesh

1. INTRODUCTION The primary objective of micronance institutions (MFIs) is to provide nancial services (credit and saving) to the poor in order to release nancial constraints and help alleviate poverty. Each MFI tries to maximize its repayment performance, whether or not it is protoriented. High repayment rates are indeed largely associated with benets both for the MFI and the borrower. 1 They enable the MFI to cut the interest rate it charges to the borrowers, thus reducing the nancial cost of credit and allowing more borrowers to have access to it. Improving repayment rates might also help reduce the dependence on subsidies of the MFI which would improve sustainability. It is also argued that high repayment rates reect the adequacy of MFIs services to clients needs. They limit the incidence of crosssubvention 2 across the borrowers. Last but not least, repayment performance is a key variable for donors and international funding agencies on which many MFIs still depend for their access to funds. The rst-best level of repayment performance is a perfect (100%) on-time repayment rate. If the maximum repayment rate the MFI can reach given its lending methodology is lower than the targeted 100%, the MFI will use

second-level strategies to increase its repayment performance. Such strategies include the allocation of larger loans to borrowers with lower default probability and attempts to reduce the delay in repayment. The MFI will develop incentive mechanisms so as to meet these objectives. The main factors inuencing repayment are either related to information asymmetries, to adverse shocks aecting the borrower, or to the low performance of institutions such as justice or education. Information asymmetries arise when gaining information on the characteristics or on the behavior of the borrower is costly for the MFI. Information asymmetries generate problems of adverse selectionallocation of loans to borrowers with undesirable characteristics such as a high level of risk or inability to take advantage of the loan- as well as moral hazardthe borrower may behave in an undesirable way (make little or insucient

* I am grateful to Sylvie Lambert, Pr. Jean-Claude lemy and Pr. Jacqueline Pradel for helpful Berthe suggestions and encouragements. I also want to thank two anonymous referees for their thoughtful comments. Final revision accepted: 26 May 2004.

1909

1910

WORLD DEVELOPMENT

eort to take advantage of his loan or used it for unproductive purposes). Adverse selection and moral hazard increase the proportion of borrowers who cannot repay their loans on time. Borrowers that have enough money to reimburse their loan might also default strategically. The cost of strategic default might indeed be low if the lending institution has low collateral requirements and if the legal system gives little power to the MFI to enforce contracts. MFIs try to restrict the occurrence of those three types of situations in designing appropriate credit schemes. In this paper, we wish to contribute to the improvement of MFI repayment performance by examining their determinants with a particular emphasis on micronance innovations such as the use of nonnancial services, group lending, and dynamic incentives. This paper also questions the adequacy of loan allocations (in terms of loan size) based on the comparison of the determinants of the repayment performance to the determinants of the loan size. This work uses an objective repayment variable, i.e., a repayment variable based on the declaration of the borrower (not the MFIs). It addresses the endogeneity problem of the principal of the loan. The results indicate that the use of nonnancial services has a positive impact on micronance repayment performance but that group homogeneity and social ties among group members are not always associated with a better repayment performance. This article is organized as follows: after a brief presentation of the conceptual framework (Section 2), Section 3 briey reviews the empirical literature and Section 4 provides details on the context of the case study. Section 5 presents the econometric methodology. The results of the regression models are discussed in Section 6 and the article concludes with implications for policy recommendations and future research. 2. IMPROVING REPAYMENT PERFORMANCE: THE CONCEPTUAL FRAMEWORK (a) The rst best level of repayment performance Credit rationing and collateral requirement are the traditional means used by banks to cope with information asymmetries in the credit market (Stiglitz & Weiss, 1981) but both meth-

ods lead to the exclusion of poor borrowers. To explain the success of micronance in providing credit to the poor, a large literature uses the principal/agent theory to demonstrate that micronance contracts lending to joint-liable groups allow the lender to bypass moral hazard (Stiglitz, 1990) and adverse selection (Ghatak, 1999) due to information asymmetries. It is also argued (Besley & Coates, 1995) that joint-liable lending groups help enforce repayment as social interactions make strategic default more costly. Social ties (Besley & Coates, 1995) and group homogeneity (Besley & Coates, 1995; Stiglitz, 1990) are also indirectly linked to repayment performance as they can facilitate peer monitoring and peer pressure or result from an eective peer selection of group members. Regular repayment schedules (Armendariz de Aghion & Morduch, 2000) or dynamic incentives 3 (Besley, 1995) are other appropriate incentive mechanisms used by MFIs to increase their repayment performance. The provision of nonnancial services as a complement to credit and saving services (Edgcomb & Barton, 1998) not only develops the economic ability of the borrower to repay but also makes the relationship with the MFI more valuable to him. The previous mechanisms are considered to be nancial innovations (Edgcomb & Barton, 1998) making it nancially sustainable for MFIs to lend to the poor. When the use of such mechanisms fails to enable the MFI to reach a perfect repayment rate and when borrowers are heterogeneous in their default probability, the MFI could also allocate loans of dierent sizes to the borrowers in order to maximize the value of outstanding debts repaid on time. In the following section, we explain why borrowers are interested in larger loans and why the MFIs should allocate larger loans to borrowers with a lower default probability. (b) The second-best perspective: increasing the value of outstanding debts repaid on time (i) The context We consider a micronance institution providing credit to joint-liable credit groups at a uniform interest rate. The MFI deals with credit applications coming from borrowers heterogeneous in their localization, lending group, ability and preferences. The aim of the MFI is to maximize the global net expected return of its borrowers under a zero prot condition. Where borrowers face high credit rationing, there is a large set of highly productive projects

MICROFINANCE REPAYMENT PERFORMANCE

1911

Expected profit

Size of the Loan (L) L max =L*

Figure 1. Optimal size of the loan (demand size).

and therefore an increasing mean capital productivity. The expected prot for the borrower will thus increase with the loan size for a given duration as shown in Figure 1. 4 The borrower may acquire the information on the cash requirement, expected return, and probability of success of the dierent projects he may manage before applying for a loan. The projects should be such that the sum of the return to the project and of the external returns is higher than each installment. This set of projects is possibly restricted by the local environmental factors such as the distance from marketable activities, the economic situation, and the pre-existing competition of similar projects in the area. The set of accessible projects for a given borrower may be further restricted by his lending group (Madajewicz, 1999) as the group members may incite the borrower to undergo a project similar to the other members projects in terms of size, activity and probability of success. Such a project would indeed be easier to monitor and would limit the deviation from the groups mean default probability and the consequent group cross-subsidization.

(ii) The behavior of the borrower: demand for credit and repayment We consider a borrower who is given the opportunity of obtaining a loan from a micronance institution. The loan application of the borrower corresponds to the size and duration that maximizes his expected return. 5 The optimal duration depends on the distribution over time of the returns of the project, on the scale of the project, on the preferences of the borrower for the present consumption, and on the allocation of a new loan conditional to the total repayment of the previous loan. There is usually little exibility in the duration of microcredit loans and the duration of the loan may be taken as an exogenous variable for a borrower. For a given duration, each loan size corresponds to a single project as the borrower will undertake the project that has the higher expected return for each loan size. As the net return is an increasing function of the size of the loan, the borrower always prefers bigger loans and therefore asks for the largest loan size he may apply for (Lmax in Figure 1) given the set of projects within his reachdened by her own characteristics, those of her environment, and those of her lending group. For a given borrower and duration of the loan, it is argued (Freimer & Gordon, 1965) that the repayment probability decreases with the size of the loan as shown in Figure 2 where Pmin represents the minimum default probability explained by external factors such as illness or accidental destruction of the borrowers productive assets. The increase in the default probability with the loan size diers across borrowers according to their initial endowments and moral hazard or strategic default associated costs.

Default probability

High risk borrower (h)

Low risk borrower (l) P min L

Figure 2. Heterogeneity in the default probability.

1912

WORLD DEVELOPMENT

Default probability

High risk borrower (h)

Low risk borrower (l) Ptarget Pmin L h* L l*

Figure 3. Optimal size of the loan (supply side).

(iii) The MFI: Increasing the value of outstanding debts repaid on time If the maximum repayment rate the MFI can reach given its methodology is lower than the targeted 100%, the MFI will have to dene a new target default probability. It will then allocate loans to borrowers only if the default probability of the loan they are asking for is inferior to the new targeted default probability. If there is observable heterogeneity in the repayment probability of borrowers, the MFI will allocate larger loans to safer borrowers as shown in Figure 3. 3. EMPIRICAL LITERATURE REVIEW Relying on the theoretical literature on microcredit we expect joint liabilityespecially through peer selection, peer monitoring, and peer pressureto be associated with a better repayment performance. Group homogeneity and social ties are also expected to increase the repayment performance through a greater eciency due to group dynamics. Group homogeneity as a result of eective peer selection (group homogeneity in terms of risks, Ghatak, 1999) and as a means to increase peer monitoring (group homogeneity in terms of interest, economic power, Stiglitz, 1990) should imply higher repayment rates. High levels of social ties should have the same impact because they facilitate peer monitoring and increase the potential social sanction of peer pressure (Besley & Coates, 1995). Dynamic incentives and the use of nonnancial services, which are micronance innovations not based on group lending, are also expected to increase the repayment performance. Addressing the question of the relative performance of group loans compared to individ-

ual loans and using data from Zimbabwe, Bratton (1986) states that group loans perform better than individual loans in years of good harvest and worse in drought years when peers are expected to default. Paxton (1996) analyzes with a mean and covariance structural model the determinants of successful group loan repayment of 140 credit groups in Burkina Faso. She draws attention to a negative domino eect 6 that can outweigh the positive effects of group lending. Zeller (1998) uses information on 146 credit groups in Madagascar and provides evidence in favor of group lending. In his article, Zeller shows that the group generates insurance, which leads to a better repayment performance. Analyzing the potential positive eects associated with group dynamics, some studies examine the impact of dierent levels of peer selection, peer monitoring and peer pressure. Wenner (1995) presents a methodology to test whether the selection mechanism has an impact on the repayment performance of 25 Costa Rican credit groups and whether group members use local information for the screening of their peers. His study shows that lending groups use private information to select their peers and that this selection mechanism increases the group repayment performance. 7 On the same issue, the above-mentioned study of Zeller (1998) conrms the positive role of peer selection (internal rules of conducts) on repayment performance. Wydick (1999) uses data from 137 Guatemalan credit groups to show how social cohesion aects group performance in terms of repayment rates, group insurance and moral hazard. He found that peer monitoring in urban groups and peer pressure in rural ones signicantly aects group performance. Contrary to the conclusions of the previous three articles on the positive impact of group

MICROFINANCE REPAYMENT PERFORMANCE

1913

dynamics, Diagne, Chimombo, Simtowe, and Mataya (2000), working on data from Malawi, found that peer monitoring, peer pressure and joint liability had little or a negative impact on repayment performance and that peer selection was found to be limited. Social ties and group homogeneity are supposed to improve the power of group dynamics. Nevertheless, the empirical studies regarding this issue produced mixed results. The study of Sharma and Zeller (1997), based on the analysis of repayment rates of 128 credit groups in Bangladesh, leads to a controversial negative impact of preexisting social ties as well as group homogeneity in terms of asset and enterprise diversity. The study of Zeller (1998) investigates the eects of intragroup pooling of risky assets or projects on repayment rates. While this analysis supports the positive role of social cohesion, it also concludes that risk diversication (up to a certain level) has a signicant positive eect on repayment performance. This could be explained by a matching problem (Paxton, 1996). The matching problem arises when credit terms and conditions are no longer appropriate to each group members needs as credit is renewed. If initial group homogeneity and prior experience of group activities were associated with better repayment performance of the rst loans, as time goes by, the negative impact of matching problems and the absence of risk diversication which limit the possibilities of intragroup insurance should outweigh the positive eect of group homogeneity on peer monitoring. Along with group lending, MFIs usually use dynamic incentives and nonnancial services. MFIs are said to use dynamic incentives when they increase the amounts lent to a specic borrower as credit is renewed, and condition the allocation of new loans to previous repayment behavior. Some micronance programs are also referred to as credit plus (Edgcomb & Barton, 1998; Zohir et al., 2001) as they provide services (such as health services or adult literacy) or training that go beyond nancial services. In contrast to group lending, these two main features of the micronance methodology have been little documented up to now. In the study of Diagne et al. (2000), the most important factor inciting lending groups to repay is the relative value they attach to access to future credit. For Sharma and Zeller (1997), credit rationing, up to a certain level, has a signicant positive eect on repayment performances. In a study on the Grameen

Bank, Khandker, Khalily, and Khan (1995) found that the longer the branch operates in an area, the higher the loan default rate. They explain this feature by the possible decreasing marginal protability of new projects. This could also be due to a deceasing power of dynamic incentives as credit is renewed over time especially if borrowers observe that credit is not systematically denied to defaulting or late borrowers. Khandker et al. (1995) also found that membership training, which relates to nonnancial services, had a positive inuence on repayment. Most of the studies on the determinants of repayment rates also introduce control variables on the characteristics of the area and of the borrower. Khandker et al. (1995) raise the question of whether default is random, inuenced by erratic behavior, or systematically inuenced by area characteristics that determine local production conditions or branch-level eciency. Their empirical test on Grameen overdue loans supports the idea of partial inuence of area characteristics. Rural electrication, road width, primary educational infrastructure and commercial bank density are positively correlated with a low default rate as well as the predicted managers pay. Paxton (1996) shows also that access to other credit sources, market selling activities and urban location were linked to a better repayment performance. Questioning the impact of the characteristics of the borrower, Zeller (1998) showed that traditional prejudices against women, young borrowers or large families should not inuence the determination of repayment ability. Matin (1997), analyzing the determinants of the repayment performance of Grameen Bank borrowers, found that multiple nongovernment organization (NGO) membership, which he associates to access to other sources of cheap nance, had a negative impact on repayment performance. He also found that education and the area of the operated land, which could be proxies for wealth of the borrower, had a positive impact on repayment performance. The membership period was positively associated with default while loan size did not have a signicant impact on repayment performance. The above-mentioned studies attribute a questionable role to mechanisms related to group lending in the explanation of repayment performance of micronance whereas the role of nonnancial services and dynamic incentives is very little documented. It is thus important to

1914

WORLD DEVELOPMENT Table 1. Other source of credit Other credit providers (353 observations) Government Krishi Bank Commercial Bank Cooperative Other NGO Relatives Friends and neighbor Shopkeeper Landlord Other 2.8% 10.2% 10.5% 2.3% 7.4% 33.1% 21.8% 3.7% 5.1% 3.8%

produce further tests of the impact of these nancial innovations on repayment performance. This article provides a test of the explanatory power of social ties and group homogeneity as well as nonnancial services and dynamic incentives. The impact of the main characteristics of the loan contract and of the borrower is also taken into account. 4. DATA The data come from a quasi-experimental survey carried out in Bangladesh in 199192 by the BIDS (Bangladesh Institute for Development Studies) and the World Bank. This survey was designed to assess the impact of micronance on poverty outcomes in Bangladesh (Morduch, 1998; Pitt, 1999; Pitt & Khandker, 1996). The survey covered 1798 households, coming from 87 villages from 29 dierent thanas (subdistricts). 8 Of these households 1,538 were eligible for micronance programs, which means that they were poor enough (possessed less than 0.5 acres of cultivable land) to benet from micronance services. Nine hundred and ve of these households actually took part in one of the three available micronance programs, namely the BRAC, the BRDB and the Grameen Bank. For the purpose of our work, we concentrate our interest on households that had actually borrowed from one of the three MFIs. We use 2,349 loan observations, 485 of which corresponded to BRAC loans, 430 to BRDB loans, 1,081 to Grameen Bank loans and 353 to other credit providers loans (see Table 1 for the description of other credit providers). Based on the due date of the loans and on the date they were fully repaid, we were able to construct dierent dummy variables for indi-

vidual repayment performance. These variables constitute real repayment indicators since they are based on the declaration of the borrowers and are not transformed by the micronance programs based on their denition of late repayment. As shown in Table 2, the on-time repayment rate is signicantly lower than the gures (95% or 98%) normally used in micronance programs communications. But when a 12-month grace period is given to the borrower, the repayment rate increased from 50% to 94%. 9 Table 2 also shows that whatever the grace period, micronance programs experience signicantly higher repayment rates than other credit providers (14% on time repayment. 10) The principal characteristics of the loan contract of micronance programs and of other lenders are also relatively dierent from each other (see Tables 3 and 4). Micronance programs allocate signicantly smaller loans than other credit providers. Around 70% of micronance loans were sizerationed 11 compared to only 30% of the other loans. We have no information on the other dimensions of rationing such as rejected or discouraged loan applications.

Table 2. Dierent measures of the repayment ratea Repayment rates ROT RTG3 RTG6 RTG12 MFIs 0.50 0.76 0.84 0.94 (1,700) (1,591) (1,479) (1,355) BRAC 0.37 0.65 0.74 0.90 (388) (341) (315) (270) BRDB 0.50 0.65 0.77 0.90 (317) (286) (254) (234) GB 0.55 0.84 0.89 0.96 (995) (964) (910) (851) Other credit providers 0.14 0.37 0.43 0.60 (303) (172) (153) (119)

ROT: Dummy = 1 if the borrower repaid his loan on time. RTG3: Dummy = 1 if the borrower repaid his loan with arrears inferior to three months. RTG6: Dummy = 1 if the borrower repaid his loan with arrears inferior to six months. RTG12: Dummy = 1 if the borrower repaid his loan with arrears inferior to twelve months. a The number of loan observations available for the calculation of the dierent repayment rates is given in parentheses.

Table 3. Descriptive statistics of the microcredit contract Variables PRIN DURATION BRAC BRDB SEX PASSET SELFEAGR NBLR AGEGP SAMEEDU SAMEAGE NFSL NFSH CRD PREVLOAN Description Size of the loan (in taka) Duration of the loan (in days) Dummy := 1 if the loan was taken from the BRAC Dummy := 1 if the loan was taken from the BRDB Dummy := 1 if the borrower is a male; 2 if the borrower is a female Value of productive assets of the household (in taka) Dummy := 1 if the borrower received income from agricultural self-employment Number of landed relatives Age of the borrowing group at the due date of the loan (in months) Dummy := 1 if the borrower and group leader have the same education level (more or less two years) Dummy := 1 if the borrower and group leader have the same age Dummy := 1 if the loan program provided the borrower with access to basic literacy Dummy := 1 if the loan program provided the borrower with access to a primary health facility Dummy := 1 if the borrower would have liked to borrow more at the same interest rate Size of the previous loan (in taka) Predicted impact on repayment Number of observations 1996 1981 1996 1996 1996 1996 1996 1996 1949 1996 1996 1996 1996 1996 1996 Mean 2931.057 400.371 0.243 0.215 1.661 35756.800 0.712 2.809 33.177 0.578 0.394 0.678 0.696 0.712 1621.738 Standard deviation 1381.296 187.873 0.429 0.411 0.474 57856.670 0.453 3.309 22.104 0.494 0.489 0.467 0.460 0.453 1700.985 Min 1,000 0 0 0 1 0 0 0 0 0 0 0 0 0 0 Max 10,000 2542 1 1 2 459,001 1 13 114 1 1 1 1 1 8000

MICROFINANCE REPAYMENT PERFORMANCE

+ + + + + +

1915

1916

WORLD DEVELOPMENT Table 4. The loan contract of other credit providers N Mean 5753.853 356.143 36.255 1.102 68135.710 0.816 3.500 0.309 SD 17961.540 350.453 53.648 0.303 105246.500 0.388 3.579 0.463 Min 1,000 0 0 1 0 0 0 0 Max 190,000 2772 300 2 609,602 1 15 1

PRIN DURATION INTEREST RATE SEX PASSET SELFEAGR NBLR CRD

353 203 353 353 353 353 354 353

We were able to reconstruct the credit history of micronance borrowers with their MFI. We found evidence of variation in the size of loans to dierent borrowers for a given loan cycle (see Table 5), which seems to prove the existence of a certain exibility in the determination of loan size by the credit agent. This might partly reect the fact that the studied MFIs use dierent loan size ranges for a given loan cycle depending on the location of the borrower in a municipality, a semi-rural or a remote area. For the three studied MFIs, however, the variance of attributed loan size increases with the loan cycle for the rst three to four cycles (the variance of the loan cycle then declines due to a signicant decrease in the number of observations),

Table 5. Variation of the size of the loan with the loan cycle Mean Std. dev. Min 1,000 1,000 1,500 1,500 3,000 1,000 1,000 2,000 3,000 5,000 1,000 1,000 1,000 1,000 2,000 4,000 4,000 Max 6,000 8,000 7,000 5,000 5,000 6,000 6,000 6,000 8,000 8,000 10,000 6,000 10,000 6,500 6,500 6,500 6,000 N 274 150 41 14 6 266 137 19 5 3 300 273 237 158 83 27 3

BRAC loan cycle 1 2048.566 1147.083 2 2888.553 1411.682 3 3707.317 1600.686 4 3785.714 1155.493 5 3750 758.287 BRDB loan cycle 1 2036.09 727.709 2 2828.467 1093.83 3 3552.632 1289.862 4 5,800 2167.948 5 6,000 1732.051 Grameen Bank loan cycle 1 2253.667 947 2 2963.736 929.047 3 3727.848 1192.662 4 4202.532 1148.3 5 4777.108 1003.813 6 5166.667 784.464 7 5333.333 1154.701

which would imply that older groups become less homogeneous with time in terms of granted size of the loan. The mean, minimum and maximum of the loan size increases with the loan cycle, reecting the eective use of dynamic incentives (the loan size increases at each loan cycle conditioned on the repayment of the previous loan). Only 0.7% of the loans provided by micronance programs had no xed due date compared to 40% for the other credit providers. There is little variation in the duration of the loans of micronance programs: 88.13% of the loans have a duration from 11 to 13 months compared to only 33% of the loans from the other credit providers. There is no variation in the interest rate of micronance programs, which was uniformly xed at 16% per annum for the three studied MFIs (20% after 1991 for the BRAC and the Grameen Bank due to an increase in bank employees payment). The interest rate of the other credit providers is variable and ranges from zero to 300% with 34% of interest-free loans 12 and 30% of the loans with an interest rate superior to 20%. Micronance programs are less exible regarding the purposes of loans they nance with around 85% of the loans obtained for nonagricultural/business purposes; 6% for agricultural purposes and 9% for personal use (respectively 24%, 41% and 35% for the other credit providers). Several studies however, point out the dierences between stated purposes of the loan and eective ones. 13 The growth of the loan portfolio in the sample was similar to the eective growth of the portfolio of the MFIs. The annual growth rate of the Grameen Banks loan portfolio was of 32.73% during 198691 in our sample compared to 37.8% according to calculations based on Khandker et al. (1995).

MICROFINANCE REPAYMENT PERFORMANCE Table 6. Provision of nonnancial services besides credit BRAC Panel A. Access to nonnancial services Access to primary health Access to basic literacy Access to marketing information Access to occupational and skill training N 56.7 66.39 6.8 32.78 485 BRDB 48.14 32.79 20.23 72.79 915

1917

GB 80.57 85.66 40.43 8.05 430

Panel B. Number of nonnancial services the borrower has access to besides credit No access to nonnancial services 0.82 1.4 Access to one NFS 25.36 42.09 Access to two NFS 31.75 5.12 Access to three NFS or more 42.06 51.4 N 485 915

2.13 18.41 31.54 47.92 430

Panel C. Type of NFS provided if borrower has access to only one NFS besides credit Access to primary health 39.84 3.87 Access to basic literacy 25.2 6.63 Access to marketing information 0 1.1 Access to occupational and skill training 0 58.01

36.68 47.24 0 0

Our ndings show few dierences among the MFIs in terms of the characteristics of the loans or types of borrowers. 14 There were, however, dierences in the provision of nonnancial services (see Tables 6, Panels AC). Four types of nonnancial services were provided: primary health, basic literacy, marketing information and occupational and skill training. Almost all the loans were associated with access to at least one nonnancial service. Less than 7% of BRAC borrowers had access to marketing information and only 8% of Grameen borrowers had access to occupational and skill training. This is why we will focus on basic literacy and primary health in the empirical strategy. 5. THE ECONOMETRIC FRAMEWORK Following the discussion on the theoretical literature, we describe the interaction between the borrower and the MFI with the following model: Stage one: The borrower applies for a loan of a given size, which corresponds to the largest one he/she can expect given the projects within his/her reach, and dened by his characteristics, the characteristics of his/her environment, and those of his/her lending group.

Stage two: Before allocating a loan to the borrower, the credit ocer of the MFI assesses the default probability of the applicant using the information he has on the borrower, the lending group, the environment, and the predicted eectiveness of the MFIs repayment incentives for this borrower. If the predicted default probability is inferior to the maximum default probability accepted by the MFI, he will allocate the loan the borrower applied for; otherwise, the credit agent will allocate a smaller loan corresponding to the maximum accepted default probability for this borrower. Stage three: The borrower reimburses the loan on time or not, given his/her environment, his/her characteristics/ability, the characteristics of his/her group and the characteristics of the loan contract. The breakdown in three steps of the microcredit relationship sheds light on the possibility of endogeneity of the principal characteristics of the microcredit contract 15 in the estimation of repayment. The determination of the size of the loan in stage two and the determination of the repayment in stage three might indeed be based on the same omitted variablesvariables observed by the MFI and the borrower but not available in our dataset (such as characteristics of the environment, credit investigation made by the credit agent on the credit history of the

1918

WORLD DEVELOPMENT

borrower with other lenders, and the localize of the borrower). We constructed an individual dummy variable for on-time repayment and use a probit model to estimate the probability for a borrower to repay his loan at the due date. 16 We used the method of Smith and Blundell (1986) to test for exogeneity in the probit model. Endogeneity of the size of the loan could not be rejected and the size of the loan was instrumented. The previous discussion results in the following estimation strategy: Step one: Estimation of the size of the loan:
p b i ep Pi P i a 4 X j1 p p rp j W ij c IVp ei

bp j X ij

2 X j1

p qp j Y ij k Z i

X
j

expected to have a positive impact on the repayment. Wj stands for the exogenous control variables. Control variables gather characteristics of the borrower and his/her household and basic information on the loan (dummy for the MFI and duration of the loan). IVp represents the instrument used for the size of the loan. We use the value of the previous loan as an instrument for loan size. The value of the previous loan is likely to be a good instrument for the loan size as it should not affect the repayment of the present loan and as it was determined on the same unobservables as the present loan. Step two: Smith and Blundells (1986) exogeneity test for principal in the determination of the repayment performance: bi Ri a x P X
j 4 X j1 2 X j1

where Xj represents the variables associated with the incentive structure of group lending, social ties and group homogeneity here. We used the age of the group at the time the loan was due (i.e., the number of months between the date the group was created and the due date) for intragroup social ties. We postulate that the knowledge of the other members characteristics and behavior as well as the level of intragroup social ties are likely to increase with the age of the borrowing group (AGEGP). This is why we expect the ability of the members of the group to monitor and pressure each other to increase with the age of the group. The variable AGEGP should thus have a positive impact on repayment performance. The variables of group homogeneity are based on shared characteristics (age, education level) of the borrower and its group leader. As discussed in the conceptual framework, we expect the variables of group homogeneity (SAMEEDU, SAMEAGE) to have a positive impact on repayment performance. Yj describes the variables of nonnancial services. We use the access to basic literacy (NFSL) and access to primary health (NFSH) as nonnancial services provided by the MFIs. Access to such services is expected to have a positive impact on the repayment as discussed in the conceptual framework. Zi is a variable for dynamic incentives proxied by credit rationing. The incentive power of dynamic incentives (allocation of new and larger loans conditional on previous repayment) is expected to be greater for credit-rationed borrowers and the variable (CRD) is

bj X ij

qj Y ij kZ i 2

rj W ij gep i ei

where Ri, the latent variable of the model, is the borrowers capacity to generate cash in excess to the amount (principal plus interests) he has to repay before the initial due date. We observe the reimbursement R i which takes the value of 1 if Ri > 0 and 0 if Ri < 0. Exogeneity is rejected if the coecient of the error (g) of the instrumented regression of the principal is signicant. This would indeed mean that the structure of the error term is the following: ei aep i l. The test could not reject the endogeneity of the loan size in the determination of the repayment and conrms that we used an appropriate instrumental variable for the size of the loan. Step three: Estimation of the repayment performance: bi Ri a x P X
j 4 X j1 2 X j1

bj X ij

qj Y ij kZ i 3

rj W ij ei

Step four: Larger loans, for whom? After a comment on the regression of the repayment probability, we compare its determinants to those of the size of the loan. If

MICROFINANCE REPAYMENT PERFORMANCE

1919

we assume that the loan size reects the MFIsperception of the capacities of the borrower, this allows us to analyze the adequacy of the loan allocation by the MFI. Because, as previously noted, the loan size is a result of both demand (stage one) and supply (stage two) factors, we cannot simply assume that the loan size reects the abilities the MFI attributes to a specic borrower. We must consider separately the following cases: (a) If the demand of the borrower in terms of loan size is higher than the nal loan size, the loan size reects the perception of the MFI. (b) If the demand is equal to the allocated loan size, it could be that the MFI would have granted the borrower a larger loan had he or she asked for it. In this case we do not know if the loan size reects the exact perception the MFI has of the abilities of the borrower. The estimation of the loan size used for this step was thus based on a sample restricted to those borrowers who were credit rationed. 17 (around 70% of the sample) Since the size of the loan is likely to increase automatically with the loan cycle (use of dynamic incentives), we completed this analysis of the determinants of the loan size by the analysis of the determinants of the variation of the loan size for dierent loan cycles. We used the deviation of the loan size from the mean loan size of the corresponding loan cycle and MFI (specication II) and the loan size for the subsample of the rst (specication III) and the second (specication IV) loan cycle. 18 6. RESULTS AND DISCUSSION (a) The repayment behavior of the borrower The results of the probit estimations of the repayment behavior are reported in Table 7.The model was implemented with a subset of the database that included only those loan observations whose due date had passed at the time of the survey and for which information on all the explanatory variables was completely available. Table 7 presents ve specications of the empirical model of Eqn. (3). The rst column provides the results of the restricted model in which the instrumented size of the loan and the exogenous control variables are introduced. Group level variables, nonnancial services and dynamic incentives

are separately added to this restricted model in columns two to four. The last column stands for the complete model. For each of the ve specications, the Smith and Blundell test could not reject the endogeneity of the size of the loan in the determination of the repayment performance and the principal was instrumented with the size of the previous loan and the variables of the specication. The size of the loan (PPRIN) showed the expected negative sign and is signicant in the ve specications as in the study by Sharma and Zeller (1997). This negative sign is theoretically explained by the fact that the loan size increases the gain associated with ex ante and ex post moral hazard. Table 2 showed however, that most of the unpaid loans at the due date were fully repaid one year after the due date. In this setting, moral hazard should be understood as the choice of a project with a longer maturity (and higher expected value) than the scheduled duration of the loan rather than a choice of a riskier project. The negative sign of the loan size of the loan could also be linked to the borrowers diculty in repaying a larger amount over a given period (usually one year). It could be that, for a given duration, large loans do not meet the borrowers needs and are not suited to the local economy. This statement is to be related to the positive and signicant sign of the duration of the loan (DURATION) throughout the specications. The negative and signicant signs of the dummies for the BRAC and the BRDB indicate that the Grameen Bank has a better repayment performance than the other two MFIs. In our sample, female borrowers (SEX) did not prove to have a signicantly better repayment performance. Even though the coecient is positive, it is not signicant. My results cannot justify the priority given to women in micronance programs based on better repayment performance of women as is sometime done. Using the same dataset, Pitt and Khandker (1998) found that female borrowing had a better impact on poverty reduction than male borrowing. This might be a greater argument in favor of female borrowing. The fact that on average women have lower default probabilities could be partly explained by their lower average loan size. The value of the productive assets of the household (PASSET), the dummy for self-employment in agriculture (SELFEAGR) and the number of landed relatives (NBLR) were used as control variables for the wealth of the household and wealth of its social

1920

WORLD DEVELOPMENT Table 7. Determinants of the micronance repayment performancea (1) (2) 0.0005 (11.94)*** 0.0018 (4.90)*** 0.8083 (8.69)*** 0.3305 (3.09)** 0.1363 (1.81) 0.0000 (3.22)** 0.2638 (3.44)*** 0.0569 (5.35)*** 0.2260 (3.01)** 0.1668 (2.05)* (3) 0.0004 (7.56)*** 0.0019 (5.23)*** 0.7788 (7.95)*** 0.3902 (3.71)*** 0.0888 (1.15) 0.0000 (3.16)** 0.2741 (3.6)*** 0.0579 (5.42)*** (4) 0.0005 (11.87)*** 0.0018 (4.91)*** 0.8855 (9.69)*** 0.4806 (4.84)*** 0.1185 (1.57) 0.0000 (3.37)*** 0.2663 (3.50)*** 0.0585 (5.49)*** (5) 0.0004 (7.72)*** 0.0018 (5.14)*** 0.7247 (7.36)*** 0.2866 (2.58)*** 0.1025 (1.32) 0.0000 (3.08)** 0.2805 (3.62)*** 0.0551 (5.09)*** 0.1957 (2.56)** 0.1215 (1.46) 0.0057 (2.65)** 0.0048 (0.07) 0.0898 (1.31) 0.0419 (0.57) 0.0672 (0.26) 1,664 1013.04 0.122

PPRIN DURATION BRAC BRDB SEX PASSET SELEAGR NBLR NFSL NFSH AGEGP SAMEEDU SAMEAGE CRD CONS N Log likelihood Pseudo R2
a *

0.0005 (11.87)*** 0.0018 (4.92)*** 0.8872 (9.72)*** 0.4776 (4.83)*** 0.1204 (1.60) 0.0000 (3.37)*** 0.2636 (3.48)*** 0.0589 (5.57)***

0.0067** (3.17) 0.0010 (0.01) 0.0860 (1.26) 0.0249 (0.34) 0.3149 (1.32) 1,664 1015.39 0.120 1,664 1022.29 0.114

0.3286 (1.39) 1,664 1022.37 0.114

0.0044 (0.02) 1,664 1018.94 0.117

Estimates obtained using a probit, t statistics are given in parentheses. 10% level of signicance. 5% level of signicance. *** 1% level of signicance.
**

network. Those variables showed a positive and signicant impact on the on-time repayment performance. This trend can be explained by a greater ability of richer households or households with richer relatives to cope with shocks. It is also likely that households with more productive assets have access to projects with higher returns or safer projects. Social ties inside the group, proxied by the age of the group (AGEGP), had a signicant and unexpected negative impact on the repayment rate. 19 If there is a little change in the composition of the borrowing group, this result can be related to the negative impact on repayment of the membership period in the study of Matin

(1997). We could explain this contrasting feature in two ways. At rst, we can refer to what Paxton (1996) called the matching problem: as the duration of membership increases, the credit needs of the members of the group evolve dierently. This circumstance could result in tensions inside the group. It is also possible think that the provision of intragroup insurance becomes more costly as the size of the loans increase and especially if borrowers that are granted a small loan are still jointly liable for borrowers that are granted larger loans. A decreasing power of social penalties can also explain this feature: as members know each other better they are more reluctant to control and sanction

MICROFINANCE REPAYMENT PERFORMANCE

1921

each other. This trend seems to overcompensate the benets of the increasing experience in the provision of intragroup insurance. The understanding of the factors that lead the age of the borrowing group to have a negative impact on the repayment performance needs further research. Should borrowing groups be renewed after several years? Should individual lending succeed group lending after several years? 20 Group homogeneity (SAMEEDU, SAMEAGE) proved to have no signicant impact on repayment performance as in the study of Wydick (1999). 21 We did not address the question of the predicted positive impact on repayment of group homogeneity in terms of risk as a result of peer selection (Ghatak, 1999). Nevertheless, both the study by Zeller (1998) and the study by Sharma and Zeller (1997) give evidence that this kind of homogeneity has a negative impact on repayment performance. Access to basic literacy services (FACL) and to health services (FACH) had a positive and signicant sign. Borrowers who have access to health services are more likely to be able to prevent health shocks and to cope with them. But this positive impact is no longer signicant in the last column where the group-related variables and dynamic incentives variable were added. Borrowers who had access to basic literacy might have access to more protable projects or might be able to generate more cash out of a project. The positive impact of access to basic literacy on the repayment performance is even higher than the impact of access to health services. The innovative features of education programs (both adult and children education) run by these MFIs can explain this feature. The BRAC education program, for example, emphasizes mainly the education of girls in rural areas, giving them the opportunity to attain higher levels of education by providing exible learning hours and scholarships depending on educational performance. Access to nonnancial services can also increase the value of the relationship with the MFI and increase the opportunity cost of strategic default. The incentive power of dynamic incentives (conditionality of the allocation of new and larger loans conditional on the loan size) was expected to be higher for credit-rationed borrowers. Credit rationing (CRD) had the expected positive impact on repayment performance but this impact was not signicant. What we observe here are only marginal impacts of micronance innovations since each of the studied MFIs uses group lending, nonnan-

cial services and dynamic incentives in its credit methodology. This is likely to lead to an underestimation of the impact of these nancial innovations. The analysis of the change in the log likelihood of the dierent specications provides support for the full model (column 5). The share of the variance explained by our variables remains small which could pledge for missing variables. These variables might be community-level variables such as the economic drive of the area, the degree of monetarization, and the exposure of the area to natural disasters 22 (Khandker et al., 1995; Zeller, 1998); program-level variables like functioning costs of the MFIs (Khandker et al., 1995) or borrower-level characteristics such as idiosyncratic shocks (illness and injuries). (b) Is the determination of the loan size ecient? Table 8 reports the results of the estimation of the size of the loan for credit-rationed borrowers (step 4). We consider that MFIs allocate larger loans for projects they anticipate to have a lower default probability. The previous analysis of the repayment performance proved that there were factors aecting the default probability and that it was rational for the MFI to use such information to lower the value of its default loans. The duration of the loan (DURATION) has a positive impact on the size of the loan. There is however, little variation in the duration of the loans and 88.13% of the loans were 1113 months in duration. The coecient of the sex of the borrower (SEX) is negative, which means that MFIs allocate smaller loans to their female borrowers. If MFIs in Bangladesh predominantly lend to women, the fact that they lend smaller amounts to women (the study of Zohir et al. (2001), using data collected in 1998 also found evidence that the micronance programs lend signicantly smaller amounts to their female members) seems to suggest a certain bias against female borrowing. We have shown, however, that females did not experience a lower repayment performance compared to their male counterparts, which conrms Zellers (1998) nding that traditional bias against female borrowing is not justied. MFIs allocate larger loans to households with bigger stocks of productive assets (PASSET) who also demonstrate a better repayment performance. The other two coecients for the wealth of the household and of

1922

WORLD DEVELOPMENT Table 8. Determinants of the size of the loana (1) (2) 1.3194 (6.87)*** 34.8472 (0.5) 79.6920 (0.98) 186.7730 (3.04)** 0.0010 (1.97)** 14.8222 (1.8)* 27.0551 (0.45) 281.3405 (4.83)*** 238.5985 (3.68)*** 2.1302 (1.73)* 174.0838 (3.18)** 180.9802 (3.38)** (3) 0.8993 (3.96)*** 202.4050 (2.12)** 175.2989 (1.66)* 199.1072 (2.51)** 0.0003 (0.4) 11.6931 (1.03) 8.9203 (0.11) 239.9620 (3.18)** 212.2208 (2.56)** 9.1766 (4.03)*** 68.7365 (0.95) 213.1835 (2.91)** (4) 1.9184 (4.31)*** 131.1337 (0.92) 78.1494 (0.52) 175.7549 (1.45) 0.0015 (1.38) 23.6703 (1.41) 62.6599 (0.55) 218.4354 (1.98)** 321.3028 (2.62)** 1.2264 (0.39) 198.3234 (1.9)* 252.3748 (2.44)**

DURATION BRAC BRDB SEX PASSET SELFEAGR NBLR NFSL NFSH AGEGP SAMEEDU SAMEAGE PREVLOAN CONS N ADJR2

1.3390 (7.68)*** 93.0995 (1.4) 68.4574 (0.92) 108.7201 (1.99)** 0.0011 (2.31)** 12.8999 (0.24) 8.7989 (1.2) 216.7072 (4.16)*** 222.4859 (3.84)*** 2.7045 (1.97)** 83.2923 (1.71)* 100.1637 (2.1)** 0.5394 (29.54)*** 1304.6500 (8.63)*** 1,386 0.6024

692.5682 (4.09)*** 1,386 0.0988

1639.9290 (7.75)*** 593 0.1443

2027.0000 (5.92)*** 392 0.1026

Dependent variables, sample: Specication I: Loan size, sample: credit rationed loans. Specication II: Deviation of the loan size from the mean loan size of the corresponding loan cycle and MFI, sample: credit rationed loans. Specication III: Loan size, sample: credit rationed loans, rst loan cycle. Specication IV: Loan size, sample: credit rationed loans, second loan cycle. a Estimates obtained using OLS. t statistics are given in parentheses. * 10% level of signicance. ** 5% level of signicance. *** 1% level of signicance.

its relatives (SELFEAGR and NBLR) show no signicant impact on the size of the loan. But, the use of both of these wealth indicators for the choice of the loan size would be likely to increase the inequalities among borrowers and it is thus preferable that micronance credit agents do not take these variables into consideration when allocating loans. The coecient of the age of the borrowing group at the due date (AGEGP) has a positive and signicant coecient. It is indeed expected that the social ties and other benets of the group, such as information-sharing, increase with the age of the group. As previously men-

tioned, however, the age of the group has a negative impact on repayment performance. If MFIs want to allocate larger loans to group members with whom they have an established relationship, they should also develop specic incentives for their experienced borrowers to have a better repayment performance. Group homogeneity in terms of age (SAMEAGE) has a positive impact on loan size and group homogeneity in terms of education (SAMEEDU) has a negative impact on the size of the loan. But, as shown in the analysis of repayment performance, group homogeneity based on age or education level did not prove

MICROFINANCE REPAYMENT PERFORMANCE

1923

to have any signicant impact on the repayment. Group homogeneity is frequently used as a methodological guideline for group formation in many micronance programs 23 and further research must be undertaken to understand what type, if any, group homogeneity has a positive impact on the borrowers repayment performance. Access to both of the nonnancial services (NFSL, NFSH) has a positive inuence on the size of loans. These services might indeed increase the borrowers capabilities and thus increase their probability of success as shown in the analysis of repayment performance. The value of the previous loan (PREVLOAN) has a positive and significant impact on the loan size. This is explained by the lending methodologies of our MFIs as the size of the loan usually increases with the loan cycle. The MFIs also use dierent loan size ranges for a given loan cycle depending on the location of the borrower in a municipality, a semi-rural or a remote area. The size of the previous loan thus incorporates this information. Since MFIs are likely to increase systematically the size of the loan with the loan cycle, analyzing loan allocation through the variation of loan size for dierent loan cycles might be more appropriate. Specications II to IV present alternative methods of evaluating loan allocation using the deviation to the mean loan size for the corresponding loan cycle and MFI (specication II) and the loan size for the subsample of the loan observations corresponding to the rst cycle (specication III) and second cycle (specication IV). These alternative specications conrm our previous ndings. The value of the previous loan (PREVLOAN) seems thus sucient to take dynamic incentives into account into the estimation of loan allocation through loan size. 7. CONCLUSION Micronance programs are now a key element of poverty alleviation strategies. The nancial innovations of their lending methodologies such as the use of group lending, nonnancial services and dynamic incentives have indeed raised the interest of policy makers and researchers as means to alleviate poverty in a self-sustainable way. In this study we test the explanatory power of theoretical models that attribute the performances of MFIs in terms of repayment to the use of such nancial innovations. We used household level data, which allowed us to compute detailed repayment performance variables. We

also draw empirical researchers attention on the endogeneity problem of the principal for future analysis on repayment. Our results suggest that the provision of nonnancial services has a positive impact on repayment performance. This provides arguments in favor of the integrated development strategies. The provision of such services is, however, costly and further research is needed to assess the costs and benets associated with the dierent types of nonnancial services. Research is also needed on the best institutional way of providing these services: should they be provided by the MFI or should the MFI operate in partnership with another NGO providing these services? The results also show that MFIs allocate larger loans to borrowers as the age of their borrowing group increases. This can be justied by the use of dynamic incentives, as the number of allocated loans is likely to grow with the age of the group. The age of the group was also found to have a negative impact on the repayment. This raises the need to develop new incentives for experienced borrowers to avoid decreasing repayment performance and negative domino eects as the clientele of the MFI becomes more mature. Another important point that emerged from the study is that MFIs tend to attribute larger loans to homogeneous groups in terms of age. Group homogeneity was not, however, found to aect the repayment performance in a significant way. We did not address here the question of the predicted positive impact on repayment of group homogeneity in terms of risk as a result of peer selection (Ghatak, 1999). Nevertheless, both the studies of Zeller (1998) and of Sharma and Zeller (1997) provide evidence from Madagascar and Bangladesh that show that this kind of homogeneity has a negative impact on repayment performance. As group homogeneity is frequently used as a methodological guideline for group formation in many micronance programs, further research must be undertaken to understand what type, if any, of group homogeneity has a positive impact on the borrowers repayment performance. Micronance programs have been successful in extending credit to the poor thanks to appropriate lending methodologies. The negative impact on the repayment performance of the size of the loan and of the age of the borrowing group could reveal the incompleteness of these lending methodolo-

1924

WORLD DEVELOPMENT

gies as the clientele of micronance becomes more mature.

NOTES
1. A high level of repayment performance is a prerequisite for nancial sustainability, but it is not a sucient condition of nancial health as high administrative costs or high borrower turnover could be the counterpart of high repayment rates. 2. As borrowers have dierent default probabilities and as it is dicult for the MFI to charge each borrower an interest rate corresponding to his default probability, borrowers who are more prone to default will be subsidized by lower risk borrowers. 3. Dynamic incentives refer to the threat of not renancing a borrower who defaults on a debt obligation. The incentive power of dynamic incentive is enhanced if the MFI allocates larger loans over time to borrowers with a good repayment performance. 4. Figure 1 stands for a constant marginal productivity of capital. We could also allow for an increasing marginal productivity of capital that would strengthen the attractiveness of larger loans for the borrowers. 5. We consider that the utility of the borrower is increasing with the return he gets from his project. 6. The domino eect occurs when at least one member of a credit group default due to the defaults of other members. 7. Wenner challenges the positive eect of this feature as further analysis indicates that costs faced by the borrowers to get this information overcompensate the induced benets in terms of repayment performance. 8. These 29 thanas were randomly chosen out of the 391 thanas of Bangladesh. 9. Zohir et al. (2001), using borrower level information, found a percentage of borrowers with current overdue of 28.7% for the Grameen Bank and 34.7% for BRAC. These lead to higher on-time repayment rates (71.3% for the Grameen Bank, 65.5% for BRAC) than the ones that we found (55% and 37% respectively). These rates are also signicantly lower than 95%, but it is dicult to compare their magnitude with the repayment rates that we found given the nine years dierence between the two surveys. 10. Khalily & Meyer (1993) found a recovery rate on rural loans by state-owned commercial banks of 18.8% in 19881989 in Bangladesh. 11. We consider a loan to be size-rationed when the borrower reports he would have liked to borrow more from the same source, at the same interest rate, for the same purpose at the time he borrowed the loan. 12. 70% of interest-free loans were provided by relatives, 85% by relatives or friends and neighbors. 13. 109 of the 1996 micronance loan observations corresponded to loans allocated before the previous loan was fully repaid. It is likely that some of these loans were partly used for renancing purposes even if renancing was not one of the eligible purposes for loans. Rahman (1999) shows that 70% of the 217 Grameen loan observations he collected were actually used for purposes other than the approved loan project, including money lending. 14. The only signicant dierence we found was that the BRDB, a government program, had a lower proportion of female borrowers (34% compared to over 70% for the other two MFIs at the time of the survey, 10% for the other credit providers), which resulted in a higher education level of its borrowers. 15. Here the size of the loan is the only characteristic of the loan for which correction for endogeneity is provided. We tried to provide correction for the possible endogeneity of the duration of the loan but we were not able to nd satisfactory instruments. This might be explained by the limited variation of the duration of MFI loans. 16. We also tried estimations using larger denitions of repayment (repayment made within three, six or 12 months after the maturity date). For such estimations, the sample had to be reduced to borrowers whose due date was three to 12 months prior to the date of the survey. 17. The borrowers were classied as credit rationed if they reported they would have liked to borrow more from the same source, at the same interest rate, for the same purpose at the time they borrowed.

MICROFINANCE REPAYMENT PERFORMANCE 18. We did not have enough observations to extend this analysis to older cycles (the adjusted R2 of the regression of the deviation of the loan size for the third loan cycle, 193 observations, was of 1.85% only). 19. We have no possibility of knowing whether some of the borrowers are new members in old groups or old members in new groups since the borrower were not asked about the date their group was created. There is now a lot of recomposition of borrowing groups, especially in the case of the BRDB where members are expected to stay within the program for 57 years. This should not be an issue however here since the survey was conducted in 1991 when micronance programs had not been in operation for a long in most of the villages. 20. One of the referees pointed that the role of the

1925

learning on the rules of the game takes place for all the parties. The group then becomes more a convenient loan collection forum and loan contracts are more individualized. But program members still have to attend the weekly meetings and repay by group to their loan ocers. 21. In Wydicks study SAMESEX is the only variable of social ties that aected repayment performance and it has a negative impact only in urban groups. Here, since the three studied micronance programs operate with same-sex group only, we did not include this variable in the analysis. 22. Village xed eects were not used because we had only a few loan observations per village.

23. The Grameen Bank Constitution reproduced in Rahman (1999, p. 161) states, for example, that All groups shall be formed with persons who are likeminded, are in similar economic condition and enjoy mutual trust and condence.

group is evolutionary. Initially, the group helps screen applicants and increase the use of social sanctions and mutual insurance. But as the group members interact among themselves and with the MFI workers, critical

REFERENCES
Armendariz de Aghion, B., & Morduch, J. (2000). Micronance beyond group lending. Economics of transition, 8(2), 401420. Besley, T. (1995). Nonmarket institutions for credit and risk sharing in low-income countries. Journal of Economic Perspectives, 9(3), 115127. Besley, T., & Coates, S. (1995). Group lending, repayment incentives and social collateral. Journal of Development Economics, 46, 118. Bratton, M. (1986). Financing smallholder production: A comparison of individual and group credit schemes in Zimbabwe. Public Administration and Development, 6, 115132. Diagne, A., Chimombo, W., Simtowe, F., & Mataya, C. (2000). Design and sustainability issues of rural credit and saving programs for the poor in Malawi: An actionoriented research project. Washington, DC: IFPRI. Edgcomb, E., & Barton, L. (1998). Social intermediation and micronance programs: A literature review. Washington, DC: USAID, Microenterprise Best Practices. Freimer, M., & Gordon, M. J. (1965). Why bankers ration credit. Quarterly Journal of Economics, 79, 397416. Ghatak, M. (1999). Group lending, local information and peer selection. Journal of Development Economics, 60, 2750. Khalily, B., & Meyer, R. (1993). The political economy of rural loan recovery: Evidence from Bangladesh. Savings and Development, 17(1), 2338. Khandker, S. R., Khalily, B., & Khan, K. (1995). Grameen Bank: performance and sustainability. World Bank Discussion Paper, 306, The World Bank, Washington, DC. Madajewicz, M., (1999). Capital for the poor: The eect of wealth on the optimal credit contract. Mimeo, Columbia University, New York. Matin, I. (1997). Repayment performance of Grameen Bank borrowers: the unzipped state. Savings and Developpment, 4, 451473. Morduch, J. (1998). Does micronance really help the poor? New evidence from agship programs in Bangladesh. Mimeo, Department of Economics, Harvard University, Cambridge, MA. Paxton, J. A. (1996). Determinants of successful group loan repayment: an application to Burkina Faso. Unpublished doctoral dissertation, The Ohio State University, OH. Pitt, M. M. (1999). Reply to Jonathan Morduchs Does micronance really help the poor? New evidence from agship programs in Bangladesh. Mimeo, Department of Economics, Brown University, Providence, RI. Pitt, M. M., & Khandker, S. R. (1996). Household and intrahousehold impact of the Grameen Bank and similar targeted credit programs in Bangladesh,

1926

WORLD DEVELOPMENT Stiglitz, J. E., & Weiss, A. (1981). Credit rationing in markets with imperfect information. American Economic Review, 17(3), 393410. Wenner, M. (1995). Group credit: a means to improve information transfer and loan repayment performance. Journal of Development Studies, 32, 263281. Wydick, B. (1999). Can social cohesion be harnessed to repair market failures? Evidence from group lending in Guatemala. The Economic Journal, 109, 463 475. Zeller, M. (1998). Determinant of repayment performance in credit groups: The role of program design, intragroup risk pooling, and social cohesion. Economic Development and Cultural Change, 46(3), 599621. Zohir, S., Mahmud, S., Sen, B., Asaduzzaman, M., Jahirul Islam, Md., Ahmed, N., & Al Mamun, A. (2001). Monitoring and evaluation of micronance institutions. Final report, Dhaka: BIDS.

World Bank Discussion Papers, 320, The World Bank, Washington, DC. Pitt, M. M., & Khandker, S. R. (1998). The impact of group-based credit programs on the poor households in Bangladesh: does the gender of participants matter? Journal of Political Economy, 106(5), 958996. Rahman, A. (1999). Women and microcredit in rural Bangladesh: An anthropological study of Grameen Bank lending. Boulder, CO: Westview Press. Sharma, M., & Zeller, M. (1997). Repayment performance in group-based credit programs in Bangladesh: An empirical analysis. World Development, 25(10), 17311742. Smith, R. J., & Blundell, R. W. (1986). An exogeneity test for a simultaneous equation tobit model with an application to labor supply. Econometrica, 54(3), 679685. Stiglitz, J. E. (1990). Peer monitoring and credit markets. The World Bank Economic Review, 4(3), 351366.