Liability Structure in Small-Scale Finance: Evidence
             from a Natural Experiment

         Fenella Carpena, Shawn Cole, Jeremy Shapiro, and Bilal Zia


   Microfinance, the provision of small individual and business loans, has experienced
   dramatic growth, reaching over 150 million borrowers worldwide. Much of the
   success of microfinance has been attributed to attempts to overcome the challenges of
   information asymmetries in uncollateralized lending. However, very little is known
   about the optimal contract structure of these loans, and there is substantial variation
   across lenders, even within a particular setting. This paper exploits a plausibly
   exogenous change in the liability structure offered by a microfinance program in
   India, which shifted from individual to group liability lending. We find evidence that
   the lending model matters: for the same borrower, the required monthly loan
   installments are 11 percent less likely to be missed under the group liability setting in
   comparison with individual liability. In addition, compulsory savings deposits are 20
   percent less likely to be missed under group liability contracts. JEL codes: D14, O12,
   O16, O17




Theory and evidence highlight financial market imperfections as a central
cause of poverty and a key impediment to growth (Banerjee and Newman,
1993; Rajan and Zingales, 1998). In theories of capital accumulation, for
example, financial market imperfections influence the ability of the poor to
borrow for investments in education and physical capital. Additionally, in
models that explain entrepreneurship, information asymmetries and transaction
costs prevent the poor from undertaking profitable entrepreneurial activities
because they often have no collateral. The lack of access to financial services

    Fenella Carpena is a PhD student in the Department of Economics at the University of California,
Berkeley; her email address is fenella@econ.berkeley.edu. Shawn Cole is an Associate Professor of
Business Administration at Harvard Business School; his email address is scole@hbs.edu. Jeremy
Shapiro is a Consultant at McKinsey & Company; his email address is jeremypshapiro@gmail.com.
Bilal Zia (corresponding author) is an Economist in the Development Research Group at the World
Bank; his email address is bzia@worldbank.org. This project is a collaborative research effort with
Saath Microfinance. The authors thank Saath, Xavier Gine   ´ , David McKenzie, Petia Topalova, and
workshop participants at the World Bank for helpful comments and suggestions. Stuti Tripathi and the
Center for Microfinance at IFMR provided excellent research assistance. This work was supported by
the World Bank Gender Action Plan, the HBS Division of Faculty Research and Development to S.C.,
and the NSF Graduate Research Fellowship Program to F.C.

THE WORLD BANK ECONOMIC REVIEW, VOL. 27, NO. 3, pp. 437 –469           doi:10.1093/wber/lhs031
Advance Access Publication December 4, 2012
# The Author 2012. Published by Oxford University Press on behalf of the International Bank
for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions,
please e-mail: journals.permissions@oup.com

                                               437
438    THE WORLD BANK ECONOMIC REVIEW



may thus leave many productive opportunities for the poor untapped and may
generate persistent income inequality and low growth (Beck et al., 2007).1
   Microfinance, which is the provision of credit, savings, and other financial
services to low-income households and entrepreneurs, has exploded in popular-
ity and coverage in recent years, particularly in its ability to meet the large
unmet demand for finance (Morduch, 1999; Armenda            ´ riz de Aghion and
Morduch, 2010). Both emerging markets and developed economies, including
the United States, now provide microfinance services through a variety of
public and private channels. The growth of microfinance has been unprecedent-
ed: between 2004 and 2008, the sector’s average annual asset growth rate was
39 percent, reaching US$60 billion in total assets by the end of 2008 (Chen
et al., 2010). A careful evaluation of microfinance in Banerjee et al. (2009)
reveals that microcredit has important effects on business outcomes and the
composition of household expenditures. The rapid growth of microfinance and
its potential for promoting development have attracted the interest not only of
governments, donors, and socially oriented investors but also of mainstream
commercial banks.
   Perhaps the most celebrated feature of microfinance is the group liability
contract, which is a lending methodology pioneered by the Grameen Bank in
Bangladesh. Under this contract, loans without collateral are extended to a
group of borrowers whose members are jointly liable for each other’s repay-
ment. Because groups form voluntarily and group members are responsible for
paying off each other’s debts, borrowers have the incentive to screen for risky
clients, monitor their peers, and enforce repayment. The success of this model
with the Grameen Bank led to its replication in many countries around the
world, reaching more than 150 million individuals by the end of 2007
(Daley-Harris, 2009). This model is particularly important because small firms
suffer the most from institutional weakness (Beck et al., 2005) and because the
structure of the banking sector may have important distributional impacts on
growth (Cetorelli and Gambera, 2001).
   Although most microfinance organizations use group liability, not all do so.
On the one hand, group liability may solve information asymmetry problems
by leveraging social ties and the borrower’s knowledge about the community,
and it may reduce monitoring costs to the lender by motivating borrowers to
monitor each other. On the other hand, social sanctions may be limited, bad
clients may free-ride on good clients, and borrower groups may collude against
the lender. In addition to group liability lending, many microfinance programs
employ a variety of approaches to maintain high repayment rates. For example,
some programs implement frequent repayment schedules and progressive
lending, or they require collateral substitutes. However, very little is known
about the efficiency of these designs in ensuring repayment.


  1. See World Bank (2008) for a literature summary.
                                                           Carpena, Cole, Shapiro, and Zia         439


   The question of an optimal loan contract structure remains largely unan-
swered in both the theoretical and the empirical microfinance literature.
Theoretical studies have primarily focused on explaining how and why the
group liability mechanism works and offering competing predictions on its
benefits, whereas the empirical literature lags behind the theory. Two impor-
tant exceptions are Gine  ´ and Karlan (2009) and Attanasio et al. (2011). Gine   ´
and Karlan (2009) report on a field experiment in the Philippines to test the
effect of individual versus group liability lending. Their analysis focuses on the
importance of peer monitoring and finds no significant difference in default
among individual and group borrowers. Gine     ´ and Karlan identify the effects of
peer monitoring, but they do not focus on the effects of joint liability. In con-
trast, our paper examines the effect of the contract structure on the group of
borrowers who are willing to borrow with either individual or group liability.
Attanasio et al. (2011) conducted a field experiment in Mongolia in which vil-
lages were randomly assigned to obtain access to group loans, individual loans,
or no loans. The main objective of Attanasio et al. (2011) is to measure the
impact of both types of microcredit on different poverty measures. The authors
find a positive impact for group liability loans on food consumption and entre-
preneurship, with no difference in repayment rates between individual and
group liability.
   The identification of the impact of group liability on outcomes such as the
default rate is complicated by the standard problems of selection and omitted
variable bias. Individuals with different financial habits might choose one form
of contract but not the other. Alternatively, lenders with different levels of
sophistication may attract different client mixes and offer different contracts.
One cannot simply compare clients across lending contracts because self-
selection or other aspects of the program may be the cause of any observed
differences.
   In this paper, we use a natural experiment to compare loan repayment and
savings discipline between individual and group lending models.2 In this
setting, group lending differs from individual lending in both the liability struc-
ture and the repayment practices. In group lending, borrowers are liable for the
scheduled payments of group members, and the loan officer interacts primarily
with the group leader, who collects payments from the other group members.
Under individual lending, the borrowers are personally liable and interact
directly with the loan officer. Our empirical strategy takes advantage of a
change in the lending policies of Saath, a non-government organization provid-
ing microfinance services in India. Saath switched from individual to group
lending. This transition was governed by a strict policy rule: after a particular
date, all of the borrowers completing an individual liability cycle were offered
a group liability loan in their next loan cycle. The individual liability loan com-
pletion dates were distributed relatively uniformly throughout the year, offering

  2. Throughout this paper, we use the terms “group liability” and “joint liability” interchangeably.
440    THE WORLD BANK ECONOMIC REVIEW



a natural variation in the timing of the loan contract transitions. Thus, in July,
for example, individual liability borrowers completing a loan cycle would
switch to group liability in the following loan, whereas those whose loan cycle
ended after July would remain under an individual contract setting until the
end of their cycle. This plausibly exogenous change, which was phased in over
time, generates natural control groups and allows us to credibly identify the
causal impact of the group liability structure in what amounts to a repeated
difference-in-difference framework. At any particular point in time, our “treat-
ment” group consists of clients who have fully repaid their individual liability
loan and who currently have a group liability loan, whereas our “control”
group consists of individual liability loan clients who will eventually convert to
a group liability loan.
   Our primary analysis focuses on loan performance and estimates the effect
of group liability on these outcomes. We find that the group liability structure
significantly improves repayment rates. In particular, clients are approximately
11 percent less likely to miss a monthly repayment in the group liability setting
relative to individual liability; this effect holds even with individual fixed
effects. We also find that there is greater discipline regarding the monthly com-
pulsory savings deposits when clients have a group liability loan. Specifically,
compulsory deposits are approximately 20 percent less likely to be missed in
the group liability setting. Our results provide the first credible evidence that
group liability contracts improve upon individual liability, particularly in en-
suring repayment and increasing savings discipline among clients.
   These results, however, are subject to some important caveats. First, the
transition to group liability lending was accompanied by other changes to the
lending structure, particularly an increase in the loan size, which may have
raised the continuation value of borrowing. Nevertheless, we argue that our es-
timates of the impact of these changes in the lending model should be in the
lower bounds. Second, because our empirical strategy focuses on clients who
chose to borrow under both individual and group liability settings, the external
validity of our results may be limited. Third, limitations in the data availability
preclude us from examining loan outcomes such as delinquency or prepayment.
Finally, our empirical strategy does not allow us to test for the specific mecha-
nisms by which group lending improves repayment. Based on our discussions
with Saath, interviews of field officers, and our reading of the evidence, we
speculate that “peer pressure” is the mechanism at work. We discuss these
caveats further in Section IV.
   From a practical and policy perspective, our results are quite timely.
Microlenders worldwide are increasingly weakening joint liability in their
lending approaches (Armenda    ´ riz de Aghion and Morduch, 2010). BancoSol in
Bolivia has shifted a significant proportion of its lending portfolio from group
to individual lending, and even the Grameen Bank has moderated its joint lia-
bility clause, allowing defaulters to get back on track without invoking group
pressure. Therefore, our results suggest a cautionary tale for microfinance.
                                                Carpena, Cole, Shapiro, and Zia   441


Many MFIs are moving away from joint liability to individual liability, but this
transition is not supported by strong empirical evidence. This finding is impor-
tant because, to our knowledge, only two other papers examine the relative
merits of joint and individual liability contracts. Our paper underscores the
fact that more research is required to provide better policy guidance for MFI
practitioners worldwide.
    The rest of this paper is organized as follows. Section I reviews the existing
literature on the liability structure in microfinance. Section II provides the back-
ground for the microfinance program that we study and explains the change in
the liability structure of its loan products. In Section III, we provide a descrip-
tion of the data and summary statistics. We discuss our empirical strategy and
results in Section IV. Finally, Section V concludes the paper.

                   I. PREDICTIONS      OF   GROUP LIABILITY

A wealth of theoretical literature in microfinance explores the mechanisms
behind group liability contracts, particularly in terms of how they mitigate in-
formation asymmetries and enforcement problems. Stiglitz (1990) shows that
the group liability structure overcomes ex ante moral hazard because it creates
incentives for group members to monitor each other’s loans. Similarly,
Banerjee et al. (1994) study credit cooperatives and underscore the role of peer
monitoring. These authors describe a model in which higher monitoring results
in higher borrower effort and, hence, a higher probability of project success.
   Even if a project succeeds, however, borrowers may refuse to repay or may
claim that the project failed so that they can avoid repayment. This type of
strategic default is captured in several theoretical studies on group liability. For
example, Besley and Coate (1995) provide a model demonstrating that joint lia-
bility may harness social capital to increase a borrower’s willingness to repay.
Likewise, Armenda   ´ riz de Aghion (1999) demonstrates that joint liability agree-
ments may reduce the incidence of strategic default because borrowers may
impose social sanctions on the defaulter.
   In addition to examining moral hazard, the theoretical literature investigates
how joint liability mitigates adverse selection. Ghatak (2000) describes a
model in a scenario in which borrowers have ex-ante information about the
riskiness of other borrowers’ investment projects, but lenders do not. Joint lia-
bility thus acts as a screening device that induces “assortative matching.”
Specifically, borrowers with safe investments partner with other safe borrowers,
leaving risky borrowers to form groups among themselves.
   These theoretical models, among others, have shown that group liability
may improve repayment rates by alleviating imperfections in the credit market.
However, whether group liability outperforms other contract structures
remains an open question in the microfinance literature. For example, Besley
and Coate (1995) point out in their model that if borrowers cannot repay as a
group, then some group members will not find it worthwhile to contribute
442    THE WORLD BANK ECONOMIC REVIEW



their share of repayment even though they would have repaid under individual
lending.
   Inconclusive empirical evidence accompanies these ambiguous theoretical
predictions. Some empirical studies support the theoretical advantages of
group liability. For instance, in Bangladesh, Sharma and Zeller (1997) show
that groups that were formed through self-selection had better repayment rates;
however, this study may suffer from omitted variable biases. Other studies
provide little empirical support for this theory. For example, Ahlin and
Townsend (2007) use Thai data to show that repayment rates are negatively as-
sociated with social ties.
   Only a handful of studies examine the merits of group liability relative to
other contract structures. Fischer (2010) conducts a series of lab experiments
with actual microfinance clients and provides evidence that the contract struc-
ture affects project selection. Specifically, he finds that group liability increases
risk-taking relative to individual liability contracts because borrowers free-ride
on the insurance provided by their partners. In a randomized experiment in
India in which borrowers were assigned to either weekly or monthly repayment
meetings, Feigenberg et al. (2010) find that more frequent repayment meetings
build social capital among borrowers, which, in turn, leads to reduced default.
   The most relevant studies on repayment rates under different loan liability
structures are Gine ´ and Karlan (2009) and Attanasio et al. (2011). Gine    ´ and
Karlan (2009) report evidence from two field experiments in the Philippines. In
the first experiment, borrowers who had signed up under a group liability
structure were converted to individual liability. Because both the joint and indi-
vidual liability groups previously underwent the same screening, the authors
can independently identify the peer monitoring effect under group liability.
However, they cannot identify or rule out any impact of screening with this
methodology. In addition, the group repayment and monitoring mechanisms
may be entrenched and difficult to undo, even with an individual liability struc-
ture. Their second experiment randomly introduced either group or individual
liability lending to new borrowers. However, the experiment was conducted at
the loan center level, and take-up was quite uneven between the group and in-
dividual loan centers, resulting in potential statistical power concerns. In both
instances, the authors find that the default rates are invariant to the contract
structure.
   Attanasio et al. (2011) conduct a field experiment in Mongolia where villag-
es were randomly assigned to group loans, individual loans, or no loans. The
authors seek to measure the impact of individual and group loans on reducing
poverty. In particular, the study finds that clients who received group loans had
higher food consumption and were more likely to operate a business than
clients in the control villages. Clients in the individual-lending villages had no
significant increases on any of these measures. For the repayment outcomes,
the authors find no significant differences in the repayment rates between indi-
vidual and group liability.
                                                  Carpena, Cole, Shapiro, and Zia   443


   Although loan default and repayment are the primary outcomes of interest
when examining group liability contracts, the economics literature on rotating
savings and credit organizations (Roscas) suggests that group liability may have
positive effects on savings. Bouman (1995) argues that participating in credit
and savings groups allows individuals to avoid demands for financial support
from their relatives because contributions to a Rosca are generally recognized
by society as senior claims. In a theoretical model, Ambec and Treich (2007)
show that Roscas can serve as a commitment device that helps people to over-
come self-control problems. Gugerty (2007) provides support for this model,
reporting that many Rosca participants in rural Kenya cite “you can’t save
alone” or “sitting with other members helps you to save” as their primary mo-
tivation for participating in a Rosca.
   Our paper complements Gine    ´ and Karlan (2009) by examining the optimal
contract structure in an alternative setting. Although the original experiment in
Gine´ and Karlan (2009) focuses on moving from group to individual liability
contracts, we explore the reverse; that is, we explore the shift from individual
to group liability. The following section describes the setting and our empirical
strategy in more detail.

                             II. EMPIRICAL SETTING

Our partner institution, Saath, is a non-government organization based in
Ahmedabad, India. Founded in 1989, Saath implements development initiatives
in slum communities, including health, infrastructure improvement, and liveli-
hood training programs. Additionally, Saath provides credit and savings servic-
es to the urban poor through its microfinance unit. In 2009, Saath
Microfinance had over 6,400 active clients in 4 branches with a savings portfo-
lio of INR 18 million (USD 390,000) and a loan portfolio of INR 19 million
(USD 410,000).3
   Although Saath has provided mentoring support to community-based credit
and savings groups since the mid-1990s, its microfinance unit was not formally
established until 2002. In that year, Saath integrated these credit and savings
groups into its organization and registered them as cooperative societies with
the Indian government. Saath also began managing these credit and savings co-
operatives at this time, evolving into the Saath Microfinance Unit. Today,
Saath Microfinance provides various financial services to slum communities, in-
cluding voluntary savings accounts, compulsory savings accounts, and group li-
ability loans.

                                    Savings Products
Since its inception in 2002, Saath Microfinance has offeredvoluntary savings
accounts to its clients. These voluntary savings accounts earn an interest rate of

  3. Based on Saath’s 2008–2009 Annual Report.
444     THE WORLD BANK ECONOMIC REVIEW



6 percent per year and do not require a minimum balance. As the name sug-
gests, members are not obliged to make regular deposits into voluntary savings
accounts. Any amount can be deposited, but only six withdrawals can be made
per year.
   In November 2007, Saath Microfinance initiated compulsory savings ac-
counts for its members. Specifically, members are required to deposit INR 100
(USD 2) every month into compulsory savings accounts for the duration of
their membership with Saath Microfinance. Clients may withdraw any amount
from their compulsory savings at any time as long as a minimum balance of
INR 3,500 (USD 70) is maintained. Similar to voluntary savings, compulsory
savings earn an interest of 6 percent per year. Any amount that the client de-
posits over the compulsory savings of INR 100 is deposited into the client’s
voluntary savings account. The goal of the compulsory savings account is to
allow clients to build a financial buffer stock against adverse shocks and to
provide low-cost capital to Saath. These compulsory deposits were mandated
for all borrowers independent of the switch to group liability loans. Hence, all
outstanding loans under both individual and group liability were required to
make compulsory deposits after November 2007. In Section IV, we compare
adherence to these compulsory deposits for the same person as she moves from
individual to group liability.
                                         Loan Products
In addition to savings products, Saath Microfinance provides loans for asset
creation (e.g., house repairs), production (e.g., business working capital), and
consumption (e.g., health, social functions). From Saath’s beginnings in 2002
until November 2007, it provided credit through individual liability loans.
Beginning in November 2007, Saath discontinued its individual liability loans,
instead offering group liability loans to members applying for credit.
   Under the individual liability loan model, a client was required to have been
a member of Saath for at least six months with a savings account to be eligible
for a loan. Members could borrow up to three times their savings account
balance at an interest rate of 18 percent per year.4 These individual-liability
loans generally require no collateral; however, each loan applicant must meet
two requirements. First, the loan applicant must have two “guarantors” who
also have savings accounts with Saath. Second, the combined savings balances
of the loan applicant and the two guarantors must be greater than or equal to
the loan amount applied for. Although the guarantors are, in principle, re-
quired to maintain these savings balances throughout the duration of the loan,
in practice, this rule is not strictly enforced. The guarantors are not eligible for

    4. Microfinance organizations typically quote interest rates in one of two forms: “declining,” the
standard used in developed markets, where the amount of interest due each period is calculated based
on the interest rate and the remaining principal, and “flat,” where the interest payments are calculated
using the original principal amount. Thus a 10 percent “flat” rate is significantly higher than a 10
percent “declining” rate. Saath quotes rates using the standard declining balance approach.
                                                Carpena, Cole, Shapiro, and Zia   445


a loan until the loan that they guaranteed has been fully repaid, but loan repay-
ment is the sole responsibility of the borrower. The borrowers are required to
make monthly installments that cover principal and interest. The monthly prin-
cipal installment is a fixed amount, and because the interest rate reflects a de-
clining balance, the total installment amount ( principal plus interest) varies
every month. If the borrower defaults, Saath reserves the right to seize the bor-
rower’s savings. If the savings are not sufficient to cover the loan, Saath re-
serves the right to take the guarantor’s savings as well. However, in practice, as
an NGO whose mission is to empower the poor, Saath has never seized any of
its individual borrowers’ or guarantors’ savings.
   With the group liability model, however, Saath extends credit to groups of
individuals at an interest rate of 24 percent per year. Four loan size categories
are available to clients: (1) Rs. 3,000–5,000, (2) Rs. 6,000–10,000,
(3) Rs. 11,000–20,000, and (4) Rs. 21,000–30,000. These groups are formed
primarily through self-selection with joint applications submitted to Saath. The
groups are composed of three to six individuals, all of whom must be Saath
Microfinance members. Within each group, several criteria must be fulfilled.
First, at least 50 percent of the group must have been Saath Microfinance
members for at least 6 months and must have at least asavings account.
Second, at least 50 percent of the group must be female. Third, relatives or in-
dividuals from the same household are not allowed in the same group. Finally,
the loan terms must be homogenous across group members; that is, the
number of installments and the monthly installment due dates must be the
same, and the loan amount must not vary widely within each group. As in
the individual liability model, group liability borrowers are required to make
monthly installment payments for both principal and interest, although in this
setting, the total installment amounts ( principal plus interest) are equal every
month. (In the individual liability model, the monthly principal installment re-
payment is fixed, but the interest and, therefore, the installment size vary each
month.) Before any loans are disbursed, the group members are also required
to sign a “mutual agreement form” stating that they are liable to pay each
other’s debts in the event of default or delinquency. Borrower groups who have
defaulted or are delinquent are not eligible to receive another loan from Saath.
                  The Shift from Individual to Group Liability
Saath’s decision to shift from offering individual liability to group liability
loans in November 2007 was driven by a change in the management’s priori-
ties. Saath wanted to lend to more people, provide larger loan amounts, and
expand its microfinance operations geographically, but its lending activities
had become stagnant under the individual liability model. In particular, the
“guarantors” requirement for individual liability loans restricted credit eligibili-
ty because Saath had already reached a point where almost all of its members
were either borrowers or guarantors. Additionally, savings clients were reluc-
tant to stand as guarantors for other clients’ loans, and the loan amounts were
446   THE WORLD BANK ECONOMIC REVIEW



limited to 3 times the total savings account balance of the borrower. Saath’s
management thus shifted to group liability loans to overcome these restrictions
in its individual liability model. In terms of the models discussed above, the
limited ability of Saath members to pledge savings as collateral prevented Saath
from expanding, and Saath saw group liability as a way to solve this problem.
In the year following this change, Saath gained almost 800 new clients and in-
creased its reach from 11 to 20 wards.
   The transition from individual liability to group liability loans was imple-
mented using the following rule. Beginning in November 2007, all new loans
disbursed were group liability loans; Saath would no longer disburse individual
liability loans. However, existing loans whose terms lasted beyond November
2007 were unaffected. For example, individual liability loan clients who com-
pleted their loan in February 2008 continued under the individual liability con-
tract until then. After February 2008, if they chose to borrow again, they
received a group liability loan. The date to switch from individual to group lia-
bility was therefore determined by the individual liability loan completion
dates. These completion dates and the subsequent conversion to group liability
loans were distributed relatively uniformly throughout the year.
   Although Saath’s loan product moved from individual to group liability be-
ginning in late 2007, the location for repayments, the frequency of loan repay-
ment collection, and the salaries for the field officers remained similar across
time in our dataset. An empirical concern for the subsequent analysis is
whether Saath’s policy shift from individual to joint liability was accompanied
by a shift in its loan collection techniques. In June 2010, we conducted short
interviews with Saath field officers regarding repayment collection. We were
able to interview 10 out of the 12 Saath field officers who collected repayment
among the sample of clients we study; the remaining field officers are no longer
with Saath. The results from this survey confirm that the location and frequen-
cy of loan repayments remained the same across the two settings. Specifically,
for both individual and group liability loans, we find that field officers collected
repayment at the client’s household every month 100 percent of the time.
Additionally, 90 percent of the time under both regimes, the lender turned
down future loan requests from defaulting clients.
   For group liability borrowers, Saath does not require a group to designate
one of its members as the “group leader.” However, in practice, all groups
have a leader who is in charge of collecting repayments from the other
members. Hence, for individual borrowers, the field officers visit each bor-
rower, whereas for group borrowers, the field officers typically visit only the
group leader’s household. If any member of a particular borrower group fails
to make a scheduled payment, the field officer assembles all of the group
members together and collects the installment amount from the other
members, as stipulated in the group liability contract. Both individual liability
borrowers and joint liability borrower groups in default are not granted future
                                                            Carpena, Cole, Shapiro, and Zia        447


loans. In addition, over the period that we study, there was no change in wages
among the field officers, who continued to receive a fixed monthly sum.5
   In summary, borrowing conditions changed in the following ways: (1) bor-
rowers were obliged to enter joint liability lending groups rather than borrow-
ing independently from Saath; (2) the loan officers collected repayment from
the group leader rather than from each individual borrower; (3) monthly in-
stallments became fixed rather than varying; and (4) the interest rate increased
from 18 percent to 24 percent. Given that multiple dimensions of the contract
structure were changing, we discuss the implications of each of these changes
on our analysis in Section IV.

             I I I . D ATA CO L L E C T I O N     AND    S U M M A RY S TAT I S T I C S

In this study, we use data from Saath Microfinance’s administrative software
systems. The loan data cover outstanding loans every month from April 2005
through March 2009. Because the change in the type of loan contract occurred
in November 2007, the data contain more than two years of monthly data on
individual liability loans and more than one year on group liability loans. Data
are available electronically from only 2 of the 4 Saath Microfinance branches,
Behrampura and Vasna; we focus on these branches. These two branches are
the largest and the oldest, accounting for the vast majority of Saath’s clients.
   The data were maintained for accounting purposes to record cash flowing in
and out of each branch. The data are therefore of very high quality. The data
do not, however, contain information on the terms of each loan, such as matu-
rity dates, installment amounts, and amounts outstanding. These data were re-
corded by the loan officers in the client passbooks and administrative ledgers.
As a result, we are unable to examine the overdue amounts, prepayments, and
similar measures.6
   The data on loans cover April 2005 to March 2009, but savings data were
only available from January 2008 to March 2009. The savings data include the
monthly aggregate deposit and withdrawal amounts for the compulsory savings
accounts.

    5. After March 2009, field officers received 1 percent of the loan interest they collect. Surveys of
field officers indicate that they were not aware of this change in compensation structure before it
occurred. Nevertheless, to isolate the focus of our study on contract structure, we exclude months after
March 2009 in the analysis.
    6. The loan data come from three software systems that Saath Microfinance has used at different
points in time. Each of the two branches in our dataset used a separate system until early 2008, when
the current system was introduced in both branches. Because client identifiers were not carried over
from one software system to another, we had to rely on using client names to track individuals over
time. These names were unique because they included first, middle, and last names. In identifying clients
across systems, 80 percent of the clients had exact name matches, while 14 percent had to be matched
by hand due to name spelling errors. The remaining 6 percent, however, could not be matched to the
current software system. It is likely that these clients have withdrawn their membership with Saath
Microfinance and therefore have not borrowed under the group liability setting because Saath migrated
information from the previous to the current software system only for existing members.
448      THE WORLD BANK ECONOMIC REVIEW



   As previously described, in our main analysis, we study borrowers who re-
ceived both individual and group liability loans to overcome the selection
problem. Hence, in our dataset, these clients begin with an individual liability
loan and, after November 2007, receive a group liability loan. In Saath’s
Behrampura and Vasna locations, we identified a sample of 276 such clients,
representing 22 percent of the loan client base in these two branches as of
March 2009.
   Table 1 provides the summary statistics for our sample. Collectively, these
clients received a total of 748 loans from Saath, of which 450 were individual
liability loans and 298 were group liability loans. The average individual liabil-
ity loan amount was approximately INR 10,000 (USD 220), and the group lia-
bility loan average was approximately INR18,000 (USD 390). Figure 1 plots
the number of group liability loans disbursed over time. As the figure shows,
the borrowers in our sample, all of whom received individual liability loans,
switched to group liability loans in varying months. Our empirical strategy
takes advantage of this staggered timing to compare individual liability loan
clients who received group liability loans to future recipients to identify the
impact of group liability on loan repayment behavior and savings discipline.


                   I V. E M P I R I C A L S T R A T E G Y   AND   AN A LY S I S

                                       Empirical Strategy
To study the effect of contract structure on lending outcomes, we exploit the
natural experiment provided by Saath’s change in policy. The presence of an
exogenous policy change is important. Without exogenous variation, it would
be very difficult to determine whether differences in outcomes were attributable
to contract structure or to any number of other unobservable characteristics of
either the borrowers or the lending institutions. Indeed, theory predicts that dif-
ferent contracts are optimal for different types of borrowers.
   To overcome the selection problem, we focus our attention on the Saath bor-
rowers who received both individual and group liability loans. We exploit the


T A B L E 1 . Summary Statistics
                                              Individual Liability           Group Liability

              Total No. of   Total No. of   No. of    Ave. Loan Amt      No. of   Ave. Loan Amt
Branch          Clients         Loans       Loans          (Rs.)         Loans         (Rs.)

Behrampura        198            512          303           9981.1        209        19081.34
Vasna              78            236          147           9927.211       89        16764.04
Full Sample       276            748          450           9963.014      298        18389.26

   Notes: This table reports the summary statistics for the borrowers in our sample. These bor-
rowers received both individual liability and group liability loans.
   Source: Authors’ analysis based on data sources discussed in the text.
                                                        Carpena, Cole, Shapiro, and Zia       449


F I G U R E 1. Group Liability Loans Disbursement in Sample




    Notes: This figure plots the number of group liability loans disbursed over time in our sample.
All of the borrowers were individual liability borrowers who subsequently received a group
liability loan after the policy change.
    Source: Authors’ analysis based on the data sources discussed in the text.



natural phasing-in of group liability in what amounts to a repeated
difference-in-difference framework. At any point in time, our “treatment”
group thus consists of clients who have fully repaid their individual liability
loan and who currently have a group liability loan, whereas our “control”
group consists of individual liability loan clients who will eventually convert to
a group liability loan. Specifically, we estimate the following equation:

                               yilt ¼ a þ bTil þ gi þ dt þ 1ilt                               ð1Þ

where the subscript i refers to individuals, l refers to loans, and t refers to
months. T is a dummy variable equal to 1 if loan l of client i is a group liability
loan and 0 if it is an individual liability loan. yilt is a measure of loan repay-
ment or savings discipline. The estimate of b then provides the effect of switch-
ing an individual who is already borrowing to the group liability loan. We
include time effects dt because the conversion to group liability loans was stag-
gered across individuals, and the individual fixed effects gi absorb the time-
invariant characteristics of each borrower.
                                         Limitations
In what follows, we note some features of our setting that may limit the gener-
alizability of the empirical results.
   Concurrent Changes in Loan and Savings Products. As discussed in Section
II, the shift from individual to group liability lending contracts was
450    THE WORLD BANK ECONOMIC REVIEW



accompanied by other changes in the contract features. Specifically, in the
group liability setting, the loan officers collected repayment from the group
leader rather than from each individual borrower, monthly installments
became fixed rather than varying, and the interest rate increased from 18
percent to 24 percent. Furthermore, simultaneous with the change to joint lia-
bility loan products, the savings rules shifted because Saath began requiring all
of its members to maintain compulsory savings accounts.
   In our view, an ideal experimental evaluation would include (1) requiring
borrowers to enter joint liability lending groups rather than borrowing on their
own and (2) loan officers collecting repayment from the group leader rather
than from each individual borrower, leaving the installment sizes and the inter-
est rate fixed. The group lending contracts offered by the majority of MFIs in
India collect repayment either from one person (an assigned leader) or from
every borrower in the group simultaneously. Hence, we believe that the change
in the mode of payment is a feature of the group liability contract.
   Although the interest rate change and the change in monthly installments
are not typical, our regression coefficients, which estimate the impact of the
contract change, are likely to be lower bounds. In theory, the increase in the in-
terest rate could have several effects: a price effect might reduce demand,
whereas a higher interest rate could increase the repayment burden and induce
default. Most evidence suggests that microfinance borrowers are not very price
elastic, so we are not overly concerned about the demand effects. The increase
in the interest rate should bias us against a finding that joint liability lending
reduces default. Furthermore, our empirical analysis only considers individuals
who borrowed under both the individual and group liability regimes, thus ac-
counting for any self-screening among clients based on the increase in interest
rates between the two loan contracts.
   The change in monthly repayment installments and the mode of payment
also warrants further discussion. The repayment schedule for individual liabili-
ty loans required fixed principal repayments along with interest. Hence, the
nominal size of the monthly payments declined over the cycle of the loan. In
contrast, the group liability repayment structure is based on a fixed monthly re-
payment throughout the term of the loan. Because our analysis focuses on the
shift from individual to joint liability loans for the same person, for the same
loan amount, we pick up the effect of a lower payment under individual liabili-
ty (because the borrower is at the end of her loan cycle) versus a relatively
higher fixed payment under joint liability. This shift should bias us against
finding a reduction in default. This bias effect is likely even greater in our case
because the average loan size and the corresponding repayment installment size
are higher under joint liability.
   The mode of payment also shifted under the individual and group liability
settings. In the former, the field officer visited each individual liability borrower
to collect repayment, whereas in the latter, the field officer only visited the
group leader, who was in charge of collecting repayment from the other
                                               Carpena, Cole, Shapiro, and Zia   451


members. Nevertheless, this change in the mode of payment is a feature of
group lending because in the setting that we study, group liability is a lending
contract that involves both joint repayment to a group leader and joint liability.
Thus, the “group” features, such as repayment to a group leader, may lower
default, whereas the higher interest rates may increase default, so the effect
that we capture may well be a lower bound.
   Saath’s savings products also changed during our study period, as discussed
in Section II. However, we note first that in our main analysis of loan repay-
ment and savings discipline, we focus only on the individuals who converted to
group liability loans, exploiting the timing of their switch. This focus allows us
to control for any changes that occurred at the microfinance institution level
under the two loan contract regimes. Thus, in the context of the MFI-wide
change in savings requirements, we are comparing a shift from individual to
joint liability for the same person (when we include individual fixed effects)
who faces mandatory savings under both liability structures. Because our em-
pirical strategy rests on the continuous, rolling changeover from individual to
joint liability after the announcement, our sample consists of borrowers who
are opening and maintaining mandatory savings accounts prior to shifting to a
joint liability loan. Although the introduction of a mandatory savings account
may have influenced the composition of borrowers, the internal validity of our
results remains unaffected because our analysis considers only those individuals
who chose to renew their loans with Saath.
   Theoretically, the imposition of mandatory savings may have two opposing
effects: (a) it may discourage borrowers from renewing their loans because the
real cost of borrowing has increased through the imposition of a mandatory
savings plan; or (b) it may encourage borrowers to renew their loans because
individuals appreciate the saving discipline provided by compulsory accounts.
This latter point is not trivial. Individuals may fail to save enough because they
consistently put off saving for their future (Laibson, 1997), they may be
tempted to spend on immediate consumption (Banerjee and Mullainathan,
2010), or they may face intra-household constraints (Ashraf et al., 2010). In a
recent field experiment, Atkinson et al. (2010) find that prompting individuals
to save at the time of loan repayment doubles the amount of savings.
   The overall effect of mandatory savings on borrower selection is therefore
ambiguous. Unfortunately, we lack any household-level data that would allow
us to empirically differentiate these effects. In addition, we only have basic
socioeconomic data from Saath records for the clients who eventually joined
the joint liability groups. Nevertheless, we run a simple regression of renewal
on the percentage payments missed and find a strong statistically significant
negative coefficient. Clearly, there is screening based on past loan performance.
However, we also find that this screening occurs even for previous
individual-to-individual renewals and therefore cannot be considered an effect
of either joint liability or mandatory savings.
452    THE WORLD BANK ECONOMIC REVIEW



EXTERNAL VALIDITY. Our study sample consists of individual borrowers who
have repaid their individual loans and choose to borrow under joint liability.
We therefore estimate the effect of joint liability on improving repayment rates
among those who choose to borrow under joint liability. Although restricting
our analysis to this sample may compromise external validity, we believe that
this sample is highly relevant; measuring the effect on those who decline to
borrow under joint liability would have little relevance for the outside world.
Of course, the sample is also selected based on individuals who chose to
borrow in an individual liability setting. However, given that most theory sug-
gests that joint liability leads to stricter screening and stricter monitoring, this
additional screen may not be that restrictive.
   The setting that we study may be anomalous in that the typical transition in
the microfinance industry is presently the reverse, but we believe that careful
study of the effects of shifts in liability in either direction is informative and
valuable. Indeed, the recent collapse of Banco del Exito (BANEX), one of the
largest microlenders in Nicaragua, highlights the importance of examining the
relative merits of group liability and individual liability contracts. Furthermore,
the main question that we ask—whether conditional on borrowing, liability
rules impact repayment performance—has important policy and theoretical im-
plications because we consider how joint liability lending may improve upon
individual liability lending. It is difficult to imagine how any single study could
capture both the compositional effects and the effect of contract structure on
those who have borrowed in individual and joint liability settings.
   By focusing on clients who borrowed under both individual and joint liabili-
ty, our study further highlights the importance of examining the impact
of lending contracts on financial inclusion. In the setting that we study, almost
80 percent of the borrowers who completed their individual liability loan and
could have borrowed under joint liability did not do so. Although this high
percentage indicates that it is possible for the shift in joint liability to have neg-
ative consequences on financial inclusion, we do not have data on why people
chose to borrow or not to borrowwith Saath. Many new clients (who had not
previously borrowed under individual liability) joined Saath after joint liability
and may have chosen to stop borrowing for a number of reasons (e.g., no
further project/investment needs, shifting business or employment status).
Without further data, we cannot determine the impact of joint liability on fi-
nancial inclusion.
   Finally, although we find strong evidence that joint liability improves upon
individual liability lending in terms of repayment behavior, we cannot be
certain that the treatment effect is similar for other MFIs. The MFI that we
study, Saath, has operations that are fairly typical of small MFIs around the
world. However,we evaluate a particular joint liability lending program in
which borrower groups have a group leader who collects repayment for the
group as opposed to public repayment (e.g., at the village center), a collection
                                                Carpena, Cole, Shapiro, and Zia   453


method that may bemore common among other MFIs. In this sense, the exter-
nal validity of our findings may be limited.

DATA LIMITATIONS. Because the data that are available were used primarily for
accounting purposes, the dataset does not contain information on the terms of
each loan, such as maturity dates, installment amounts, and outstanding loan
amounts, all of which were recorded by loan officers in paper ledgers. Thus,
our analysis is limited to observing whether a client made a loan repayment or
a compulsory savings deposit for a particular month, and we are unable to con-
sider outcomes on overdue loan amounts, prepayments, and other measures.
Furthermore, aside from data on gender and the client’s neighborhood, the
data do not contain information on other demographic or household character-
istics of the clients.
    Because of data limitations, our ability to understand why borrowing
amounts increase is limited. Under individual liability, borrowers were limited
to loan amounts that were a fixed proportion of their savings and their guaran-
tors’ savings. These restrictions were removed under joint liability and replaced
with a strict appraisal process for group members. Ideally, if we had deposit
data under individual liability, we could determine whether these borrowing
constraints were binding under individual liability. Unfortunately, Saath did
not keep good historical records of these data; hence, it is not possible to statis-
tically distinguish demand and supply effects on loan size.
                Effect of Lending Structure on Loan Repayment
We now turn to the critical question of loan repayment. We note that the joint
liability structure will, in theory, induce not only better screening but also
greater monitoring efforts. Our empirical design does not distinguish between
the two potential causes of improved repayment but rather estimates the com-
bined causal effect.
   Table 2 presents the OLS estimates of Equation 1, where the outcome of in-
terest is a dummy variable for a missed payment. This dummy variable indi-
cates whether the client failed to make a repayment for a particular month.
Saath Microfinance clients are required to make monthly repayments until the
principal balance is paid in full, beginning 30 days after disbursement. Hence,
the dependent variable takes the value of 1 for a particular month if the total
amount repaid by the borrower for that month is nil and 0 otherwise. We use
this dummy variable as our first measure of monthly loan repayment discipline.
   In our sample, almost 20 percent of the individual liability but only
0.1 percent of the group liability monthly loan repayments were not made.
Our main results are presented in Table 2. Column (1) presents the regression
results with no fixed effects, controlling only for which branch the borrower
uses. Taken at face value, group lending reduces the probability of missing a
payment by 17.5 percentage points. In columns (2)-(4), we add individual fixed
effects, calendar month fixed effects, and both sets of fixed effects, respectively.
                                                                                                                                                              454
T A B L E 2 . Dependent Variable: Missed Payment Dummy




                                                                                                                                                              THE WORLD BANK ECONOMIC REVIEW
                                          (1)                  (2)                  (3)                  (4)                  (5)                 (6)

Group Liability Loan Dummy             2 0.175***           2 0.163***          2 0.152***           2 0.112***          2 0.159***            2 0.081**
                                        (0.014)              (0.012)             (0.042)              (0.026)             (0.015)               (0.033)
Behrampura Branch                        0.067***                                 0.072***                                 0.067***
                                        (0.016)                                  (0.015)                                  (0.016)
Number of Previous Loans                                                                                                 2 0.008                0.045***
                                                                                                                          (0.007)              (0.017)
Loan Age in Months                                                                                                       2 0.028***           2 0.014**
                                                                                                                          (0.007)              (0.007)
Loan Age Squared                                                                                                           0.003***             0.002***
                                                                                                                          (0.001)              (0.001)
Loan Age Cubed                                                                                                           2 0.000***           2 0.000***
                                                                                                                          (0.000)              (0.000)
Constant                                0.128***             0.091***             0.217**             0.181***             0.167***             0.272***
                                       (0.013)              (0.004)              (0.087)             (0.086)              (0.020)              (0.086)
Control for Calendar Month            No                   No                   Yes                 Yes                  No                   Yes
Individual FEs                        No                   Yes                  No                  Yes                  No                   Yes
R-squared                               0.082                0.229                0.098               0.246                0.095                0.253
N                                    6055                 6055                 6055                6055                 6055                 6055
Mean of Dep Var                         0.105                0.105                0.105               0.105                0.105                0.105

    *p , 0.10, **p , 0.05, ***p , 0.01.
    Notes: This table reports the results from the OLS regressions using panel data from April 2005 to March 2009, where the dependent variable is a
dummy for missing a payment. This dummy takes the value of 1 for a particular month if the total amount repaid by the borrower for that month is nil
and 0 otherwise. The observations are at the loan-month level. The sample includes the loans of clients who received both an individual and a group lia-
bility loan. The variable Group Liability Loan Dummy is a dummy variable equal to 1 if the loan is a group liability loan and 0 if it is an individual lia-
bility loan. Behrampura Branch is a dummy for one of the two MFI branches in the sample. Number of Previous Loans is the number of loans that the
client received before the current loan. Loan Age in Months is the difference between the month of the observation and the month when the loan was dis-
bursed, plus 1. Standard errors, clustered at the individual level, are given in parentheses beneath each point estimate.
    Source: Authors’ analysis based on data sources discussed in the text.
                                                             Carpena, Cole, Shapiro, and Zia        455


Finally, in columns (5) and (6), we add controls for the age of the loan: repay-
ments may be higher early in the cycle, when borrowers are flush with cash, or
later in the cycle, when borrowers seek to repay a loan to obtain a new one.
The coefficient drops, although only coefficients (1) and (6) have 95 percent
confidence intervals that (barely) do not overlap. Because only overdue loans
last over 12 months, the loan age coefficients may “soak up” some of the treat-
ment effect, particularly when individual and month fixed effects are present.7
   Our preferred point estimate is column (4), which indicates that group
lending reduces the probability of a missed payment by 11.2 percentage points.
This effect is large and meaningful and may have significant implications for
the profitability of a lender.

INTERNAL VALIDITY. We conduct a number of robustness tests. First, we report a
direct “falsification” test of our analysis by using data from our clients’ previ-
ous individual-to-individual loan renewals. Specifically, we focus on the clients
in our sample who had at least two individual liability loans. The sample is
reduced because many clients did not have multiple loans in the past. Among
these clients, we study whether the borrowing experience with the microlender
is related to loan repayment; in other words, does having a second loan lead
borrowers to repay more reliably? Table 5 presents these results and shows no
significant effect on missed payments.8
   A second concern regarding the internal validity of our analysis is the fact
that joint liability contracts are a completely new contractual arrangement. In
particular, because the arrangement is new, clients may be in a “honeymoon”
period.9 During this period, clients may be on their best repayment behavior
while they are learning the rules of the game. As they gain a betterunderstanding
of the consequences of missing a payment, they may begin to behave more stra-
tegically. To test for this “honeymoon” effect, we compare the default rates of
new clients (i.e., first-time Saath borrowers) under the individual and joint liabil-
ity regimes over our sample period. Specifically, we find that first-time Saath bor-
rowers make late payments 49 percent of the time in the individual liability
setting and 2 percent of the time under joint liability. Because we are comparing
clients who are borrowing for the first time in either setting, this result suggests
that the “honeymoon” effect does not drive clients’ repayment behavior.


   7. A simple way to address the relationship between loan age and repayment status is to to restrict
the sample to the first twelve months of repayment data: doing so with the same specifications as
reported in Table 2 yields point estimates ranging from 2 .074 to 2 .153, statistically indistinguishable
from each other, but all statistically different from 0 at the one percent level.
   8. The specific date (November 2007) for the falsification test was not chosen arbitrarily (rather, it
was precisely one year prior to the actual change in date), but we have conducted the analysis for all
months at least one year before the policy change and find that our effect is dramatically larger than at
any other date. Specifically, we re-ran our specification with each of the previous twelve months as our
placebo date and found only two cases to be significant, but of much lower magnitude.
   9. We thank an anonymous referee for raising this point.
456    THE WORLD BANK ECONOMIC REVIEW



   A third concern is that a client’s propensity to repay may be correlated with
the time in the loan cycle. Specifically, clients may be more likely to make re-
payments toward the end of their last individual liability cycle to ensure their
eligibility for a group liability loan in the future. We note that this situation
would bias estimates against finding that group liability improves borrower per-
formance. Nevertheless, we investigate this possibility using an event-time re-
gression with the dependent variable for missed payment as previously
described, where the event is the conversion from an individual liability loan to
a group liability loan.
   Figure 2 plots the coefficients for each event-time dummy. The first month
of repayment in the group liability setting is time ¼ 0, the final repayment
month for the individual liability loan is time ¼ 2 1, the second-to-last individ-
ual liability loan repayment month is time ¼ 2 2, and so on. Thus, the figure
describes loan repayment behavior under the individual liability contract
before switching to group liability. Saath requires its borrowers to pay their
current loan in full before they are given their next loan. Therefore, by defini-
tion, all clients in our sample made a repayment at time ¼ 2 1. Examining the
periods in which time  7 2 2 reveals no pattern to support the idea that
clients strategically repaid their individual liability loans so that they could
borrow under the group liability setting.
   Alternatively, clients may be more likely to make repayments early in the
loan cycle because they may be flush with cash from a recent loan disbursal.
We examine this possibility, again using an event-time regression, as shown in
Figure 3. We estimate how repayment rates change around loan renewal times
when a client pays off an individual liability loan and renews another individu-
al liability loan (blue line) as well as cases in which a client pays off a group li-
ability loan and renews a second group liability loan (red line). Note that the
first month of repayment in the second loan cycle is time ¼ 0, and the final re-
payment in the first loan cycle is time ¼ 2 1. Similar to Figure 2, at time ¼ 2 1,
all of the clients made a repayment by definition, so the missed payment
dummy must mechanically equal zero. Figure 4 shows that the prior missed
payments are uncorrelated with the number of months since loan origination.
   Our study sample consists of individual liability clients who chose to renew
their borrowing under the group liability setting. These clients may be better at
repayment than borrowers who did not want to enter into a group liability
loan contract. However, our analysis focuses exclusively on those who renew
and includes individual fixed effects. Hence, an interpretation of our results is
that even “good” clients exhibited higher repayment discipline under the group
liability setting in comparison with the individual liability setting. However,
we acknowledge that by focusing only on those clients who borrowed under
both types of contracts, we limit the external validity of our results.
   The outcome that we have considered thus far, whether a client missed a
loan installment for a particular month, is a rough measure because repay-
ments may be partial. That is, a client may have repaid an amount greater than
                                                       Carpena, Cole, Shapiro, and Zia      457


F I G U R E 2. Event Time Regression: Missed Payment on Event Time Dummies
of Switching from Individual to Group Liability Loan




    Notes: This figure plots coefficients for the event-time dummies where the event is the
conversion from an individual liability to a group liability loan. The dependent variable is a
dummy for missing a monthly repayment, which takes the value 1 for a particular month if the
total amount repaid by the borrower for that month is nil and 0 otherwise. The first month of
repayment in the group liability setting is time ¼ 0; the final repayment month in the individual
liability setting is time ¼ 2 1. The dashed lines indicate the 95 percent confidence interval.
    Source: Authors’ analysis based on the data sources discussed in the text.



zero but less than the required installment amount. Thus, another measure of
repayment discipline is the standard deviation of the principal amount repaid
for individual liability loans and the total amount repaid for group liability
loans. As described in Section 2, only the principal installment amount is fixed
in the individual liability setting, whereas in group liability, the required total
installment amount ( principal plus interest) is equal every month. If the
458     THE WORLD BANK ECONOMIC REVIEW



F I G U R E 3. Event Time Regression: Missed Payment on the Event Time
Dummies for Switching from the First to the Second Loan Cycle




    Notes: This figure plots the coefficients for the event-time dummies where the event is the shift
from the client’s first loan cycle to the second loan cycle for each of the individual and group
liability loans. The dependent variable is a dummy for missing a monthly repayment, which takes
the value 1 for a particular month if the total amount repaid by the borrower for that month is
nil and 0 otherwise. The first month of repayment in the second loan cycle is time ¼ 0, and the
final repayment month in the first loan cycle is time ¼ 2 1. The dashed lines indicate the 95
percent confidence interval.
    Source: Authors’ analysis based on the data sources discussed in the text.


required amount is repaid each month, then the standard deviation would be
zero. However, if there are many months in which people pay less or more
than the required amount, then the standard deviation would be higher.
Table 6 provides OLS estimates in which the dependent variable is the standard
deviation of repayment. Again, our estimates show that there is greater loan
                                                          Carpena, Cole, Shapiro, and Zia       459


F I G U R E 4. Calendar Month of Loan Origination and Missed Payments




   Notes: This figure plots the percentage of monthly repayments that were missed using the first
three repayments from the loan disbursement of the client’s most recent individual liability loan.
The calendar month of loan origination refers to the calendar month when the loan was
disbursed. The size of the bubbles represents frequencies. The red line represents the best-fit line.
The sample includes clients who received both an individual and a group liability loan.
   Source: Authors’ analysis based on the data sources discussed in the text.



repayment discipline in the group liability setting relative to individual liability,
although this effect is not statistically significant.
   Finally, we note that the average loan size increased from 10,000 INR under
individual liability to 18,000 INR under joint liability (see Table 1). This in-
crease is large and warrants further discussion. In some ways, the increase can
be thought of as an effect of the group lending model because Saath was
willing to extend larger loans to borrowers on the strength of social collateral.
Indeed, Saath management told us that they were willing to give larger loans
precisely because of the joint liability framework.
   However, one may reasonably wonder whether the increase in loan size
itself affects repayment rates. For example, if borrowers invest in convex pro-
duction technologies, higher credit limits could increase repayment. We test for
this possibility in two ways. First, we split the sample into four quartiles based
on the percentage increase in the credit limit from which a borrower benefited
when she or he converted to joint liability lending. We do not observe any sys-
tematic variation in the treatment effect estimate along this dimension (results
not reported). Second, because our dataset includes the reported purpose of
each group liability loan, we can conduct separate analyses for loans taken for
the purpose of consumption, productive activities, and asset creation. Again,
we find no evidence that the treatment effect varies across these three categories
(results not reported).
460    THE WORLD BANK ECONOMIC REVIEW



DISCUSSION. Given these results, a natural question that arises is why group lia-
bility outperforms individual liability for clients who are already borrowing.
Although the guarantors requirement in individual liability contracts provided
incentives for guarantors to monitor loans and enforce repayment, in practice,
these incentives were quite weak; the microlender rarely seized the savings of
the guarantors of defaulting clients and did not strictly enforce that the guaran-
tors maintain their savings account balance. The microlender collected repay-
ment from the guarantors only if all other options (e.g., seizing the defaulting
borrower’s savings, revisiting the defaulting borrower, threatening to charge
penalties, rescheduling the loan, and having the branch manager intervene) had
been exhausted. In contrast, the group liability structure strengthens coopera-
tion and trust among the group members, as indicated by the fact that almost
all of the joint liability borrower groups designate a group leader, even though
it is not required. It is possible that having a group leader increases incentives
for monitoring and enforcing repayment. For example, having a group leader
may create a person of authority who can enforce repayment or impose sanc-
tions in the event of default.
    We also note that the borrowing requirements for individual liability loans
suggest that the effects that we find would likely be smaller in magnitude if
these requirements were not in place. Because individual liability loan borrow-
ers must have two guarantors and because Saath may use the guarantors’
savings in the event of default, it is possible that these requirements encourage
the screening of potential individual liability loan borrowers. We are not able
to control for such effects.
                               Savings Discipline
We now turn our attention to the compulsory savings deposits required by
Saath. As discussed in Section I, we might expect savings to be higher in the
joint liability setting because participating in a borrower group may allow indi-
viduals to avoid financial demands from their families or to overcome self-
control problems. Furthermore, compulsory savings are required to continue as
a member in good standing with Saath; those who do not meet the compulsory
savings requirements are not allowed to borrow until these requirements have
been met. Thus, the same form of peer pressure that applies to loan repayment
may apply to compulsory savings.
   Although Saath initiated both a shift to group liability lending and compul-
sory deposits in November 2007, we can separately identify the effect of group
liability on compulsory savings by exploiting the time-series variation in loan
renewals. Specifically, although the compulsory savings were mandated across
the board for all borrowers after November 2007, the shift from individual to
group liability was staggered depending on when each individual loan term
expired. As explained previously, these renewals were distributed relatively uni-
formly throughout the year, resulting in variations in loan contracts at a time
when compulsory savings were uniform. Hence, we can study adherence to
                                                           Carpena, Cole, Shapiro, and Zia        461


compulsory savings for the same person who borrowed under an individual lia-
bility contract after November 2007 and who eventually converted to a group
liability contract.10
   Table 3 presents the OLS estimates in which our dependent variable is a
dummy for missing a compulsory savings deposit. The dependent variable
takes the value of 1 if the client deposited less then INR 100 and 0 otherwise.
The point estimate in column (4), which includes month and individual fixed
effects, indicates that the same borrower is 20.5 percentage points less likely to
miss a compulsory deposit in a group lending arrangement than when borrow-
ing individually.
   This finding suggests that one possible mechanism through which group lia-
bility reduces loan delinquency may be increased savings; a greater savings
balance may provide a buffer for borrowers hit with liquidity shocks.

                                    Heterogeneous Effects
We test for heterogeneous effects along two dimensions. As before, our depen-
dent variable is a dummy for whether a client missed a loan repayment for a
particular month. Columns (1)-(3) indicate a significantly larger impact of
group liability in reducing missed payments among men, although the effect
for women remains negative and significant. It is important to note, however,
that the control group means for females are also significantly lower, with
missed payment rates at 16 percent for females and 23 percent for males.11
The results suggest that group liability effectively neutralizes this gender differ-
ential in missed payments.
   In columns (4)-(6), we examine whether group lending improves repayment
behavior more for clients who initially exhibited poor repayment discipline
under individual liability. We define ‘borrower quality’ as the percentage of
missed payments in the client’s first individual liability loan in the data, and we
divide the sample in two along this measure. Note that because our definition
of borrower quality makes use of a client’s repayment behavior in her first indi-
vidual loan, the regression columns (4)-(6) are restricted to the subsample of
clients who had at least two individual loans and include repayment data only
from the client’s second individual loan onwards.
   In terms of past missed payments, we find that group liability has a larger
impact on borrowers who, at the outset, were of poor quality. Clients who
missed 10 percent of their first individual loan monthly payments were 2.4
percent less likely to miss repayments under the group liability regime. Hence,
similar to the gender results, the introduction of group liability is effective in
reducing missed payments among those with inconsistent payment records.

   10. because we only have savings data from January 2008 onwards, we cannot study the effect of
compulsory savings under the individual liability setting, as we have no pre-period data (i.e. savings
data pre-November 2007).
   11. A difference in means test is significant at the 1 percent level.
                                                                                                                                                                462
T A B L E 3 . Dependent Variable: Missed Compulsory Deposit




                                                                                                                                                                THE WORLD BANK ECONOMIC REVIEW
                                           (1)                  (2)                  (3)                  (4)                  (5)                  (6)

Group Liability Loan Dummy             2 0.234***           2 0.338***            2 0.074*             2 0.205**           2 0.230***            2 0.198***
                                        (0.035)              (0.032)               (0.041)              (0.045)             (0.038)               (0.069)
Behrampura Branch                        0.003                                    2 0.008                                  2 0.000
                                        (0.036)                                    (0.034)                                  (0.035)
Number of Previous Loans                                                                                                   2 0.001                0.124
                                                                                                                            (0.017)              (0.098)
Loan Age in Months                                                                                                         2 0.040**              0.050***
                                                                                                                            (0.019)              (0.019)
Loan Age Squared                                                                                                             0.004              2 0.006**
                                                                                                                            (0.003)              (0.003)
Loan Age Cubed                                                                                                             2 0.000                0.000**
                                                                                                                            (0.000)              (0.000)
Constant                                 0.389***             0.698***             0.607***             0.807***             0.461***             0.623***
                                        (0.039)              (0.027)              (0.045)              (0.040)              (0.049)              (0.123)
Control for Calendar Month             No                   No                   Yes                  Yes                  No                   Yes
Individual FEs                         No                   Yes                  No                   Yes                  No                   Yes
R-squared                                0.054                0.489                0.103                0.511                0.059                0.514
N                                     2929                 2929                 2929                 2929                 2929                 2929
Mean of Dep Var                          0.205                0.205                0.205                0.205                0.205                0.205

    *p , 0.10, **p , 0.05, ***p , 0.01.
    Notes: This table reports the results from the OLS regressions with panel data from January 2008 to March 2009, where the dependent variable is a
dummy for missing a compulsory deposit. This dummy takes the value of 1 for a particular month if the client deposited less than the required amount of
Rs. 100 and 0 otherwise. The observations are at the individual-month level. The sample includes clients who received both an individual and a group lia-
bility loan. The variable Group Liability Loan Dummy is a dummy variable equal to 1 if the client had a group liability loan and 0 if it is an individual li-
ability loan. Behrampura Branch is a dummy for one of the two MFI branches in the sample. Number of Previous Loans is the number of loans that the
client received before the current loan. Loan Age in Months is the difference between the month of the observation and the month when the loan was dis-
bursed, plus 1. Standard errors, clustered at the individual level, are given in parentheses beneath each point estimate.
    Source: Authors’ analysis based on data sources discussed in the text.
T A B L E 4 . Heterogeneous Effects by Gender and Borrower Quality
                                        (1)            (2)               (3)           (4)           (5)           (6)

Group Liability Loan Dummy           2 0.224***     2 0.212***        2 0.132***    2 0.137***    2 0.122***    2 0.075
                                      (0.034)        (0.032)           (0.042)       (0.028)       (0.025)       (0.048)
Female                               2 0.041
                                      (0.039)
Group Liability Loan*                  0.063*         0.065*            0.070**
Female                                (0.037)        (0.035)           (0.032)
Borrower Quality                                                                      0.223**
                                                                                     (0.111)
Group Liability Loan*                                                               2 0.234*      2 0.240*      2 0.248**
Borrower Quality                                                                     (0.119)       (0.123)       (0.121)
Behrampura Branch                      0.061***                                       0.079***
                                      (0.017)                                        (0.024)
Number of Previous Loans                                                0.045***                                  0.025
                                                                       (0.017)                                   (0.023)




                                                                                                                              Carpena, Cole, Shapiro, and Zia
Loan Age in Months                                                    2 0.014*                                    0.012
                                                                       (0.007)                                   (0.008)
Loan Age Squared                                                        0.002***                                2 0.001
                                                                       (0.001)                                   (0.001)
Loan Age Cubed                                                        2 0.000***                                  0.000
                                                                       (0.000)                                   (0.000)
Constant                              0.165***        0.108***          0.296***      0.088***      0.075***      0.147
                                     (0.038)         (0.011)           (0.087)       (0.022)       (0.010)       (0.144)
Individual FEs                      No              Yes               Yes           No            Yes           Yes
Month FEs                           No              No                Yes           No            No            Yes
R-squared                             0.084           0.231             0.255         0.095         0.239         0.268
N                                  6055            6055              6055          2579          2579          2579




                                                                                                                              463
                                                                                                               (Continued )
                                                                                                                                                              464
                                                                                                                                                              THE WORLD BANK ECONOMIC REVIEW
TABLE 4. Continued
                                                  (1)                (2)                 (3)                 (4)                (5)                (6)

Mean of Dep Var in Control (Males)              0.226               0.226               0.226
Mean of Dep Var in Control (Females)            0.157               0.157               0.157

   *p , 0.10, ** p , 0.05, ***p , 0.01.
   Notes: This table reports the results from the OLS regressions where the dependent variable is a dummy for missing a payment. This dummy takes the
value of 1 for a particular month if the total amount repaid by the borrower for that month is nil and 0 otherwise. The regressions in columns 1 to 3
provide the heterogeneous effects by gender and include the full sample of clients. The regressions in columns 4 to 6 provide the heterogeneous effects by
borrower quality, defined by the percentage of missed payments in the client’s first individual liability loan in the data. The regressions in columns 4 to 6
include only the subsample of clients who had at least two individual liability loans and only loans from the 2nd loan onwards. All of the observations
are at the loan-month level. The variable Group Liability Loan Dummy is a dummy variable equal to 1 if the loan is a group liability loan and 0 if it is
an individual liability loan. Borrower Quality is the percentage of missed payments in the client’s first individual liability loan in the data. Behrampura
Branch is a dummy for one of the two MFI branches in the sample. Number of Previous Loans is the number of loans that the client received before the
current loan. Loan Age in Months is the difference between the month of the observation and the month when the loan was disbursed, plus 1. Standard
errors, clustered at the individual level, are given in parentheses beneath each point estimate.
   Source: Authors’ analysis based on data sources discussed in the text.
TA B L E 5 . Falsification Test
                                         (1)                  (2)                  (3)                  (4)                  (5)                 (6)

False Treatment                      2 0.004              2 0.011                 0.049                0.002              0.021                 0.058
                                      (0.023)              (0.026)               (0.041)              (0.025)            (0.025)               (0.050)
Behrampura Branch                      0.076***                                   0.089***                                0.066***
                                      (0.027)                                    (0.027)                                 (0.029)
Number of Previous Loans                                                                                                2 0.028**              0.046
                                                                                                                         (0.014)              (0.035)
Loan Age in Months                                                                                                      2 0.025              2 0.014
                                                                                                                         (0.019)              (0.022)
Loan Age Squared                                                                                                          0.003                0.004
                                                                                                                         (0.003)              (0.003)
Loan Age Cubed                                                                                                          2 0.000              2 0.000
                                                                                                                         (0.000)              (0.000)
Constant                               0.095***             0.058***             0.222                0.139***            0.158***             0.319**
                                      (0.025)              (0.013)              (0.142)              (0.020)             (0.043)              (0.129)
Control for Calendar Month           No                   No                   Yes                  No                  No                   Yes
Individual FEs                       No                   Yes                  No                   No                  No                   Yes




                                                                                                                                                            Carpena, Cole, Shapiro, and Zia
R-squared                              0.011                0.140                0.042                0.000               0.016                0.175
N                                   1584                 1584                 1584                 1584                1584                 1584
Mean of Dep Var                        0.140                0.140                0.140                0.140               0.140                0.140

    *p , 0.10, **p , 0.05, ***p , 0.01.
    Notes: This table reports the results from a falsification test, where the dependent variable is a dummy for missed payment. This dummy variable
takes on the value of 1 for a particular month if the total amount repaid by the borrower for that month is nil and 0 otherwise. The regressions corre-
spond with those in Table 2. The sample contains the observations of the individual liability loans among clients who received both an individual and a
group liability loan and who obtained at least 2 individual liability loans, the most recent of which was disbursed after November 2006. Observations are
at the loan-month level. False Treatment is a dummy variable equal to 1 if the loan was disbursed after November 2006 and 0 otherwise. Behrampura
Branch is a dummy for one of the MFI branches in the sample. Number of Previous Loans is the number of loans that the client received before the
current loan. Loan Age in Months is the difference between the month of the observation and the month when the loan was disbursed, plus 1. Standard
errors, clustered at the individual level, are given in parentheses beneath each point estimate.
    Source: Authors’ analysis based on data sources discussed in the text.




                                                                                                                                                            465
466     THE WORLD BANK ECONOMIC REVIEW



T A B L E 6 . Dependent Variable: Standard Deviation of Repayment
                                              (1)                    (2)                     (3)

Group Liability Loan Dummy              2 288.666              2 306.367*              2 220.794
                                         (185.204)              (185.391)               (251.901)
Behrampura Branch                                                839.101***
                                                                (149.442)
Constant                                 1352.427***             798.894***              281.277***
                                         (100.308)              (120.412)                (83.967)
Individual FEs                            No                     No                      Yes
R-squared                                   0.005                  0.041                   0.472
N                                         689                    689                     689
Mean of Dep Var                          1243.496              1243.496                 1243.496

   *p , 0.10, **p , 0.05, ***p , 0.01.
   Notes: This table reports results from the OLS regressions using cross-sectional data, where
the dependent variable is the standard deviation of monthly repayment. If the required amount is
repaid every month, then the standard deviation would be zero. If there are many months where
people pay less or more than the required amount, then the standard deviation would be higher.
The observations are at the loan level. The sample includes the loans of clients who received both
an individual and a group liability loan. The variable Group Liability Loan Dummy is a dummy
variable equal to 1 if the loan is a group liability loan and 0 if it is an individual liability loan.
Behrampura Branch is a dummy for one of the two MFI branches in the sample. Standard errors,
clustered at the individual level, are given in parentheses beneath each point estimate.
   Source: Authors’ analysis based on data sources discussed in the text.



                         V. D I S C U S S I O N     AND   CONCLUSION

Microfinance has reached over 150 million borrowers worldwide and is
growing at a cumulative 40 percent average growth rate. Recent initial public
offerings (IPOs), which valued the Mexican microfinance institution
Compartamos at $2 billion and SKS in India at $1.5 billion, have attracted the
attention of the global financial markets. However, there have also been spec-
tacular failures, such as the collapse of Banco del Exito (BANEX), which was
recently the largest micro, small, and medium enterprise lender in Nicaragua,
with a $125 million dollar loan portfolio. Suffering from a 45 percent delin-
quency rate, BANEX was ordered into liquidation.12
   As many microlenders around the world weaken their group liability ap-
proach and shift toward individual lending, understanding the role of group li-
ability in enhancing performance has become critical in microfinance
programs. However, the empirical literature provides little guidance for policy
makers and microfinance practitioners because few empirical studies have com-
pared group liability contracts with other lending strategies.
   In this paper, we exploit an exogenous change in the liability structure in an
Indian microfinance program, in which the program shifted from an individual
liability structure to a group liability structure. We find evidence that for the

   12. See: http://financialaccess.org/node/3547
                                                Carpena, Cole, Shapiro, and Zia   467


same borrower, the shift to group liability reduces default rates and improves
savings discipline. Under the group liability setting, the required monthly loan
installments are 11 percent less likely to be missed, and compulsory savings de-
posits are approximately 20 percent less likely to be missed relative to individu-
al liability. Thus, our findings indicate that group lending outperforms
individual lending in loan repayment and savings discipline.
   We see our study as an important piece of evidence rather than a definitive
answer to the question of the optimal lending structure for microfinance. The
microlender that we study, Saath, has operations that are fairly typical of MFIs
that lend in urban areas. Moreover, the management, infrastructure, and stated
goals of Saath are not markedly different from other lenders throughout India
or in other low-income settings. Saath’s most remarkable characteristic is prob-
ably its small size; the microfinance industry includes an important right tail of
very large lenders. However, small institutions such as Saath (an NGO with
$410,000 dollars in its total loan portfolio as of 2009) make up a non-trivial
portion of the industry.
   Second, as with any natural experiment, we caution that there are limita-
tions to our study. Our sample consists of only those who elected to continue
from the individual to the group lending model and thus may not be represen-
tative of the entire population that would be effected by changes in lending
models. We are unable to clearly identify the mechanisms through which group
lending improves repayment, and our data are not sufficient to allow us to pre-
cisely calculate the effect of the lending structure change on lender profitability.
An ideal experiment to answer these questions might have randomly assigned
individuals to a range of different lending models, such as group lending with
self-selected group members, group lending with randomly assigned group
members, and group lending without group liability.
   Nevertheless, it may be useful to discuss our view of the mechanisms at
work based on our reading of the evidence and on numerous conversations
with Saath clients, staff, and management. We believe that peer pressure is im-
portant. The Saath operations manual states, “The concept of peer pressure
must be executed properly in favor of the organization and concept of micro
finance,” arguing that the group leader should take responsibility for ensuring
that the members repay and stressing the importance of joint liability. In the
case of repeated missed payments, the manual instructs staff to ensure that
“the members take the responsibility of closing the loan amount of” a delin-
quent borrower. We also interviewed 10 Saath field officers regarding their
views of changes to the loan collection process with group lending. All ten field
officers mentioned that collection under joint liability was easier because
clients were more disciplined about paying on time, and several others men-
tioned the importance of “collective responsibility” within the group. These
qualitative reports are consistent with a recent paper on microfinance in
Andhra Pradesh, which finds that peer pressure is an important determinant of
loan repayment (see Breza (2012)).
468     THE WORLD BANK ECONOMIC REVIEW



   Taken as a whole, we believe that our results provide a cautionary tale for
policy-makers and microfinance institutions that are eager to convert from
group to individual lending models. Although most microfinance organizations
around the world have reported repayment rates that are impressively high, the
industry has also witnessed both idiosyncratic failure and widespread collapse,
such as the recent crisis in Andhra Pradesh. Our results highlight the impor-
tance of the group lending structure in facilitating the sustainable provision of
credit to the poor. Our findings suggest a need for further research to identify
the precise mechanisms of how joint liability affects borrower behavior, the
role of savings as a mechanism for improving repayment, and the relative
merits of peer screening versus peer monitoring under group liability contracts.
Finally, it would be useful to understand whether the group lending structure
itself could be improved upon, perhaps by introducing formal group or sub-
group level repayment incentives or insurance mechanisms.


                                         REFERENCES

Ahlin, C., and R. M. Townsend. 2007. “Using Repayment Data to Test Across Models of Joint Liability
   Lending.” The Economic Journal 117 (517): F11–F51.
Ambec, S., and N. Treich. 2007. “Roscas as Financial Agreements to Cope with Self-Control
  Problems.” Journal of Development Economics 82 (1): 120 –37.
      ´ riz de Aghion, B. 1999. “On the Design of a Credit Agreement with Peer Monitoring.”
Armenda
  Journal of Development Economics 60 (1): 79– 104.
      ´ riz de Aghion, B., and J. Morduch. 2010. The Economics of Microfinance. Cambridge, MA:
Armenda
  MIT Press.
Ashraf, N., D. Karlan, and W. Yin. 2010. “Female Empowerment: Further Evidence from a
   Commitment Savings Product in the Philippines.” World Development 38 (3): 333– 44.
Atkinson, J., A. de Janvry, C. McIntosh, and E. Sadoulet. 2010. “Prompting Microfinance Borrowers to
   Save: A Field Experiment from Guatemala.” Economic Development and Cultural Change
   (forthcoming).
Attanasio, O., B. Augsburg, R. D. Haas, E. Fitzimons, and H. Harmgart. 2011. “Group Lending or
   Individual Lending? Evidence from a Randomized Field Experiment in Mongolia.” Working Paper
   W11/20. Institute for Financial Studies, London, UK.
Banerjee, A., T. Besley, and T. Guinnane. 1994. “Thy Neighbor’s Keeper: The Design of a Credit
   Cooperative with Theory and a Test.” Quarterly Journal of Economics 109 (2): 491–515.
Banerjee, A., E. Duflo, R. Glennerster, and C. Kinnan. 2009. “The Miracle of Microfinance? Evidence
   from a Randomized Evaluation.” Working Paper. Massachusetts Institute of Technology,
   Department of Economics, Cambridge, MA.
Banerjee, A., and S. Mullainathan. 2010. The Shape of Temptation: Implications for the Economic
   Lives of the Poor. Working Paper. Massachusetts Institute of Technology, Department of
   Economics, Cambridge, MA.
Banerjee, A. V., and A. F. Newman. 1993. “Occupational Choice and the Process of Development.”
   Journal of Political Economy 101 (2): 274– 98.
                   ¨c
Beck, T., A. Demirgu ¸ -Kunt, and V. Maksimovic. 2005. “Financial and Legal Constraints to Growth:
   Does Firm Size Matter?” The Journal of Finance 60 (1): 137– 77.
Beck, T., A. Demirgu ¸ -Kunt, and M. S. Martinez-Peria. 2007. “Reaching Out: Access to and Use of
                    ¨c
   Banking Services across Countries.” Journal of Financial Economics 85 (1): 234 –66.
                                                          Carpena, Cole, Shapiro, and Zia       469


Besley, T., and S. Coate. 1995. “Group Lending, Repayment Incentives and Social Collateral.” Journal
   of Development Economics 46 (1): 1 – 18.
Bouman, F. J. A. 1995. “Rotating and Accumulating Savings and Credit Associations: A Development
   Perspective.” World Development 23: 371– 84.
Breza, E. 2012. “Peer Effects and Loan Repayment: Evidence from the Krishna Default Crisis.”
   Working Paper. Massachusetts Institute of Technology, Department of Economics, Cambridge, MA.
Cetorelli, N., and M. Gambera. 2001. “Banking Market Structure, Financial Dependence and Growth:
   International Evidence from Industry Data.” The Journal of Finance 56 (2): 617 –48.
Chen, G., S. Rasmussen, and X. Reille. 2010. “Growth and Vulnerabilities in Microfinance.” CGAP
   Focus Note No. 61.
Daley-Harris, S. 2009. State of the Microcredit Summit Campaign Report 2009. Washington, DC:
   Microcredit Summit Campaign.
Feigenberg, B., E. Field, and R. Pande. 2010. “Building Social Capital through Microfinance.” Working
   Paper 16018. National Bureau of Economic Research, Cambridge, MA.
Fischer, G. 2010. “Contract Structure, Risk Sharing and Investment Choice.” Working Paper EOPP/
    2011/23. London School of Economics, London, UK.
Ghatak, M. 2000. “Screening by the Company You Keep: Joint Liability Lending and the Peer Selection
  Effect.” Economic Journal 110 (465): 601– 31.
   ´ , X., and D. Karlan. 2009. “Group versus Individual Liability: Long Term Evidence from
Gine
   Philippines Microcredit Lending Groups.” Policy Research Working Paper 4008. World Bank, Policy
   Research Department, Washington, DC.
Gugerty, M. K. 2007. “You Can’t Save Alone: Commitment in Rotating Savings and Credit
  Associations in Kenya.” Economic Development and Cultural Change 55 (2): 251– 82.
Laibson, D., 1997. “Golden Eggs and Hyperbolic Discounting.” Quarterly Journal of Economics 112,
   443–477.
Morduch, J. 1999. “The Microfinance Promise.” Journal of Economic Literature XXXVII (4):
  1569–1614.
Rajan, R., and L. Zingales. 1998. “Financial Dependence and Growth.” American Economic Review
   88 (3): 559–86.
Sharma, M., and M. Zeller. 1997. “Repayment Performance in Group-Based Credit Programs in
   Bangladesh: An Empirical Analysis.” World Development 25 (10): 1731–42.
Stiglitz, J. E. 1990. “Peer Monitoring and Credit Markets.” World Bank Economic Review 4 (3):
    351–66.
World Bank. 2008. Finance for All? Policies and Pitfalls in Expanding Access. Washington, DC: World
  Bank.