Microfinance and Moneylender Interest Rate: Evidence from

Bangladesh

Debdulal Mallick* School of Accounting, Economics and Finance

Deakin University 221 Burwood Highway, Burwood, Victoria 3125

Australia Email: dmallic@deakin.edu.au

Phone: (61 3) 925 17808 Fax: (61 3) 9244 6283

March, 2011

*The author thanks Aldo Benini, Shyamal Chowdhury, Matthew Clarke, Chris Doucouliagos, Cong Pham, Bazle Razee, and Mehmet Ulubasoglu for their helpful comments, and Research and Evaluation Division, BRAC for making the data set available. All errors and omissions are the sole responsibility of the author, and the conclusions do not necessarily reflect the views of the organization with which he is associated.

2

Microfinance and Moneylender Interest Rate: Evidence from Bangladesh

Abstract

The linkage between the formal and informal credit markets in developing countries has

largely been unexplored. This paper addresses one important aspect of the linkage by

empirically investigating the impact of the microfinance program intervention on the

moneylender interest rates in northern Bangladesh, and finds that moneylender interest

rates increase with microfinance program coverage. Higher microfinance program

coverage increases moneylender interest rates in the villages in which more loans are

invested in productive economic activities than consumption. Borrowers resort to

moneylenders for additional funds probably because of inadequate supply, unavailability

of seasonal working capital from MFIs, and tight repayment schedule, which in turn

increases demand for moneylender loans.

Keywords: Microfinance, moneylender, interest rate, informal sector.

JEL Classification codes: G21, O17, O12, C26

3

Microfinance and Moneylender Interest Rate: Evidence from Bangladesh

I. Introduction

Microfinance program1 was devised to mitigate credit market imperfections in

developing countries. The poor do not have access to the formal financial institutions

mainly due to the lack of collateral and are forced to resort to informal credit market such

as moneylenders who then charge exorbitantly higher interest rate on loans. One of the

arguments given for expanding the microfinance industry is to substitute for the informal

credit market2

Recent theoretical development points out that the response of the interest rates in

the informal sector to the expansion of the formal credit depends on the characteristics of

both sectors such as market structure in the informal sector as well as repayment schedule

and help poor people escape the clutches of “evil moneylenders” who

allegedly charge usurious rates of interest (Meyer, 2002). This has been a persistent

problem encountered by the policymakers in developing countries. As early as the 1950s,

the Indian government tried to provide positive institutional alternatives to the

moneylenders (Bell, 1990). However, the effect of the expansion of microfinance on the

informal credit market is unexplored although the linkage between formal and informal

credit markets has been of great interest to development economists. Morduch (1999, p.

1595) raises the question, “how will the existence of a subsidized program affect the

profitability of both formal and informal institutions operating nearby?” This paper

addresses one important aspect of the linkage by empirically investigating the impact of

the microfinance program intervention on the moneylender interest rates in northern

Bangladesh. This is worth mentioning that Bangladesh has the largest operation of the

microfinance program in the world with about 12.4 million active borrowers and over

629 microfinance institutions (MFIs) engaged in microfinance (World Bank 2003, p. 21).

1 This paper does not differentiate “microfinance” from “microcredit”, and defines microfinance operation narrowly as lending activities only. 2 Informal sector includes friends, relatives, neighbors, moneylenders, employers, input suppliers, and shopkeepers. Microfinance institutions (MFIs), government institutions, commercial banks, cooperatives, and other NGOs are defined as the formal sector (McKernan, Pitt and Moskowitz, 2005).

4

of the formal sector. Hoff and Stiglitz (1993, 47-48; 1998) outline the conditions under

which increasing access to formal credit may increase or decrease interest rates in the

informal sector. If some borrowers can satisfy all their borrowing needs from the formal

sector at lower interest rates, there will be less demand for informal credit. Under perfect

competition and perfect information, this would exert downward pressure on interest

rates. But in a monopolistically competitive market3

Jain and Mansuri (2003) follow an entirely different route that resembles the

present scenario in Bangladesh that a microfinance program may well have a “crowding-

in” effect on informal lenders. Under certain circumstances, this “crowding-in” effect

might be strong enough to raise the interest rates in the informal sector. For example, the

with free entry and one moneylender

being an imperfect substitute for another, a subsidy in the formal sector may cause the

interest rates in the informal sector to rise because the induced new entry drives up the

marginal enforcement cost of lending in the informal sector (Hoff and Stiglitz, 1998).

Bose (1998) extends Hoff and Stiglitz (1998) by including heterogeneous agents who

differ in their probability of default. The subsidized programs diminish optimal scale and

siphon off the best borrowers. As a result, the informal lenders are left with a riskier pool

of borrowers that leads to higher enforcement costs, and consequently they charge higher

interest rates as risk premium. Jain (1999) reaches similar conclusion by considering a

case in which scale advantages of the formal sector outweigh the informational

advantages of local moneylenders, while Floro and Ray (1997) consider the case in which

an expansion of formal credit may strengthen the ability of informal lenders to collude

among themselves.

3 There is evidence that rural informal credit markets are not perfectly competitive. One celebrated study is by Aleem (1990) who studied services, costs, and charges of 14 informal market moneylenders and their clients in Chambar, Pakistan. He estimated the resource costs incurred by informal lenders for screening, pursuing delinquent loans, overhead, and cost of capital and found that lenders charge interest rates that are equal to their average cost of lending but exceed their marginal cost. The findings suggest that the informal credit market is characterized by monopolistic competition in the presence of imperfect information. Iqbal (1988) found evidence of the reduction of moneylender monopoly power in rural India as a result of increased competition from formal lending agencies. Kochar (1997, p. 366) obtained a negative but insignificant (statistically) coefficient of per capita formal credit on the probability of borrowing from informal sector, and concluded that a more thorough analysis of the relationship between formal and informal credit is warranted.

5

tight repayment schedule of MFIs (in Bangladesh installment of MFI loans starts almost

immediately after loan disbursement) forces many borrowers to borrow from

moneylenders to repay MFI loans. Borrowers also find difficult to finance long-gestation

projects and even seasonal working capital needs for agricultural production by MFI

loans.

Although there are numerous theoretical models that explain the interest rates in

the informal sector as a result of MFI program expansion, empirical evidence is scant. To

the best of our knowledge, this is the first paper that investigates this linkage in the

context of Bangladesh. This paper analyzes village level survey data from 156 villages in

three districts in northern Bangladesh, and finds that moneylender interest rates increase

with MFI program coverage. The interest rates have been found to be higher in the

villages where higher percentage of households borrows from MFIs after controlling for

uses of MFI and moneylender loans, adoption of modern agricultural production

technology, quality of cultivable land, village level physical infrastructure, incidence of

poverty, and a proxy for competition among moneylenders.

We also find that use of moneylender loans in productive purposes than

consumption lowers interest rates. But higher MFI program coverage increases

moneylender interest rates in the villages in which more loans are invested in productive

activities. The explanation of the results may be the following. Productive use of loan

decreases the default risk which in turn lowers the interest rate. On the other hand, since

MFIs do not supply adequate amount for productive investment or for seasonal working

capital and repayment is required immediately after borrowing, borrowers resort to

moneylenders for additional funds to sustain their projects. The increased demand raises

moneylender interest rates. This argument supports Jain and Mansuri (2003) hypothesis

although we cannot directly test it. The reason is that there is no variation in the

repayment schedule among the MFIs in Bangladesh; maturity of loan is usually one year

and borrowers are required to begin repayment immediately after loan disbursement

(usually after one or two weeks). Interest rates are also higher in the villages located

away from place of commercial activities. We also employ heteroskedascity based

identification recently developed by Lewbel (2007) that does not rely on exclusion

restrictions, and find that the results are also robust to correction for endogeneity.

6

The results have important policy implications. MFIs can enhance productive

investment by allowing more flexibility in loan disbursement and repayment schedule.

Borrower projects will also be more profitable if MFIs expand their program of loan-only

to loan-plus such as credit with skill development training, information, and provision of

inputs. On the other hand, competition between formal and informal lenders will increase

borrowers’ access to fund at competitive interest rates.

The rest of the paper is organized as follows. Section II provides a brief literature

review of the formal-informal linkages in the rural credit market in Bangladesh. Section

III discusses in detail the data collection methodology. The results are presented in

Section IV that include descriptive statistics, and the results by OLS and system

estimation. Finally, Section V concludes.

II. Formal-informal linkages in credit market in rural Bangladesh

Very little is known about the response of the informal lenders to the MFI

program expansion. However, there is widespread evidence that MFI clients borrow from

informal sources as part of their financial management strategy. Sinha and Matin (1998)

report that about 87% of rural households in northern Bangladesh, many of them are MFI

borrowers, borrow from informal sources. Husain et al. (1998) find that 11.6% of

borrowers from BRAC also borrowed from moneylenders. Zeller et al. (2001) report

similar practices among the rural households in Bangladesh they surveyed in 1994. MFI

borrowers received 20% of their total debt outstanding from friends and relatives, and

another 18% from shopkeepers and other informal sources.

McKernan, Pitt and Moskowitz (2005) document that during the 1998-99 time

period, the informal sector was at least as important as the formal sector in Bangladesh.

About 29% of the rural households received money (in the form of loan and gift) from

the formal sector compared to 24% of the households receiving money from the informal

sector. Average amount of money received from the informal sector was significantly

larger than average amount received from the formal sector. An average household

received 2,125 Taka in formal loans, 1,440 Taka in informal loans, and 3,040 Taka in

informal gifts. Only 6% of households received money from both formal or informal

sectors and 49% from either sector. About 2% of rural households received credit from

7

moneylenders, accounting for 23% of all non-relative credit and 9% of the informal

credit.

Household consumption and payment of other loans are the two most frequently

reported uses of the informal loans. Sinha and Matin (1998) point out that internal cross-

financing takes place due to the large increase in MFI lending in which the proceeds from

one loan were being used to repay another. Clients borrow from informal sources so they

could maintain high repayment rates with MFIs and become eligible for larger future

loans. Atique Rahman (1992) also provides evidence of similar practices. Aminur

Rahman (1999) observes that MFI employees are expected to increase disbursement of

loans among their members and press for high recovery rates to earn profit necessary for

economic viability of the institution. To ensure timely repayment in the loan, MFI

employees and borrowing peers inflict an intense pressure on women clients. Many

borrowers maintain their regular repayment schedules through a process of loan recycling

that considerably increases the debt-liability on the individual households, and increases

tension and frustration among household members. Zeller et al. (2001) report that 9% of

the funds borrowed from informal sources were used to pay existing debt. Todd (1996)

also describes the complex financial management practices of women members of the

Grameen Bank. Clients borrow to pay moneylender and other informal loans, and borrow

small sums informally to meet loan installments.

III. Data

Data were collected in 2002 from 156 villages in three districts in northern

Bangladesh (Kurigram, Rangpur and Nilphamari) as part of the baseline survey for

BRAC’s (one of the largest NGOs in the world, formerly known as Bangladesh Rural

Advancement committee) CFPR/TUP (Challenging the Frontiers of Poverty Reduction–

Targeting the Ultra Poor) program.4

4 The CFPR/TUP program is designed to target only the extreme poor whose poverty lasts long or throughout their entire life, and who even lack the opportunities for upward mobility through regular microfinance programs (Matin and Halder, 2004, p. 5).

Before launching the CFPR/TUP program on a pilot

basis, BRAC first needed to select the villages and the extreme poor therein. In the first

step of the selection, BRAC’s Research and Evaluation Division (RED) conducted

8

Participatory Rural Appraisal (PRA) in the villages randomly selected to determine the

extent of poverty and also to identify the extreme poor in the selected villages.5

Households initially selected by PRA were surveyed to cross-check the information

gathered by PRA, and also to create a baseline for future program evaluation. Along with

the baseline household survey, comprehensive village level information was collected by

Focus Group Discussion (FGD).6

There were several preparatory stages before the FGD. The first stage was to

build rapport with the community members. This was not difficult because in most

situations the community members were already familiar with BRAC programs and staff.

BRAC’s Program Organizers walked around the community inviting people of all walks

of life to the PRA meeting. The PRA sessions and the next household surveys enabled the

research team to build good rapport with the community and to have a good idea about

who to invite for the FGD. These are the people who are the most knowledgeable about

the village. Given the type of issues to be covered in the FGD, members of different

socio-economic background, age, and occupation were selected. The group of invitees

generally included school teachers, elected Union Parishad (smallest local government

unit) members, health workers, students, and clients of different MFIs. Local bazaar,

school or the well traveled place in the village was selected as the venue for the FGD.

The size of the group ranged between 8 and 10.

This paper uses village level survey data for these

villages. Note that the household survey was conducted only among the extreme poor

selected by PRA and therefore does not represent the village population.

7

5 For a detailed description of data collection method, see BRAC (2004).

6 Stewart and Shamdasani (1990), and Litosseliti (2003) provide nice discussions on Focus Group Discussion. 7 The group seated on chairs around a table so that maximum opportunity for eye contact is possible with both the moderator and other group members. If chairs were not available for everyone, the group sat on the ground making a circle. The moderator, who was a researcher well trained in qualitative research and data collection, was fully aware of the fact that MFI clients are poor women and being borrowers they are less powerful in the community. The moderator encouraged them to freely express their opinion by explicitly asking (not insisting to give a particular answer) them and by providing verbal rewards for such opinions. This was necessary because their responses are crucial for some information. They borrow from different sources including MFIs and moneylenders and

9

A structured questionnaire was used to collect data because of the type of

information sought and also for reliability and possibility of replication. At the beginning

of the session, the moderator clearly specified the objective of the FGD and the type of

information to be sought. It was also made clear at the outset that the FGD will continue

for about two hours. However, in several occasions, all issues were not possible to cover

in the stipulated time so that discussions had to discontinue. One reason is that

information was recorded only if there was either a consensus or agreement among

majority of the participants. Groups were not kept beyond schedule because it was

perceived that impatience of the participants may lead to inaccurate answers. Therefore,

all information could not be collected from many villages; there are 89 villages for which

data are available for all variables included in the regression.

It is important to mention that more than one FGD were conducted in large

villages because participants residing in one part of the village may not have the most

accurate information about those residing in another part.

IV. Results

In this section, we first present descriptive statistics of the villages and then

regression results of the determinants of the moneylender interest rates.

A. Descriptive statistics

Descriptive statistics of the villages are presented in Table 1. In the sample

villages, on average 3.9 MFIs have loan operation. The minimum and maximum numbers

of MFIs working in a village are one and nine respectively. Average number of big

MFIs8

have also close contacts with other MFI clients, and therefore they are better informed about the rural credit market. However, their responses were also discussed by other participants in the group and enumerated only after verification.

is 3.6 with the minimum and maximum numbers being one and eight respectively.

This indicates competition among the MFIs for clients even among the big ones. On

average 33% of all households borrow from any MFI with a standard deviation of 19.

8 These are Grameen Bank, BRAC, ASA, Proshika, BRDB, Bangladesh Agricultural Bank, RDRS, and Thengamara. The last two are big regional MFIs working only in northern Bangladesh.

10

Insert Table 1 here

In 32.7% of the villages, higher percentages of clients utilize MFI loans for any

productive investment9

In about 27% of the villages, poverty situation improved and in 17% of the

villages it remained the same in last five years. In about 35% of the villages the situation

deteriorated severely. Average daily male and female wage rates (wage rates are

averaged over three cropping seasons) are Taka 44 and 30, respectively. On average,

above 50% of cultivable land grows two crops a year, and 32% grows three crops a year.

, and agriculture has been the predominant sector for investment.

Conversely, in only 10% of the villages, the primary use of moneylender loans was any

productive investment.

The moneylender interest rates vary considerably with a mean of 103.33% and

standard deviation of 59.06. The minimum and maximum interest rates are 10% and

240%, respectively. These are the averages of annual moneylender interest rates in a

village.

B. Determinants of moneylender interest rates

We now report the results for multivariate regressions. The dependent variable is

the annual moneylender interest rate. This is average interest rate in a village; it does not

contain information about the interest rate variations among the borrowers.

9 Participants in the FGD reported the sectors where (both moneylender and MFI) loans are used. Three main sectors of loan use were recorded for each village. Ten broad sectors of loan use were reported in the FGDs. These are: i) investment in small businesses such as rural trading, ii) purchasing assets such as rickshaw/van, iii) children’s education, and bribery for salary-paid job or work, iv) agricultural inputs (seed, fertilizer, irrigation water), v) cultivation (renting power tiller, hiring labor), vi) purchasing bullocks for tilling, vii) purchasing water pump machine for irrigation, viii) consumption (food and other households items), ix) medical treatment, and x) daughter’s marriage/dowry. The first seven uses (i-vii) were recoded as productive investment (= 1), and the last three uses (viii-x) as non-productive use or consumption (= 0). The main use of loan in a village was considered as productive investment if at least two of the three uses were “1”. Conversely, the main use was considered as consumption if at least two of the three uses were “0”.

11

OLS estimation

We first present benchmark results for the OLS regressions. The estimation

equation is

= _i i iInterestML MFI Coverα β ε+ + +iγX , --- (1)

where iInterestML is the average moneylender interest rate in village i, _ iMFI Cover is

MFI program coverage defined as the percentage of households borrowing from MFIs in

village i,10iX is a vector of other control variables, and iε is the error term.

The variables in the iX vector include uses of moneylender and MFI loans,

current incidence of and past changes in poverty, opportunity for productive activities,

and village level physical infrastructure and marketing opportunities. Interest rate on MFI

loans is excluded from equation (1) because MFIs charge the same interest rate to all

borrowers, and there is almost no variation in interest rates among MFIs.

Average daily male and female wage rates are used as proxy for current poverty

incidence in the village. Ghatak (1983, p. 32) finds that interest rates on informal loans

depend on the poverty level. Average rural interest rates for different classes such as

casual laborers, tenants, and agricultural laborers varied between 36-84% per year in a

relatively more prosperous district like Burdwan in West Bengal in India, while it

averaged between 72-120% in a relatively poor district like Nadia.

We include percentage of cultivable land that grows one, two, and three crops a

year as proxies for soil quality, and percentage of land irrigated using electricity as a

proxy for adoption of modern agricultural production technology. The latter variable

reflects the opportunity for productive investment in agriculture. Main use of both MFI

and moneylender loans (1 = investment, 0 = consumption), on the other hand, capture

realization of any such opportunity. Iqbal (1988) finds in the context of India that

moneylender interest rates are influenced by a host of borrower and environmental

characteristics. Borrower’s profit potential as measured by farm size, soil quality, and

even farmer’s education are important consideration for moneylenders. Farmers residing

10 Since both variables are in percentage, β is interpreted as percentage points change in moneylender interest rates for one percentage point increase in MFI client households.

12

in areas characterized by the use and/or provision of new technology face lower interest

rates.

The other controls in iX are several measures of village level infrastructure that

include distances of the center of the village from the nearest bank, bazaar, bus stop, and

all-weather (paved) road, percentage of households with electricity connection, and

number of shops in the village. It is argued that MFI program placement is not random

(de Aghion and Morduch, 2005, p. 223; Mallick and Nabin, 2010). Village level

infrastructure and other controls previously mentioned are intended to account for this

non-randomness. A complete list of variables is provided in Table 1.

We estimate equation (1) with different combinations of the variables in iX . To

account for the village level heterogeneity, we report the White (1980) heteroskedasticity

corrected standard errors. The results are reported in Table 2. Column 2 reports the

results for the simple regression when iX is excluded from equation (1). The coefficient

of MFI coverage is 1.1 and significant at any conventional level implying that one

percentage point increase in MFI borrowers in a village increases the average

moneylender interest rates by 1.1 percentage points. This variable alone can explain 14%

variation in the moneylender interest rates. In columns 3, uses of moneylender and MFI

loans are included. In column 4, village level infrastructure and incidence of poverty are

included. The magnitude and significance of MFI coverage remains the same in all

specifications. In column 5, we report the results for a non-linear specification in which

the square of MFI coverage is added. The coefficient of MFI coverage is positive and that

of its square is negative and both are highly significant indicating that moneylender

interest rate increases with MFI coverage up to a certain level of coverage after which

interest rate decreases.11

11 The marginal effect of MFI coverage, evaluated at its mean value, is 1.723 with a standard error of 0.275.

In column 6, past changes in poverty are included. This is a

categorical variable; three dummies—considerable increase, marginal increase and

unchanged incidence of poverty (decreased incidence is the base category)—are

included. The main conclusion does not change although the coefficient of MFI coverage

slightly decreases to 0.87. In all regressions, the coefficients of both moneylender and

MFI loan use enter negatively but are not significant. We also find that moneylender

13

interest rates are higher in the villages located away from bazaar (place of commercial

activities) and also in the villages where higher percentage of land is irrigated using

electricity.

Insert Table 2 here

Competition among the moneylenders can also influence interest rates in both

directions. If the informal credit market is characterized by perfect competition,

competition among moneylenders drives down interest rates. On the other hand, loss of

economies of scale can result in higher interest rates. We do not have any direct measure

for such competition. Moneylending activities in rural areas, to a large extent, are

controlled by large landowners (Bell, 1990, p. 306; Hoff and Stiglitz, 1998, p. 488).

Information on the number of moneylenders or large landowners in a village is not

available. Hossain, Sen and Rahman (2001) document that landownership in rural

Bangladesh is very unequal and the inequality has been increasing overt time. Based on

two censuses in 62 villages from 57 (out of 64) districts in 1988 and 1995, they document

that both marginal and large landowners have been increasing over time. Percentage of

landless and functionally landless households (owning up to 0.2 hectare) increased from

46% to 49.6% during 1988-1995. Percentage of large owners (owning 3 hectares or

more) also increased from 3.3% to 3.5% during the same period (their Table 3 in p.

4632). The concentration coefficient in land ownership, which was already high at 0.67 in

1988, further increased to 0.69 in 1995. Griffin, Khan and Ickowitz (2002, Table A.1 in

p. 323) also report a Gini coefficient of 0.65 of land ownership in rural Bangladesh. This

implies that the higher the number of marginal landowners and landless in a village, there

are also more large landowners and hence more moneylenders in the village. Therefore,

percentage of marginal landowners and landless in the village is used as a proxy for the

number of moneylenders (large landowners) which in turn implies competition among

moneylenders. Information about the number of households owning no more than 10

decimal of land was collected by PRA (discussed in Section III) and the follow-up survey

to select the extreme poor. We exploit this information to calculate percentage of landless

and marginal landowners who own no more than 10 decimal of land. We find that

14

competition among moneylenders lowers interest rate on their loans (column 7). The

other conclusions do not change in this specification.

To comprehend whether the opportunity for and actual productive investment in a

village influence the effect of MFI coverage on moneylender interest rates, we include

interaction of MFI coverage with use of MFI and moneylender loans, and percentage of

land irrigated using electricity.

1 2

3 4

= _ _ * _ * _ *

i i i i

i i i i i

InterestML MFI Cover MFI Cover UseMLMFI Cover UseMFL MFI Cover ElectIri

α β ββ β ε

+ ++ + + +iγX

,

--- (2)

where iX has been defined previously. The results are reported in Table 3 for different

combinations of the explanatory variables. The coefficient of MFI coverage ranges

between 0.9 and 1.4 and is robustly significant as previously. One important change in

the results is that the negative coefficient of moneylender loan use is now robustly

significant. This implies that if loans are utilized in productive purposes, moneylenders

lower interest rates because of lower default risk of the borrowers. The interaction of MFI

coverage with moneylender loan use enters positively and is significant at 1% level. This

result in conjunction with the previous result implies that although productive use of loan

lowers moneylender interest rates, such use increases interest rates in the villages with

greater MFI coverage. In addition to inadequate supply of loan, MFIs allow little or no

flexibility in loan disbursement so that seasonal working capital is difficult to finance by

MFI loans. Repayment schedule is also very rigid as first installment is due immediately

after loan disbursement so borrowers in many instances save a portion of loan for

installment. As a consequence, more loans are demanded from moneylenders to sustain

the project, which in turn raises moneylender interest rates.

We also find that moneylender interest rates are higher in the villages with greater

opportunity for productive activities (higher percentage of land irrigated using electricity)

but its interaction with MFI coverage is negative. This suggests that moneylenders

charge lower interest rates if MFIs provide large coverage in a village having greater

opportunity for productive investment. The reason for this result may be the following. In

rural Bangladesh, large-scale irrigation facilities are controlled by large landowners, who

15

are also the moneylenders. They charge higher interest rates to their tenants who purchase

water from them. However, higher MFI coverage forces the moneylenders/large

landowners to lower interest rates due to competition.

Insert Table 3 here

Estimation of heteroskedasticity based identification

If higher moneylender interest rates cause good borrowers to switch to MFIs,

OLS estimation may give rise to biased and inconsistent estimates in the presence of

reverse causality. Instrumental variable (IV) estimation that relies on exclusion

restrictions solves the problem but in our case finding appropriate instrument(s) is

extremely difficult and therefore, IV estimation is not feasible.

To address the endogeneity issue, we employ an alternative approach to

identification proposed by Lewbel (2007) that does not rely on any exclusion restrictions

but exploits heteroskedasticity for identification. In the following, we first provide a brief

intuitive discussion of the approach.12

ε̂

Identification can be achieved without imposing

any exclusion restrictions if there is a vector of exogenous variables Z and the errors are

heteroskedastic. The Z vector can be a subset of X in the regression or even Z = X. In the

first stage, each endogenous variable is regressed on all the control variables (including

the Z vector), and the vector of residuals is retrieved. These estimated residuals are

then used to construct instruments ( ) ˆ−Z Zε , where Z is the mean of Z. Identification

requires that the error terms in the first stage regressions are heteroskedastic (Lewbel,

2007).

For equation (1), in the first stage we run a regression of MFI_cover, the

endogenous variable, on all other variables and then retrieve the residual. The Breusch-

Pagan test rejects the null of homoskedasticity at 5% level. In the second stage, equation

(1) is estimated by IV with ( )ε̂−Z Z as the instruments, where the Z vector includes all

continuous variables (we include only continuous variables because of practicality as the 12 For a detailed discussion of the approach, please see Lewbel (2007). For application of this approach, please see Emran and Hou, 2008; Rashad and Markowitz, 2007; Sabia, 2007.

16

mean of a discrete variable is not well defined). The instruments also satisfy the

exogeneity condition; the null of exogeneity cannot be rejected with a p-value equal to

0.362 for Hansen's J-statistic. Shea's Partial R2 is 0.42. Column 2 in Table 4 presents the

results. The coefficient of MFI_cover is 1.834 with a large t-statistic of 4.66 supporting

the OLS results that MFI coverage increases moneylender interest rates. This result is

comparable to column 7 in Table 2. The magnitude of the coefficient is now 59% higher

than that estimated by OLS.

In equation (2), there are four endogenous variables—MFI_cover and its

interaction with UseML, UseMFL and ElectIri. In the first stage, each of the four

endogenous variables is regressed on all control variables, and the corresponding four

residual are retrieved. In each case, Breusch-Pagan test rejects the null of

homoskedasticity at any conventional level. In the second stage, equation (2) is estimated

by IV with ( ) ˆ−Z Zε as the instruments, where the Z vector has been constructed defined

as before. Again the instruments also satisfy the exogeneity condition; p-value for

Hansen's J-statistic is 0.166. The lowest value for Shea's Partial R2 is 0.866. The results

are presented in column 3 in Table 4, which are comparable to the results in column 6 in

Table 3. The coefficient of MFI_cover is 1.31 with a large t-statistic of 4.75. This

magnitude is very close to what we have found in OLS estimation. We also find that the

sign and significance of other coefficients do not qualitatively change. Use of

moneylender loan enters negatively and significantly, and its interaction with MFI

coverage positively and significantly. These results reconfirm our previous results that

moneylenders charge higher interest rates if loans are used in productive purposes in the

villages where higher percentage of households borrow from MFIs. We also find that

moneylender interest rate is higher in the villages located away from place of commercial

activities (distance from bazaar). Finally, competition among moneylenders lowers their

interest rates.13

13 We also bootstrap 500 times the two-step estimation to know if the results suffer from small sample bias. The results do not qualitatively change although standard errors slightly increase. The new standard errors for MFI_cover, the main the explanatory variable, are also reported in Table 4 .

17

Our results are based on village level data that lack information at the individual

borrower level. Moneylenders may charge different interest rates to different borrowers

in the same village based on individual borrower characteristics and loan size. It is worth

mentioning that there may be possible multi-level sample selection bias in the household

data (Iqbal, 1988). For example, X number of total households are surveyed out of which

1X X< are found to be borrowers, and only 2 1X X< borrow from moneylenders. If the

analysis is restricted to the 2X borrowers only, the results suffer from two sources of

selection bias. If borrowing households are not randomly selected from the overall

sample, which may not be unlikely, OLS regression does not distinguish between the

behavioral function relating the interest rate to its determinants and the sample selection

function relating the probability of borrowing to its determinants. Heckman (1979)

correction can be employed to account for the bias, but there is another level of sample

selection bias. If moneylender clients are not randomly selected from the sample of

borrowers, which may not again be unlikely, a similar statistical problem occurs. Village

level data can avoid this econometric difficulty but at a cost. Nonetheless, it would be

more informative to investigate using household data with duly addressing the

econometric issues.

V. Concluding remarks

This paper explores the impact of the MFI program intervention on the

moneylender interest rates in northern Bangladesh. It has been found that moneylender

interest rates increase with the percentage of households borrowing from MFIs.

Productive investment of loan lowers moneylender interest rates. But greater coverage of

MFI program increases moneylender interest rates in the villages in which more loans are

invested in productive economic activities. The explanation of the results may be the

following. As loans are utilized in productive purposes, the likelihood of repayment

increases so that moneylenders are able to charge lower interest rates. But if the overall

demand for fund goes up as indicated by higher percentages of households borrowing

from MFIs, and if MFI loan is inadequate or seasonal working capital is unavailable from

MFIs or repayment schedule is tight, borrowers will resort to moneylenders for additional

18

fund to sustain their projects. Increased demand for fund will drive up moneylender

interest rates. The results are robust to correction for endogeneity.

The results may be potentially important for the policymakers. Borrowers can

make more productive investments if MFIs meet their demand for loan by allowing more

flexibility in loan disbursement and repayment schedule. Borrower projects will also be

more profitable if MFIs expand their program of loan-only to loan-plus.14

In that case,

moral hazard problem faced by moneylenders will be greatly reduced. Active presence of

the moneylenders is not necessarily harmful and can even be beneficial if increasing

competition between formal and informal lenders increases borrowers’ access to funds at

competitive interest rates.

14 BRAC follows a loan-plus approach, where loans are accompanied by various forms of assistance for the borrowers such as skill development training, provision of higher quality inputs, and technical assistance as well as marketing for finished products.

19

References

Aleem, I. 1990. “Imperfect Information, Screening, and the Costs of Informal Lending: A Study of a Rural Credit Market in Pakistan.” World Bank Economic Review 4, no. 3:329-349. Bell, C. 1990. “Interactions between Institutional and Informal Credit Agencies in Rural India.” World Bank Economic Review 4, no. 3:297-327. Bose, P. 1998. “Formal-Informal Sector Interaction in Rural Credit Markets.” Journal of Development Economics 56, no. 2:265-280. BRAC. 2004. “Stories of Targeting: Process Documentation of Selecting the Ultra Poor for CFPR/TUP Programme.” CFPR-TUP Working Paper Series No. 1, Research and Evaluation Division, BRAC: Bangladesh. De Aghion, B. A., and J. Morduch. 2005. “The Economics of Microfinance.” MIT Press: Massachusetts. Emran, M. S., Z. Hou. 2008. “Access to Markets and Household Consumption: Evidence from Rural China.” forthcoming, Review of Economic and Statistics. Floro, M. S., and D. Ray. 1997. “Vertical Links Between Formal and Informal Financial Institutions.” Review of Development Economics 1, no. 1:34-56. Ghatak, S. 1983. “On Interregional Variations in Rural Interest Rates.” Journal of Developing Areas 18, no. 1:21-34. Griffin, K., A. R. Khan, and A. Ickowitz. 2002. “Poverty and Distribution of Land.” Journal of Agrarian Change 2, no. 3: 279-330. Heckman, J. J. 1979. “Sample Selection Bias as a Specification Error.” Econometrica 47, no. 1:153-161. Hoff, K., and J. E. Stiglitz. 1993. “Imperfect Information and Rural Credit Markets: Puzzles and Policy Perspectives.” In The Economics of Rural Organization: Theory, Practice, and Policy, ed. K. Hoff, A. Braverman, and J. E. Stiglitz, Oxford University Press: New York, 33-52. Hoff, K., and J. E. Stiglitz. 1998. “Moneylenders and Bankers: Price-Increasing Subsidies in a Monopolistically Competitive Market.” Journal of Development Economics 55, no. 2: 485-518.

20

Hossain, M., B. Sen, and H.Z. Rahman. 2001. “Growth and Distribution of Rural Income in Bangladesh: Analysis Based on Panel Survey Data.” Economic and Political Weekly 35, no. 52/53 (December 30): 4630-4637. Iqbal, F. 1988. “The Determinants of Moneylender Interest Rates: Evidence from Rural India.” Journal of Development Studies 24, no. 3:364-378. Jain, S. 1999. “Symbiosis vs. Crowding-out: The Interaction of Formal and Informal Credit Markets in Developing Countries.” Journal of Development Economics 59, no. 2:419-444. Jain, S., and G. Mansuri. 2003. “A little at a time: the use of regularly scheduled repayments in microfinance programs.” Journal of Development Economics 72, no. 1:253-279. Kochar, A. 1997. “An Empirical Investigation of Rationing Constraints in Rural Credit Markets in India.” Journal of Development Economics 53, no. 2:339-371. Lewbel, A. 2000. “Using Heteroskedasticity to Identify and Estimate Mismeasured and Endogeneous Regressor Models.” Working Paper, Boston College. Litosseliti, L. 2003. “Using Focus Groups in Research.” Continuum Research Methods Series, Continuum: New York. Mallick, D., M. H. Nabin. (2010). “Where NGOs Go and Do Not GO.” Deakin University. Matin, I., and S. R. Halder. 2004. “Combining Methodologies for Better Targeting of the Extreme Poor: Lessons from BRAC’s CFPR/TUP Programme.” CFPR-TUP Working Paper Series No. 2, Research and Evaluation Division, BRAC: Bangladesh. McKernan, S-M., M. M. Pitt, and D. Moskowitz. 2005. “Use of the Formal and Informal Financial Sectors: Does Gender Matter? Empirical Evidence from Rural Bangladesh.” World Bank Policy Research Paper 3491: World Bank, Washington DC. Meyer, R. L. 2002. “The Demand for Flexible Microfinance Products: Lessons from Bangladesh.” Journal of International Development 14, no. 3:351–368. Morduch, J. 1999. “The Microfinance Promise.” Journal of Economic Literature 37, 4:1569-1614. Rahman, Aminur. 1999. “Micro-credit Initiatives for Equitable and Sustainable Development: Who Pays?.” World Development 27, no. 1:67-82. Rahman, Atique. 1992. “The Informal Financial Sector in Bangladesh: An Appraisal of its Role in Development.” Development and Change 23, no. 1:147-168.

21

Rashad, I., S. Markowitz. 2007. “Incentives in Obesity and Health Insurance.” NBER Working Paper No. W13113. Sabia, J. J. 2007. “The Effect of Body Weight on Adolescent Academic Performance,” Southern Economic Journal 73, 871-900. Sinha, S., and I. Matin. 1998. “Informal Credit Transactions of Microcredit Borrowers in Rural Bangladesh.” IDS Bulletin 29, no. 4:66-80. Stewart, D. W., and P. N. Shamdasani. 1990. “Focus Groups: Theory and Practice.” Applied Social Research Methods Series, Vol. 20, Sage Publications: New York. Todd, H. 1996. “Women at the Center: Grameen Bank after One Decade.” University Press Limited: Dhaka. World Bank. 2003. “Development Policy Review.” Technical Report No. 26154-BD, World Bank: Washington DC. White, H. 1980. “A Heteroskedasticity-Consistent Covariance Matrix Estimator and A Direct Test for Heteroskedasticity.” Econometrica 48, 4:817-838. Zeller, M., M. Sharma, A. U. Ahmed, and S. Rashid. 2001. “Group-based Financial Institutions for the Rural Poor in Bangladesh: An Institutional- and Household-level Analysis.” Research Report No 120, IFPRI: Washington DC.

22

Tables

Table 1: Descriptive statistics of the villages

Variables Notation Mean (Standard Deviation)

Average number of MFIs working in a village MFIno 3.9 (1.4) Average number of big MFIs working in a village 3.6 (1.3) Average number of households in a village 548.3 (426.5) Average number of households in a village with MFI membership 179.7 (148.0) Percentage of households in a village with MFI membership MFI_cover 33.0% (19.0) Average annual moneylender interest rate (%) InterestML 103.33% (59.06) Daily average male wage rate (in Taka) MaleWage 44.37 (8.26) Daily average female wage rate (in Taka) FemWage 30.09 (6.88) Percentage of villages where main use of MFI loan is productive investment

UseMFL 32.7%

Percentage of villages where main use of moneylender loan is productive investment

UseML 10.0%

Percentage of cultivable land that grows single crop a year Crop1 12.8% two crops a year Crop2 50.5% three crops a year Crop3 32.4% four crops a year 4.1% Percentage of households owning less than 10 decimal of land Landcon_10 0.081% In the last 5 years, percentage of villages where poverty incidence increased considerably PovCon 34.6% increased marginally PovSli 21.6% remained the same PovSam 17.0% decreased 26.8% Distance (in kilometer) of center of the village from the nearest bank DisBank, 6.5% (4.1) bazaar DisBazaar 1.9% (1.6) all-weather (paved) road DisRoad 1.7% (1.6) bus stand DisBus 5.7% (4.7) Number of shops in the village* shop 22.4% (39.0) Percentage of households with electricity connection Electr 0.10% (0.14) Percentage of agricultural land irrigated using electricity ElectIri 28.9% (34.1)

* small shops such as temporary grocery shops and tea stalls.

23

Table 2: OLS regression: Moneylender interest rate (InterestML) is the dependent variable) Explanatory variables

(2) (3) (4) (5) (6) (7)

MFI_cover 1.109*** (0.241)

1.092*** (0.256)

1.006*** (0.298)

4.301*** (0.951)

0.874** (0.364)

1.010*** (0.334)

Square of MFI_cover

-0.039*** (0.011)

UseML -23.389 (26.208)

-24.158 (32.151)

-39.193 (30.471)

-12.839 (29.366)

-21.121 (32.565)

UseMFL -14.273 (12.591)

-9.371 (16.457)

-11.775 (14.829)

-8.294 (15.253)

-3.070 (14.697)

Shop 0.056 (0.082)

0.068 (0.091)

0.067 (0.082)

0.067 (0.092)

DisBank 0.203 (2.227)

0.338 (2.024)

0.388 (2.129)

0.259 (2.108)

DisBazaar 7.060 (4.807)

10.513** (4.943)

7.940* (4.683)

9.862** (4.659)

DisBus 0.639 (1.519)

-0.003 (1.471)

-0.310 (1.532)

-0.334 (1.630)

DisRoad -2.529 (3.545)

-0.582 (3.315)

-3.197 (3.187)

-2.612 (3.192)

Electr 27.885 (40.803)

59.267 (44.67)

-0.046 (46.083)

-13.078 (44.988)

ElectIri 0.454** (0.183)

0.380** (0.188)

0.444** (0.181)

0.527*** (0.178)

Crop1 -0.686 (1.331)

-0.971 (1.526)

-0.339 (1.341)

-0.288 (1.356)

Crop2 -0.709 (1.097)

-0.784 (1.291)

-0.428 (1.147)

-0.458 (1.194)

Crop3 -0.638 (1.245)

-0.426 (1.438)

-0.234 (1.272)

-0.198 (1.325)

MaleWage 0.144 (1.451)

-0.181 (1.441)

-0.208 (1.456)

-0.419 (1.452)

FemWage 1.124 (1.620)

0.650 (1.526)

0.557 (1.676)

0.769 (1.641)

PovCon -30.239 (18.625)

-24.259 (17.933)

PovSli 11.943 (16.266)

15.968 (15.344)

PovSam 2.776 (21.401)

2.636 (20.675)

Landcon_10 -198.312* (115.011)

R-square 0.144 0.172 0.310 0.389 0.356 0.391 Sample size 106 106 89 89 89 89

Figures in the parentheses are White (1980) corrected robust standard errors. All regressions include a constant. *** Significant at the 1% level, ** Significant at the 5% level, * Significant at the 10% level.

24

Table 3: OLS regression: Moneylender interest rate (InterestML) is the dependent variable)

Explanatory variables

(2) (3) (5) (6)

MFI_cover 0.930*** (0.255)

1.398*** (0.353)

1.263*** (0.329)

1.085*** (0.307)

UseML -137.760*** (42.422)

-182.270*** (44.691)

-172.904*** (44.834)

-197.799*** (44.437)

UseMFL -18.808 (27.101)

-16.857 (30.681)

-29.760 (29.004)

-30.276 (26.473)

MFI_cover* UseML

3.519*** (1.080)

4.768*** (0.961)

4.766*** (1.011)

5.308*** (1.029)

MFI_cover* UseMFL

0.082 (0.584) -0.014 (0.664) 0.340 (0.718) 0.576 (0.620)

Shop 0.121 (0.087) 0.165* (0.094) 0.154 (0.104) DisBank -0.394 (2.118) -0.523 (2.014) -0.662 (1.973) DisBazaar 7.926 (4.920) 9.405** (4.714) 11.144**

(4.657) DisBus 0.953 (1.428) -0.474 (1.531) -0.087 (1.444) DisRoad -1.830 (3.378) -1.300 (2.946) -1.267 (2.872) Electr 51.631 (40.690) 30.615 (46.246) 13.631 (41.795) ElectIri 0.969***

(0.284) 1.186*** (0.308)

1.046*** (0.282)

MFI_cover* ElectIri

-0.016** (0.007)

-0.021*** (0.007)

-0.014** (0.007)

Crop1 -0.695 (1.379) -0.302 (1.386) -0.208 (1.399) Crop2 -0.701 (1.154) -0.418 (1.220) -0.448 (1.253) Crop3 -0.569 (1.292) -0.057 (1.327) -0.103 (1.363) MaleWage 0.589 (1.292) 0.106 (1.252) 0.117 (1.199) FemWage 0.027 (1.488) -0.645 (1.511) -0.244 (1.430) PovCon -41.126**

(17.949) -32.519* (17.897)

PovSli 10.982 (14.747) 15.509 (14.371) PovSam -14.474

(20.383) -12.472 (19.350)

Landcon_10 -235.068** (90.007)

R-square 0.228 0.408 0.474 0.514 Sample size 106 89 89 89

Figures in the parentheses are White (1980) corrected robust standard errors. All regressions include a constant. *** Significant at the 1% level, ** Significant at the 5% level, * Significant at the 10% level.

25

Table 4: System estimation by heteroskedasticity based identification

Explanatory variables

(2) (3)

MFI_cover 1.834*** (0.393) [0.518] 1.306*** (0.275) [0.413] UseML -25.923 (28.565) -187.510*** (42.816) UseMFL -0.041 (13.126) -26.430 (20.640) MFI_cover* UseML 4.970*** (1.074) MFI_cover* UseMFL 0.463 (0.507) Shop 0.072 (0.086) 0.155* (0.092) DisBank 0.789 (1.948) -0.541 (1.702) DisBazaar 10.649** (4.379) 11.248*** (4.078) DisBus -0.960 (1.532) -0.287 (1.312) DisRoad -0.705 (3.155) -0.868 (2.523) Electr -20.135 (49.521) 12.511 (38.184) ElectIri 0.439** (0.177) 1.072*** (0.255) MFI_cover* ElectIri -0.016*** (0.006) Crop1 -0.386 (1.148) -0.241 (1.191) Crop2 -0.337 (1.019) -0.432 (1.073) Crop3 -0.087 (1.116) -0.070 (1.168) MaleWage -0.301 (1.262) 0.057 (1.049) FemWage 0.987 (1.521) -0.224 (1.243) PovCon -13.597 (16.480) -31.173** (15.310) PovSli 15.065 (15.395) 15.103 (12.288) PovSam 16.120 (20.843) -10.280 (17.120) Landcon_10 -252.138** (102.504) -230.078*** (75.860) Marginal effect of MFI_cover 2.479*** (0.3696) p-value of Hansen J-statistic 0.362 0.166 Sample size 89 89

Figures in the parentheses are White (1980) heteroskedasticity corrected standard errors. Figures in the brackets are White (1980) heteroskedasticity corrected standard errors obtained by bootstrapping. All regressions include a constant. *** Significant at the 1% level, ** Significant at the 5% level, * Significant at the 10% level. Columns 4a and 4b replicate columns 3a and 3b, respectively, but exclude the additional instruments (changes in the incidence of females going out to work).