microfinance and investment: a comparison with bank and informal lending

World Development Vol. 39, No. 6, pp. 882–897, 2011� 2011 Elsevier Ltd. All rights reserved

0305-750X/$ - see front matter

www.elsevier.com/locate/worlddevdoi:10.1016/j.worlddev.2011.03.002

Microfinance and Investment: A Comparison with Bank and

Informal Lending

LUCIA DALLA PELLEGRINA *

University of Milan-Bicocca, ItalyPaolo Baffi Center, Bocconi University, Italy

Summary. — Comparing the impact of different types of credit on households’ investment in Bangladesh, we find that loans from micro-finance institutions are likely to be channeled toward non-agricultural activities while both informal and bank lending are associated to ahigher expenditure in agricultural inputs. Estimated effects are net of the differences in the amount borrowed, interest rates, and collat-eral. Results suggest that features which are specific to microfinance—such as tight repayment schedules and land-based eligibilityrules—may reduce the suitability of this source of funds for the farming sector.� 2011 Elsevier Ltd. All rights reserved.

Key words — microfinance, banks, informal lending, investment

* I am grateful to two anonymous referees, Gani Aldashev, Jean-Marie

Baland, Vittoria Cerasi, Lisa Crosato, Asif Dowla, Hadi Esfahani, Chri-

stopher Flinn, Eliana La Ferrara, Robert Lensink, Niels Hermes, Matteo

Manera, Jean-Philippe Platteau, Hans Seibel, and to seminar participants

at University of Namur (FUNDP), University of Milan-Bicocca, Bocconi

University, University of Lecce, and 2007 Conference on Microfinance at

the University of Groningen. I am particularly grateful to Shahidur

Khandker for allowing me to use the World Bank dataset. The usual

disclaimer applies. Final revision accepted: October 30, 2009.

1. INTRODUCTION

A large part of the literature on microcredit has been de-voted to the analysis of its effectiveness in terms of povertyreduction. Several applied studies have investigated the impactof different programs operating on the basis of group lendingon the behavior of households and firms, such as per capitaconsumption, labor supply, children school enrollment(Morduch, 1998; Pitt & Khandker, 1998), and business perfor-mance (Madajewicz, 2003a; McKernan, 2002), providingevidence of success.

Standardized microfinance (hereafter MF) agreements,however, have been recently criticized since they are consid-ered not properly suited to fulfill the needs of all sectorsof the economy. Agriculture, in particular, seems to sufferfrom the absence of contractual flexibility (Christen &Pearce, 2005; Llanto, 2007; Meyer, 2002; Murray, 2001).In fact, it is a common practice for microfinance institu-tions (from now on MFIs) to lend to landless householdsand require reimbursement of the loan soon after it hasbeen granted.

In particular, tight repayment schedules may preclude bor-rowers from undertaking long-term investments, as is oftenthe case in agriculture where the production cycle is longerthan in several other activities. In addition, farmers mayencounter difficulties to commit to regular installments dueto the risk related to climate conditions (see, e.g., Caldwell,Reddy, & Caldwell, 1986). This might also bias MFIs in favorof the non-agricultural sector as consequence of reducing therisk of default more commonly associated with natural disas-ters. 1 Such a situation could be exacerbated by the eligibilityrules for MF programs which rely on the lack of land owner-ship in order to identify poor borrowers. As opposite to otherproducers, in fact, farmers are more likely to have all theirwealth concentrated in small land plots, and are thus less enti-tled to obtain loans.

This paper examines the role of financing mechanisms onhousehold investment decisions. In particular, we aim atverifying what kind of investment—in agricultural and non-agricultural activities—is promoted by different kinds oflending: micro-lending, informal lending, and bank lending.The article, in particular, adds to previous work by concentrating

882

on a comprehensive definition of investment—expenditure inboth working capital and fixed assets—and by including allavailable sources of credit. 2

The reason why we focus on investment is that, as oppositeto consumption behavior, the former is more suited to provideinsights on programs’ long-term effect on growth. For exam-ple, practitioners (Bage, 2004) stress that in order to achieveself-sustainability households in low income contexts shouldnot myopically consume borrowed funds but rather investthem in productive activities. 3 Ahlin and Jiang (2008) alsoclaim that the key to the success of MF long-term objectivesrests in the fact of promoting the gradual accumulation ofaverage returns in self-employment. These recommendations,though, might be ineffective when addressed to poorest people,since, having a higher time preference rate compared to richerindividuals, they might be tempted to raise present consump-tion instead of acquiring production inputs (Lawrance,1991). Hence—at least at first glance—it becomes importantto verify to what extent borrowers from MF programs per-form better than non-borrowers, and this is essentially whathas been done by most of the empirical studies dealing withthe impact of MF.

However, this approach seems somehow restrictive since itclassifies borrowers from other sources as non-borrowers. Infact, in the same vein as ignoring non-treated groups (see,e.g., Heckman, 1976), disregarding those who are treated withdifferent types of credit may provide biased estimates whenevaluating the impact of MF programs. Moreover, compari-son with other types of lending is important to our purposessince it allows verifying whether there are features of MF

http://dx.doi.org/10.1016/j.worlddev.2011.03.002

MICROFINANCE AND INVESTMENT: A COMPARISON WITH BANK AND INFORMAL LENDING 883

contracts which are likely to penalize some categories of bor-rowers, such as farmers. If this is the case, we should observethat these categories more frequently apply to other credit pro-viders having different characteristics. 4

We use data from a survey of the World Bank carried outduring the years 1991 and 1992 in Bangladesh. The surveycontains information about credit from MFIs and non-governmental organizations providing group lending, loansobtained from landlords, input suppliers, shopkeepers,employers, relatives and friends (which we define as the infor-mal lending channel), and banks. There are a number of rea-sons why it is still useful to investigate data related to aperiod when group lending was predominant in the contextanalyzed in the paper. First, although the Grameen Bank(GB) has significantly reduced the focus on the traditionalfive-person group in favor of the village organization, it stillmaintains the group structure (Barua & Dowla, 2006), so thatborrowers’ incentives should not be significantly affected bythese changes (Armendariz & Morduch, 2005, p. 101). Second,traditional group loans are even now the core of credit servicesprovided by other institutions included in our dataset, such asthe Bangladesh Rural Development Board. 5 Third, the grouplending model as used by GB in the 1990s is still the dominantone in many developing countries, and especially in Africa(see, e.g., Basu, Blavy, & Yulek, 2004, on a variety of Africancounties, Brandsma & Chaouali, 1998, on North Africa andthe Middle East; Hermes, Lensink, & Mehrteab, 2005, onEritrea).

The empirical analysis has been carried out through tech-niques that use instrumental variables to reduce the endogene-ity bias generated by the correlation between the choice of thecredit source and non-measurable characteristics affectinginvestment, such as for example, household members’ ability.Moreover, we concentrate on a core relationship involvinginvestment, on the one hand, and the probability of borrowingfrom each source, on the other hand, while the impact of theamount borrowed, interest rate, and collateral is analyzed sep-arately. This helps isolating the effect of other characteristicsof credit contracts such as, among others, the repayment sys-tem.

Results show that, conditional on measurable features ofcredit agreements, borrowers from MFIs are likely to investmore in non-agricultural activities while both informal andbank lending are associated to a higher investment in agricul-tural inputs. According to what has been discussed so far, thisseems to provide evidence in favor of our hypothesis concern-ing the lower suitability of MF standardized lending programsfor the agricultural sector.

It is worth stressing, however, that there may be other fac-tors driving our empirical findings. Joint or individual liability(Ghatak & Guinanne, 1999; Hermes & Lensink, 2007; Hermeset al. 2005; Madajewicz, 2003b; Paxton, Graham, & Thraen,2000; Sharma and Zeller, 1997; Wydick, 1999), the quality ofmonitoring (Armendariz, 1999; Banerjee, Besley, & Guinnane,1994; Madajewicz, 2003a; Stiglitz, 1990; Varian, 1990, amongothers) and the pattern of sanctions (Besley & Coate, 1995)may induce different attitudes toward investment dependingon the contract chosen. Threat of future credit denial, whichis typically used in MF, could also work in a similar way.Nevertheless, there seems to be weaker evidence, as well as aless intense debate, as to whether these elements are likely tounevenly affect different sectors of the economy.

The rest of the paper is organized as follows. In Section 2 weillustrate the dataset. Section 3 concentrates on the estimationtechniques and instruments adopted. In Section 4 we discussthe results. Section 5 concludes.

2. DATA

Data were collected in a survey carried out on 1,798 house-holds in rural Bangladeshi villages by the Bangladesh Instituteof Development Studies at the World Bank in 1991–92. Thesurvey was conducted in three rounds, approximately corre-sponding to the harvesting of the rice crop. The first round(November 1991–February 1992) corresponds to the Amanseason, the second (March–June 1992) to the Boro, and thethird (July–October 1992) to the Aus. The original sampleconsists of three randomly selected villages from each of the29 districts (thanas) surveyed. In 24 of these districts, a micro-credit program had been in operation for at least three years.A total of 20 households in each village were surveyed. Wemainly concentrate on the first round since a great deal ofinformation went missing during the remaining two. In partic-ular, we work under a cross-sectional setup, 6 using the otherrounds to gather information on investment in fixed assets(i.e., to compute the difference in the stock), since data donot provide a direct measure of this variable. 7

Households engaged in self employed activities number1,276. Almost all of them (1,192) are farmers or fishermen,797 are non-farmers, and of these 713 are engaged in bothfarming and non-farming activities. Investment in workingcapital corresponds to total operating costs in the year preced-ing the survey. 8 In the case of farming, these include expendi-ture on seeds, fertilizers, pesticides, water, tillage, rented labor,and veterinary costs. Non-farmers’ operating costs 9 are con-stituted by raw materials, rented labor, fuel for transport,and other expenses for equipment maintenance. Investmentin fixed assets is computed as the incremental value of physicalcapital between the first and the third rounds of the survey.Capital consists of bullocks, cows, sheep, poultry, and agricul-tural equipment if the household engages in farming activities.As for non-farmers it is mainly constituted by buildings,machinery, rickshaws, sewing machines, and other durables.We exclude inherited assets since they are not bought by thehousehold, neither through credit, nor by self-financing. Landis also omitted—it is instead used as a control variable for rea-sons that will be discussed in the following sections—whilerent paid for land is accounted for as part of operating expen-diture.

The importance of distinguishing between fixed assets andworking capital is a pure statistical issue. It lies in the fact thatthe distributions of these two variables differ considerably (seethe following section). More precisely, fixed assets may be dis-missed—hence a negative investment is possible—and alsoshow a considerably higher variance as compared to workingcapital. Jointly considering the two variables would not ac-count for the heterogeneity of the latter, since all the effectswould be driven by the relationship between credit and themore highly-volatile component of investment. 10

Table 1 reports summary statistics on investment. Workingcapital expenditure is on average 1,297 taka for farmers com-pared to 1,191 for non-farmers. Also investment in fixed assetsis higher for farmers (766 taka, against 69 for non-farmers ifnegative values are excluded; 350 taka, against �3,958 fornon-farmers if negative values are included). 11 T-tests of meancomparison reported in Table 1 suggest that there is no signif-icant discrepancy between the operating costs of farmers andnon-farmers. The difference in means is instead weakly signif-icant for investment in fixed assets.

We consider all loans granted to household members in theyear preceding the survey. The sources of microcredit are theGB, the Bangladesh Rural Advancement Committee (BRAC),and the Bangladesh Rural Development Board (BRDB). All

Table 1. Investment in working capital and fixed assets (in Taka): farming and non farming activitiesa

Obs. Mean Std. Dev. Min Max

Farmers

Investment in working capitalb 1,192 1,297 2,263.745 0 31,420Investment in fixed assets after 12 monthsd 1,192 350 2,589.825 �20,000 30,000Investment in fixed assets after 12 months correctede 1,007 766 2,325.911 0 30,000

Non-farmers

Investment in working capitalc 797 1,191 6,472.574 0 108,461.5Investment in fixed assets after 12 monthsd 797 �3,958 13,228.07 �225,000 7,950Investment in fixed assets after 12 months correctede 224 69 638.5657 0 7,950

T test of farmers and non-farmers mean comparison: Diff. in means Std. Dev. T statistic Degr. of freedom

Investment in working capital 105 203.8573 0.5151 1,987Investment in fixed assets after 12 monthsd 4,307 393.9166 10.9338f 1,987Investment in fixed assets after 12 months correctede 697 481.6863 4.4484f 1,229

First round of interviews: 1991/11–1992/2. Second round of interviews: 1992/3–1992/6. Third round of interviews: 1992/7 to 1992/10.a Based on respondents’ statements about the main activity carried out by the household.b Corresponds to the average expenditure of one production cycle in the year preceding the (first round of the) survey.c Corresponds to the expenditure one year preceding the (first round of the) survey.d Corresponds to the difference between fixed assets in the third round of the survey and fixed assets in the first round of the survey, including negative

values.e Corresponds to the difference between fixed assets in the third round of the survey and fixed assets in the first round of the survey, excluding negative

values.f Denotes significant difference in means.

884 WORLD DEVELOPMENT

three institutions operated through group lending at the timeof the survey. The class of informal lending, instead, is repre-sented by suppliers and merchants, landlords, relatives, andneighbors. On the one hand, loans from suppliers and mer-chants are almost exclusively short-term based and collat-eral-free, and are recovered through the purchase of theoutput at a price agreed in advance. On the other hand, land-lords are typically wealthy persons who set very high interestrates and often require physical collateral, as opposite to rela-tives and neighbors who lend at lower interest rates and rarelyrequire collateral. In general, the main characteristic of bor-rowing from the informal channel consists in the advantageof obtaining immediate approval and flexible amounts ofmoney (Timberg & Aiyar, 1984). Finally, the banking sectorconsists of commercial banks, specialized banks, and coopera-tive banks.

Statistics on credit are reported in Table 2. The number ofhouseholds borrowing from MFIs under group lending duringthe year previous to the survey is 297, while those borrowingfrom the informal sector and banks number 111 and 40,respectively. Average micro-loans are 6,386 taka. A slightlylower principal is accorded by informal lenders (5,445 taka),while banks provide substantially higher amounts (11,795taka). Interest rate is 16% on both MF and bank contracts, 12

compared to a mean of 52% on informal ones. In particular,informal rates differ considerably across the sample with astandard deviation of 62%. 13 Informal moneylenders andbanks require collateral on 10% and 45% of loans respectively,whereas MFIs never require any guarantee. We build a dum-my variable taking the value of 1 in case collateral is requiredby the lender, although there are no observations concerningits value.

Loan duration at the time the loan is granted 14 is almost thesame for MF and informal credit (392 and 384 days, respec-tively), while the duration of bank contracts is considerablyhigher. At the time they were surveyed, the percentage of bor-rowers having repaid the loan ranges between 12% and 15%regardless the type of contract, and should presumably corre-spond to loans obtained at the beginning of 1991. As we con-sider loans that have been obtained earlier in time, that is 1.5

and 2 years previously, the repayment rate increases. As onecan see, despite having the same duration of informal lending,MF loans seem to perform better in this regard. Banks seem toperform the worst, but since bank contracts are of longerduration they cannot be compared with other sources.

As for the timing of loans and investment, we assume thatworking capital expenditure takes place within 12 monthsfrom the loan, while disbursement for fixed assets is assumedto occur up to twenty-four months from the loan. The choiceof time lags relies on the fact that buying working capital doesnot require a lengthy and detailed evaluation process, and nor-mally occurs at the time—or right after—the loan is obtained.Instead, longer periods may elapse from the time the loan isgranted to its actual disbursement for purchasing physical cap-ital. We explain this with the fact that fixed assets normallyrepresent important expenditures, 15 and, therefore, buyersand sellers are likely to engage in lengthier negotiations—which often involve cash advances—before the transfer ofownership occurs. Sometimes agreements may even fail tocomplete, so that entrepreneurs must further delay purchasessince they need to find other potential sellers.

From a breakdown of the main loan characteristics in termsof borrowers’ activity it emerges that typically MFIs and infor-mal lenders grant similar sums to households engaged in farm-ing and non-farming activities, while banks providesubstantially higher amounts to farmers. This may be due tothe fact that farmers own land, which, according to the state-ments of respondents, is almost the only asset accepted as col-lateral. Therefore, farmers seem to have a better access to banklending but also appear to be precluded from accessing the MFsector, perhaps because of land eligibility rules (see below).

For the purposes of this paper, it may be useful to combineinformation on credit and investment discussed so far. Arough measure of the ability of different credit sources to raiseinvestment, in fact, is the ratio of expenditure to the amountborrowed for each type of loan. This is a proxy of the shareof invested funds, that is, not used for consumption pur-poses. 16

Table 3 reports computed values of this measure, separatingworking capital from fixed assets. At first glance, what seems

Table 2. Credit: amount borrowed, interest rates, and collateral

Number of households Mean Std. Dev. Min Max

Group lending (MFIs) 297Amounta (taka) 6,386 6,262 1,000 54,014Interest rate (%) 16 0.63 16 20Collateral (% loans requiring-) 0 0 0 0Duration (days) 393 192 0 2,338

Repayment rate

Date loan 1991–2011 to 1991–2012 0.12 0.33 0 1Date loan 1990–2007 to 1991–96b 0.45 0.49 0 1date loan 1989/1–1989/12c 0.81 0.38 0 1

Informal lending 111Amounta (taka) 5,445 5,924 1,000 40,000Interest rate (%) 52 62.10 0 240Collateral (% loans requiring-) 0.09 0.30 0 1Duration (days) 385 509 12 3,710

Repayment rate

Date loan 1991/1–1991/12 0.13 0.34 0 1Date loan 1990/7–1991/6b 0.17 0.38 0 1Date loan 1989/1–1989/12c 0.44 0.51 0 1

Bank lending 40Amounta (taka) 11,795 27,707 1,000 175,000Interest rate (%) 16 0 16 16Collateral (% loans requiring-) 0.45 0.503 0 1Duration (days) 511 407 0 2,190

Repayment rate

Date loan 1991/1–1991/12 0.15 0.36 0 1Date loan 1990/7–1991/6b 0.22 0.42 0 1Date loan 1989/1–1989/12c 0.34 0.47 0 1

T-test of mean comparison on the amount borrowed: Diff. in means Std. Dev. t statistic Degr. of freedom

Group lending versus informal lending 940.865 686.6809 1.3702 406Group lending versus bank lending �5,409.36 1,875.675 �2.8840 335Informal lending versus bank lending �6,350.23 2,777.574 �2.2863 149

All data refer to credit obtained in the period 1991/1–1991/12 (one year preceding the survey). In order to compute the lending interval the central time ofthe first round of the survey (1991/12/31) has been considered as a point in time to move one year backward.

a Cumulative amount borrowed by the household.b Credit obtained 1.5 years preceding the survey.c Credit obtained 2 years preceding the survey.

Table 3. Ratio of investment to the amount borrowed

Farmers Non-farmers

t test of meancomparisona

Group lending

Working capital 0.143 0.153 0.2049 (592)Fixed assets after12 months

0.024 0.142 �2.6885b (443)

Informal lending

Working capital 0.523 0.091 2.8943b (220)Fixed assets after12 months

0.178 0.000 2.1193b (178)

Bank lending

Working capital 0.276 0.081 1.9487b (78)Fixed assets after 12 months 0.098 0.000 1.6148 (58)a t statistics are reported. Degrees of freedom in parentheses.b Denotes significant difference in means.


interesting (see figures in italics) is that farmers who are fi-nanced by informal lenders tend to show the highest expendi-

ture ratio in both working capital and fixed assets, while thesame occurs to non-farmers when borrowing from MFIs.However, univariate statistics may not account for importantcomponents such as the fact that borrowers have several differ-ent characteristics like, for example, wealth. More relevantlythey do not account for causality.

Among the other variables that may affect investment deci-sions, own wealth is particularly relevant since it may allowself-financing. It has been recognized that in less developmentcontexts like the one we are analyzing, land seems the mostreliable proxy of wealth. In particular, we consider cultivableland following the de facto approach described in Pitt(1999), although a measure of cultivated land is also accountedfor. For most of the households the latter is different from landowned since many of them are sharecroppers or rent land forfarming. In addition, it is also important to account for landtenure since the latter may affect incentives to increase invest-ment (a typical example is the traditional sharecropping inef-ficiency described in Eswaran & Kotwal, 1985). To thispurpose we build two dummies each one taking the value of1 when the household head is either a sharecropper or rentsland.

Table 4. Descriptive statistics: dependent variables

Variable Mean Std. Dev. Min Max

HH head: age (years) 40.24027 12.42312 16 85HH head: education (years) 2.377642 3.404029 0 16HH spouse: age (years) 29.00111 14.46732 0 70HH spouse: education (years) 1.161846 2.317328 0 14HH head: father lives in HHa 0.023916 0.152828 0 1HH spouse: father lives in HHa 0.001669 0.040825 0 1HH spouse: mother lives in the HHa 0.154060 0.361107 0 1HH spouse: mother lives in the HHa 0.008899 0.093939 0 1HH head is malea 0.943270 0.231390 0 1Religiona (1 = Islam) 0.883760 0.320602 0 1Feminine ratio 0.493146 0.172192 0 1No. of persons in HH 5.226363 2.293305 1 19No. of parents alive 1.838710 1.288886 0 4No. of siblings alive 8.120133 3.839145 0 25No. of other relatives alive 6.739711 5.771005 0 36Father: same activitya 0.756396 0.429375 0 1Total area cultivated (acres) 0.608766 1.138134 0 15Tenure: fixed renta 0.130701 0.337167 0 1Tenure: sharecroppera 0.263626 0.440722 0 1Land (acres) 0.694768 2.296784 0 52.5House value (taka) 1033.122 6192.850 0 160,000Transport value (taka) 29.42158 459.3392 0 15,000Injury last yeara 0.334263 0.471862 0 1Medical expenditures last year (taka) 169.4914 677.1430 0 15,600No. of days not working last year 5.903226 10.11567 0 80Transfers last year: cash value (taka) 168.2864 1485.269 0 33,000Transfers last year: other value (taka) 16.19244 148.2298 0 5,000Transfers last year: food value (taka) 10.49194 105.6856 0 2,000

Observations: 1798.a Dummy variables.


The survey also provides information on household mem-bers, such as the age and education of the household headand spouse, the gender of the household head, his/her religion,his/her father’s activity, the number of family members,femininity ratio, and several other factors which are used ascontrols. Finally, since networking can also affect both invest-ment choices and the form of financing (Wydick, Karp, &Hilliker, 2007), a set of controls aimed at capturing the rela-tionship network of the household is included. These concernthe number of various types of relatives of the household headand spouse who are alive, and in particular those living in thehousehold. All variables are summarized in Table 4, while amore extensive definition is provided in Table 12 in theAppendix.

3. ESTIMATION

One crucial step toward the identification of the impact ofdifferent credit channels on investment is to address the prob-lem of the endogenous nature of the former, which wouldcause estimates to be inconsistent otherwise. Therefore, wefirst investigate the mechanism underlying the selection pro-cess in each type of credit channel in order to find suitableinstruments. Then we test for the effect of the predicted valueof credit on investment.

We estimate both equations of investment in working capi-tal (4) and fixed assets (5), conditional on borrowing fromeach type of lender (Eqns. (1)–(3)) and on a set of control vari-ables, representing credit features, household preferences, andtechnology.

The complete set of reduced form equations is the following:

CMij ¼ a0M þ X ijaM þ ZC

ijbMljM þ eijM ð1ÞCI

ij ¼ a0I þ X ijaI þ ZCijbIljI þ eijI ð2Þ

CBij ¼ a0B þ X ijaB þ ZC

ijbBljB þ eijB ð3ÞEij ¼ a0E þ X ijaE þ ZijbE þ CM

ij cE þ CIijdE

þ CBijkE þ ljE þ eijE ð4Þ

Aij ¼ a0A þ X ijaA þ ZijbA þ CMij cA þ CI

ijdA

þ CBijkA þ ljA þ eijA ð5Þ

where i identifies the household, and j refers to the village.Eij and Aij are investment in working capital and in fixed as-

sets, respectively; CMij , CI

ij, CBij are dummy variables taking the

value of 1 if the household borrows from a MFI, an informallender or a bank, respectively. Credit is treated as binary inthis context. A separate variable reporting the amount ob-tained from whatever type of lender is exploited to isolatethe money-value contribution of loans to investment. Twoother measurable components of credit, such as the interestrate and collateral, are controlled for. CM

ij , CIij, CB

ij are the fittedvalues of CM

ij , CIij, CB

ij, respectively, representing the probabilityof being financed from each type of lender. The associatedparameters, which are our main concern, can be interpretedas the additional investment induced by borrowing from eachcredit source compared to non-borrowing.

X ij is a vector of general characteristics of the householdcommon to all equations, such as religion, age, genderand education of the household head, the feminine ratio, 17


variables seizing on the relationship network of the household,as well as technological features (total area cultivated and landtenure) and proxies of idiosyncratic shocks, such as illness andmedical expenditures. Measures of wealth (land and other as-sets), transfers, and elements capturing the possibility that rel-atives interfere with investment decisions are also included inthe vector X ij for reasons that will be explained further on.

ZCij are characteristics of the household that affect credit

transactions but not other household estimated behaviors(instruments), while Zij includes measurable characteristics ofcredit contracts which may influence the propensity towardinvestment, such as the amount borrowed, interest rates, andcollateral requirements. 18 Note that these variables are notto be classified as instruments since we observe values thatare different from zero only when loans actually take place.If we were to do so, we would observe, for example, that thecorrelation between interest rates and the likelihood of obtain-ing credit is positive, which is counter-intuitive. The sameproblem occurs when using whatever information availableonly for households participating in credit transactions.

a0M , a0I , a0B, a0E, and a0A are constant terms; ljM , ljI , ljB,ljE, and ljA are village specific-effects, while eijM , eijI , eijB, eijE,and eijA are idiosyncratic errors, such as EðeijjX ij; ZC

ij ; ljÞ ¼ 0in Eqns. (1)–(3), and EðeijjX ij;Cij; Zij; Zij; ljÞ ¼ 0 in Eqns. (4)and (5). The following is a covariance matrix of error termsof the reduced form of the model:

r2M rMI rMB rME rMA

rIM r2I rIB rIE rIA

rBM rBI r2B rBE rBA

rEM rEI rEB r2E rEA

rAM rAI rAB rAE r2A

0BBBBBB@

1CCCCCCA

We assume that rME, rMA, rIE, rIA, rBE, and rBA are zero. Bydoing this we allow for both unmeasurable household featuresthat simultaneously affect all credit equations, and unmeasur-able household features that simultaneously affect expendi-ture. There are several reasons for these assumptions.

Regarding credit equations, one is the possibility that thereare individual characteristics of the households which inducesome of them to borrow from multiple sources. Jain andMansuri (2003), for example, explain this practice with the factthat some of them get refinanced by informal lenders in orderto meet a more rigid structure of installments of other out-standing contracts. This frequently occurs in case of severenegative shocks. The presence of these shocks, in case theycannot be controlled for, falls in the error term causing rMI ,rMB, and rIB to be different from zero.

The hypothesis of rEA different from zero, instead, stemsfrom the possibility that there are unmeasurable characteris-tics of the household, such as longer experience in carryingout some activities, which may raise investment in both work-ing capital and fixed assets. We instead exclude there beingunmeasurable characteristics common to credit and invest-ment equations. 19

Finally, we tackle the problem of outliers by using a percen-tile approach. On the one hand, the distribution of workingcapital is—not surprisingly—skewed to the right and trun-cated in zero. Therefore, we gradually trim 20 the right tail,which corresponds to larger activities, since these are morelikely to include outliers. 21 In particular, we carry out esti-mates first on the full distribution (including the zeros), thenon observations below the 90th percentile, and finally onobservations below the 70th percentile. On the other hand,as discussed previously in the paper, fixed assets involve both

positive and negative values, while zeros lay in the center ofthe support. Hence, possible outliers are eliminated by gradu-ally trimming both tails of the distribution. 22 In this case, weperform regressions on the full distribution, and then on 90%and 80% of it, eliminating observations in correspondence ofthe tails. Differences in the portion of observations trimmedfrom the distributions of working capital and fixed assets aredue to the non-equal number of zeros (more frequent for thelatter) along their support.

(a) Sources of bias

As pointed out by Pitt and Khandker (1998), biases thatmay arise when treating program effects—particularly whendealing with cross sectional data—can be summarized intothree major classes. The first originates from nonrandomplacement of credit programs. Treating placement as random,when instead programs are most frequently allocated inpoorer districts, can lead to a downward bias of program ef-fects, as discussed in Pitt, Rosenzweig, and Gibbons (1993)and Heckman (1990). This problem mainly concerns MF,although a similar argument holds for banks, which may notbe uniformly distributed across the sample.

The second class of bias is related to unmeasured villageattributes that affect both credit transactions and householdbehavior. Climate conditions and a high propensity to naturaldisasters, among others, are important characteristics affectingboth these variables, especially when dealing with agriculture.

The last source of bias concerns unmeasured household fea-tures that affect both credit transactions and households’behavior (see Heckman, 1990, for a general discussion onselection bias). These are—typically time-invariant—intrinsiccharacteristics or personal qualities, like ability and individualaptitudes. Problems of this kind are traditionally solved usinginstrumental variables. In particular, the selection system orig-inated by eligibility rules, and other factors which do not de-pend on borrowers’ aptitudes, are often exploited to correctfor selection mechanisms.

(b) Techniques and instruments

Eqns. (1)–(5) are estimated by means of Two-Stage LeastSquares (2SLS) with village Fixed-Effects, 23 where the latterare exploited to correct for both nonrandom allocation ofcredit and unmeasured village characteristics mentionedabove. Instruments are instead used to further correct forselection biases that are induced by unmeasured householdfeatures.

The selection mechanism for MF programs has been thor-oughly investigated in the literature. The ownership of lessthan 0.5 acres of cultivable land, which is the eligibility rulefor MF programs, is normally assumed to be exogenous withrespect to households’ decision making in contexts where theland market is rather static. The absence of an active landmarket is in fact the rationale for the treatment of land own-ership as an exogenous instrument for identifying the impactof credit in almost all the empirical work on household behav-ior in South Asia (see Pitt & Khandker, 1998; Rosenzweig,1980; Binswanger & Rosenzweig,1986; Rosenzweig & Wolpin,1985; see also Pitt, 1999, for a discussion on de jure and de fac-to eligibility rules, i.e., the enforcement of the half-acre rule inpractice, and for addressing the concerns raised by Morduch(1998), on land purchases by program households).

Following the previous literature, we use a dummy variable(Target) that takes the value of 1 if the household owns lessthan 0.5 acres of land as an instrument for group lending,

Table 6. Credit market participation: FIRST STAGE with villagefixed-effects

Grouplending (MFIs)

Informallending

Banklending

Targeta 76.365*** 3.040 �0.894(25.221) (18.164) (10.281)

Spouse: distance from familya 0.223 0.356** 0.161


while controlling for continuous measures of land owned andland cultivated in X ij.

24 We also add other exogenous mea-sures of wealth—ownership of an inherited house and meansof transport—in order to minimize the presence of othercomponents of wealth in the error term. As can be observedin Table 5, target households are more frequently locatedwithin the set of MF borrowers (80% against 60% and 53%of households borrowing from informal lenders and banks,respectively).

Turning to bank lending, the candidate instrument relates toother collateral requirements that do not directly involvehouseholds’ characteristics. By analyzing the answers to thesurvey one can observe that two main types of guaranteesare accepted by banks, physical collateral (mainly own landand, to a lesser extent, valuables) and personal guarantees,typically provided by wealthier persons (others’ land). 25 Sincethe former represents an included instrument, we rely upon thelatter as an exogenous one. In particular, we use the number ofthe household head and spouse’s relatives owning land as aproxy of the likelihood of obtaining co-signed guarantees. Inour data (see again Table 5) bank borrowers have indeed thehighest number of non-close 26 relatives owning land com-pared to other borrowers.

Again, the possibility that the instrument captures house-hold wealth is minimized by including land ownership andother inherited assets as controls. An additional issueoriginates from the fact that relatives owning land may repre-sent important sources of transfers (Pitt & Khandker, 1998),which are likely to affect investment directly inducing violationof the exclusion restrictions. To avoid this possibility we con-trol for transfers in cash, food, and other commodities re-

Table 5. Descriptive statistics: instruments

Variable Mean Std. Dev. Min Max

Borrowers from MFIsa

Target 0.80 0.40 0 1Spouse: distance from family 12.11 33.18 0 400No. of relatives own land 3.10 4.66 0 36Dowry 1,220 3,126 0 37,500Marriage gifts to spouse 195 886 0 10,500

Borrowers from informal sectorb 0Target 0.60 0.46 0 1Spouse: distance from family 21.72 55.58 0 450No.of relatives own land 3.18 4.29 0 21Dowry 2,198 4,823 0 35,500Marriage gifts to spouse 382 1,392 0 20,500

Borrowers from banksc 0Target 0.53 0.51 0 1Spouse: distance from family 21.02 77.41 0 478No. of relatives own land 3.95 2.43 0 15Dowry 1,461 4,093 0 21,428Marriage gifts to spouse 1,671 6,820 0 40,000

Non-borrowersd 0Target 0.70 0.44 0 1Spouse: distance from family 10.44 33.33 0 450No. of relatives own land 3.27 4.60 0 27Dowry 1,353 4,193 0 60,000Marriage gifts to spouse 380 2,184 0 45,000

Note: 46 households borrowing from multiple sources.a Obs. 297.b Obs. 111.c Obs. 40.d Households not borrowing in the twelve months preceding the survey,

Obs. 1,369.

ceived by the household in the last year (see Table 4 fordescriptive statistics). 27

The spatial distance between in-law relatives is used as aninstrument for informal lending. The literature dealing withthe topic of distance and inter-household financial supportin South Asia suggests that there is a positive relationship be-tween the two. Park (2006), for example, finds that credittransactions more frequently occur between relatives livingeither in different clusters or in distant villages, rather than be-tween neighbors. An explanation of this phenomenon can befound in Caldwell et al. (1986), and Rosenzweig (1988), whosuggest that a higher spatial distance between households re-duces the correlation of incomes, mostly in the agriculturalsector where climatic conditions cause a high geographicalvariance of returns. A longer distance should, therefore, en-hance households’ advantage to engage in mutual credit trans-actions for income-smoothing purposes, 28 since the furtherthe households, the lower the likelihood of being hit by thesame natural event. In particular, Rosenzweig and Stark(1989) find that this kind of lending pattern mainly occursamong relatives in-law. 29 Consequently, we use the distanceof the household head’ spouse from her birthplace as a mea-sure of the flow of informal loans accruing to the household.

(0.244) (0.176) (0.099)No. of relatives own landa �1.771 �1.422 3.366*

(4.571) (3.292) (1.864)Dowryb �0.075 0.048** �0.003

(0.237) (0.023) (0.097)Marriage gifts to spouseb �0.316 �0.126 0.589***

(0.411) (0.296) (0.168)HH head: education �0.001 0.005** 0.001

(0.003) (0.002) (0.001)Religion 0.022 0.078*** 0.011

(0.040) (0.028) (0.016)Area cultivated 0.004 0.009 0.008*

(0.012) (0.009) (0.005)Fixed rent 0.047* 0.011 0.025**

(0.027) (0.019) (0.011)Sharecropper �0.038 0.039** �0.007

(0.023) (0.017) (0.010)Partial R2 0.19 0.15 0.17

Joint significance of all parameters in simultaneous estimationequations (1)–(3): v2(119) = 301.16 (P-val = 0.003)Joint significance of instruments in simultaneous estimationequations (1)–(3): v2(15) = 34.12 (P-val = 0.000)Weak identification test: F-statistic 32.78Anderson canon. corr. LR statistic: v2(3) 167.923

Least squares estimates; obs.: 1,798; robust standard errors in parentheses.Other regressors: age of HH head and spouse, spouse’s education, parentsof the HH head and spouse living in the HH, gender of the HH head, Nr.HH members, feminine ratio, Nr. relatives alive, illness, medical expen-diture, non-working days due to illness, land owned, house, transport,cash and non-cash transfers, household head conducts the same activity ofhis/her father.

a Parameters multiplied by 1000.b Parameters multiplied by 10000.* Significant at 10%.

** Significant at 5%.*** Significant at 1%.

Table 7. Investment in working capital: agricultural activities: SECOND STAGE with village fixed-effects

Full distribution 90% of obs.a 70% of obs.b

2SLS 2SLS 2SLS OLS

Group lending (MFIs) �2011.023 �1253.929 �158.891 �19.779(2063.347) (885.358) (131.388) (79.382)

Informal lending 1064.533 1029.210 1009.266 613.294*

(1065.704) (1110.217) (1617.42) (353.058)Bank lending 2391.461* 1414.608 305.359** 172.169**

(1222.182) (1902.079) (138.447) (83.544)Loan amountc 0.010* 0.003 �0.000 �0.000

(0.006) (0.005) (0.001) (0.001)Landd 94.100** 33.059 �8.507 �1.773

(46.785) (26.483) (6.707) (4.863)Transfers: cashe 0.043 0.066** 0.025 0.011

(0.031) (0.028) (0.016) (0.015)Father: same activityf 242.748** 183.248*** 26.313*** 19.086***

(97.160) (61.284) (9.933) (4.620)

R2 0.72 0.73 0.46 0.46Hausman–D–Wg v2(3) 8.19 (P-val = 0.042) 26.83 (P-val = 0.000) 1.15 (P-val = 0.764)Hansen-J v2(2) 1.37 (P-val = 0.503) 0.76 (P-val = 0.681) 1.14 (P-val = 0.589)

Robust standard errors in parentheses.All variables related to those reported at points (c)–(f) are rarely significant. Full estimation report is available upon request. Other regressors: age andeducation of HH head and spouse, religion, gender of the HH head, Nr. HH members, feminine ratio, Nr. relatives alive, illness, medical expenditure, non-working days due to illness, total area cultivated, land tenure.

* Significant at 10%.** Significant at 5%.

*** Significant at 1%. Full distribution: 1,798 obs.a Below 90th percentile, obs.: 1,618.b Below 70th percentile, obs.: 1,258.c Other measurable determinants of credit included: interest rate and collaterald Other measures of wealth included: house and transport ownership.e Non-cash transfers also includedf Parents of the HH head and spouse living in the HH also included.g Computed forcing non-robust standard errors.

Table 8. Investment in working capital: non-agricultural activities: SECOND STAGE with village fixed-effects

Full distribution 90% of obs.a 70% of obs.b

2SLS 2SLS 2SLS OLS

Group lending (MFIs) 15659.929 224.853*** 25.236*** 31.644***

(12136.07) (65.176) (5.879) (5.504)Informal lending 10246.625 -289.50 20.36 76.467

(16773.968) (318.204) (93.32) (53.840)Bank lending 60534.664 166.672 17.098 29.703**

(73082.03) (103.161) (10.678) (12.081)Loan amountc 0.043* 0.000 0.000 0.000

(0.022) (0.000) (0.000) (0.000)Landd 745.588 2.000 0.280*** 0.125*

(553.627) (1.437) (0.105) (0.071)Transfers: cashe 0.390 0.000 0.000* 0.000**

(0.350) (0.001) (0.000) (0.000)Father: same activityf 282.002 29.919*** 3.662*** 4.643***

(362.352) (5.173) (0.471) (0.356)R2 0.18 0.18 0.28 0.29Hausman–D–Wg v2(3) 12.85 (P-val = 0.004) 6.43 (P-val = 0.092) 12.55 (P-val = 0.005)Hansen-J v2(2) 0.49 (P-val = 0.782) 1.22 (P-val = 0.541) 0.07 (P-val = 0.961)

Robust standard errors in parentheses.All variables related to those reported at points (c)–(f) are rarely significant. Full estimation report is available upon request. Other regressors: age andeducation of HH head and spouse, religion, gender of the HH head, Nr. HH members, feminine ratio, Nr. relatives alive, illness, medical expenditure, non-working days due to illness, total area cultivated, land tenure.


*** Significant at 1%. Full distribution: 1,798 obs.a Below 90th percentile, obs.: 1,618.b Below 70th percentile, obs.: 1,258.c Other measurable determinants of credit included: interest rate and collateral.d Other measures of wealth included: house and transport ownership.e Non-cash transfers also included.f Parents of the HH head and spouse living in the HH also included.g Computed forcing non-robust standard errors.



Statistics reported in Table 5 provide evidence of a positiverelationship between the distance of spouses from theirfamilies and participation to the credit market. Correlation be-tween the two variables is higher for both households borrow-ing from informal sources and banks, compared to thoseparticipating to MF programs or not borrowing. The vari-ance, though, is lower for informal transactions, which shouldbe indicative of the fact that the instrument is more suitablefor this type of credit.

It might be possible, however, that distance between house-holds was part of a marital agreement serving to mitigate in-come risk (Rosenzweig & Stark, 1989). Typically in ruralcontexts such agreements are carried out by parents of groomsand brides. Therefore, problems may arise if, at the time of thesurvey, these lived sufficiently close to the household as to takepart to investment decisions. In this case, in fact, parents’unmeasurable characteristics may interact with both credittransactions—through marriage arrangements—and invest-ment choices. In order to mitigate this effect we include a setof dummy variables controlling for both the presence of par-ents in the household and the household head’s engagementin his/her father’s activity (Table 4). 30

Dowries are also contemplated among instruments for bothinformal and bank lending. In general, a dowry is intended asthe money, goods, or estate that a woman brings to her hus-band in marriage. To the purposes of this paper, however, itis useful considering also other forms of payments that tookplace at the time of marriage of the household head andspouse. The same culture, for example, may simultaneouslypractice dowries and other transfers, such as gifts to thebride. 31 Typically, dowries and other gifts are not decided

Table 9. Investment in fixed assets after twelve months: agricul

Full distribution 90%

2SLS

Group lending (MFIs) �2032.27 �(1846.737) (1

Informal lending 6783.825** 4(3173.856) (2

Bank lending �14314.559 2(23610.896) (3

Loan amountc 0.003(0.010)

Landd �28.707 �(39.401) (

Transfers: cashe �0.063(0.047)

Father: same activityf 356.047** 2(174.058) (1

R2 0.10Hausman–D–Wg v2(3) 0.40 (P-val = 0.940) 8.92 (PHansen-J v2(2) 1.46 (P-val=0.4817 0.29 (P

Robust standard errors in parentheses.All variables related to those reported at points (c)–(f) are rarely significant. Feducation of HH head and spouse, religion, gender of the HH head, Nr. HH meworking days due to illness, total area cultivated, land tenure.


*** Significant at 1%. Full distribution: 1,798 obs.a Ten per cent of the tails dropped, obs.: 1,618.b Twenty per cent of the tails dropped, obs.: 1,438.c Other measurable determinants of credit included: interest rate and collaterd Other measures of wealth included: house and transport ownership.e Non-cash transfers also included.f Parents of the HH head and spouse living in the HH also included.g Computed forcing non-robust standard errors.

by grooms and brides themselves, but are instead settled bytheir families. It is a matter of culture that these are passeddown the family line as something personal that others cannoteasily claim (Anderson, 2007) and it is very infrequent thatthey are sold to buy productive inputs. This provides an argu-ment in favor of the exogeneity of such variables since their ef-fect on investment is not likely to be direct but may insteadpass through credit.

In fact, as Anderson points out, both dowries and marriagegifts to the spouse are often pledged as collateral against loansfrom moneylenders (see also Aleem, 1990; Bhattacharyya,2005). It is also frequent that women are financed by formalinstitutions, normally cooperative banks—which are part ofthe banking sector in our sample—acting as pawnbrokersand accepting valuables as collateral. 32 From statistics in Ta-ble 5 it emerges that more generous dowries are associated to ahigher probability of accessing informal loans, while marriagegifts received by the spouse are considerably higher for bankborrowers. As in the case of the other instruments, marriagetransfers can be considered exogenous to investment, providedthat suitable controls are included to avoid violation of exclu-sion restrictions. In particular, we refer to measures of wealth,transfers, and other variables capturing relatives’ interferencewith household investment decisions, as discussed above. Allinstruments and controls mentioned in this section are listedin Table 12 in the Appendix.

In conclusion, it is worth stressing that despite the inclusionof several controls, there is sill possibility that the instrumentsmay have some direct impact on investment. For this reason,at the end of the following section we provide robustnesschecking through regressions of investment on instruments

tural activities: SECOND STAGE with village fixed-effects

of obs.a 80% of obs.b

2SLS 2SLS OLS

1804.435 �1776.115 �1293.399634.776) (1977.570) (1139.227)730.050* 3473.591* 1489.511*

808.193) (1884.595) (887.498)597.857 1595.794 2044.014192.838) (2373.350) (2971.352)0.002 0.002 �0.023

(0.009) (0.005) (0.043)22.656 �36.623* �32.227**

34.874) (20.986) (16.184)�0.055 �0.028 �0.012(0.041) (0.026) (0.009)81.847* 300.609*** 149.093***

53.987) (90.656) (57.774)

0.12 0.13 0.13-val = 0.030) 13.00 (P-val = 0.004)-val = 0.862) 0.26 (P-val = 0.874)

ull estimation report is available upon request. Other regressors: age andmbers, feminine ratio, Nr. relatives alive, illness, medical expenditure, non-

al.


and controls. In any case, however, since instrumentation isstill open to question, a certain degree of caution is used inthe following section while interpreting the results.

4. RESULTS

In this section we present estimates of Eqns. (1)–(5). Resultsof the first stage Eqns. (1)–(3) are reported in Table 6. Tables7–10, instead, refer to the second stage Eqn. (4) and (5) forinvestment in working capital and in fixed assets, respectively.OLS parameters are also reported for the main regressions forcomparison.

(a) Credit market participation

Results reported in Table 6 show that instruments are signif-icant in determining credit access. As expected, participationin MF programs considerably increases with the eligibility sta-tus measured by the ownership of less than half an acre ofland, a result that is in line with the previous empirical litera-ture. Next, having more landed relatives gives a higher proba-bility of accessing bank lending. Hence it seems that when aproject is planned, households are more likely to resort toMFIs or banks in order to obtain funds, and they enter thesemarkets based on land status and land collateral availability.Instead, the non-significance of both the eligibility thresholdand the number of relatives owning land in the informal mar-ket equation indicates that this type of credit behaves differ-ently from the other two.

Table 10. Investment in fixed assets after twelve months: non-agri

Full distribution 90%

2SLS 2

Group lending (MFIs) 24507.309* 58(14153.903) (46

Informal lending �17027.124 33(33295.945) (22

Bank lending �13735.815 �25(16388.338) (156

Loan amountc �0.251 �(0.240) (0

Landd 353.584 15(320.630) (27

Transfers: cashe �0.168 �(0.581) (0

Father: same activityf �4241.267*** �308(1012.939) (71

R2 0.18Hausman–D–Wg v2(3) 9.41 (P-val = 0.024) 8.60 (P-Hansen-J v2(2) 0.12 (P-val = 0.937) 0.05 (P-

Robust standard errors in parentheses.All variables related to those reported at points (c)–(f) are rarely significant. Feducation of HH head and spouse, religion, gender of the HH head, Nr. HH meworking days due to illness, total area cultivated, land tenure.


*** Significant at 1%. Full distribution: 1,798 obs.a Ten per cent of the tails dropped, obs.: 1,618.b Twenty per cent of the tails dropped, obs.: 1,438.c Other measurable determinants of credit included: interest rate and collaterd Other measures of wealth included: house and transport ownership.e Non-cash transfers also included.f Parents of the HH head and spouse living in the HH also included.g Computed forcing non-robust standard errors.

In particular, as previously found in the literature (seeabove), the parameter measuring the effect of distance fromthe spouse’s original birthplace is positive and significant inthe equation of informal lending, providing support to the factthat in-law relatives located far from the household are likelyto extend a large portion of credit and smooth income andinvestment. Dowries also enter the equation of informal lend-ing with a positive sign. Conditional on other measures ofhousehold’s wealth, this possibly suggests that marriage valu-ables are used as collateral in informal credit transactions.Furthermore, the significance of marriage gifts to the spousemay also be indicative of the fact that women pledge theirown assets in order to obtain cooperative credit.

Econometric tests may provide some idea as to whetherthese instruments are relevant. First, Hausman–Durbin–Wustatistics reported in Tables 7–10 often indicate that there isendogeneity in the relationships we are estimating conditionalon the controls. We choose to use 2SLS estimation techniquesalso when tests reject the hypothesis that OLS and 2SLS donot provide substantially different output. This is the safest ap-proach one could adopt, since treating a behavior as endoge-nous when it is in fact exogenous still yields consistent,although less efficient, estimates (Pitt, 1999). The Hansen-Jtest for overidentifying restrictions virtually every time sug-gests that the instruments are valid. The Kleibergen–Paap F-statistic reported in Table 6 for the first stage estimates alsoindicates that instruments are likely to be non-weak.

In conclusion, land related variables, including tenure (fixedrent), and the village fixed effects are the most significantfactors explaining participation to group lending and bank

cultural activities: SECOND STAGE with village fixed-effects

of obs.a 80% of obs.b

SLS 2SLS OLS

00.231 3513.601* 4846.695**

89.378) (2130.521) (2088.774)99.026 3654.011 �1897.05201.167) (3555.580) (2010.292)23.209 �2795.600 �3913.30490.349) (3832.159) (4035.651)0.015 �0.013 �0.055.012) (0.011) (0.038)9.536 86.242 58.6837.821) (55.935) (39.785)0.007 �0.015 �0.006.035) (0.011) (0.010)0.130*** �1389.622*** �1148.477***

1.609) (193.635) (111.843)

0.19 0.20 0.20val = 0.035) 15.64 (P-val = 0.001)val = 0.973) 0.20 (P-val = 0.900)

ull estimation report is available upon request. Other regressors: age andmbers, feminine ratio, Nr. relatives alive, illness, medical expenditure, non-

al.


lending. Education, religion, and the sharecropper status ofthe household head (on the latter see Braverman & Stiglitz,1982) seem instead to be good drivers of the access to informallending, as it turns out from parameters reported in Table 6.

(b) Investment

Estimates of the impact of the three different types of crediton investment in working capital are reported in Tables 7(agricultural activities) and 8 (non-agricultural activities), 33

whereas estimates of the impact of credit on investment infixed assets are reported in Tables 9 (agricultural activities)and 10 (non-agricultural activities).

Moving from the left to the right columns of Tables 7 and 8we trim higher levels of expenditure corresponding to the lefttail of the working capital distribution. According to thediscussion in the previous section, this should manage theproblem of outliers. Moreover, dropping the left tail of the dis-tribution also allows isolating the effect of credit on the small-est activities—presumably associated to the pooresthouseholds—in the last column on the right. Similarly, movingfrom the left to the right columns of Tables 9 and 10 impliesconcentrating on observations involving less volatile invest-ment activities located in the center of the distribution of fixedassets, that is, in the area around zero. This may also isolate

Table 11. Significance of instrumen

Investment in working capit

Agricultural activities Non-agr

Without controlsa

Target �172.003 71(138.708) (92

Spouse: distance from family 1.250 0(0.998) (1

No. of relatives own land 9.392 5(9.065) (3

Dowry 0.021* 0(0.012) (0

Marriage gifts to spouse 0.003 �0.004) 0

With controlsb

Target �76.409 1,1(127.120) (1,0

Spouse: distance from family 1.091 0(1.002) (1

No. of relatives own land 10.378 5(8.826) (3

Dowry 0.013 0(0.009) (0

Marriage gifts to spouse 0.007 �(0.014) (0

Land 138.116*** 64(48.565) (48

Transfers: cash 0.058** 0(0.027) (0

Father: same activity 122.369** 20(50.554) (20

Observations: 1798; robust standard errors in parentheses; fixed-effects regress* Significant at 10%.

** Significant at 5%.*** Significant at 1%.

a All the variables used in second stage (Tables 7–10) except credit, land, househave been included in the regressions.

b All the variables used in second stage (Tables 7–10) except credit have bee

poorer households, since their asset buying/dismissal activityis likely to be smoother as compared to that of richer families.

For each regression we report parameters associated to themain variables. 34 In particular, we display results concerningpredicted participation to each type of lending transaction,which should account for non-measurable determinants ofcontracts, as discussed above. We also report the parametersassociated to the amount borrowed and to the main variableswhich are critical to the instrumentation argument. 35 Remain-ing controls are listed in the bottom of each table.

As for working capital, we find that farmers’ investment ispositively associated to bank credit (Table 7), while MF islikely to be more beneficial for non-farmers (Table 8). In par-ticular, estimated parameters for bank lending are significantfor farmers who reach either very high levels of investmentor relatively low ones. We interpret the former evidence asbeing a collateral/wealth effect, namely larger activities aremanaged by richer households who are also endowed withmore valuable collateral, while the latter possibly reveals thepractice of pledging valuables in cooperative banks, as exten-sively discussed above. Parameters associated to MF becomeinstead more significant for observations corresponding tomid-low non-agricultural expenditure levels, that is, afterdropping 10% of the distribution corresponding to the largestactivities.

ts in the second stage equations

al Investment in fixed assets

icultural act. Agricultural activities Non-agricultural act.

4.126 �37.315 2,304.669*

9.490) (175.668) (1,204.774).256 1.444 �3.823.241) (1.243) (8.001).750 �7.153 -113.7396.150) (21.783) (98.975).014 0.019 �0.018.049) (0.015) (0.066)0.045 �0.015 0.042.043) 0.023) 0.105)

87.496 41.614 1,765.15819.296) (173.201) (1,294.166).214 1.130 �4.576.097) (1.251) (8.003).215 �9.025 �116.8196.930) (21.755) (97.971).014 0.018 �0.025.037) (0.015) (0.060)0.028 �0.015 0.013.028) (0.023) (0.100)3.806 �25.603 199.9212.257) (33.233) (239.393).227 �0.049 �0.109.206) (0.035) (0.379)5.704 193.435** �2,648.067***

7.620) (93.468) (415.189)

ions.

, transport, parents of HH head and spouse living in the HH, and transfers

n included in the regressions.


Fixed assets almost behave like operating costs. In this case,informal lending rather than bank lending is associated to ahigher investment in farming activities (Table 9), while MFstill performs better in the non-farming sector (Table 10). Itis interesting to note that within each sector the significantparameters associated to fixed assets are larger than those ofworking capital, particularly with regard to MF. Furthermore,in the case of fixed assets there is no clear pattern that links amore intensive use of credit with the dimension of borrowers’activity.

Measurable components of credit, such as the amountborrowed, interest rates, and collateral requirements, arerarely significant. The amount seems to matter for very largeexpenditure only, whereas there is no clear-cut pattern forthe incidence of the other credit conditions. Control variablesthat have been exploited to curb the possible violation ofexclusion restrictions have the expected signs and are often sig-nificant. Among other variables associated with investment—which are not reported in the tables—there is evidence of astrong relationship between agricultural expenditure and otherfeatures connected to farming, such as cultivated land andland tenure. Finally, the traditional lack of incentives in share-cropping seems evident in agriculture, but only for largeractivities, otherwise sharecroppers behave well.

(c) Robustness to exclusion restrictions

In order to check the robustness of instruments to exclusionrestrictions we perform regressions without—previouslyinstrumented—credit variables included in the specificationbut with instruments substituted in their place. In the upperpart of Table 11, parameters refer to the specification notincluding controls for household wealth, transfers of varioustypes, and variables connected to the possibility that the headof the household and spouse’s parents may interfere withinvestment activities. The lower part of the table shows param-eters of the regressions including these controls (only the mostsignificant ones are reported).

Without the presence of controls, the instruments which inprinciple appear to suffer more from violation of exclusionrestrictions seem to be both the eligibility rule for MF pro-grams and dowries. The former may have several explana-tions. In principle, one should expect either a positive ornegative sign of the parameter. For example, target house-holds may invest less because they are poorer or the landlessare more likely to invest in activities other than agriculture.Our interpretation of the results is that the latter effect prevailsover the former. As for dowries, we interpret their significantpositive relationship with investment as a wealth effect, sincehigher dowries may correspond to both richer householdsand richer extended families providing more transfers to thehousehold.

However, including all the controls, and particularly all themeasures of assets and transfers—lower part of the table—theparameters associated with the eligibility rule and dowries be-come not significant, suggesting that part of the significanceseen above is likely to pass through the credit channel.

5. CONCLUSIONS

A considerable number of works dealing with the impact ofMF programs have been written so far. Many assess the suc-cess of these programs on several household behaviors such asincreasing consumption, labor supply, and children schoolenrollment. This paper, in particular, concentrates on invest-

ment in productive activities since this is likely to representan important way toward capital accumulation and growthfor households in less developed countries.

Using data from a World Bank survey documenting loansfrom different types of lenders in Bangladesh, we investigatewhich kind of lending—micro-lending, informal lending, andbank lending—is better suited to promote investment in eitherfarming or non-farming activities.

Distinguishing between two sectors and comparing threetypes of credit is thus important for the purpose of verifyingwhether MF contracts are equally suitable for all sectors ofthe economy. In fact, standardized lending agreements, whichare typical of MF programs, have been recently criticized asnot being suited to match the needs of the agricultural produc-tion. In particular, borrowers seem forced into a situationwhere they have to produce rapidly in order to meet a matu-rity date that has perhaps been fixed too close to the date whenthe loan was granted. This, besides being an empiricallycontroversial way to foster financial education (see, e.g., Field& Pande, 2008; McIntosh, 2008) may push farmers, whoseproduction cycle is longer than in other activities, towardmore flexible—but sometimes more expensive—credit chan-nels, such as the informal one.

Results from the empirical analysis show that there is nosuperior credit agreement in terms of efficiency, since a posi-tive relationship between all types of credit and investment isobserved. However, there are interesting differences whichare likely to support the hypothesis of a low suitability ofMF programs for some sectors. In particular, we find thathouseholds whose members belong to a group lending pro-gram invest more in non-agricultural inputs compared bothto households who are borrowing from other sources andnon-borrowers. The situation is reversed in agriculture, whereit emerges that only households borrowing from informalsources and banks invest more.

A key point of the paper is that the emerging evidence isnot ascribable to measurable features of contracts such asthe amount borrowed, interest rates, and collateral require-ments. This makes it more likely that the explanation ofour results rests in the repayment system required by differ-ent lenders. The observed pattern, however, could also bethe outcome of the MFIs’ bias in favor of non-farming activ-ities since the risk of default related to climate conditions islower and the production cycle is considerably shorter thanin agriculture, so that repayments are made effortlessly ona regular basis. The eligibility criterion based on land owner-ship may represent another factor excluding a number offarmers—who are more likely to be endowed with land—from the MF network. Also in this case, however, it is pos-sible to interpret things in the opposite way, since it may bethat collateral availability allows farmers to access creditchannels different from MF.

Other features, such as “joint” compared to “individual” lia-bility systems, “peer” versus “lender” monitoring technologies,and sanctions, can be considered some of the factors that ac-count for the evidence we observe. Non-credit services(McKernan, 2002), such as financial education and health-programs associated to credit, can also help stimulating dis-tressed households’ willingness to care about the future, andby this, foster investment. However, we do not strictly relyon these elements to justify our results since, as discussed inSection 1, there is apparently no specific reason why theseshould favor non-agricultural activities more than farming.

Finally, while tackling the issues of endogeneity betweencredit variables and investment, we obtain interesting infor-mation on credit market selection mechanisms. We find that


the principle of targeting landless households seems enforcedby MFIs, while the access to bank lending depends on col-lateral—land and other valuables—endowment. Besides thepossible use of dowries as collateral in some type of infor-

mal credit transactions, the informal channel seems workingfor households linked by long-distance in-law relationshipswho benefit from lending as a means to reduce incomecovariance.

NOTES

1. There is evidence that inviting farmers to plant hybrids, for instance, isanother measure adopted by some MFIs in this regard.

2. Madajewicz (2003a and b) and McKernan (2002), while analyzing theimpact of MF on business profits, have also been dealing with investmentand credit, but they either ignored expenditure in fixed assets ordisregarded other credit sources.

3. It is no surprise that we find this recommendation in the GrameenBank program, which states: “We shall grow vegetables all the year round.We shall eat plenty of them and sell the surplus.”

4. In the absence of this evidence, instead, we would not be in theposition to infer that MF is not suitable for these borrowers. For example,they may not need loans because of higher self-financing capabilities.

5. The Bangladesh Rural Development Board operates by organizingsmall and marginal farmers, asset-less men and women in cooperativesocieties and informal groups (http://www.brdb.gov.bd/gen-eral_Info.htm).

6. Although the problem of missing data is particularly severe, regres-sions have also been performed on the three-rounds panel using householdfixed-effects. In terms of significance of the parameters results do not bringadditional information compared to the cross-sectional setup.

7. A question on the difference between the ownership of fixed assets isactually present in the survey, but these data cannot be used due to theconsiderable amount of missing information. Moreover, this measure isavailable for farmers only, and over a period that is not clearly specified.

8. The period goes from January 1991 to December 1991.

9. Due to the fact that the production cycle is likely to be shorter thanone year in non-agricultural activities, we observe that non-farmersfrequently buy raw materials during a one-year period. Hence, we consideroperating costs for one average cycle only. The assumption lies in the factthat if credit is available, it is used to immediately raise input demand andnot saved for further cycles, since proceeds generated by the loan are morelikely to be used for that purpose (i.e., rolled over subsequent cycles). Thiseliminates overlapping and possible overestimates of variable inputexpenditure.

10. Note that, in any case, if we were aggregating investment theestimated parameters of the aggregate equations would result in the sumof the parameters separately estimated in two different equations, oneworking capital and one for fixed assets. We omit this redundant step.Results are available on request.

11. Data not enable a clear separation between assets dismissal andmissing answers (some households declared a positive value of assets in thefirst round of the survey and then did not answer this question in thefollowing rounds). This problem is partially managed through thepercentile analysis carried out in the empirical section.

12. Anti-usury ceilings were 16% at the time of the survey. A handful ofreports of slightly higher rates in MF may include fees.

13. Such high interest rates (up to 240% in our sample) may either reflectmoneylender’s usurious behavior (see, e.g., Basu, 1984; Rahman, 1979;Blitz & Long, 1965, and Bhaduri, 1977) or be a result of high monitoringcosts as argued by Aleem (1990). See also Timberg and Aiyar (1984) for adiscussion.

14. The actual duration (including delays) cannot be computed sincemost of the loans we consider are still outstanding at the time thehousehold is surveyed.

15. In Table 1, average investment in fixed assets apparently looks lowerthan working capital expenditure. This is due to the fact that asubstantially low number of self-employed households invest in fixedassets, while all of them make use of working capital, conditional on thefact of belonging to a specific sector. By looking at individual householdsrecording a value of investment in fixed assets which is greater than zerothe average expenditure is considerably higher.

16. Note that we cannot use the ratio of investment to the amount ofborrowed funds as a dependent variable since this measure is availableonly for entrepreneurs that are actually borrowing. However, truncatingthe sample at positive values of borrowing would cause a selection bias(Heckman, 1976).

17. This variable cannot represent an instrument for MF since besidesaffecting access to MF, females may also have a different attitude towardrisk and investment. There is, however, no consensus in the literature onthis point (see, e.g., Schubert, Brown, Gysler, & Brachinger, 1999, for adiscussion).

18. For the specification of the system (1)–(5) see Park (1974) andKhazzoom (1976).

19. These assumptions are also a consequence of testing the hypothesis ofno correlation between the residuals of the system (1)–(5). Tests suggestthat the error terms are correlated between credit equations, andparticularly between group lending and bank lending. The correlationbetween credit and investment is instead negligible, while it is weakbetween Eqns. (4) and (5).

20. On Least Trimmed Squares techniques see, for example, Rousseeuw(1997) and Bassett (1991). For an overview of the properties of LeastTrimmed Squares in linear and nonlinear regression see Cızek and Vısek(2000), and Stromberg (1993).

21. Zeros cannot be trimmed all at once without incurring a selection-bias (Heckman, 1976).

22. Tails are dropped in such a way to keep a stable ratio of non-zeroobservations in each tail.

23. Due to the possible bias embedded in the first stage LinearProbability model (see, e.g., Greene, 2000), we also performed the firststage using a Multinomial Logit model. The predicted probabilities ofborrowing from each type of lender have then been plugged in the second-stage Eqns. (4) and (5). Estimates do not provide substantial changes ascompared to 2SLS. In particular, correcting for negative predicted values

http://www.brdb.gov.bd/general_Info.htm

http://www.brdb.gov.bd/general_Info.htm


of the probability of accessing each credit market in the 2SLS estimationprocedure returns second-stage parameters which are closer to thoseobtained by the use of the Multinomial Logit.

24. According to the discussion in the previous section it is plausible thatcontrolling for land removes a considerable part of wealth from the errorterm. This should enhance the argument in favor of the validity of theinstrument.

25. Besides agricultural land (required in 29% of bank loans), we findthat respondents declare that banks also accept guarantees from the landregistration book, such as barga certificates (see Bhattacharyya, 2005, formore details) or other property documents (35% of cases). Other buildingsare required as collateral in only 4% of the cases.

26. Note that we do not account for close relatives—parents, siblings,sons, and daughters living outside the household—since these are morelikely to provide direct contribution to household investment and,therefore, violate exclusion restrictions.

27. Past transfers may not account for the incentive effect stemmingfrom being assured future assistance from wealthy relatives, so that thelatter remains embedded in the instrument. Hence, the problem ofviolating the exclusion requirement may not be completely ruled outunless there is positive correlation between past and potential transfers.This, however, seems a reasonable hypothesis since a household relies onthe amount of aid received in the past in order to calculate theprobability of a relative’s helping in the future. Therefore, our choice ofcontrolling for past transfers appears sufficient to preserve the quality ofthe instrument.

28. In particular, Park finds that credit is mainly used for investmentsmoothing purposes—especially to buy cattle, which is the main bufferagainst hardships—while the transfers in nature occurring among neigh-bors are instead used for consumption.

29. This large flow of lending transactions seems consistent with statisticsreporting that approximately a very high portion of rural-to-ruralmigration in East Asia is represented by women who move for marriagereasons. For example, according to the 2001 Indian Population Census,156 millions females against 2 millions males move from their birthplacefor marriage reasons.

30. Furthermore, controlling for land ownership is again crucial sinceland has been recognized as being one the major determinants of marriageagreements. This is driven by an assortative matching argument, sincelanded (richer) families tend to mutually insure more frequently (Rosen-zweig & Stark, 1989).

31. Bride prices and Groom prices are other forms of transfers which wedo not consider since they accrue to brides’ and grooms’ families. Formore details see Anderson (2007).

32. A detailed description of this practice is provided, for example, byBouman and Houtman (1988) who document the experience of thePeople’s Bank in India.

33. Note that if we were aggregating investment, the estimated param-eters of the aggregate equations would result in the sum of the parametersseparately estimated in two different equations, one for farmers and onefor non-farmers. We omit this redundant step. Results are available onrequest.

34. Full results available on request.

35. In particular, we report the coefficient of land as a representativemeasure of wealth, while we omit house and transport ownership sincetheir coefficients are rarely significant. We do the same for transfers incash, neglecting transfers in food and other goods. For the same reasons,only the presence of the household head father in the house is displayedwhile leaving out his/her mother and the spouse’s parents.

REFERENCES

Ahlin, C., & Jiang, N. (2008). Can micro-credit bring development?Journal of Development Economics, 86(1), 1–21.

Aleem, I. (1990). Imperfect information, screening and the costs ofinformal lending: A study of a rural credit market in Pakistan. WorldBank Economic Review, 4(3), 329–349.

Anderson, S. (2007). The economics of dowry and brideprice. Journal ofEconomic Perspectives, 21(4), 151–174.

Armendariz, B. (1999). On the design of a credit agreement with peermonitoring. Journal of Development Economics, 60(1), 79–104.

Armendariz, B., & Morduch, J. (2005). The economics of microfinance.Cambridge, MA: MIT Press.

Bage, L. (2004). Savings: the forgotten half of microfinance,Statement delivered by the president of the International Fund forAgricultural Development. <http://www.ifad.org/events/microcredit/op.htm>.

Banerjee, A. V., Besley, T., & Guinnane, T. (1994). Thy neighbor’s keeper:The design of a credit cooperative with theory and a test. The QuarterlyJournal of Economics, 109(2), 491–515.

Barua, D., & Dowla, A. (2006). The poor always pay back: The Grameen IIstory. Bloomfield, CT: Kumarian Press Inc..

Bassett, G. W. (1991). Equivariant, monotonic, 50% breakdown estima-tors. American Statistician, 45(2), 135–137.

Bhattacharyya, S. (2005). Interest rates, collateral and (de-)interlinkage: Amicro-study of rural credit in West Bengal. Cambridge Journal ofEconomics, 29(3), 439–462.

Basu, A., Blavy R., & Yulek M. (2004). Microfinance in Africa: Experienceand lessons from selected African Countries. IMF Working Paper 04/174.

Basu, K. (1984). Implicit interest rates, usury and isolation in backwardagriculture. Cambridge Journal of Economics, 8(2), 145–159.

Bouman, F. J. A., & Houtman, R. (1988). Pawnbroking as an instrumentof rural banking in the third world. Economic Development andCultural Change, 37(1), 69–89.

Besley, T., & Coate, S. (1995). Group lending, repayment incentives andsocial collateral. Journal of Development Economics, 46(1), 1–18.

Bhaduri, A. (1977). On the formation of usurious interest rates inbackward agriculture. Cambridge Journal of Economics, 1(4), 341–352.

Binswanger, H. P., & Rosenzweig, M. R. (1986). Behavioural and materialdeterminants of production relations in agriculture. Journal of Devel-opment Studies, 22(2), 503–539.

Blitz, R. C., & Long, M. F. (1965). The economics of usury regulation.Journal of Political Economy, 73(6), 608–619.

Brandsma, J., & Chaouali R. (1998). Making microfinance work in theMiddle East and North Africa. World Bank Discussion Paper,Washington DC.

Braverman, A., & Stiglitz, J. E. (1982). Sharecropping and theinterlinking of agrarian markets. American Economic Review, 72(4),695–715.

Caldwell, J. C., Reddy, P. H., & Caldwell, P. (1986). Periodic high risk as acause of fertility decline in a changing rural environment: Survivalstrategies in the 1980–1983 south Indian drought. Economic Develop-ment and Cultural Change, 34(4), 677–701.

Christen, R. P., & Pearce, D. (2005). Emerging lessons in agriculturalmicrofinance: Selected case studies. CGAP Occasional Paper, 11,Washington, D.C.


Cızek, P., & Vısek, J. A. (2000). Least trimmed squares. In W. Hardle, Z.Hlavka, & S. Klinke (Eds.). XploRe application guide (pp. 49–64).Heidelberg: Springer.

Eswaran, M., & Kotwal, A. (1985). A theory of contractual structure inagriculture. American Economic Review, 75(3), 352–367.

Field, E., & Pande, R. (2008). Repayment frequency and default in micro-finance: Evidence from India. Journal of European Economic Associ-ation Paper and Proceedings, 6(2–3), 501–509.

Ghatak, M., & Guinanne, T. N. (1999). The economics of lending withjoint liability: Theory and practice. Journal of Development Economics,60, 195–228.

Greene, W. (2000). Econometric analysis (4th ed.). New York: PrenticeHall.

Heckman, J. (1976). The common structure of statistical models oftruncation, sample selection, and limited dependent variables and asimple estimator for such models. Annals of Economic and SocialMeasurement, 5(4), 120–137.

Heckman, J. (1990). Varieties of selection bias. American EconomicReview, 80(2), 313–318.

Hermes, N., Lensink, R., & Mehrteab, H. T. (2005). Peer monitoring,social ties and moral hazard in group lending programs: Evidence fromEritrea. World Development, 33(1), 149–169.

Hermes, N., & Lensink, R. (2007). The empirics of microfinance: What dowe know?. The Economic Journal, 117(1), F1–F10.

Jain, S., & Mansuri, G. (2003). A little at a time: The use of regularlyscheduled repayments in microfinance programs. Journal of Develop-ment Economics, 72(1), 253–279.

Khazzoom, J. D. (1976). An indirect least squares estimator foroveridentified equations. Econometrica, 44(4), 741–750.

Lawrance, E. C. (1991). Poverty and the rate of time preference: Evidencefrom panel data. Journal of Political Economy, 99(1), 54–77.

Llanto, G. M. (2007). Overcoming obstacles to agricultural microfinance:Looking at broader issues. Asian Journal of Agriculture and Develop-ment, 4(2), 23–40.

Madajewicz, M. (2003a). Does the credit contract matter? The impact oflending programs on poverty in Bangladesh. Working Paper. ColumbiaUniversity, New York.

Madajewicz, M. (2003b). Capital for the poor: The effect of wealth on theoptimal credit contract. Working Paper. Columbia University, NewYork.

McIntosh, C. (2008). Estimating treatment effects from spatial policyexperiments: An application to Ugandan microfinance. Review ofEconomics and Statistics, 90(1), 15–28.

McKernan, S. M. (2002). The impact of microcredit programs on self-employment profits: Do noncredit program aspects matter?. TheReview of Economics and Statistics, 84(1), 93–115.

Meyer, R. L. (2002). The demand for flexible microfinance products.Lessons from Bangladesh. Journal of International Development, 13(1–18), 351–368.

Morduch, J. (1998). Does microfinance really help the poor? Evidence fromFlagship Programs in Bangladesh. Harvard University, Cambridge,MA.

Murray, I. (2001). Cultivating client loyalty. MicroBanking Bulletin, 6(16),20–24.

Park, C. (2006). Risk pooling between households and risk-copingmeasures in developing countries: Evidence from rural Bangladesh.Economic Development and Cultural Change, 54(2), 423–457.

Park, S. (1974). On indirect least squares estimation of a simultaneousequation system. Canadian Journal of Statistics, 2(1), 75–82.

Paxton, J., Graham, D., & Thraen, C. (2000). Modeling group loanrepayment behavior: New insights from Burkina Faso. EconomicDevelopment and Cultural Change, 48(3), 639–655.

Pitt, M. M., Rosenzweig, M. R., & Gibbons, D. M. (1993). Thedeterminants and consequences of the placement of governmentprograms in Indonesia. World Bank Economic Review, 7(3), 319–348.

Pitt, M. M., & Khandker, S. (1998). The impact of group-based creditprograms on poor households in Bangladesh: Does the gender ofparticipants matter? Journal of Political Economy, 106(5), 958–996.

Pitt, M. M. (1999). Reply to Jonatan Murdoch’s “ Does microfinancereally help the poor? Evidence from Flagship Programs in Bangladesh”.Brown University, Providence, RI.

Rahman, A. (1979). Usury capital and credit relations in Bangladeshagriculture. Bangladesh Development Studies, 12(2), 1–46.

Rosenzweig, M. R. (1980). Neoclassical theory and the optimizingpeasant: an econometric analysis of market family labor supply in adeveloping country. Quarterly Journal of Economics, 94(1), 31–55.

Rosenzweig, M. R., & Stark, O. (1989). Consumption smoothing,migration, and marriage: Evidence from rural India. The Journal ofPolitical Economy, 97(4), 905–926.

Rosenzweig, M. R. (1988). Risk, implicit contracts and the family inrural areas of low-income countries. Economic Journal, 98(393),1148–1170.

Rosenzweig, M. R., & Wolpin, K. I. (1985). Specific experience, householdstructure and intergenerational transfers: Farm family land and laborarrangements in developing countries. Quarterly Journal of Economics,100, 961–987.

Rousseeuw, P. J. (1997). Introduction to positive-breakdown methods. InG. S. Maddala, & C. R. Rao (Eds.), Handbook of statistics(pp. 101–121). Elsevier: Amsterdam.

Schubert, R., Brown, M., Gysler, M., & Brachinger, H. W. (1999). Genderand economic transactions. financial decision-making: Are womenreally more risk-averse?. The American Economic Review, 89(2),381–385.

Sharma, M., & Zeller, M. (1997). Repayment performance in group-basedcredit programs in Bangladesh: An empirical analysis. World Devel-opment, 25(10), 1731–1742.

Stiglitz, J. E. (1990). Peer monitoring and credit markets. The World BankEconomic Review, 4(3), 351–366.

Timberg, T. A., & Aiyar, C. V. (1984). Informal credit markets in India.Economic Development and Cultural Change, 33(1), 43–59.

Stromberg, A. J. (1993). High breakdown estimation of nonlinearregression parameters. Journal of American Statistical Association,88(421), 237–244.

Varian, H. (1990). Monitoring agents with other agents. Journal ofInstitutional and Theoretical Economics, 146(2), 153–174.

Wydick, B., Karp, H., & Hilliker, S. (2007). Social networks, neighbor-hood effects, and credit access. In Paper presented at the 1st Centre forInternational Banking, Insurance and Finance Conference on “Micro-finance: What Do We Know?”, Groningen, The Netherlands.

Wydick, B. (1999). Can social cohesion be harnessed to repair marketfailures? Evidence from group lending in Guatemala. EconomicJournal, 109(457), 463–475.

APPENDIX A

See Table 12.

Table 12. List of the variables used in the regressions

Variable description Name Contents

Investment Investment in working capital(agricultural activities)

Expenditure in seeds, fertilizers, pesticides during one production cycle(taka)

Investment in working capital (non-agricultural activities)

Expenditure in raw materials, fuel, equipment maintenance, etc. duringone production cycle (taka)

Investment in fixed assets(agricultural activities)

Increase/decrease of the stock of agricultural assets in a one-year period(taka)

Investment in fixed assets (non-agricultural activities)

Increase/decrease of the stock of non-agricultural assets in a one-yearperiod (taka)

Credit market participation Group lending (MFIs) Informallending Bank lending

Binary variables (1= the household borrows from each one of the threesources)

Other characteristics of loans AmountInterest rateCollateral required

Loan principal (taka)PercentageBinary (1 = yes)

Characteristics of the household HH head: age Age of the household head and spouse (years) at the time of the surveyHH spouse: ageHH head: education Years of schooling achieved by the household head and spouse at the

time of the surveyHH spouse: educationHH head is male Gender of the household head. Binary (1 = male)Religion Religion of the household head. Binary (1 = Islam, 0 otherwise)Feminine ratio Ratio of females to males in the householdNo. of persons in HH Number of household membersFather: same activity Binary (1 = household head carries out the same activity of his/her

father)Head: father lives in HHSpouse: father lives in HHHead: mother lives in HHSpouse: mother lives in HH

Binary variables stating whether the head of the household andspouse’s parents reside in the household (1 = yes)

Relationship network No. of parents aliveN. siblings aliveN. other relatives alive

Number of the head of the household and spouse’s relatives who arealive

Illness/Injuries Injury last year The household head suffered any injury or disease last yearBinary (1 = yes)

Medical expenditures last year Amount spent last year for medical expenditures (taka)No. of days not working last year Number of days that the household head has lost during the last year

due to injuryMeasures of wealth Land

House valueTransport value

Land ownership (acres). Value (taka) of the house and means oftransport owned by the household

Transfers Transfers: cashTransfers: otherTransfers: food

Amount of transfers in cash, food or other commodities received by thehousehold from relatives, friends, and neighbors during the last year(taka)

Farming related variables Total area cultivated Acres, included those received for sharecroppingTenure: fixed rentTenure: sharecropper

Binary (1 = yes)

Instruments Target Eligibility rule for MFIs. Binary (1 = the household owns less thanhalf an acre of land)

Spouse: distance from family Distance (in miles) between the household and the spouse’s birthplaceNo. of relatives own land Number of the head of the household and spouse’s relatives owning

landDowry Amount of money, goods, or estate given by the woman’s family at

marriageMarriage gifts to spouse Amount of money, goods, or estate received by the woman at marriage


Available online at www.sciencedirect.com

microfinance and investment: a comparison with bank and informal lending

Documents